EPECs 2013 Papers

CONFERENCE COMMITTEES
The 3rd International Conference on Electric Power and Energy Conversion Systems, Yildiz Technical University, Istanbul, Turkey, October 2-4, 2013
Power Electronics Key Technology for Renewable Energy Systems Status and Future
Frede Blaabjerg, Yongheng Yang, Ke Ma
Center of Reliable Power Electronics, Department of Energy Technology Aalborg University Pontoppidanstraede 101, Aalborg DK-9220, Denmark fbl@et.aau.dk, yoy@et.aau.dk, kema@et.aau.dk
Abstract The energy paradigms in many countries (e.g. Germany and Denmark) have experienced a significant change from fossilbased resources to clean renewables in the past few decades. The scenario of highly penetrated renewables is going to be further enhanced. This requires that the production, distribution and use of the energy should be as technological efficient as possible and incentives to save energy at the end-user should also be strengthened. In order to realize the transition smoothly and effectively, energy conversion systems, currently based on power electronics technology, will again play an essential role. Using highly efficient power electronics in power generation, power transmission/ distribution and end-user application, together with advanced control solutions, can pave the way for renewable energies. In view of this, some of the most emerging renewable energies, e.g. wind energy and photovoltaic, which by means of power electronics are changing character as a major part in the electricity generation, are explored in this paper. Issues like technology development, implementation, power converter technologies, control of the systems, and synchronization are addressed. Special focuses are paid on the future trends in power electronics for those systems. Keywords power electronics; renewable energy systems; photovoltaic systems; wind turbine systems; reliability; advanced control; ancillary services
700
600
500 Wind Power
400 300 200 100

0
Solar Photovoltaic Power
...
...
Fig. 1. Global cumulated wind power and solar photovoltaic capacity (Gigawatts) from 1996 to 2012, and an Estimation (E) by 2020 [1], [2].
I.
INTRODUCTION
With an imperative demand of reliable and environmentally friendly electricity generation from Renewable Energy Systems (RESs), the total power generation of RESs is continuously booming and is going to be tripled within the next few decade [1]-[4]. Consequently, great efforts have been made by many countries (e.g. Germany, Spain, and Denmark) to introduce more renewable energies such as wind power, PhotoVoltaic (PV) power, hydropower, and biomass power, etc. to be integrated into the electric grid. As it is shown in Fig. 1, among various renewable energies, Wind Turbine System (WTS) and PV system technologies are still the most promising technologies, accounting for a large portion of renewable energy generation [4]-[14]. However, the increasing adoption of RESs poses two major challenges, which are in urgent need to be coped with. One is the change of electrical power production from the conventional and fossil-based energy sources to renewable energy resources. The other one is the wide-scale use of power electronics in the power generation, the power transmission/ distribution and the end-user application. The power electronics systems should be highly efficient and exceedingly reliable. As this technology has been the key to the
energy conversion from the most emerging renewable energy sources, e.g. WTS and PV systems, it should be able to transfer the renewable energies to the power grid, and capable to exhibit advanced ancillary functions (e.g. Low Voltage Ride-Through, LVRT, grid support with reactive power injection). A wide-scale adoption of power electronics technology makes those completely weather-based energies more controllable, but increasingly intricate. Underpinned by intelligent control strategies, the power electronics technology can fulfill the requirements imposed by the distribution/ transmission system operators as well as specific demands from the end-customers, especially when more advanced power devices and more accurate knowledge of the mission profiles are available. In this paper, the status and the future trends in power electronics technology, which enables a clean and reliable power conversion from WTSs and PV systems, are discussed. In Section II, the basic demands of RESs are firstly presented, and followed by the WTS and PV technologies including main power converter topologies for both PV systems and WTSs. Then, typical control strategies for PV systems and wind turbines are presented by considering the grid demands. Due to the increasing complexity of the future power systems integrated with a large amount of RESs, in Section III, the focuses are paid on the trends in power electronics technology as the key to renewable energy integration. Finally, the conclusions and perspectives are given for the two main renewable energies. II. POWER ELECTRONICS FOR RESS
A. Demands of Renewable Energy Systems Fig. 2 demonstrates the architecture of a modern RES based power generation system, where the power electronics unit is the
(E )
core of the system. An increasing penetration level of RESs results in more stringent grid demands. As it is shown in Fig. 2, the tasks of a power electronics based RES are as varied as they are demanding by the local operators or the end-customers [4], [5]. A very common demand of a RES is to transfer the energy to the grid based on the renewable energy characteristics. Other specific demands can be summarized as: a) reliable/secure power supply, b) high efficiency, low cost, small volume, and effective protection, c) control of active and reactive power injected into the grid, d) dynamic grid support (ride-through operation), and e) system monitoring and communication.
Appliances, Industry, etc. Bi-directional Power Flow
Voltage Source Converter (2L-VSC), featuring with simple structure and few components, as it is shown in Fig. 4. However, since the power capacity of an individual wind turbine keeps growing up to even 10 MW [13], [14], the 2L-VSC is not very feasible due to lower efficiency. In view of this, the multi-level converter technology, which can achieve more output voltage levels, higher voltage and larger output power, gains much more popularity in the wind turbine applications [9], [10], [13]-[16]. The most commercialized multi-level converter is the threelevel Neutral Point diode Clamped (3L-NPC) topology shown in Fig. 5. Compared to the 2L-VSC, 3L-NPC can achieve one more output voltage level, thus leading to a smaller filter. The major drawback of this converter is the unequal loss distribution between the inner and outer switching devices, which may contribute to a de-rated converter power capacity [14], [15]. To solve this problem, multi-cell converter topologies (i.e. parallel/ series connection of converter cells) are developed and widely adopted by the industries (e.g. Gamesa and Siemens) [17], [18].
Load/ Generator
2/3 Power Electronics
2/3
Renewable Energies (PV, Wind Turbines, etc.) References (Local/Centralized)
Power Grid Intelligent Control Communication
Fig. 2. Advanced modern power electronics technologies and intelligent control techniques for renewable energy sources and loads.
Generator Filter
To Grid
B. RES Technologies and Power Converter Topologies The design and operation of power electronics converters for both wind turbine and PV systems strongly rely on the grid requirements and the energy demand. It can be seen from the evolution of wind turbine power converters, which has changed from non-power-electronics-based topologies to full-scale power converters with increasing power ratings of individual wind turbine (tens-kW to several MW) [4], [8]-[10], [12]. As the demand of higher power ratings and efficiency increases for PV systems, the PV power converters also had an obvious change, and they are mostly transformerless nowadays [4], [6], [7]. For WTSs, the most commonly used design concepts can be categorized into four types: 1) fixed speed wind turbine systems, 2) partial variable speed wind turbine with variable rotor resistance, 3) variable speed wind turbine with partial-scale frequency converter, and 4) variable speed wind turbine with full-scale power converter. Among those WTS configurations, the latter two types are currently dominant in the markets and they are shown in Fig. 3. Moreover, the two concepts are expected to be even more widely adopted in the future.
DFIG Transformer Grid
2L-VSC
2L-VSC
Fig. 4. 2L-VSC Back-to-Back (2L-VSC BTB) converter.
Generator Filter Filter
To Grid
3L-NPC
3L-NPC
Fig. 5. 3L-NPC Back-to-Back (3L-NPC BTB) converter.

Multiple PV Strings PV Strings PV String Source: Infineon, SMA
DC DC
DC DC
PV Modules
DC bus
DC DC AC AC DC AC DC AC DC AC
Gear
AC
DC
AC bus Power Rating Applicaitons 3 phase
1 or 3 phase 30 kW~10 kW Commercial/ Residential Type B
1 phase 10 kW~1 kW Residential Type C
1 phase 300 W~ Small System Type D
AC DC (a) Partial-scale back-to-back power converter based wind turbine system AC Gear DC DC AC Transformer Grid
~30 kW Commercial/ PV Plants Type A
Fig. 6. Grid-connected PV systems: Type A- central inverter; Type B- multistring inverter; Type C- string inverter; Type D- module inverter [6].
Generator (b) Full-scale back-to-back power converter based wind turbine system
Fig. 3. Two dominant power electronics based WTS configurations.
As it is indicated in Fig. 3, the power electronics technology plays an essential role in those two concepts. Regarding the power converter topologies for wind power applications, the most commonly adopted three-phase converter is the two-level
For PV systems, a general classification of grid-connected PV inverters is shown in Fig. 6. A common central inverter can be used in a PV plant larger than tens kWp with higher efficiency and lower cost. Its major disadvantages are high voltage DC cables, common Maximum Power Point Tracking (MPPT) and module mismatch [6]. Compared to central inverters, the string inverter can achieve MPPT separately, leading to better total
energy yield. However, there are mismatches in the PV panels connected in series. Thus, the module inverter is developed, which acts on a single PV panel with a single MPPT. The main disadvantage of a module inverter is the low overall efficiency. Another PV technology is an intermediate solution between the string inverter and the module inverter, being multi-string inverter. This configuration is flexible with a high overall efficiency because each PV string is controlled separately. In contrast to WTSs, the PV systems are still dominant in residential applications with much lower power ratings (e.g. several kW), as it is shown in Fig. 6. Thus, at present, singlephase topologies are more common for the PV applications. Normally, in those cases, DC/DC converters are adopted to boost up the PV voltage within an acceptable range of the PV inverter. The boost converter also offers the flexibility of extracting the maximum power. However, several PV power plants have come into service recently using central inverters (e.g. SMA Sunny Central CP XT inverter) and more are under construction, due to an intense energy demand and carbon dioxide emission reduction. The power converter technology for this is similar to the grid side converter technology in WTSs. When it comes to the design of PV inverters as well as their related control methods, the efficiency and leakage current are two main considerations. Connecting the PV inverters to the grid through isolation transformers can solve the safety issues due to the leakage current, but leading to a lower efficiency and a more bulky system. Thus, transformerless PV inverters are developed [4], [6], [7], [19]-[25] by considering the leakage current issue and they have gained much more popularity especially in the European markets. A widely adopted single-phase PV inverter is the Full-Bridge (FB) topology as shown in Fig. 7. Two main modulation schemes are available for the FB inverter - unipolar modulation and bipolar modulation. In the light of safety issues, the FB with bipolar modulation is more feasible in single-phase transformer-less PV applications. However, the conversion efficiency is not very satisfied.
iPV
S1 D1 S3 D3
SD5 D7
SD1
SD3
SD1
SD3
A Vpv
D8 SD6 SD2 SD4
A B Vinv Vpv
SD2 SD4
SD5
B
SD6
Vinv
DC Bypass
Full-Bridge
Full-Bridge
AC Bypass
(a) H6 inverter
(b) HERIC inverter
Fig. 8. Two transformerless PV inverters (H6 and HERIC) [20], [22].

SD1 SD3
Vpv B
SD4
A
Vinv
SD2
Fig. 9. Neutral point clamped transformerless topology for PV application.
LCL- Filter
RESs continues growing, many specific grid requirements have been imposed on those systems. It is further better for the RESs to provide ancillary services, such as LVRT, reactive power control and frequency control through active power control, in order to ensure a reliable and efficient power conversion from such renewable energies. For example, in Germany, the medium- and/or high-voltage systems should have LVRT capability with reactive power injection [4], [5], [10], [26], [28], and in Italy, similar requirement has been published and acted on low power rating systems (6 kW) [27]. Hence, it can be seen that the distribution/transmission system operators (DSO/TSO) have given priority to finding a solution in order to guarantee stable operation of RESs and accept more renewable energies. This consideration should be taken into account, which makes the control systems of RESs multi-functional, as shown in Fig. 10 and Fig. 11. The basic controls like current regulation, DC bus stabilization and grid synchronization have to be quickly performed by the power converter, where a Proportional-Integral (PI) controller and Proportional-Resonant (PR) controllers are typically used [5]. Some advanced control functions of RESs, e.g. riding through operation of the grid faults and providing grid-support functions, are needed for both WTSs and PV systems as shown in Fig. 10 and Fig. 11. In the variable speed wind turbine concept, the current in the generator will typically be changed by controlling the generator side converter, and thereby the rotational speed of turbine can be adjusted to achieve maximum power production based on the available wind power. In respect to operation under grid fault, coordinated control of several subsystems in the wind turbine such as the generator/grid side converters, braking chopper/ crowbar and pitch angle controller, is necessary in order to handle the situation properly. For the reactive power injection during LVRT operation in WTS applications, there are at least four major strategies available [4], [5], [28]-[32]: 1) unity power factor control, 2) positive and negative sequence control, 3) constant active power control and 4) constant reactive power control. Unbalanced grid faults are one of the most observed faults in three-phase systems. Since there is an interaction between voltage sequences and current sequences under grid faults, either the controlled active power or the controlled reactive power will give oscillations.
PV Panels/ Strings
CPV
S2
A B
D 2 S4 D4
vg
O Ground Current: iCMV CP vCMV
Full-bridge Ground
Fig. 7. Single-phase full-bridge PV inverter with an LCL-filter.
Many other transformerless PV inverters available on the markets are derived from the FB topology. For instance, the H6 inverter patented by Ingeteam [20] shown in Fig. 8 disconnects the PV panels/strings from the inverter using four extra devices to realize the isolation; while the Highly Efficient and Reliable Inverter Concept (HERIC inverter) by Sunways [22] provides an AC bypass. There have been other topologies reported in the literature, [7], [23]-[25]. An example shown in Fig. 9 is based on the Neutral Point Clamped (NPC) technology. C. Control of Renewable Energy Systems The first priority of the RES control is to extract as much energy as possible for the renewable energies in normal operation, which is known as MPPT. As the penetration level of
Pin
D DFIG Wind Gear S DC SG/PMSG I Ambient Temperature IGen. IG Vdc PWM DC PWM Xf Igrid Vgrid Pmeas.,Qmeas. AC AC
Po Q
Filter
of the grid faults and the extraction of positive and negative sequence currents are of importance for the control of RESs in LVRT operation modes. III. FUTURE TRENDS IN POWER ELETRONICS FOR RESS
Grid
Vdc Current/Voltage Grid Control Synchronization Control

Basic Control Functions
gen.
Power Maximization & Limitation Inertia Emulation
Fault ride through & Grid support Power Quality
WTS Specific Functions
Mission profiles Communication
Energy Storage
Ancillary Services
Monitoring and Control
Supervisory command from DSO/TSO
Fig. 10. General control function blocks for modern wind turbine systems.
Solar Irradiance
PV Panels/ Strings PPV
CPV
Boost DC
Cdc
Inverter DC
Filter
Po Q Grid
2/3
As the heart of every renewable energy generation system, the power electronics converter is responsible for the power generation from wind and solar energy efficiently and reliably. Thus, to realize a widespread adoption of such renewables, the power electronics technology will be more active into the grid in the future. Together with advanced control strategies, it can fulfill the upcoming stringent requirements regarding the efficiency, the controllability, the cost and the reliability. A. More Power Electronics and Intelligent Control In the last few decades, the power electronics technology has become more and more advanced and brought significant improvements for the renewable energy generation [8]-[10], [37]- [39]. Together with intelligent control strategies, modern power electronics technology makes RESs more controllable and as active as the conventional power plants. There will more advanced power electronics systems in the future RESs in order to enable a better and flexible integration with the power grid. For instance, it can be seen from the evolution of the wind turbine technology in Fig. 12 that the power electronics converter has already achieved 100 % power coverage in the wind turbine system since 2005, while the PV systems have already been complete power electronics based systems. Fully power-electronics-based wind turbine technology will be further enhanced by advanced control strategies. Actually, in most of the newly established WTSs, power electronics converters have become essential components carrying all of the generated power up to multi-MW [5]-[9].
7~8 MW D 164 m 5 MW D 124 m 2 MW D 80 m 500 kW D 40 m 600 kW D 50 m 10 MW D 190 m
DC
Ambient Temperature PWM
AC vdc
PWM
Grid Synchronization PV Panel/Plant Monitoring
iPV
vPV
Current/Voltage Control Maximum Power Point Tracking Grid Support (V, f, Q control)
Vdc Control Anti-Islanding Protection
Xfilter
Basic Control Functions
vg ig
Mission Profiles
PV System Specific Functions
Fault Ride Through
Energy Storage
Communication
Harmonic Compensation Constant Power Generation Control

Ancillary Services
Supervisory command from DSO/TSO
Monitoring and Control
Fig. 11. General control function blocks of a typical PV system with a DC/DC boost stage.
As the PV systems are still at residential level in respect to single-phase systems, there are less control freedoms (grid voltage and grid current) under grid faults. Typically, they are required to cease energizing local loads under grid faults, known as anti-islanding protection. However, the penetration level increases the necessity of LVRT with reactive power injection [33]-[35]. By considering the over-current protection of PV inverters and the reactive current injection requirements under grid faults, possibilities for reactive power injection of singlephase PV systems can be [36]: 1) constant peak current strategy, 2) constant average active power strategy, 3) constant active current strategy, and 4) thermal optimized strategy. As for the three-phase PV systems, the control under grid faults is similar to that of the grid side converter in a wind power system, and thus the above strategies can be adopted and implemented in the different reference frames [4], [5]. D. Synchronizations in Renewable Energy Systems The injected current into the grid has to be synchronized well with the grid voltage, as standards require that in the field [4][6]. Therefore, the grid synchronization issue plays an important role for both WTSs and PV systems. To address this problem, Phase Locked Loop (PLL) based synchronization methods stand out of various reported solutions [5]. Evaluating criterions for synchronization methods are the dynamic response speed and the disturbance rejection capability. The Second Order Generalized Integrator based PLL (SOGI-PLL) presents a better performance compared to other methods, especially for singlephase systems [4], [5]. It can be a good candidate for the synchronization for RESs and used in industrial applications. Moreover, in respect to the aforementioned control methods for WTSs and PV systems, a fast and accurate synchronization system will strongly contribute to the dynamic performance and the stability margin of the whole control systems. The knowledge of grid conditions significantly affects the control systems in different operation modes. For example, the detection
50 kW D 15 m
100 kW D 20 m
1980 1985 1990 0% Power Rating: Electronics Role: Soft starter
1995
2000 2005 10% 30% Rotor Rotor resistance power
2011 2018 (Estimated) 100% Full generator power
Fig. 12. Evolution of wind turbine size and the power electronics seen from 1980 to 2018 (Estimated), where the blue circle indicates the power coverage by power electronics.
B. More Stringent Grid Requiements In order to accept more renewables in the grid, the conventional power grid, which is normally based on centralized and large power plants, have to be modified to be more distributed and smaller generation units. Thus, new demands for grid integration standards, communication, power flow control, and protection are needed [10]-[12]. Power electronic converters together with dedicated and intelligent control strategies again play an important role in this technology transformation. Taking the PV systems as an example, typically, MPPT is required during the operation. However, recent studies showed that a limitation of the maximum feed-in power from PV systems only contributes a limited energy reduction, as it is
shown in Fig. 13. Thus, it is reasonable to avoid upgrading power infrastructure by limiting the maximum feed-in power from PV systems. This may be included in the future grid demands at a very high penetration level. The same philosophy may be imposed on other renewables (e.g. fuel-cell system). In that case, the control of power electronics should be ready.
120
Reduction Percentage of Annual Energy Production
100
80 60 40 20 0
Yearly Solar Irrandiance
20 % Reduction of Feed-in Power 6.23 % Reduction of Energy Yield

0 20 40 60 80 100 120 Feed-in Power Percentage of PV Peak Power
Yearly Ambient Temperature
Mission Profiles
PV Power Conversion
Energy Production
Fig. 13. Energy reduction due to the limitation of maximum feed-in power.
D. High Efficiency and High Reliability Achieving high efficiency and high reliability are always of intense interest in order to reduce energy losses and to extend service time, and it will be further strengthened in the future RESs. Improvements of efficiency can be achieved by integrating more power electronics systems with intelligent control strategies and developing more advanced power electronics devices (e.g. SiC-based modules). For example, transformerless PV inverters will be even more widely adopted since they can achieve high efficiency. As the devices and components that comprise the power electronics system in a RES, the behavior of the power electronics devices will impose constraints on the system conversion performance [37]-[39]. Thus, for the future wind power systems with high power ratings, using advanced power electronics devices can improve the whole performance in terms of efficiency and reliability. Notably, the dramatic growth of total installations and the individual capacity make the failures of wind power or PV system costly or even unacceptable. In view of this, the reliability is another critical requirement for the future RESs [9], [10], [14], [37]. According to previous research and field experiences, the control and power electronics systems in a RES have higher failure rate than the other subsystems. Therefore, possible improvements of the reliability can be achieved by means of: a) proper component selection (e.g. considering rated power, the most stressed situations, and the severe users, using advanced device packaging technologies, and choosing new power electronics devices), b) effective thermal management, c) robustness design and validation with the knowledge of mission profiles [9], [10], [37], [41]. This consideration should be taken into during the design and operation of a RES. It also leads to possible activities for reliability analysis and improvement as it is illustrated in Fig. 15.
Stress Analysis
x x x x Mission profile Converter design Load estimation ... x x x x
C. Lower Cost of Energy Reducing the cost of energy is one of the most important considerations, which may affect the installation capacity of the energy technologies. Generally, a Levelized Cost of Energy (LCOE) index is adopted to quantify and compare the cost for different renewables [40]-[42], and it can be expressed as,
LCOE
CDev CCap CO & M E Annual
(1)
in which CDev is the initial development cost, Ccap represents the capital cost, CO&M denotes the cost for operation and maintenance, and EAnnual is the average annual energy production in the whole lifetime. It is shown in (1) that the possibilities to lower the cost of energy are: a) to reduce the cost for development, capital, operation and maintenance and b) to extend the energy production or increase the lifetime of the generation system. As it is shown in Fig. 14, the onshore wind power technology is currently competitive with the fossil-based power generation in terms of the cost, while offshore wind power and solar PV technologies are still more expensive than the onshore wind power. The cost advantage is the main contribution to the significant adoption of onshore wind power systems in the past few decades. As it is indicated in Fig. 14, there is large potential to reduce the cost of offshore wind and solar PV technologies in the future. Since the power electronics technology is the key technology for RESs with higher power ratings, special cost considerations should also be taken into account for the design and control of power electronics converters.
Typical LCOE ranges USD / kWh
Strength Modeling
Failure mechanism Accelerating test Field feedback ...
Mission Profile to Component Stress
Design Tools for Reliability

Monitoring & Control
x x x x Lifetime prediction Control for reliability Efficient maintenance ...
Lifetime Model
Probability & Statistics
IGBT Module
Infrared Camera
Thermal Thermal Picture Picture
x x x x
Six sigma Devices variation Production robustness ...
Temp. Measurement
Reliability Engineering
Fig. 15. Multi-disciplinary approaches for more reliable power electronics in renewable energy systems.
IV.
CONCLUSIONS
Cost of fossil fuel generation
Fig. 14. Estimated LCOE for several renewable energy technologies for entering service in 2018 [41].
In this paper, the status and future of the key technology for renewable energy systems power electronics have been discussed. Currently, underpinned by the power electronics technology, the electrical power generation from renewable energy sources is continuously booming in many countries. An overview of the mainstream power converter topologies for a few kW residential PV applications and for several MW WTSs has also been given, together with basic operation principles and control strategies. It can be concluded that the power electronics technology is playing an important role in the electricity generation, and also it will play an even more active role for both wind energy and PV applications. In the future, it is expected that there will be more advanced power electronics systems
associated with intelligent control strategies for the renewable energy systems in order to increase the power capacity. Moreover, high efficiency and high reliability of the power conversion systems will be further strengthened on the purpose to reduce the cost of energy and also to extend the service period. REFERENCES
[1] [2] [3] REN21, "Renewables 2013: Global Status Report (GSR)," [Online]. Available: http://www.ren21.net/, Jun. 2013. REN21, "Global Futures Report (GFR) - Scenario Profiles Report (Draft)," [Online]. Available: http://www.ren21.net/, Jan. 2013. E. J. Coster, J. M. A. Myrzik, B. Kruimer, and W. L. Kling, Integration issues of distributed generation in distribution grids, Proc. IEEE, vol. 99, no. 1, pp. 2839, Jan. 2011. R. Teodorescu, M. Liserre, and P. Rodriguez, Grid Converters for Photovoltaic and Wind Power Systems. Hoboken, NJ, USA: Wiley, 2011. F. Blaabjerg, R. Teodorescu, M. Liserre, and A.V. Timbus, "Overview of control and grid synchronization for distributed power generation systems," IEEE Trans. Ind. Electron., vol. 53, no. 5, pp. 1398-1409, Oct. 2006. S.B. Kjaer, J.K. Pedersen, and F. Blaabjerg, "A review of single-phase grid-connected inverters for photovoltaic modules," IEEE Trans. Ind. Appl., vol. 41, no. 5, pp. 1292-1306, Sept.-Oct. 2005. D. Meneses, F. Blaabjerg, O. Garca, and J.A. Cobos, "Review and comparison of step-up transformerless topologies for photovoltaic ACmodule application," IEEE Trans. Power Electron., vol. 28, no. 6, pp. 2649-2663, Jun. 2013. F. Blaabjerg, Z. Chen, and S.B. Kjaer, "Power electronics as efficient interface in dispersed power generation systems," IEEE Trans. Power Electron., vol. 19, no. 5, pp. 1184-1194, Sept. 2004. F. Blaabjerg, M. Liserre, and K. Ma, "Power electronics converters for wind turbine systems," IEEE Trans. Ind. Appl., vol. 48, no. 2, pp. 708719, Mar.-Apr. 2012. F. Blaabjerg and K. Ma, "Future on power electronics for wind turbine systems," IEEE Journal of Emerging and Selected Topics in Power Electronics, in press, 2013. J.M. Carrasco, L.G. Franquelo, J.T. Bialasiewicz, E. Galvan, R.C.P. Guisado, Ma.A.M. Prats, J.I. Leon, and N. Moreno-Alfonso, "Powerelectronic systems for the grid integration of renewable energy sources: a survey," IEEE Trans. Ind. Electron., vol. 53, no. 4, pp. 1002-1016, Jun. 2006. M. Liserre, T. Sauter, and J.Y. Hung, "Future energy systems: Integrating renewable energy sources into the smart power grid through industrial electronics," IEEE Ind. Electron. Mag., vol. 4, no. 1, pp. 18-37, Mar. 2010. M. Liserre, R. Cardenas, M. Molinas, and J. Rodriguez, "Overview of multi-MW wind turbines and wind parks," IEEE Trans. Ind. Electron., vol. 58, no. 4, pp. 1081-1095, Apr. 2011. K. Ma and F. Blaabjerg, "Multilevel converters for 10 MW wind turbines," in Proc. of EPE11, pp. 1-10, Aug. 30 2011-Sept. 1 2011. J. Rodriguez, S. Bernet, P.K. Steimer, and I.E. Lizama, "A survey on neutral-point-clamped inverters," IEEE Trans. Ind. Electron., vol. 57, no. 7, pp. 2219-2230, Jul. 2010. F. Blaabjerg, K. Ma, and D. Zhou, "Power electronics and reliability in renewable energy systems," in Proc. of ISIE, pp. 19-30, 28-31 May 2012. B. Andresen and J. Birk, "A high power density converter system for the Gamesa G10x 4.5 MW Wind turbine," in Proc. of EPE, pp. 1-7, 2007. R. Jones and P. Waite, "Optimised power converter for multi-MW direct drive permanent magnet wind turbines, " in Proc. of EPE, pp. 1-10, 2011. S.V. Araujo, P. Zacharias, and R. Mallwitz, "Highly efficient single-phase transformerless inverters for grid-connected PV systems," IEEE Trans. Ind. Electron., vol. 57, no. 9, pp. 3118-3128, Sept. 2010. R. Gonzalez, J. Lopez, P. Sanchis, and L. Marroyo, "Transformerless inverter for single-phase photovoltaic systems," IEEE Trans. Power Electron., vol. 22, no. 2, pp. 693-697, Mar. 2007. T. Kerekes, R. Teodorescu, P. Rodriguez, G. Vazquez, and E. Aldabas, "A new high-efficiency single-phase transformerless PV inverter topology," IEEE Trans. Ind. Electron., vol. 58, no. 1, pp. 184-191, Jan. 2011.
[4] [5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14] [15]
[16] [17] [18] [19]
[20]
[21]
[22] H. Schmidt, S. Christoph, and J. Ketterer, "Current inverter for direct/alternating currents, has direct and alternating connections with an intermediate power store, a bridge circuit, rectifier diodes and a inductive choke," German Patent DE10 221 592 A1, 4 Dec. 2003. [23] I. Patrao, E. Figueres, F. Gonzalez-Espin, and G. Garcera, "Transformerless topologies for grid-connected single-phase photovoltaic inverters," Renewable and Sustainable Energy Reviews, vol. 15, no. 7, pp. 34233431, Sept. 2011. [24] L. Zhang, K. Sun, L. Feng, H. Wu, and Y. Xing, "A family of neutral point clamped full-bridge topologies for transformerless photovoltaic grid-tied inverters," IEEE Trans. Power Electron., vol. 28, no. 2, pp. 730-739, Feb. 2013. [25] B. Gu, J. Dominic, J.-S. Lai, C.-L. Chen, T. LaBella, and B. Chen, "High reliability and efficiency single-phase transformerless inverter for gridconnected photovoltaic systems," IEEE Trans. Power Electron., vol. 28, no. 5, pp. 2235-2245, May 2013. [26] E. ON GmbH, "Grid Code - High and extra high voltage." [Online]. Available: http://www.eon-netz.com/. [27] Comitato Elettrotecnico Italiano, "CEI 0-21: Reference technical rules for connecting users to the active and passive LV distribution companies of electricity." [Online]. Available: http://www.ceiweb.it/. [28] P. Rodriguez, A.V. Timbus, R. Teodorescu, M. Liserre, and F. Blaabjerg, "Flexible active power control of distributed power generation systems during grid faults," IEEE Trans. Ind. Electron., vol. 54, no. 5, pp. 25832592, Oct. 2007. [29] G.M.S. Azevedo, G. Vazquez, A. Luna, D. Aguilar, and A. Rolan, "Photovoltaic inverters with fault ride-through Capability," in Proc. of ISIE09, pp. 549-553, 5-8 Jul. 2009. [30] C.H. Benz, W.-T. Franke, and F.W. Fuchs, "Low voltage ride through capability of a 5 kW grid-tied solar inverter," in Proc. of EPE/PEMC, pp. T12-13-T12-20, 6-8 Sept. 2010. [31] X. Bao, P. Tan, F. Zhuo, and X. Yue, "Low voltage ride through control strategy for high-power grid-connected photovoltaic inverter," in Proc. of APEC13, pp. 97-100, 17-21 Mar. 2013. [32] H.-C. Chen, C.-T. Lee, P.T. Cheng, R.Teodorescu, F. Blaabjerg, and S. Bhattacharya, "A flexible low-voltage ride-through operation for the distributed generation converters," in Proc. of PEDS13, pp. 1354-1359, 22-25 Apr. 2013. [33] N.P. Papanikolaou, "Low-voltage ride-through concept in flyback inverterbased alternating current photovoltaic modules," IET Power Electron., vol. 6, no. 7, pp. 1436-1448, Aug. 2013. [34] Y. Bae, T.-K. Vu, and R.-Y. Kim, "Implemental control strategy for grid stabilization of grid-connected PV system based on german grid code in symmetrical low-to-medium voltage network," IEEE Trans. Energy Conv., vol. 28, no. 3, pp. 619-631, Sept. 2013. [35] Y. Yang, F. Blaabjerg, and Z. Zou, "Benchmarking of grid fault modes in single-phase grid-connected photovoltaic systems," IEEE Trans. Ind. Appl., vol. 49, no. 5, in press, Sept./Oct. 2013. [36] Y. Yang, F. Blaabjerg, and H. Wang, "Low voltage ride-through of singlephase transformerless photovoltaic inverters," in Proc. of ECCE13, Sept. 2013. [37] H. Wang, M. Liserre, and F. Blaabjerg, "Toward reliable power electronics: challenges, design tools, and opportunities," IEEE Ind. Electron. Mag., vol. 7, no. 2, pp. 17-26, Jun. 2013. [38] J.D., van Wyk and F.C. Lee, "On a future for power electronics," IEEE Journal of Emerging and Selected Topics in Power Electronics, , vol. 1, no. 2, pp. 59-72, Jun. 2013. [39] J.G. Kassakian and T.M. Jahns, "Evolving and emerging applications of power electronics in systems," IEEE Journal of Emerging and Selected Topics in Power Electronics, vol. 1, no. 2, pp. 47-58, Jun. 2013. Aschenbrenner, [40] M. Campbell, J. Blunden, E. Smeloff, and P. "Minimizing utility-scale PV power plant LCOE through the use of high capacity factor configurations," in Proc. of IEEE PVSC, pp. 421-426, 712 Jun. 2009. [41] E. Koutroulis and F. Blaabjerg, "Design optimization of transformerless grid-connected PV inverters including reliability," IEEE Trans. Power Electron., vol. 28, no. 1, pp. 325-335, Jan. 2013. [42] U.S. Energy Information Administration, "Levelized Cost of New Generation Resources in the Annual Energy Outlook 2013, " [Online] Tech. Rep., Jan. 2013. Available: http://www.eia.gov/.
EPECS13 Conference Agenda Wednesday, October 2, 2013

07.30-09.30
Registration and Reception (Registration is open during the conference)
Main Hall
09.30-10.00 Opening Ceremonies (Turkish Music Concert) Opening Speech1 - Chair (Mugdesem Tanrioven, YTU) 10.00--10.30 Opening Speech3 - Co Chair (Celal Kocatepe, YTU) Opening Speech3 - Co Chair (Rached Dhaouadi, AUS) Opening Speech3 - Honorary Chair (Ismail Yuksek, Rector, YTU) 10.30-11.15 11.15-11.30 11.30-12.15 12.15-12.45 12.45-13.00 13.00-14.00 14.00-14.45 14.45-15.00 15.00-15.45 15.45-16.00 16.00-17.40 17.40-19.00 19.00-23.00 23.00-23.30 Keynote Speaker1- Prof. Ralph M. Kennel, TUM (Predictive Control - The Powerful Method to Control Power Converters and Drives in the Future) Coffee Break Keynote Speaker2- Prof. Saifur RAHMAN, VA (Opportunities and Challenges of Deploying the Smart Grid) Mehter Band Concert (Ota Humayun Building) Break at Ota Humayun Building (Sightseeing) Lunch Break Keynote Speaker3- Prof. Frede Blaabjerg, AU (Key Technology for Renewable Energy Systems Status and Future) Coffee Break Keynote Speaker4- Prof. Stanislaw M. Gubanski, TUW (Dielectric Response Diagnoses for Transformer Windings) Coffee Break Parallel Session-Main Hall Parallel Session-Hall A Electric Machines&Drives Renewable Energy Parallel Session-Hall B Power Distribution Parallel Session-Hall C Power Electronics
Transportation to Gala Dinner Gala Dinner Transportation to Hotels
Thursday, October 3, 2013

09.00-11.00 11.00-11.20 11.20-13.20 13.20-14.20 14.20-16.20 16.20-16.40 Parallel Session-Hall A 16.40-18.40
Renewable Energy/ Power Electronics
Parallel Session-Hall A Electric Machines&Drives
Parallel Session-Hall B High Voltage Coffee Break
Parallel Session-Hall C Power Distribution
Parallel Session-Hall A
Power Electronics/ Electric Machines&Drives
Parallel Session-Hall B Power Systems Lunch Break
Parallel Session-Hall C High Voltage/ Power Systems
Parallel Session-Hall A Power Distribution/Smart Grid
Parallel Session-Hall B
Energy Harvesting / Renewable Energy
Parallel Session-Hall C Electric Machines&Drives
Coffee Break Parallel Session-Hall B Renewable Energy/ Smart Grid Parallel Session-Hall C Power Electronics
Friday, October 4, 2013

09.00-10.40 10.40-11.00 11.00-12.40 12.40-14.00 14.00-16.00 16.00-16.20 16.20-17.30 17.30-20.00 Parallel Session-Hall A
Electric Machines&Drives
Parallel Session-Main Hall Smart Grid
Parallel Session-Hall A Power Systems
Parallel Session-Hall C
Power Electronics/ Renewable Energy
Power Systems
Coffee Break
Parallel Session-Main Hall Renewable Energy
Parallel Session-Hall A Power Systems
Power Systems
Parallel Session-Hall C Smart Grid
Coffee Break and Lunch Break Parallel Session-Hall B

Electric Machines&Drives
Parallel Session-Hall C
Power Systems/ Renewable Energy
Parallel Session-Hall D
Electric Machines&Drives/ Renewable Energy
Coffee Break Transportation to Bosphorus Tour (On-site Registration Required) Bosphorus Tour
Saturday, October 5, 2013

10.00-17.00 Istanbul City Tour (On-site Registration Required)
EPECS 2013 PROGRAM DETAIL

Parallel Session-Main Hall (Electric Machines & Drives), October 2, 2013
Session Chair: Erkan Mese
Time
16.0016.20 16.2016.40 16.4017.00 17.0017.20 17.2017.40
Paper Title
A New Approach for Temperature Rising Test of an Induction Motor Loaded by a Current Controlled Eddy Current Brake A Novel Feed-Forward Control of PMSM
Authors
Mehmet Onur GULBAHCE, Handan Nak, Ahmet Derya Kocabas, Istanbul Technical University Yusuf ALTUN, Balikesir University; Kayhan Gulez, Tarik Veli Mumcu, Yildiz Technical University; M. zgr Kizilkaya, Turkish Air Force Academy Altan Gencer, HURRIYET EML
A Novel Microcontroller Based Speed Control Method Employing Current Controller for TWUSM
A Study of Hybrid Excitation Axial Flux Motor Topology for Traction Electric Vehicle
Ademir Pelizari, Ivan Eduardo Chabu, University of Sao Paulo
Adaptive PID Controller Design by Using Adaptive Interaction Approach Theory
Tayfun Gundogdu, Istanbul Technical University; Guven Komrgoz, ITU
Parallel Session-Hall A (Renewable Energy), October 2, 2013

Session Chair: Haluk Gorgun
Time Paper Title

16.0016.20 16.2016.40 16.4017.00 17.0017.20 17.2017.40
A Comparison among Reactive Power Compensation Strategies in Wind Farm Power Plant A Complementary Hybrid System for Electricity Generation Based on Solar and Wind Energy Taking into Account Local Consumption - Case Study A Dynamic Model of a Grid Connected PV System based on Outdoor Measurment using Labview
Authors
Vito Calderaro, University of Salerno; Gaspare Conio, Italian Vento Corporation Group; Roberto Fusco, ArcerolMittal; Vincenzo Galdi, University of Salerno Ajla Merzic, Mustafa Music, Elma Redzic, Public Enterprise Electric Utility of Bosnia and Herzegovina Imane Drouiche, USTHB; Aissa Couder, Centre de Developpement des Energies Renouvelable; Samia Harrouni, USTHB
Farid Benhamida, UDL university of Sidi Bel Abbes; Slimane Souag, Irecom laboratory, dept. of electrotechnics; Yacine Salhi, UDL university of Sidi Bel Abbes; Amel GRAA, Department of Business Sciences, Faculty of economis,; Abdelber Bendaoud, UDL university of Sidi Bel Abbes; Djehaf Mohamed, University of Sidi bel abbes
A Fuzzy Logic Controller for Reactive Power and PQ Bus Voltage Control
A Novel Single-Phase Soft Switching Microinverter for Photovoltaic Applications
Saad Pervaiz, Muhammad Muneeb Ur Rehman, Ahmed Bilal Asghar, Nauman Zaffar, Lahore University of Management Sciences
Parallel Session-Hall B (Power Distribution), October 2, 2013

Session Chair: Walid ElKhattam
Time Paper Title

16.0016.20 16.2016.40 16.4017.00 17.0017.20 17.2017.40
A combined approach to multi-objective optimization of capacitor placement in radial distribution networks A novel control method in Grid Interconnection of DG Based on Pulse Adaptive VSI in order to Harmonic Compensation and Power Quality Improvement A Novel Energy Management System using Ant Colony Optimization for Micro-grids A Novel Fuzzy Cause-and-Effect-Networks AND Fuzzy logic Based Methodology for a Distribution Systems Fault Diagnosis Overhead Lines Maintenance Scheduling in Power Distribution System Considering Feeders Ranking
Authors
Igor Trach, Institute of Electrodynamics NAS Ukraine; Yuri Zubiuk, Electrosphere Ltd. Reza Galandary, K.N.Toosi university of Technology; S. M. T. Bathaee, K. N. Toosi University of Technology Ayman Esmat, AAST; Amr Magdy, Walid ElKhattam, Ain Shams University; Ahmed M. ElBakly, Arab Academy for Science and Technology Muhammad Mustafa, AASTMT; Walid ElKhattam, Ain Shams University; Yasser Galal, Arab Academy for Science and Technology Hadi Ghasemabadi, Ali Peiravi, Ferdowsi University of Mashhad
Parallel Session-Hall C (Power Electronics), October 2, 2013

Session Chair: Faruk Bakan
Time
Paper Title
Authors
Saeed Ouni, Sharif University of Technology; Negar Noroozi, Sharif University of Thechnology; Mahmoud Shahbazi, Mohammad Reza Zolghadri, Hashem Oraee, Sharif University of Technology Subhajyoti Mukherjee, Sunit Kumar Saxena, Honeywell Technology Solutions Lab; Tarak Saha, Honeywell Technology Solutions Zerouali Sakina, University of Batna; Hadri Hamida Amel, University of Constantine; Mimoune souri mohamed, University of Biskra Huseyin Yesilyurt, Yildiz Technical University; Hatice Ozel, Askaynak; HacI Bodur, YTU
16.00-16.20 A new Fault Tolerant Scheme for Cascaded H-Bridge Multilevel Converter 16.20-16.40 A Novel AC-AC Converter With Minimum Snubber Requirement 16.40-17.00 Adaptive Control with Tuning Function Control Design Applied to Class-E/F Inverter 17.00-17.20 An Improved Lossless Passive Snubber Cell for PFC Boost Converter 17.20-17.40 Finite-Control-Set Strategy
Analysis of a Modular Multilevel Inverter Under the Predicted Current Control Based on
Razieh Nejati Fard, Hamed Nademi, Lars Norum, Norwegian University of Science and Technology (NTNU),
Parallel Session-Hall A (Electric Machines&Drives), October 3, 2013

Session Chair: Ahu Ece Hartavi
Time Paper Title

09.0009.20 09.2009.40 09.4010.00 10.0010.20 10.2010.40 10.4011.00
Analysis of Heteropolar Inductor Machines Based on Space Vectors
Authors
Jingzhe Wu, Martin J Balchin, University of Bath Benamimour Tariq, LGEC, Dept. Electrotechnique, Universit Constantine 1; Amar Bentounsi, Hind Djeghloud, Electrotechnics Laboratory, University Mentouri of Constantine Claudio Bruzzese, University of Rome
CAD of Electrical Machines Using Coupled FEMM-MATLAB Softwares Validation of Sequence Circuits Useful for Split-Phase Current Signature Analysis (SPCSA) and Diagnosis of Eccentric-Rotor Traction Cage Motors Command Filtered Adaptive Bacstepping Control of Linear Servo
Jacek Kabzinski, Institute of Automatic Control, Lodz University of Technology Yusuf Yaa, Hicret Yeti, Hande Boztepeli, YIldIz Technical University; Erkan Mese, Yildiz Technical University Claudio Bruzzese, University of Rome
Comparative Design of Direct Drive PM Synchronous Motors in Gearless Elevator Systems Diagnosis-Oriented Sequence Circuit-Based Modeling of Eccentric Rotor Traction Induction Motors Accounting for Cage Damping and Split-Phase Currents
Parallel Session-Hall B (High Voltage), October 3, 2013

Session Chair: Essam A. Al-Ammar
Time
09.0009.20 09.2009.40 09.4010.00 10.0010.20 10.2010.40 10.4011.00
Paper Title
Accurate Partial Discharge Classification from Acoustic Emission Signals
Authors
Mustafa Harbaji, Dr. Ayman El-Hag, American University of Sharjah; Dr. Khaled Shaban, Qatar University
Analysis of the Coupling of a Lightning Wave with an Overhead Line
Labed Djamel, Boumaiza Mustafa, Constantine University ZIANI Abderrahmane, MOULAI Hocine, USTHB, University of Science and Technology Houari Boumediene, Algiers
Thair Ibrahim Mustafa, Hugo D. Almaguer, University of Blumenau; Luiz Henrique Meyer, University of Blumenau Fund Universidade Regional de Blumenau - FURB; Sergio Henrique Cabral, Johann Eduardo Baader, University of Blumenau; Jose de Anchieta J. Jr, CEEE-GT
Arc quenching Equations Solving in HV Circuit Breakers by a General Diagram Bloc Model Dynamic Analysis of a Telecommunication Station Grounding System, Applying the TLM Method Improvement of Thermal Performance of 36 kV DAF-30 Type Bushing
Ilim ERDEN, Aydogan OZDEMIR, Suat ILHAN, ITU Aytu Font, Istanbul Technical University; zcan Kalenderli, Istanbul Technical University Electrical Engineering Department
Insulator Contour Optimization Using Intelligent Systems & Soft Computational Methods
Parallel Session-Hall C (Power Distribution), October 3, 2013

Session Chair: Magdy Salama
Time Paper Title

09.00Energy Storage for Microgrids Islanding Operation 09.20 09.20Flexible DG Interface Using Repetitive Control 09.40 09.40- Fuzzy-Based Control of On-Load Tap Changers under High 10.00 Penetration of Distributed Generators 10.00- Power Quality Measurement and Evaluation of a Wind Farm 10.20 Connected to Distribution Grid 10.20- Design and Analysis of Composite Filter for Power Quality 10.40 improvement of Electric Arc Furnace 10.40- Voltage Control Using Coordinated Control of Heat Pump Water 11.00 Heater with Large Penetration of Photovoltaic Systems
Authors
Ahmed Awad, University of Waterloo; Tarek EL-Fouly, Natural Resources Canada; Magdy Salama, University of Waterloo
Amr Said, E. F. El-Saadany, University of Waterloo
Maher Abdelkhalek, University of Waterloo; Hany Farag, York University; E. F. El-Saadany, University of Waterloo Fatih KECECIOGLU, University of Kahramanmaras Sutcu Imam; mustafa ekkeli, university of Kahramanmaras sutcu imam; Hakan ACIKGOZ, University of Kilis 7 Aralik; Ceyhun YILDIZ, Electrical Power Resources Survey and Development Deepak Bhonsle, R B Kelkar, Maharaja Sayajirao University of Baroda
Mohamed Mufaris, The University of Tokyo
Parallel Session-Hall A (Power Electronics/ Electric Machines&Drives), October 3, 2013

Session Chair: Seref Naci Engin
Time
11.2011.40 11.4012.00 12.0012.20 12.2012.40 12.4013.00 13.0013.20
Paper Title
Analysis of Waveform Approximation for the AC Current of a Line-Commutated Converter Comparison between Reliability Prediction Methods for Power Electronic Circuits
Authors
Mehdi Daryabak, Shaahin Filizadeh, University of Manitoba Ayoub Balador, Sharif University of Technology; Negar Noroozi, Sharif University of Thechnology; Mohammad Reza Zolghadri, Sharif University of Technology; Saeed Ouni, Sharif University of Thechnology Amel Lachichi, ABB Corporate Research Centre
DC/DC Converters for High Power Application: A Survey
Power Converters Variable Structure Control surface design Investigating Converter Options for Automotive Grade Permanent Magnet Sychronous Generators Experimental Verification of Torque Ripple Calculation for Induction and Synchronous Motor
Bashar khasawneh, Maha Sabra, M.A Zohdy, Oakland University Erkan Mese, Yildiz Technical University; Yusuf Yaa, Hakan Aka, YIldz Technical University; Mustafa Gurkan Aydeniz, YILDIZ TECHNICAL UNIVERSITY; Murat Ayaz, Kocaeli University; Murat Tezcan, YTU
Tetsuya Kojima, Mitsubishi Electric; Haruyuki Kometani, Mitsubishi Electric Co.; Mitsuhiro Kawamura, Yuichi Tsuboi, Ritaka Nakamura, Hiroyuki Masuda, Yoshihiro Ohgashi, Toshiba Mitsubishi-Electric Industrial Systems Co.
Parallel Session-Hall B (Power Systems), October 3, 2013

Session Chair: Aslan nan
Time Paper Title

11.2011.40 11.4012.00 12.0012.20 12.2012.40 12.4013.00
A New Approach for optimal reactive power flow solution of multiterminal hvdc systems using genetic algorithm A Reliability Study Of Medium Voltage Grid With Private Sector Power Plants Analysis of Various Transmission Line Switching Overvoltage Limitation Techniques Congestion Effects on Regional & System Emission and Consumers Allocated Cost Design of an IIR Digital Filter for Phasor Measurement Improvement Based on a Hybrid Optimization Algorithm
Authors
Ula Kl, University of Mehmet Akif Ersoy; Krat Ayan, University of Sakarya
Levent Kl, Trkiye ie ve Cam FabrikalarI A..
Shahab Tanhaeidilmaghani, Semko Golabi, Heresh Seyedi, University of Tabriz Mahdi Pourakbari Kasmaei, Marcos Julio Rider Flores, Jose Roberto Sanches Mantovani, Universidade Estadual Paulista (UNESP) Abdelmadjid Recioui, Abderrahmane Ouadi, Hamid Bentarzi, University of Boumerdes
Parallel Session-Hall C (High Voltage/Power System), October 3, 2013

Session Chair: Luiz Henrique Meyer
Time Paper Title

11.20Breakdown Voltage Analysis of Insulating Oils Under Different Conditions 11.40 11.40- Multi-Purpose RF Antenna for Partial Discharge and Oil Quality 12.00 Monitoring 12.00- Parametric Analysis of Overvoltages Caused by Back-flashover In "Siah12.20 bishe" 400kV GIS Substation 12.20Salt Fog Testing of Glass Insulators with Different Surface Conditions 12.40 12.40- Extended Pruning-Grafting Operators for Efficient Distribution Network 13.00 Reconfiguration 13.00- Design of Power System Stabilizers Based on -Controller for Power 13.20 System Stability Enhancement
Authors
Celal Kocatepe, Oktay ArIkan, Eyp TASLAK, Celal FadIl Kumru, Yildiz Technical University Dr. Ayman El-Hag, Naser Qaddoumi, American University of Sharjah; Esam Al Murawwi, TRANSCO; Rida Mourtada, American University of Sharjah Mehdi Babaei, Faraniroo Company; Maziar Babaei, Mohsen Niasati, Semnan University Luiz Henrique Meyer, University of Blumenau - Fund Universidade Regional de Blumenau - FURB; Wagner Werner Beyer, University of Blumenau; Fernando Hidalgo Molina, CELESC Distribuio S.A. Saeed RamezanJamaat, Youhei Akimoto, Hernan Aguirre, Kiyoshi Tanaka, Shinshu University
Dr. Khaled Ellithy, Samer Said, Osama Elkahlout, Qatar University
Parallel Session-Hall A (Power Distribution/Smart Grid), October 3, 2013

Session Chair: Ehab El-Saadany
Time Paper Title

14.20Probabilistic Modeling of PHEV charging load in distribution systems 14.40 14.40Residential Load Harmonics in Distribution Grid 15.00 15.00- Self-healing Restoration of a Distribution System Using Hybrid Fuzzy 15.20 Control/Ant-Colony Optimization Algorithm 15.20A Non-Intrusive Load Monitoring System Based on A Cascaded Method 15.40
Authors
Mostafa Shaaban, E. F. El-Saadany, University of Waterloo
Jaan Niitsoo, Ivo Palu, Jako Kilter, Paul Taklaja, Tallinn University of Technology Mohamed Zaki, Arab Academy For Science and Technology; Walid ElKhattam, Ain Shams University; Dr. Rania metwally ElSharkawy, The Arab Academy for Science and Technology and Maritime Transport Kuo Lung Lian, Kuan Sheng Tung, National Taiwan University of Science and Technology
15.40- Determination of Worst Case Loading Margin of Droop-Controlled Islanded Morad Abdelaziz, E. F. El-Saadany, University of Waterloo 16.00 Microgrids 16.00- Feasibility Study Of An Intentional Islanding Operation With A New 16.20 Adaptive Load Shedding
Hasmaini Mohamad, University Technology MARA; Hazlie Mokhlis, Ab Bakar Abu Bakar, Mazaher Karimi, University of Malaya; Nofri NY Dahlan, University of Technology MARA
Parallel Session-Hall B (Energy Harvesting/Renewable Energy), October 3, 2013

Session Chair: Hakan Pasa Partal
Time Paper Title

14.20- Battery Characterization for Vehicular Applications Using 14.40 Hardware-in-Loop Real-Time Simulation 14.40- Human Motion Energy Harvesting by Design of Handheld 15.00 Linear PM Synchronous Generator 15.00- Investigation on the performance of direct borohydride fuel 15.20 cell by using statistical approach 15.20- Research on Electric Energy Harvesting from High Voltage 15.40 Transmission Line 15.40- Hardware-in-Loop Real-Time Simulation of a Battery Storage 16.00 System in a Wind Generation Scheme 16.00- Experimental Verifications of Offshore Marine Substation for 16.20 Grid-Connection of Wave Energy Farm
Authors
Damon Bazargan, Shaahin Filizadeh, Garry Bistyak, University of Manitoba
Mohammad Hossain Mohammadi, Majid Poshtan, American University in Dubai Fatma Gl BOYACI SAN, Osman OKUR, idem YGN KARADA, Il IIK GLSA, Emin OKUMU Fatma, TUBITAK Marmara Research Center Min Yao, Min Zhao, Nanjing University of Aeronautics and Astronautics
Damon Bazargan, Shaahin Filizadeh, University of Manitoba
Rickard Ekstrm, Senad Apelfrjd, Mats Leijon, Uppsala University
Parallel Session-Hall C (Electric Machines&Drives), October 3, 2013

Session Chair: Aydn Yesildirek
Time Paper Title
Authors
14.20- Design of a Mechanical Load Simulator Having an Excitation Current Mehmet Onur GULBAHCE, Handan Nak, Ahmet Derya Kocabas, Istanbul Technical University 14.40 Controlled Eddy Current Brake 14.40- Design and Implementation of Current based Vector Control Model 15.00 of Brushless Doubly Fed Induction Generator 15.00- Feature Form Extraction and Optimization of Induction Machine 15.20 Faults Using PSO Technique 15.20- Elimination The Centrifugal Switch in Single-Phase Induction 15.40 Motor(SPIM) to Increase Torque and Efficiency 15.40IRFOC vs DTC Performance Comparison Analysis 16.00 16.00Experimental Validation on Stator Fault Detection via Fuzzy Logic 16.20
Farid Khazaeli, Hamed Gorginpour, Saeed Ounie, Amirhosein Hajbabaei, Hashem Oraee, Sharif University of Technology
Ammar Medoued, Universite du 20 Aout 1955-Skikda; Abdelaziz Laifa, universite du 20 Aout 1955-Skikda Sobhan Sobhani, University of mazandaran, Iran; Mohammad Ebrahim Talebian, Managing Director (MD) of neka power plant, iran; Hoseyn Vafai Nejad, University of Mazandaran, Iran
Salah eddine Rezgui, University of Constantine 1; Adel MEHDI, unversity of Constantine 1; Said Legrioui, Houssem Meddouce, Abdelmalek Boulahia, University of Constantine 1; Hocine Benalla, Electrotechnics Laboratory, University Mentouri of Constantine
Hamid Fekri Azgomi, Iran University of Science and Technology; Dr. Javad Poshtan, Iran University of Sciense and Technology; Majid Poshtan, American University in Dubai
Parallel Session-Hall A (Renewable Energy/Power Electronics), October 3, 2013

Session Chair: Bnyamin Tamyrek
Time
16.4017.00 17.0017.20 17.2017.40 17.4018.00 18.0018.20
Paper Title
Active Damping Control of DFIG Wind Turbines during Fault Ride Through
Authors
Hao Xu, IEE,CAS
Comparative Study of Adding PV/Wind Energy Systems to Autonomus Micro Grid Coordinated Control of On-Load Tap Changer and D-STATCOM for Voltage Regulation of Radial Distribution Systems with DG Units Dynamic Fault Analysis for Type-3 and Type-4 Generic Wind Turbine Generator Models using Matlab/Simulink Using the Input Characteristic Impedance of a Static Converters for MPPT PV Systems Design and Simulation of a High Power Quality Regenerative PWM Rectifier System for 1 MW
Eman Beshr, Arab Academy for science and Technology B. Bakhshdeh Zad, j. Lobry, and F. Valle Bashir, Department of Electrical Power Engineering, University of Mons Tareq Saber Abuaisha, TU Darmstadt; Oliver Scheufeld, Julian Langstaedtler, FGH e.V. Aachen; Albert Moser, IAEW, RWTH-Aachen; Jutta Hanson, E5, TU Darmstadt Jefferson Zanotti, Federal University of Santa Catarina; Walbermark Marques dos Santos, University of Santa Catarina; Denizar C. Martins, Federal University of Santa Catarina Koray SEPETCI, Bnyamin Tamyrek, Eskisehir Osmangazi University
18.20-18.40 Electric Locomotives
10
Parallel Session-Hall B (Renewable Energy/Smart Grid), October 3, 2013

Session Chair: Magdy Salama
Time Paper Title

16.4017.00 17.0017.20 17.2017.40 17.4018.00 18.0018.20 18.2018.40
Modeling of PV Based Distributed Generation in LV Feeders for Diverse Load Patterns considering Reverse Power Flow Control A New Power Line Communication Modem Design with Applications to Vast Solar Farm Management Performance of a power plant with Fresnel Concentrator that has Average Power in Algerian sites
Authors
Mehmet Hazar Cintuglu, Osama A. Mohammed, Florida International University Fawzi Al-Naima, Nahrain University; Ramzy S. Ali, Ahmed J. Abid, Basrah University; Z. Ghassemlooy, Zhiwei Gao, Northumbria University, UK H. BELTAGY, D. Semmar, Saad Dahlab University; N. Said, Centre of development of the renewable energies EPST/CDER Bouzarah Abdulrahman Kalbat, Columbia University in the City of New York Mohammad Ebrahim Talebian, Managing Director (MD) of neka power plant, iran; Sobhan Sobhani, University of mazandaran, Iran; Arjang Borzooi, University of Mazandaran, Iran Mahmoud Ismail, University of Malaya; Mahmoud moghavvemi, Center of research and applied electronics; T.M. Indra, Department of Mechanical Engineering, University of Malaya
Linear Quadratic Gaussian (LQG) Control of Wind Turbines
New Hybrid System of Fuel Cell Power Plant and Wind Turbine for Household Consumption
Optimization of a PV/microturbine hybrid system for tropical climates
Parallel Session-Hall C (Power Electronics), October 3, 2013

Session Chair: Evren sen
Time Paper Title

16.4017.00 17.0017.20 17.2017.40 17.4018.00 18.0018.20 18.2018.40
Harmonic Filtration by Current Injection and Shunt Capacitors Technique
Authors
Mohamed Tarnini, Beirut Arab University Michael Jordan, Trung Do Thanh, Hauke Langkowski, Detlef Schulz, Helmut-Schmidt-University Hamburg Onur Satlm, YIldIz Teknik Universitesi; Erkan Mese, Yildiz Technical University
Harmonic Network Impedance Identification with Pulse Response Analysis
Investigating DC Link Current Ripple and PWM Modulation Methods in Electric Vehicles
Reliability Assessment of Some High Side MOSFET Drivers for Buck Converter
Vahid Javadian, Sharif University of Technology; Shahriyar Kaboli, Assistant Professor
Sliding mode control applied to UPS inverter using norm of the state error Stability Analysis of Adaptive method for speed sensorless associated with the IFOC Control of Induction Motor
Hamza Makhamreh, Osamn Kukrer, Eastern Mediterranean University Berrissoul Somia, Laboratory Department of Electrical Engineering, ENSET oran; Abderrahmane BOUHENNA, the laboratory of automatic and analysis systems at ENSET
11
Parallel Session-Main Hall (Smart Grid), October 4, 2013

Session Chair: Ehab El-Saadany
Time
Paper Title
High Frequency Power Assessment on Overhead Transmission
Authors
BOUDJEMAA Abdelkrim, NACER Azzeddine, BERBAR Tarik, USTHB, University of Science and Technology Houari Boumediene, Algiers Ahmed Elgargouri, College of Electrical and Electronics Technology, Benghazi, Libya; Mohamed Elfituri, Barrett Broadband Wireless Networks; Mohammed Elmusrati, University of Vaasa Mostafa Shaaban, University of Waterloo; Elham Akhavan, 1- Islamic Azad University, South of Tehran Branch; E. F. El-Saadany, University of Waterloo Afan Bahadur Khan, KING SAUD UNIVERSITY; Imthias Ahamed, King Saud University; Syed Qaseem Ali, Saudi Aramco Chair in Electrical Power; Essam A. Al-Ammar, King Saud University Heloisa H. Mller, Carlos A. Castro, University of Campinas
09.00-09.20 Lines
09.20-09.40 IEC 61850 and Smart Grids 09.40-10.00 Multi-stage PHEV smart coordination in distribution 10.00-10.20 Novel Diverse Scheme to Enhance Demand Response (DR) 10.20-10.40 State Estimator
Optimal Substation PMU Placement Method for the Two-Level
Parallel Session-Hall A (Power Systems), October 4, 2013

Session Chair: Ayman El-Hag
Time
09.0009.20 09.2009.40 09.4010.00 10.0010.20 10.2010.40
Paper Title
Hybrid GM(1,1)-NARnet One Hour ahead Wind Power Prediction Identification of Key Transmission Lines in Power Grid Using Modified Kcore Decomposition Impact of Distributed Generation in Transmission Network Expansion Planning Problem Impedance Based Fault Location Method for HVDC Transmission Line
Authors
Fatemeh Marzbani, Ahmed Osman, Mohamed Hassan Mohamed , American University of Sharjah; Aboelmagd Noureldin, Royal Military College of Canada
Xiangyu Zhang, Dept. of Electrical Engineering, Tsinghua University; Feng Liu, Rui Yao, Tsinghua University; Xuemin Zhang, Mei Shengwei, Dept. of Electrical Engineering, Tsinghua University; Zhen'an Zhang, Xiaomeng Li, HAEPC Electric Power Research Institute
Chandrakant Rathore, S.V.national Institute of Technology, Surat; Dr Ranjit Roy, National Institute of Technology, Surat
Mohammadmehdi Kamali, Javad Sadeh, Ferdowsi University of Mashhad; Yaser Damchi, Ferdowsi University Mashhad
Matrix converter based unified power flow controllers(UPFC)
Atheena A, Amritha S, AMRITA VISHWA VIDYAPEETHAM
12

Session Chair: Ahmed Osman
Time
09.0009.20 09.2009.40 09.4010.00 10.0010.20 10.2010.40
Paper Title
Integrated planning model of generation and transmission expansion for offshore oil-field power system Intelligent reserve management with high penetration of wind power
Authors
David Sun, Xiaorong Xie, Tsinghua University; Jianfeng Wang, Che Wei, Guofeng Liu, Qiang Li, China National Offshore Oil Corporation
Liu Jin, Yu Jilai, Liu Zhuo, Harbin Institute of Technology; Liu Guangyi, China Electric Power Research Institute Chandrakant Rathore, S.V.national Institute of Technology, Surat; Dr Ranjit Roy, National Institute of Technology, Surat Hamid Khorasani, Universidade Estadual Paulista Ilha Solteira-SP, Brazil
Load Uncertainty Based Transmission Network Expansion Planning Transmission Expansion Planning Via a Constructive Heuristic Algorithm in Restructured Electricity Industry Transmission Swtiching to Relieve Voltage Violations
G.J. Liu, Shandong University
Parallel Session-Hall C (Power Electronics/Renewable Energy), October 4, 2013

Session Chair: Blent Vural
Time
09.0009.20 09.2009.40 09.4010.00 10.0010.20 10.2010.40
Paper Title
Stable Operation of Grid Connected Cascaded H-Bridge Inverter under Unbalanced Insolation Conditions Steady-State and Dynamic Performance of Asynchronous Back-To-Back VSC HVDC Link
Authors
Abbas Eskandari, University of Tehran; Vahid Javadian, Sharif University of Technology; Milad Yadollahi, University of Tehran Djehaf Mohamed, University of Sidi bel abbes; Zidi Sid Ahmed, Djillali Liabes University of Sidi BelAbbes; Djilani Kobibi Y, Intelligent Control and Electrical Power System Laboratory; Slimane Souag, UDL university of Sidi Bel Abbes Vitaly Gelman, VG Controls
Thyristor Controlled Rectifiers (TCR) for traction - problems and solutions
Analytical Comparison between Two Topologies of a Series Resonance Inverter from Class D
Labiod Mokhtar, IKHLEF Nabil, ALLAG Hichem, University of JIJEL
PSCAD Simulation of Grid-Tied Photovoltaic Systems and Total Harmonic Distortion Analysis
Abdulrahman Kalbat, Columbia University in the City of New York
13
Parallel Session-Main Hall (Renewable Energy), October 4, 2013

Session Chair: Ugur Savas Selamogullari
Time
11.0011.20 11.2011.40 11.4012.00 12.0012.20 12.2012.40
Paper Title
Overview of Maximum Power Point Tracking Control Methods for PV Systems Renewables and Plug in Electric Vehicles Modeling on Electricity and Gas Infrastructures Scheduling in Presence of Responsive Demand Simulation and Dynamic Control of a Hybrid System for Controlling the Voltage and Frequency of an Island Network
Authors
Saleh Babaa, Dr. Matthew Armstrong, Newcastle University; Volker Pickert, University of Newcastle Mahmoud-Reza Haghifam, Samaneh Pazouki, Islamic Azad University (IAU), South Tehran Branch; Samira Pazouki, Islamic Azad University (IAU), Qazvin Branch Behrooz Ghahremani, Najaf Abad Branch, Islamic Azad University; Saed Abazari, Shahrekord University; Ghazanfar Shahgholian, Islamic Azad University - Najaf Abad Branch
Sustainability Assessment for an Isolated Micro-Grid using Hybird-DG System (Case Study in Marsa Ibtihal Zahran, Arab Academy for Science and Technology; Eman Beshr, Arab Academy for science Alam, Egypt) and Technology Techno-Economic Analysis on Using Battery Storage System with Household PV Panels Hamed Naderi Rad, Javad Sadeh, Ferdowsi University of Mashhad
Parallel Session-Hall A (Power Systems), October 4, 2013

Session Chair: Majid Poshtan
Time
11.0011.20 11.2011.40 11.4012.00 12.0012.20 12.2012.40
Paper Title
Multi-Gene Genetic Programming for Short Term Load Forecasting New Algorithm to Detect Balanced Three Phase Fault during Power Swing on Transmission Lines with Communication Links New Reliability Index for Power System Protection based on Multi-Agent Technique
Authors
Wael Ghareeb, E. F. El-Saadany, University of Waterloo
Mohammed H. Qais, Saad M. Alghuwainem, King Saud University
Nadheer A.Shalash, Unversity Malaysia Pahang; Abu Zaharin Bin Ahmad, University Malaysia Pahang
Online Voltage Stability Monitoring Based on PMU Measurements and System Topology
Dinh Thuc Duong, Kjetil Uhlen, Norwegian university of science and technology
Optimal FACTS Location to Enhance Voltage Stability Using Multi-objective Harmony Search
Abdelaziz Laifa, universite 20 Aout 1955-Skikda; Ammar Medoued, Universite 20 Aout 1955-Skikda
14

Session Chair: Gven Kmrgz
Time
11.0011.20 11.2011.40 11.4012.00 12.0012.20 12.2012.40
Paper Title
Optimal PID Tuning for Load Frequency Control Using Lvy-Flight Firefly Algorithm
Authors
Iraj Faraji, Mahmoud Reza Shakarami, University of Lorestan; Iraj Asghari; Mohsen Akbari, University of K.N. Toosi Dr Ranjit Roy, National Institute of Technology, Surat
Profit Based Unit Commitment Using Gbest Artificial Bee Colony Algorithm Real Time Implementation of H Loop Shaping Robust PSS for Three-Machine Power System using dSPACE Sensitivity Analysis for the IEEE 30 Bus System using Load-Flow Studies
Jayapal Reddy, R.V.College of Engineering; J K. Mendiratta, MVJ College of Engineering
Ayman Faza, Princess Sumaya University of Technology; Ibrahim Totonchi, PSUT Paramet Wirasanti, South Westphalia University of Applied Sciences/Campus Soest; Egon, Ortjohann; Marius Hoppe, Hakam Saffour, South Westphalia University of Applied Sciences/Campus Soest; Danny Morton, University of Bolton
Sequence Hybrid Calculation Technique - An Approach for Clustering Power Systems Analysis
Parallel Session-Hall C (Smart Grid), October 4, 2013

Session Chair: Mustafa Baysal
Time
Paper Title
Short Term Economical Scheduling in an Energy Hub by Renewable and Demand
Authors
Samaneh Pazouki, Islamic Azad University (IAU), South Tehran Branch; Mahmoud-Reza Haghifam, Tarbiat Modares University; Samira Pazouki, Islamic Azad University (IAU), Qazvin Branch Muhammad Ansari, Ali T. Al-Awami, KFUPM; Eric Sortomme, Alstom Grid; Mohammad Abido, King Fahd University of Petroleum and Minerals Ozan Erdinc, Joao P. S. Catalao, University of Beira Interior; Mehmet Uzunoglu, Ali RIfat Boynuegri, Yildiz Technical University Sryang Tera Sarena, Kuo Lung Lian, National Taiwan University of Science and Technology
11.00-11.20 Response
11.20-11.40 Smart Charging of Electric Vehicles using Adaptive Fuzzy Logic 11.40-12.00 Smart Insular Grids: Opportunities and Challenges 12.00-12.20 based on Hybrid of Support Vector Regression and Grey Theory 12.20-12.40 Distribution Network
Very Short Term Solar Irradiance Prediction for A Microgrid System in Taiwan
A Novel Algorithm for Aggregated Demand Response Strategy for Smart
Muhammad Babar, Saudi Aramco Chair In Electrical Power, King Saud University
15
Parallel Session- Hall A (Electric Machines&Drives), October 4, 2013

Session Chair: Sibel Zorlu Partal
Time Paper Title

14.0014.20 14.2014.40 14.4015.00 15.0015.20 15.2015.40 15.4016.00
Lifecycle-Based Design and Optimization of Electrical Motor-Drives Challenges and Possibilities Lyapunov Function-Based Control of Active Power Filters to Reduce the Adverse Effects of Harmonics on Induction Motors Modeling and Simulation of Rotor Bar Skew in Induction Motor Based on Modified Winding Function Method Optimal design of geometrical and physical parameters of permanent magnet machines in the purpose to reduce its vibratory behavior Robust Sliding Mode Observer/Controller of Wind Energy Conversion System Based on DFIG Fed by a Matrix Converter Effect of the Tap Winding Configurations on the Electromagnetic Forces Acting on the Concentric Transformer Coils
Authors
Toomas Vaimann, Ants Kallaste, Aleksander Kilk, Tallinn University of Technology; Anouar Belahcen, Aalto University Ibrahim Aliskan, Bulent Ecevit University; Kayhan Gulez, Yildiz Technical University; Gurkan Tuna, Trakya University; M. zgr Kizilkaya, Turkish Air Force Academy
Harir Miloud, University of Sciences and Technology of Oran (USTO) Algeria; Chaouch Abdellah, University of Mostaganem; Bendiabdellah Azeddine, Benouzza Noreddine, University of Sciences and Technology of Oran (USTO)
Nassira Ferkha, M.Rachid Mekideche, Jijel university; abdellatif Miraoui, Franois Peyraut, Abdoul Ousman, University of Belfort Abdelhak DJOUDI, Centre de developpement des energies renouvelables; H CHEKIREB and EM BERKOUK, Ecole nationale polytechnique Guven Komrgoz, ITU; Tayfun Gundogdu, Istanbul Technical University
Parallel Session-Hall B (Electric Machines&Drives), October 4, 2013

Session Chair: Rached Dhaouadi
Time
14.0014.20 14.2014.40 14.4015.00 15.0015.20 15.2015.40 15.4016.00
Paper Title
Speed Control of Induction Motor Based on Model Reference Adaptive Control Using Experimental Approach Transient Model Parameters Identification of Transformer Based on PSO Algorithm
Authors
Hakan ACIKGOZ, University of Kilis 7 Aralik; Fatih KECECIOGLU, University of Kahramanmaras Sutcu Imam; Ceyhun YILDIZ, Electrical Power Resources Survey and Development; mustafa ekkeli, university of Kahramanmaras sutcu imam Mohammad Valii, Islamic Azad University, Zanjan; Mehdi Bideli, Farid Hojjatiparast, Islamic Azad University Hossein Karimi, Sadegh Vaez-zadeh, University of Tehran; Farzad Rajaei Salmasi, UNIVERSITY OF TEHRAN Irfan Ullah Khan, Rached Dhaouadi, American University of Sharjah Salha Ali Al Disi, Manager System Analysis - Dubai Electricity and Water Authority; Ahmed Osman, American University of Sharjah; Awad Ibrahim, DEWA M. Gkay Unutulmaz, Istanbul Technical University; Lale T. Ergene, ITU
Combined Vector and Direct Thrust Control of Linear Induction Motors
Vibration Suppression in Elastic Drive Systems Using the Immersion and Invariance Methodology
Voltage Stability Assessment of a Real Power System Using a Detailed Dynamic Load Model
Excitation Current Control of A Claw Pole Automotive Alternator
16
Parallel Session-Hall C (Power Systems/Renewable Energy), October 4, 2013

Session Chair: Arif KARAKAS
Time Paper Title
Authors
14.00Solution to Economic Emission Load Dispatch Problem Using Modified Artificial Bee Colony Algorithm Dr Ranjit Roy, National Institute of Technology, Surat 14.20 14.20The Application of Simple Thermal Models to Improve Dynamic Load Models 14.40 14.40Transient stability constrained optimal power flow solution of ac-dc systems using genetic algorithm 15.00 15.00Performance Analysis and Economic Evaluation of a Solar Power Tower in Algeria 15.20 15.20- Techno-economic Optimization of Hybrid PV/Wind/Battery/Diesel System Considering Effects of 15.40 Wind Turbine Parameters on Optimal Wind Turbine Selection Using Genetic Algorithm 15.40Voltage Drifting Method as an Islanding Detector for Grid Connected PV System 16.00
Colin Christy, Yongli Zhu, Iowa State University
Ula Kl, University of Mehmet Akif Ersoy; Krat Ayan, University of Sakarya boudaoud Salah, University of Boumerdes; Khellaf Abdallah, CDER, Route de lobservatoire, Bouzareah, Alger.; Kamal Mohammedi, MESOteam/LEMI, M. Bougara University Boumerdes. Abdullrahman Al-Shamma'a, Khaled E. Addoweesh, King Saud University
Ebadollah Kamyab, Javad Sadeh, Ferdowsi University of Mashhad
Parallel Session-Hall D (Electric Machines&Drives/Renewable Energy), October 4, 2013

Session Chair: Oktay Arkan
Time
14.0014.20 14.2014.40 14.4015.00 15.2015.40
Paper Title
Battery/UC Hybridization for Electric Vehicles via a Novel Double Input DC/DC Power Converter Servo Motor Driver Design for High Performance Applications
Authors
Furkan Akar, University of Duzce; Blent Vural, Yildiz Technical UnIversity Yusuf Yaa, YIldIz Technical University; Ergin Sahin, Cilem Acar, Aybuke Gozutok, Ecem Firat, YTU; Erkan Mese, Yildiz Technical University Mugdesem Tanrioven, Yildiz Technical University; Bedri Kekezolu, A. Erduman, YIldIz Technical University; Hasan Can KILIKIRAN, Yildiz Technical University; Ali Durusu, YIldIz Technical University Gokhan Cakir, Ghadir Radman, Tennessee Tech University
Wind Turbine Effects on Power System Voltage Fluctuations Placement and Performance Analysis of STATCOM and SVC for Damping Oscillation
Probabilistic Modeling of PHEV charging load in distribution systems

M. F. Shaaban
Department of ECE University of Waterloo Waterloo, Canada mshaabanuw@gamil.com
E. F. El-Saadany
Department of ECE University of Waterloo Waterloo, Canada ehab@uwaterloo.ca EVs and advanced battery components that will increase these vehicles efficiencies and driving ranges [3]. From this fund, a tax credit up to 7,500 USD can be claimed for purchasing any new EV. This is a part of the USA government plan to have more than 1 million EVs on the roads by 2015. Among the different types of EVs, the plug-in hybrid electric vehicles (PHEV) are the most promising in the near future. This is because this type of EV is equipped with a gasoline engine to extend the driving range and to be used as a fossil fuel vehicle if the battery is depleted, which makes the vehicles owners feel more secure. Studied show that uncoordinated PHEV charging can have significant impacts on distribution systems. This is because, it is expected that PHEV will likely to be concentrated in residential areas, which pose significant stress on distribution system equipment. According to [4-8] PHEVs charging may lead severe consequences, such as: energy loss increase, voltage limits violation, power quality issues, feeders loading limits violation, and local distribution transformers degradation, and circuit breaker and fuse blowouts. Many work modeled the PHEV charging load either to quantify their impacts on electrical systems or to coordinate their charging. However, most of the presented work adopted some assumptions and ignored some aspects, that can significantly affect the results. In [4], a certain percentage of PHEVs is assumed to be operated in pure electric mode. A certain daily mileage is assigned to all vehicles in [9] and a single charge per day is assumed. In [10-12], the results are based on rigid recharging schedule, which is reasonably assumed. In [13-15], the percentage of annual trips below certain mileage is assumed to be driven in electric mode. A fixed percentage of daily miles are assumed to be driven in electric mode in [16]. In [17], all PHEVs are assumed to consume and charge all their batteries each day. Moreover, although the ambient temperature can have significant effect on the electric range, which can be reduced
Abstract This paper proposes an annual probabilistic model for the energy consumed by a fleet of plug-in hybrid electric vehicles (PHEV) based on Monte Carlo simulation (MCS). This model presents the PHEV charging load scenarios and their probabilities for each hour of the year, which can be easily combined with the normal load models. This facilitates the utilization of this model by local distribution companies to quantify the impacts of PHEV charging on their systems and to define the penetration limit of PHEV in each territory. Index Terms-- Electric vehicles, Emissions, Monte Carlo methods, Probability density function.
I.
INTRODUCTION
Greenhouse gas emissions have become one of the major global concerns in the last few years. Therefore, the global direction now is to utilize low or zero emissions vehicles in transportation. Electric vehicles (EVs) have gained a lot of interest due to availability of electric power systems and the recent development in renewable sources. The history of EVs goes back hundred years ago, when EVs on the roads were more than the fossil fuel vehicles. During the early 1900s, the EVs out-sold the gasoline vehicles [1]. However, due to batteries technology limitations and lack of power electronics technologies, the EVs speeds and ranges were limited. Therefore, the gasoline engine offered superior performance, which lead to the disappearance of the EVs industry by the late 1930s. However, ten years ago, the EVs starts to appear back again after development of batteries and power electronics technologies, which allows comparable speeds and ranges to the fossil fuel vehicles. Thousands of EVs are running on roads in many countries in North America, Europe, and Asia. Some governments are initiating incentives programs to increase EVs market. In Canada, Ontario government established a rebate up to 8,500 Canadian dollars for purchasing or leasing a new EV, starting July 2010 [2]. Ontario government is planning to have one EV out of every 20 vehicles driven in Ontario by 2020. This is part of the government's climate change plan to reduce gas emissions and carbon footprint. In the USA, the government is putting a fund of 2.4 billion USD for the development of next generation
by 35% [18], the aforementioned publications did not consider this effect on energy consumption of PHEV. From the above discussion, it is obvious that there is no unique model to describe the extra loading on the distribution systems due to PHEVs charging. Moreover, the current proposed models adopted some assumptions that may result in misleading outcomes. Therefore, this paper proposes to develop a probabilistic model for the energy and power consumption of uncontrolled charging of light duty fleet (LDF) of PHEVs. This model takes into consideration: Driver habits by incorporating trip purpose (e.g. commuting, education, holiday trip, etc.) Different trip mileage according to the purpose of trip Unlimited Charging events per day. Ambient temperature effect
battery characteristics. On the other hand, the grid side converter maintain constant DC link voltage and power factor on the grid side, which is typically unity. Therefore, it is obvious that the delivered power to the battery pack is independent on the grid voltage due to the isolation via the power electronics converters. Thus, the PHEV charging load is modeled as constant power load at unity power factor in the load flow analysis. Probabilistic model for the charging power for a fleet of PHEV is explained in details in the next section.
Figure 1. PHEV battery charger.
The remainder of the paper is organized as follows: Section II describes the problem description. Sections III presents the modeling. Section IV describes the case study, and section V offers conclusions. II. PROBLEM DESCRIPTION The work presented in this paper tackles the PHEV modeling in power systems based on the current situation of the grid, where smart signals to charging locations are not yet available. Therefore, the near future PHEVs charging mechanism is expected to be uncontrolled. For the electricity providers, the best time to charge the vehicles in their territory is typically at night, when normal load demand in minimum and low cost generating units are the marginal producers. Therefore, the extra load due to vehicles charging is met at low cost and without straining the existing transmission and distribution systems [11]. On the other hand, the preferred time for consumers is as soon as they return home from a trip, when charging is most convenient, as the driver is already at the vehicle. In addition, the drivers will want their vehicles batteries to be as fully charged as possible whenever needed. This situation of the consumers is denoted as the uncontrolled charging scenario. As charging process takes hours, it is not likely to be performed in charging stations. Thus, PHEVs charging is assumed to occur at home in this work, where the charger have the following characteristics: Can only charge the vehicle, discharging is not allowed Control the charging current and voltage of the battery pack in the vehicle, which should match the battery specifications. Supply electric energy to the vehicle battery directly as soon as it is plugged- in.
III.
PHEV CHARGING LOAD MODEL
In this section, the PHEV load model is described, where the following assumptions are considered: The time step is one hour, and any changes within the hour are excluded. A starting period of 10 minutes is assigned to air conditioner and heater [11]; this consumed energy is averaged over the hour.
The model mainly composed of 9 stages. Due to unavailability of historical data for the energy consumption of a LDF of PHEV, a probabilistic model based on Monte Carlo simulation (MCS) along with transportation survey data are used in this work to generate virtual PHEV charging scenarios. This model is explained in details in the following steps. The output of this model can be used to model the PHEV energy consumption as a multi-state variable in the same way as for the normal load. The proposed model can be described in steps as follows: Step 1: The total annual trips of a light duty fleet of PHEV are distributed among K purposes, using the (1) and (2), where the categorical probability distribution function in (2) represents the probability of a trip of a certain purpose to occur.
M ( u ) = Vtot M AVG f1 ( u )
[ x1 = r ] f 1 ( x1 ) = p1 (r) r =1 UP
u = 1,2,..., U P
(1) (2)
The PHEV charging system consists of two converters: AC/DC grid side converter, and DC/DC battery side converter, as shown in Fig. 1. The DC/Dc converter controls the delivered power to the battery pack, which depends on
where p1(r) is the probability of occurrence of trip of purpose r; [x1=r] is the Iverson bracket (evaluates to 1 if x=r, 0 otherwise); x1 in an integer variable indicating the purpose of the trip; UP is the number of purposes; M(u) is the annual number of trips for purpose u; MAVG is the average annual trips per vehicle; Vtot is the total number of vehicles in the considered fleet. Step 2: Each purpose annual trips are further distributed on 12 months using (3) and (4). The categorical distribution
function in (4) represents the probability of a purpose trip to occur in a certain month.
M ( u ,m ) = M ( u ) f 2 ( u ) ( m ) u, m
The probability of having daily trips z per vehicle in month m and day d. Step 5: For each purpose, the daily trips are distributed among 24 hours indicating the time of the finished trip as in (10). The categorical pdf defined in (11) describes the probability of a purpose trip to be finished in a certain hour.
(3) (4)
f 2 ( u ) ( x 2 ) = p 2 ( u ,a )
a =1
12
[ x2 = a ]
where x2 is an integer variable indicating the month of the trip; p2(u,a) is the probability of a trip of purpose u to be in month a; M(u,m) is the number of trips of purpose u in month m for the entire fleet. Step 3: The monthly purpose trips are further distributed daily using (5) and (6). The categorical distribution function in (6) represents the probability of a purpose trip to occur in a certain day. In this work, each month is modeled by 2 days: weekday and weekend. In other words, f3(u)(x3) is the same for x3=1, 2, 3, 4 and 5 (weekdays), and it has another value for both x3=6 and 7 (weekend).
M ( u ,m ,d ,h ) = M ( u ,m ,d ) f 5( u ) ( h ) u, m, d , h
(10) (11)
f 5 ( u ) ( x 5 ) = p ( u ,c )
c =1
24
[ x5 =c ]
where x5 is an integer variable indicating the hour at which the trip is finished; p(u,c) is the probability of finishing a trip of purpose u in hour c; M(u,m,d,h) is the total number of trips of purpose u in month m, and day d which are finished in hour h. Step 6: For every trip finished in a certain hour, Ny equiprobability virtual scenarios are generated. Each scenario has a probability of occurrence of 1/Ny. MCS is used to generate these scenarios of trip distance as in (12).
M ( u ,m , d ) =
M ( u ,m ) f 3 ( u ) ( d ) N weeks ( m )
7 [ x3 = b ]
u, m, d
(5)
f 3( u ) ( x3 ) = p( u ,b )
b=1
(6)
S(u ) = CDF1 6( u ) ( , ( u ) )
(12)
where x3 is an integer variable indicating the day of the trip; p(u,b) is the probability of a trip of purpose u to occur in day b; Nweeks(m) is the number of weeks in month m; M(u,m,d) is the total number of trips of purpose u in month m and day d. Step 4: The total number of daily trips is distributed among the group of vehicles according to a lognormal probability density function (pdf) with mean defined by the average daily trips per vehicles and standard deviation of one trip, as in (79).
M ( m ,d , z ) =
where, S(u) is the distance in miles for a trip under purpose u; is a vector of length Ny of normally distributed random variable between zero and one; CDF-16(u) is the inverse of the cumulative density function, which describes the probability of a trip of purpose u to be less than certain distance; (u) is the set of parameters for CDF6(u). Step 7: The daily trips are assigned to the vehicles according to the following assumption; the trips with the earliest finish time are assigned to the vehicles with the largest number of remaining trips per day. Thus, S(m,d,h,v) represents the finished trip distance for vehicle v in month m, day d, and hour h. Step 8: After generating the finished trip distances for the Ny annual scenarios, the following model is used to describe the energy consumption from the grid, assuming all vehicles to operate with lithium ion batteries. In this step, the PHEV batteries are discharged by the energy consumed for every finished trip and in the next step, they will be charged from the grid. A vector (SE(t)) represents the useful portion of the stored energy in kW.h of the fleet batteries at the end of each hour and is updated hourly in this step. It is assumed that this vector is shared between the four classes of the considered fleet according to the percentage sharing of each class, which is given in [4] and shown in Table I. This vector is defined as:
M
u
( u ,m ,d )
Pr ob ( m ,d ) ( z )
m, d , z
(7)
0 .5 z = 0 f 4 ( m ,d ) ( x 4 ) x4 = 0 (8) Pr ob( m ,d ) ( z ) = z + 0.5 f (x ) z = 1,2,3,... 4 ( m ,d ) 4 x4 = z 0 .5 (ln x 4 ( u ,m ,d ) 2 1 x>0 u exp x 4 2 2 2 f 4 ( m ,d ) ( x 4 ) = x0 0 (9)
where z is a positive integer number indicating the number of daily trips per vehicle, which is assumed to be the nearest integer to variable x4; Y(u,m,d) is the mean daily trip per vehicle, which is defined as M(u,m,d) / Vtot ; is the standard deviation, which is assumed to be one trip; M(m,d,z) is the number of vehicles with daily trips z in month m and day d; Prob(z,m,d)
SE ( t ) = [estored (1,t ) estored (1,t ) ... estored ( v ,t ) ]
(13)
where, estored(v,t) represents the battery useful stored energy in kW.h for vehicle v at the end of hour t. The initial value of the stored energy in each battery which represents the useful battery charge is defined as:
estored ( v ,0) =
(CL) AER( v ) overall
(14) (15)
overall = BT PE M A
where AER(v) is all electric range (AER) in miles for vehicle v; (CL) is the average tractive effort required per mile (kW.h/mile) for class CL; overall is the overall efficiency of the vehicle; BT, PE, M and A are the efficiency of battery, onboard power electronics, motor and accessory load respectively.
TABLE I. VEHICLE CLASS PERCENTAGE SHARING [4]
CAR
VAN
SUV
PICK-UP
Urban Rural
56.9% 42.9%
9.2% 8.0%
19.6% 19.1%
14.3% 30.0%
where e1(v,t) is the total tractive energy required in kW.h for the trip finished at hour t for vehicle v to overcome vehicle inertia, road resistance, and aerodynamics drag; e2(v,t) is the energy in kW.h required for the trip finished at hour t by vehicle v to maintain the cabin temperature comfortable for the vehicle driver and passengers; Tmax, Tmin are the maximum and minimum temperature limits for battery usage respectively; TAC, THT are the average thresholds for the air conditioner operation and heater operation respectively; eAC(v,t), eHT(v,t) are the energies required in kW.h for cooling and heating vehicle v cabin at hour t respectively; PAC-st, PACcont are the powers in kW consumed by the air conditioner during starting and continuous operation respectively; PHT-st, PHT-cont are the powers in kW consumed by heater during starting and continuous operation respectively; ttrip(v,t) is the duration in minutes of the trip finished in hour t for vehicle v; AVG(q) is the average vehicle speed in mile/h for purpose q.
Step 9: A vector (CE) represents the consumed power during each hour by the LDF of PHEV is updated hourly in All 53.0% 8.9% 19.4% 18.7% this step. Its length is 8760 and is initially set to zeros. In this work, as soon as a vehicle finishes a trip, it starts charging, as For each hour, the batteries stored energy vector (SE(t)) is the work focus on uncoordinated charging, where the updated by the consumed energy for finished trips, which can preferred time for consumers is as soon as they return home be represented by (16) and (17). The consumed energy by from a trip, when charging is most convenient. In addition, each finished trip is evaluated as a sum of two variables: 1) the the drivers will want their vehicles batteries to be fully total tractive energy required to overcome vehicle inertia, road charged as possible whenever needed. Therefore, the set of resistance, and aerodynamics drag as defined in (18), and 2) the energy required to maintain the cabin temperature charging vehicles at each hour contains all the plugged vehicles. The consumed energy vector (CE) is updated each comfortable as in (19-22). hour for each charging vehicle using (23-24). The batteries e cons ( v ,t ) = e1( v ,t ) + e 2 ( v ,t ) (16) characteristics are approximated as in [11], where the batteries are charged by full capacity except for the final hour, as in (24). e cons ( v , t ) < e stored ( v , t ) e stored ( v , t ) e cons ( v , t ) e stored ( v , t ) = (17) Finally, the battery charges of the plugged vehicles are e cons ( v , t ) e stored ( v , t ) updated using (25). 0
e1( v , t ) S ( v ,t ) ( C ) = overall 0 T min temp elsewhere
(t )
T max
(18)
CE ( t ) = CE ( t 1 ) +
v CV ( t )
e stored
( v ,0 )
e stored
( v ,t )
CH t CH ( v , t )
(23)
e 2 ( v ,t )
e AC ( v ,t ) = e HT ( v ,t ) 0
T AC temp ( t ) T max T min temp ( t ) T HT elsewhere
(19) tCH ( v ,t )
t trip ( v ,t ) =
S ( u , v , t ) 60
e stored ( v , 0 ) e stored ( v ,t ) PCH ( v ) = 1
e stored ( v ,0 ) e stored ( v ,t ) PCH ( v ) e stored ( v , 0 ) e stored ( v ,t ) PCH ( v )
>1 1
(20)
ttrip ( v , t ) 10
(24)
e stored ( v ,0 ) e stored ( v ,t ) e stored ( v ,t +1) = e stored ( v ,t ) + tCH ( v ,t ) v CHV H (25)
ANG ( u )
e AC ( v , t )
PAC st ttrip ( v , t ) 60 BT PE AC = 10 PAC st + PAC cont ( ttrip ( v , t ) 10 ) BT PE AC 60

PHT st ttrip ( v , t ) 60 BT PE AC = 10 PHT st + PHT cont ( ttrip ( v , t ) 10 ) BT PE AC 60
ttrip ( v , t ) > 10
e HT ( v , t )
ttrip ( v , t ) > 10
where CE(t) is the tth element in vector CE, and represents the average consumed power in kW during hour t; CH is the charger efficiency; tCH(v,t) is the required charging time in hours of the battery of vehicle v at time t; PCH(v) is the charger ttrip ( v , t ) 10 continuous rating in kW for vehicle v; CHV(t) is the set of (22) vehicles required to be charged at hour t.
(21)
IV.
CASE STUDY
PHEV consumption (kW)
250 200 150 100 50 0
Mainly, three types of chargers are commonly used for PHEVs charging. The specifications of each charger are shown in Table II [6]. However, the expected most used charger in Canada is level 2R charger, as recommended by vehicles manufacturers [19], which is adopted in this paper for simulation results. The data for the pdfs utilized in the proposed model is provided by [20]. This data results in the shown parameters in Table III. The work presented in this paper considers only the uncontrolled at home charging scenario, as it represents the current situation of the grid as it does not require intelligent control. Hence, the set of the vehicles that are required to be charged (CVH) includes all the vehicles in the fleet under study. The results presented in this section are based on the values shown in Table IV. Fig. 2 and Fig. 3 show five scenarios for uncoordinated charging for 100 vehicles in a typical weekday and weekend in March, respectively. The effect of considering the trip purposes is salient in the consumed energy. As shown in figures, the consumption during weekend is higher compared to that in weekdays. Moreover, the peaks of the consumption occur in later time in weekends around 6 pm compared to 5 pm in weekdays. In addition, it can be observed from the figures that relatively higher consumption in the period between 7 pm and 10 pm. This is due to longer daily trips, which have higher probability to occur in weekends. Also, March PHEV consumption trend is less than that in August, due to long daily trips in summer vacations in August compared to educational trips in March. Thus, it is shown that the consumption varies significantly from weekend to week day, and also from month to month. These variations can only be tackled by considering trip purposes as shown in the presented model.
TABLE I. PHEV CHARGERS RATINGS [6]
Scenario 1 Scenario 4
Time of day Scenario 2 Scenario 5
Scenario 3
Figure 2. PHEV consumption in a typical weekday in March.

PHEV consumption (kW) 350 300 250 200 150 100 50 0
Time of day Scenario 1 Scenario 4 Scenario 2 Scenario 5 Scenario 3
Figure 3. PHEV consumption in a typical weekend in March. TABLE III. MSC PARAMETERS
PURPOSE Commuting Education Shopping Visit friends Day trip Business Escort education Personal business and other escort Holiday trip Other trips
PDF Lognormal Lognormal Lognormal Weibull Weibull Weibull Lognormal Lognormal Weibull Lognormal
PARAMETER 1 =3.27 =2.48 =2.76 c=83.81 c=79.63 c=111.75 =2.16 =3.02 c=176.47 =3.42
PARAMETER 2 = 1.02 = 1.16 = 1.18 k=0.93 k=1.19 k=1.27 = 1.38 = 1.32 k=2.67 = 1.29
TYPE Level 1R Level 2R Level 3R
SPECIFICATIONS 110/120V, AC, 15 -20 amps Does not require installation and can use standard 120v electrical outlet Typical charge times: 8-12 hours 208-240V, AC, 15-30 amps Requires special installation Typical charge times: 3-8 hours Known as DC fast charging 440V, DC, 125 amp, 55kW or higher Requires special installations in the station and the vehicle DC Fast Charging typically returns 50% of a PHEV batterys capacity in under 30 min.
where, and are mean and standard deviation of the Lognormal pdf respectively; k and c are the shape and scale parameters of the Weibull pdf respectively.
TABLE IV.
PHEV SIMULATION PARAMETERS
PARAMETER min HT AC max BT PE M
VALUE -10oC 5 oC 27 C 40 oC 92% [4] 95% [4] 92% [4]

o
PARAMETER A CH AST Acont HST Hcont Xv
VALUE 88% [4] 95% [4] 2.99 kW [18] 2.1 kW [18] 4 kW [18] 2 kW [18] 50-mile
V.
CONCLUSIONS
In this paper, a probabilistic model for PHEV charging load is proposed. The proposed model utilizes MCS to generate virtual scenarios for the consumed power by a fleet of PHEV taking into consideration driver habits, different trip mileage, unlimited Charging events, and ambient temperature effect. This model provides outcomes that are more accurate for the PHEV charging consumption. Moreover, the paper presented an annual PHEV model, which describes the different scenarios of PHEV charging consumptions and their probability of occurrence. This model can be easily combined with knows normal load models, which facilitates the LDC to quantify the impacts of PHEV charging on their systems, and the PHEV penetration limits in each territory. A sample case study of the proposed model is provided, where the effect of considering the driver habits is significant on the consumption. REFERENCES
[1] Canadian national campaign for electric vehciles. [Online]. Available: http:// www.evcanada.org/Default.aspx [2] Government Available: of Ontario. [Online].
Laboratory, Nov 2007. [10]J. C. Gomez, M. M. Morcos, "Impact of EV battery chargers on the power quality of distribution systems," IEEE Transactions on Power Delivery, vol. 18, no. 3, pp. 975- 981, July 2003. [11]S.W. Hadley, A.A. Tsvetkova, "Potential Impacts of Plug-in Hybrid Electric Vehicles on Regional Power Generation," The Electricity Journal, vol. 22, no. 10, pp. 56-68, Dec. 2009. [12]S. Letendre, R. A. Watts, "Effects of plug-in hybrid electric vehicles on the Vermont electric transmission system," University of Vermont Transportation Research Center, Vermont, 2009. [13]A. Hajimiragha, C. Canizares, M. W. Fowler, A. Elkamel, "Optimal transition to plug-in hybrid electric vehicles in Ontario, Canada, considering the electricity-grid limitations," IEEE Transactions on Industrial Electronics, vol. 57, no. 2, pp. 690-701, Feb. 2010. [14]P. Denholm, W. Short, "An evaluation of utility system impacts and benefits of optimally dispatched plug-in hybrid electric vehicles," National Renewable Energy Laboratory, NREL/TP-620-40293, 2006. [15]G. Strbac, C.K. Gan, M. Aunedi, V. Stanojevic, P. Djapic, J. Dejvises, P. Mancarella, A. Hawkes, D. Pudjianto, S. Le Vine, J. Polak, "Benefits of Advanced Smart Metering for Demand Response based Control of Distribution Networks," Imperial College, London, 2010. [16]K. Parks, P. Denholm, T. Markel, "Costs and emissions associated with plug-in hybrid electric vehicle charging in the xcel energy Colorado service territory," National Renewable Energy Laboratory, Technical report NREL/TP-640-41410, 2007. [17]J.A.P. Lopes, F.J. Soares,P.M.R. Almeida, "Identifying management procedures to deal with connection of Electric Vehicles in the grid," in IEEE Bucharest Power Tech Conference, Bucharest, Romania, 2009, pp. 1-8. [18]R. A. Barnitt, A. D. Brooker, L. Ramroth, J. Rugh, K. A. Smith, "Analysis of Off-Board Powered Thermal Preconditioning in Electric Drive Vehicles," in the 25th World Battery, Hybrid and Fuel Cell EV Symposium and Exposition (EVS-25), Shenzhen, China, 5-9 Nov. 2010. [19]Chevrolet,. (2011) Chevrolet electric vehicle specifications. [Online]. http://www. chevrolet.com/volt/ [20]"National Travel Survey 2009," Department of Transportation, Great Britain, London, 2009.
http://www.ontario.ca/
[3] White house official website. [Online]. Available: http://www.whitehouse.gov/ the_press_office/President-ObamaAnnounces-24-Billion-in-Funding-to-Support-Next-GenerationElectric-Vehicles [4] D. Wu, D. Aliprantis, K. Gkritza, "Electric energy and power consumption by light-duty plug-in electric vehicles," IEEE Transactions on Power Systems, vol. 26, no. 2, pp. 738-746, May 2011. [5] U. Eberle, R. von Helmolt, "Sustainable transportation based on electric vehicle concepts: a brief overview," The Royal Society of Chemistry, vol. 3, pp. 689-699, May 2010. [6] Commonwealth Edison Company, "Initial Assessment of the Impact of the Introduction of Plug-in Electric Vehicles on the Distribution System," ILLINOIS COMMERCE COMMISSION, Dec. 2010. [7] J. T. Salihi, "Energy requirements for electric cars and their impact on electric power generation and distribution systems," IEEE Transactions on Industry Applications, vol. IA-9, no. 5, pp. 516-532, Sept. 1973. [8] R. Liu, L. Dow, E. Liu, A Survey of PHEV Impacts on Electric Utilities, in Proc. 2011 IEEE Innovative Smart Grid Technologies (ISGT) Conf., PP. 1-8. [9] M. Kintner-Meyer, K. Schneider, R. Pratt, "Impacts assessment of plug-in hybrid vehicles on electric utilities and regional US power grids part 1: Technical analysis," Pacific Northwest National
Design of Power System Stabilizers Based on Controller for Power System Stability Enhancement
K. Ellithy, Senior Member, IEEE S. Said, Student Member, IEEE O. Kahlout, Student Member, IEEE
Electrical Engineering Department Qatar University Doha, Qatar k.ellithy@qu.edu.qa
Abstract The problem of the poorly damped low-frequency (0.1-0.4 Hz) oscillations of power systems has been a matter of concern to power engineers for a long time, because they limit the power transfer capability in power systems. The power systems stability is also affected by these poorly damped oscillations and can lead to the system instability. The paper presents a design PSSs based on -controller to enhance power systems stability and improve power transfer capability. MATLAB dynamic model was developed for a power system and lead-lag PSS structure is considered in the model. Damping torque technique is applied to tune the PSS parameters. The results of this technique have been verified by eigenvalue analysis and timedomain simulations. The optimal sampling time was determined for transferring the s-domain of PSS model to digital (z-domain) model and then it was implemented on -controller chip. The peripheral interface controller (PIC) -controller type was used and the developed MATLAB model was interfaced with the controller. The simulations results show that the system time responses under different operating conditions are well damped with the designed PSS. Moreover, the proposed PSS based controller is relatively simple and suitable for real-time applications in the future smart power grid where the stabilizing signals to the PSSs will be provided by wide-area measurement signals using the new technology of synchrophasors. Keywords-dynamic Stability, Damping Torque, Eigenvalues, Power System Stabilizers, -Controller.
In the evaluation of stability, the concern is the behavior of the power system when subjected to disturbance. The disturbance may be small or large. Small disturbances in the form of load changes take place continually, and the system adjusts itself to the changing conditions. The system must be able to operate satisfactory under these conditions and successfully supply the maximum amount of load. On the other hand, the system must be capable of surviving numerous disturbance of a severe nature, such as short-circuit, loss of large generator or load, or loss of a tie between two subsystems [3]. Small disturbances may result in oscillations that increase in amplitude exponentially and lead the power system to instability [1, 3, 4]. In a correctly designed and operated system, these oscillations are damped toward a new non-oscillatory operating condition which leads the power system to stability. Early investigations considered attention in the literature of the excitation system and its ability in enhancing stability of the power system. Researchers have found that the negative damping of large interconnected coupled system introduced by voltage regulators with high gain was the main reason to experience oscillations [4]. A solution to improve the damping in the system was achieved by introducing a stabilizing signal into the excitation system. This signal should be taken from the Power System Stabilizer (PSS) [2, 4 ,5]. Interest in digital control has grown rapidly in the last several decades since the introduction of microcontrollers. A microcontroller is a single-chip computer, including most of a computers features, but in limited sizes. Today, there are hundreds of different types of microcontrollers, ranging from 8-pin devices to 40-pin, or even 64-pin or higher pin devices [13, 14, 17]. The serial communication capability of PIC microcontrollers with the MATLAB software along with the Simulink, which is a graphical design tools in MATLAB, contributes in developing many programs which are used in simulating and controlling electrical & mechanical systems [15, 17, 18]. The aim of this paper is to design and implement a PSS based on microcontroller for power system stability enhancement.
I.
INTRODUCTION
Power system stability may be generally defined as the characteristic of a power system that enables it to remain in a state of operating equilibrium under normal operating conditions and to regain an acceptable state of equilibrium after being subjected to a disturbance [1, 3, 4]. Dynamic stability can be defined as the behavior of the power system when subjected to small disturbances. It is usually concerned as a problem of insufficient or poorly damping of system oscillations. These oscillations are undesirable even at lowfrequencies, because they reduce the power transfer in the transmission lines. The most important types of these oscillations are the local-mode, which occurs between one machine and the rest of the system, and the inter-area mode oscillations that occurs between interconnected machines [3].
This work was sponsored by Academic Research Office at Qatar University
II.
Power System Modelling and PSS Design
A. Power System Model A single machine connected to an infinite bus through transmission line (SMIB) is considered in this study. The system consists of synchronous generator connected to a power grid (infinite bus) through a transmission line. The generator is equipped with exciter-automatic voltage regulator (AVR) [8]. Fig. 1 shows the single line diagram of the singlemachine infinite bus (SMIB) system. Turbine-governor dynamics is neglected. The parameters of the studied SMIB power system are taken from literature [1]. B. PSS Design Power System Stabilizer (PSS) is a device which provides additional supplementary control loops to the AVR. PSSs are often used as an effective means to add damping to the generator rotor oscillations [2, 7, 8, 9, 10, 11]. Adding supplementary control loops to the generator AVR is one of the most common ways of enhancing both dynamic and transient stability [1]. To provide damping for the generator rotor oscillations, PSS must produce a component of electrical torque in phase with rotor speed deviations. The basic functions of the PSS is to add a stabilizing signal that compensates the oscillations of the voltage error of the excitation system during the dynamic/transient state, and to provide a damping component when its on phase with rotor speed deviation of machine. The theoretical basis for a PSS may be illustrated with the aid of the block diagram shown in Fig. 2.
As shown in Fig. 2 the PSS block diagram representation is composed of three blocks: a gain block, a signal washout block and phase compensation block. The stabilizer gain (KPSS) function is to determine the amount of damping introduced by the PSS [4]. The function of the phase compensation block is to provide the appropriate phase-lead characteristic to compensate for the phase lag between the exciter input and the generator electrical (air-gap) torque. In Fig. 2 a single first-order phase compensation block were used to represent the phase compensation circuit. However, in practice two or more first-order blocks may be used to achieve the desired phase compensation. In some cases, second-order blocks with complex roots have been used. Normally, the frequency range is 0.1 to 2 Hz, and the phase-lead network should provide compensation over this entire frequency range. The phase characteristic to be compensated changes with system conditions; therefore, a compromise is made and a characteristic acceptable for different conditions is selected. Generally some under-compensation is desirable so that the PSS, in addition to significantly increasing the damping torque, results in slight increase of the synchronizing torque [1,3,4]. The overall linearized state-space model of the power system (SMIB including PSS) has been developed using the state-space equations [12]. The overall linearized state-space model of the system is given by:
x = AX + BU
(1)
y = CX
The system state-variables and A, B, C matrices were developed using component connection technique [6] and are given by
Fig. 1 Single Line Diagram of SMIB Power System
C. Design of PSS Based on Damping Torque Technique Stability of the power system depends mainly on the damping of the torque. If any lack of sufficient damping torque occurs in the system, this would lead the system to oscillatory instability [1, 3]. The damping torque expression has been derived for SMIB power system and is given by:
Fig. 2 Block Diagram of SMIB System with PSS
Table 1: System Eigenvalues and Damping Torque under Different Operating Points with and without PSS [6]
Eigenvalues i and Damping Torque KD 1 2 3,4 5,6 KD Without PSS Nominal Operating Point (136 MW, 83.2 MVAR) -12.9502 -6.932 -0.3348 + j9.4219 3.1744 Unstable Operating Point (112 MW, -32 MVAR) -17.3355 -3.737 0.2604 + j8.4500 -2.4675 With PSS Nominal Operating Point (136 MW, 83.2 MVAR) -18.7753 -0.1001 -0.9374 + j10.0523 -6.6175 + j3.4306 8.7570 Unstable Operating Point (112 MW, -32 MVAR) -21.0756 -0.1001 -0.4095 j9.2194 -7.2586, -4.7316 3.8822
It can be seen that the damping torque KD is mainly a function of the constants K2, K3 and K5, where K2 is the change in the electrical torque for a change in the flux linkages in the d-axis with considering a constant rotor angle, K3 is the impedance factor in which the external impedance is a pure reactance, and K5 is the change in the terminal voltage with respect to the change in the rotor angle with considering a constant flux linkages in the d-axis. The K3 is constant, is only function system reactance. The results show that under all system operating points, K2 is positive as seen in while K5 can be a negative value under certain operating points, as seen in Fig. 3 [6]. The negative value of K5 gives a negative damping (KD = -2.4675). The negative KD is the main cause of the system instability. The instability can also be observed from the positive value of the eigenvalues mode (3,4= 0.2604 8.4500i) shown in Table 1. The modes 3,4 are corresponding to electro-mechanical oscillations of synchronous generator. The instability can also be observed from the time-domain simulation shown in Fig. 4. According to equation (1), KD results in a positive damping for K5 > 0, but a negative damping for K5 < 0. Our concern here is when K5 is in negative, and KA is large enough to lead the system to instability. The constant K5 becomes positive for considering low value of external system reactance and low generator outputs, whereas it is negative for high system reactance and high generator outputs. In this case, this may offset the inherent machine damping torque. Therefore, a power system stabilizer (PSS) is introduced to eliminate the effect of the negative damping torque and lead the system to stability. After adding the PSS to the system, the damping torque equation due to the PSS is shown below:
Fig. 3a Variation of K2 at Different Operating Points
Fig. 3b Variation of K5 at Different Operating Points
The system is stable with the designed PSS. The stable condition can be seen from the system eigenvalues of the system with PSS as listed in Table 1 [6] and from the blue line of system response shown in Fig. 4. Also the damping torque KD is positive (KD = + 3.1744) [6]. III. Design of PSS Based on Microcontroller and Implementation
As shown in previous equation, the second part of the equation is due to the PSS. Therefore, the PSS will help in improving the damping and thus the stability of the system.
A. Digital Controllers Overview Nowadays, digital computers and microcontrollers are mostly used in control engineering applications. Fig. 5 illustrates an example of a typical microcontroller controlled system. The
error signal in this system is analog, and an analog-to-digital (A/D) converter is used to convert the signal and make it digital so that the computer can read it.
Rotor Speed Deviation 0.08 0.06 Deviation of Rotor Speed (Rad/s) 0.04 0.02 0 -0.02 -0.04 -0.06 -0.08 With PSS Without PSS
The microcontroller is considered as a self-contained computer system, which is designed to have, in general, a microprocessor, memory and I/O connections in one small chip. This is considered as an advantage where it doesnt need other specialized external components for its applications, B. Design of Digital Power System Stabilizer In this project, the designed s-domain power system stabilizer have been transformed into digital domain (zdomain) which can be implemented into the microcontroller and thus it can sense the rotor speed deviation through the A/D and then it can provide the aux. signal to the D/A which then it is introduced into the AVR. Bilinear Transformation (Tustin Transformation) has been used to transfer the PSS transfer function from the s domain to z domain using Bilinar transformation, the s-domain PSS transfer function show in equation (5) has been transferred to the digital domain (zdomain) as shown in equation (5).
4 Time (sec)
Fig. 4(a) System Response with and without PSS at stable operating points under 1% change in
where Kpss is the power system stabilizer gain, Tw is the washout time constant, T1 is the lead time constant, T2 is the lag time constant and T is the sampling time. The selection of the sampling time is the key for matching the s-domain PSS to the z-domain PSS frequency response. According to the literature [16], Astrom and Wittenmart (1984) have developed a guideline for selecting the sampling interval. According to them the value of T in seconds should be in the range of 0.15/ m to 0.5/ m, where m is the zero dB frequency (rad/s) of the magnitude frequency response curve for the cascaded analog compensator [13, 14, 16, 18]. Bode plot should be drawn in order to determine the zero dB frequency of the cascaded connection between the PSS and the SMIB power system. According the derived bode plot the zero dB frequency is approximately equal to 8.73 rad/s, so the limits of the sampling interval are TMax = 0.0573sec and Tmin = 0.01718 sec so the sampling frequency should between 17.45 Hz and 58.21 Hz. As mentioned earlier that the sampling frequency or the sampling time is the key for matching the s-domain PSS to the z-domain PSS frequency response, since the range of sampling frequency has been determined it is important to see the effect of sampling frequency on the frequency response of the system and to choose the best sampling frequency based on closest frequency response of digital PSS to the s-domain PSS frequency response. Fig. 6 shows the effect of sampling frequency on the frequency response of the digital PSS, it is observed that as the sampling frequency increases in the range of the sampling frequency mentioned above, the frequency response of the digital filter gets closer to the s-domain filter.
Fig. 4(b) System Response with and without PSS at Stable & Unstable Operating Points Under 1% Change in
The process of (A/D) conversion is by sampling the input signal periodically and covert these samples into a digital code so that the computer can process it. Then, the digital code is being run by software for executing the given actions in the code [13, 14, 15, 16]. Converting the digital signal that comes from the computer into an analog signal is normally done by using a digital-to-analog (D/A) converter. The operation of this converter is usually approximated by zero-order hold transfer function.
Fig. 5 Block Diagram of -controller Based Digital Control System
In this study the sampling frequency fs have been selected as 55 Hz, and this is because at this sampling frequency, the digital PSS has the closest frequency response the s-domain PSS. In order to convert the digital PSS transfer function into C-code which can be implemented in microcontroller to perform the digital PSS transfer function, the selected sampling time, which is T= 1/fs = 1/55 =0.01818 sec [6] should be substituted in the equation (4), which can be then transformed into the C code. This code can be implemented on microcontroller using C language programming.
10
magnitude
the difference in the voltage level between the microcontroller and the computer, MAX232 is used to match the voltage level between the microcontroller (05V) and the computer (-10 10V). The overall system is represented in Fig. 8 [14]. IV. Simulations of the SMIB Power System with MCU PSS and Results
5 0 0 1
z-domain at 25 Hz z-domain at 35 Hz z-domain at 45 Hz z-domain at 55 Hz s-domain
As a final stage, a microcontroller based digital PSS has been implemented and it is had successfully interfaced with MATLAB/Simulink, moreover it has also stabilized the simulated SMIB power system at different operating points. Fig. 9 shows photos of the hardware of the designed microcontroller based digital PSS interfaced with MATLAB/Simulink workstation.
10
20
30
40
50
60
70
80
z-domain at 25 Hz z-domain at 35 Hz z-domain at 45 Hz z-domain at 55 Hz s-domain
90
frequency (Hz)
angle (rad)
0.5
0 0
10
20
30
40
frequency (Hz)
50
60
70
80
90
Fig. 6 Effect of Sampling Frequency on the Frequency Response of the Digital PSS In order to plot the frequency response of the designed digital PSS and compare it to the designed s-domain PSS, a MATLAB code was applied. Figure 7 shows the frequency response of the designed digital PSS compared to the sdomain PSS.
Fig. 8 MATLAB (Laptop) & PIC18F4520 -controller Interfacing Circuit
15 z-domain s-domain
magnitude
10
20
40
60
80 100 frequency (Hz)
120
140
160
180
1 z-domain s-domain 0.5
The time domain simulations of the SMIB power including s-domain PSS at nominal operating point. Moreover, the robustness of the designed digital PSS has been evaluated by comparing the SMIB power system timedomain response with digital PSS and with s-domain PSS, at different operating points, however the rotor speed deviation under the most significant points are shown in Fig. 10.
angle (rad)
20
40
60
80 100 frequency (Hz)
120
140
160
180
Fig. 7 Comparison between the Frequency Response of the z-domain and the Response of the s-domain PSS
C. MATLAB and Microcontroller Interfacing The interfacing between the MATLAB and microcontroller is done using serial communication. It is performed using serial COM port1 in the computer and TX (RC6) and RX (RC7) pins in the microcontroller. Because of
Fig. 9(a) Photos from the -controller Based Digital PSS Hardware
V.
Conclusion
Fig Photos from the -controller Based Digital PSS Hardware

Rotor Speed Deviation 0.08 Digital PSS s-domain PSS 0.06 Deviation of Rotor Speed (Rad/s)
The major contribution of this paper lies in designing and implementing a -controller based digital power system stabilizer (PSS). The s-domain PSS have been transformed to z-domain using bilinear (Tustin) transformation. The developed MATLAB/Simulink model of the studied power system has been interfaced with -controller using serial UART interfacing. The designed -controller PSS has been applied to the power system to show the effectiveness of the designed PSS on the system stability enhancement under different operating conditions. The results have shown that the designed PSS enhanced the system stability under all studied operating conditions. The work can be extended to design PSS based on -controller using remote synchronized stabilizing signals via phasors measurement units (PMUs). References
[1] [2] P. M. Anderson and A. A. Fouad, Power System Control and Stability, IEEE Press, 1994. C. L. Chen and Y. Hsu, "Coordinated Synthesis of Multi-machine Power System Stabilizer Using an Efficient Decentralized Modal Control Algorithm," IEEE Trans. on Power Systems, Vol. 2, No. 3, August 1987. P. Kundur, Power System Stability and Control, McGraw-Hills, 1994. F. P. Demello and C. Concordia, "Concepts of Synchronous Machine Stability as Effected by Excitation Control," IEEE Trans. on power apparatus and systems, Vols. PAS-88, no. 4, April 1969. C.M. Ong, Dynamic Simulation of Electric Machinery Using MATLAB/SIMULINK, Prentice-Hall Press, 1998. First prize award received by Qatar University at IEEE PES 2012 Student Poster Contest at T&D Conference, Orlando, FL, May 2012, http://ewh.ieee.org/soc/pes/sasc/awards.html G. Gurrala and I Sen ,Power System Stabilizers Design for Interconnected Power Systems, IEEE Transactions on Power Systems, Vol.25, No. 2, May 2010. IEEE Std. 421.5, IEEE Recommended Practice for Excitation System Models for Power System Stability Studies, IEEE Power Engineering Society, 2005. L. Cai and I. Erlich , "Simultaneous coordinated tuning of PSS and FACTS damping controllers in large power systems", IEEE Trans. on Power Systems, Vol. 20, No. 1, pp. 294-300, 2005. Wang S., Chiou J., Liu C., "Parameters tuning of power system stabilizers using improved ant direction hybrid differential evolution", International Journal of Electrical Power & Energy Systems, Vol. 31, No. 1, pp. 34-42. 2009. Babaei, F. Golestaneh, M. Shafiei, and S. Galvani, Design an optimized power system stabilizer using NSGA-II based on fuzzy logic principle, Canadian Conference on the 24th Electrical and Computer Engineering (CCECE), pp. 683-686, May 2011. N. Nise, Control Systems Engineering, John Wiley & Sons, 2011. M. Verle, PIC Microcontrollers, mikroElektronika, 2008. M. Verle, PIC Microcontrollers-Programmingin C, mikroElektronika, 2009. Ibrahim, Microcontroller Based Applied Digital Control, John Wiley & Sons, 2006. K. J. Astrom and B. Wittenmark, Computer Controller Systems, Longman Higher Education, 1984. C. L. Phillips and H. T. Nagle, Digital Control System Analysis and Design, Prentice-Hall Press, 1995. S. Lee, Y. Li and V. Kapila, "Development of a Matlab-Based Graphical User Interface Environment for PIC Microcontroller Projects," Proceedings of the 2004 ASEE, Education Annual Conference & Exposition, 2004.
0.04
0.02
[3] [4] [5]

0 1 2 3 4 Time (sec) 5 6 7 8
-0.02
-0.04
-0.06
[6] [7]
Fig. 10(a) System Response with -controller PSS at Different Operating Points Under 1% Change in Tm
Rotor Speed Deviation 0.08 0.06 Deviation of Rotor Speed (Rad/s) 0.04 0.02 Digital PSS s-domain PSS
[8] [9] [10]
0 -0.02
[11]
-0.04 -0.06 -0.08
4 Time (sec)
[12] [13] [14] [15] [16] [17] [18]
Fig. 10(b) System Response with -controller at Different Operating Points Under 1% Change in Tm
From Fig. 10, it is concluded that the designed controller PSS is robust at different operating points. Moreover, there is a high level of agreement between the Digital (-controller) PSS compared to s-domain PSS SMIB Power System rotor speed deviation responses, and this proves the effectiveness of the designed Digital PSS.
Coordinated Control of On-Load Tap Changer and D-STATCOM for Voltage Regulation of Radial Distribution Systems with DG Units
B. BAKHSHIDEH ZAD, J. LOBRY, and F. VALLE
Department of Electrical Power Engineering University of Mons Mons, BELGIUM
AbstractThis paper presents a new technique for voltage regulation of a radial medium voltage (MV) distribution grid in presence of distributed generation (DG) units. The proposed technique consists of the coordinated actions of on-load tap changer (OLTC) of transformer and reactive power compensation by distribution static synchronous compensator (D-STATCOM). Managing the system voltage using the action of OLTC is one of the most common ways for voltage regulation of MV systems. However, OLTC cannot be used for voltage regulation of long radial distribution feeders as it changes the sending point voltage of the feeder. In this study, the problem of using OLTC for the voltage regulation of a radial distribution feeder will be solved by using reactive power compensation at the DG connected bus. Simulation results reveal that the proposed control method is capable of maintaining the system voltages within the permitted range. Thanks to the coordinated control method, the drawbacks of each voltage control method are covered by the other one that results in minimization of the system losses as well. Keywords-distributed generation units; coordinated control of voltage; reactive power compensation; distribution STATCOM.
units), voltage decreases towards the end of the feeder, as the impedance of lines causes a voltage drop. Thus, the biggest voltage drop happens at the end of the feeder based on the amount of load demand. With the presence of DG, if its power exceeds the local demand of loads, the power flow direction will be inversed and a voltage rise will appear at the DG connected bus. Therefore, with DG units, the voltage violations depend on the amount of loads as well as on the amount of DGs power. If the integration of DG units has to be increased, designing a new voltage control method will be necessary in order to guarantee safe system operation. Traditionally, distribution system operators have managed their system at the planning stage based on the fit and forget policy using deterministic load flow studies (considering the critical cases) in order to meet the load demands and to verify lines capacity and voltage regulation issues. With DG units, as their output power varies during the day, the uncertainty in distribution system management is increased and the safe operation of the system becomes more complicated. In this situation, implementing an on-line control system based on the active network management policy becomes more crucial. Theoretically, different methods can be applied for voltage regulation of distribution systems as presented in [1] but the most applicable methods are based on using on-load tap changer (OLTC) mechanism of the transformer [2], reactive power compensation [3], curtailment of DG active power [4] and network reinforcement [5]. Since the voltage rise problem is caused by the injected DG power, curtailment of its power is one possible method but it does not allow to maximize the benefits of integrating DG units. The voltage profile along the feeder is strictly dependent on the impedance of lines. Therefore, network reinforcement is another possible method but it is expensive, it needs long delays and DSOs normally consider it as the last possible option. Generally, OLTC action and reactive power compensation are the best possible methods but each of these methods has its own advantages and drawbacks which are explained in the following sections. Therefore, the most optimal solution is to design a coordinated voltage control method based on the OLTC action and reactive power compensation. In reference [6], the coordinated concept of OLTC and static synchronous compensator (STATCOM) based on the artificial neural network has been investigated in
I.
INTRODUCTION
During the recent years, the conventional structure of electrical power systems has been changed by the presence of distributed generation (DG) units. Previously, the electric power was generated in large generating stations at a small number of locations (called central generation). In these stations, voltage was stepped up to high voltage (HV) to be transmitted through HV transmission networks. The voltage was then stepped down to medium voltage (MV) and low voltage and distributed through radial distribution networks to the end users. In recent years, there has been a considerable growth in the amount of decentralized generation connected to the distribution systems. As a result, currently, power systems are in a state of transition from conventional systems with unidirectional power flows to active networks with bidirectional power flows. Therefore, new technical challenges have emerged for distribution system operators (DSOs). Presence of distributed generation units alters the conventional voltage regulation schemes of distribution systems. In the conventional distribution feeder (without DG
order to maintain voltage within the limits while minimizing tap changing operations and increasing reactive power capability margin of STATCOM. Also, coordination of OLTC and STATCOM for improvement of steady state and transient voltage responses of a wind park connected to HV grid has been presented in [7]. In this paper, a new technique is proposed for the voltage regulation of a medium voltage distribution grid in presence of DG units. The proposed technique is based on the coordinated actions of OLTC and reactive power compensation by D-STATCOM. The main idea is to use each voltage control method in the predefined working conditions which correspond to its merits in order to cover the drawbacks of the other voltage control method. II. ON-LOAD TAP CHANGER ACTION
Using OLTC action is the most popular method in the steady state voltage regulation of distribution systems because it is easy to implement and design. In this method, the turn ratio of the transformer winding is changed by a tap changer mechanism when voltage of the system exceeds the specified range. The tap changer action is normally adjusted by an automatic voltage control (AVC) relay which continuously monitors the system voltage and controls the action of the tap changer. The AVC relay works based on two controlling parameters which are the reference voltage of the regulated point and a defined dead band. This dead band is designed to limit the unnecessary actions of the tap changer. The drawback of OLTC is that it cannot be used in voltage regulation of long radial distribution systems because it changes the sending point voltage of the feeder while the biggest voltage violation occurs at the end of the line (ending point of the feeder). In this situation, in order to return the ending point voltage inside the permitted range, OLTC must change noticeably the sending point voltage and it can lead to voltage violation at the sending point of the feeder. Also, the tap changing operation is done with a time delay due to the slow dynamic response of the OLTC mechanism. As the maintenance costs of OLTC depend on the number of tap changing operations, there is a tendency to set a long time delay for its action that results in dealing with some unexpected voltage violations during the time that OLTC is starting to act. III. REACTIVE POWER COMPENSATION
A. Limitations of DG Units in Providing Reactive Power The needed reactive power of the system can be provided by synchronous machine-based DG units that are able to adjust their output reactive power in order to affect the system voltage. Conventional control systems for reactive power control of synchronous machines are automatic power factor control (APFC) system and automatic voltage regulation (AVR) system [8]. In the automatic power factor control mode, the reactive power of DG (QDG) follows any variation of the active power of DG (PDG); therefore, the PDG/QDG ratio is maintained constant in order to keep the system voltage within the limits. This method is not applicable in voltage regulation of distribution systems with a high ratio of R/X. Also, this method is not an effective approach as the load variations of the system are not taken into consideration. In the automatic voltage control mode, the difference between the actual bus voltage and a set reference voltage defines the needed reactive power of the system. This action can be explained by a droop characteristic; this droop shows the relationship of the needed reactive power of DG in accordance with the voltage of the system. In reference [9], a dead band for this characteristic (QDG=f (V)) has been defined in order to limit the exchanged reactive power of DG in the unnecessary range. In reference [10], a new voltage control method has been proposed which combines the advantages of AVR and APFC control systems. It must be noted that the operation of DG units in AVR mode can cause some problems like high field currents, overheating and triggering of over current protection systems. There are also some limitations in reactive power compensation by asynchronous machine-based DG units. It is not applicable in self-excited induction generators and squirrel cage induction generators. In case of doubly fed induction generators (DFIGs), reactive power compensation is possible through control of rotor current but still the physical, thermal and converter power limitations must be considered. Because of these abovementioned problems, dedicating an external source of reactive power can be a better solution for voltage regulation of the system. Power electronics based compensators like D-STATCOM can be used to tackle the current limitations of DGs in reactive power compensation. B. Reactive Power Compensation by D-STATCOM D-STATCOM is a member of FACTS (Flexible AC Transmission Systems) devices at the distribution level. It is a voltage source converter (VSC) based device which converts a DC input into a set of three-phase sinusoidal voltages with a fast controllable amplitude and phase angle. D-STATCOM can provide superior solutions for voltage regulation, flicker elimination and improvement of power quality. In voltage regulation mode, thanks to its fast response, the voltage violations can instantly be removed and the system voltage can quickly bring back to the targeted values. In this paper, DSTATCOM is used as a source for reactive power compensation. It controls the voltage of the regulated point by adapting the amount of injected or absorbed reactive power. As long as the exchanged reactive power stays within the maximal and minimal limits, the voltage is regulated at the target voltage value. Fig. 1 shows a typical V-I characteristic of DSTATCOM.
Reactive power compensation is a useful method for voltage regulation of distribution systems. Traditionally in distribution systems, capacitor banks have been used to keep the power factor close to 1 and to compensate voltage drop in the heavy load situations. In the DG connected distribution systems, as we must deal with both voltage drop and voltage rise problems, we need a source of reactive power with the ability to work in inductive and capacitive modes. Drawbacks of this method are that the exchanged reactive power can increase system losses, decrease network capacity and influence loading capacity of the system. Moreover, reactive power compensation is not an effective way for the voltage regulation of distribution systems with a high ratio of R/X.
manages 0.03 pu of the voltage violation and the rest of the voltage violation is removed by D-STATCOM. Normally, D-STATCOM regulates the system voltage based on a defined reference voltage (VREF) and keeps the voltage at this value as shown in Fig. 1. Since in the permitted range of voltage variations, there is no need for D-STATCOM response, two different reference voltages must be defined for D-STATCOM based on the upper and lower limits of voltage (1.03 and 0.97 pu). However, in this situation, there is no chance for the participation of OLTC action because DSTATCOM always quickly removes the voltage violations and OLTC cannot work. In order to solve this problem, another reference voltage must be considered for D-STATCOM. When the voltage of the regulated point (VT) is between the predefined limits, then VREF of D-STATCOM is considered to be equal to the voltage of the regulated point. In this way, DSTATCOM keeps its exchanged reactive power unchanged and as there is no fixed reference value for D-STATCOM, OLTC with its time delay can start to act in order to lessen the output of D-STATCOM. The working conditions of D-STATCOM based on the proposed reference voltages are summarized below.

Figure 1. Typical V-I characteristic of D-STATCOM
As it can be seen, if the voltage of the system is lower than the reference voltage (VREF), STATCOM works in capacitive mode and if it is higher than VREF, STATCOM works in inductive mode. In Fig.1, D-STATCOM is not used as a perfect voltage regulator and the presented regulation slope (droop) defines deviation of the D-STATCOM terminal voltage from the set reference voltage at the maximum inductive or capacitive current. IV. PROPOSED VOLTAGE CONTROL METHOD
In this paper, a new technique is proposed for the voltage regulation of a medium voltage distribution grid in presence of DG units. The proposed technique is based on the coordinated actions of OLTC and reactive power compensation by DSTATCOM. The main idea is to use each controller in the predefined working conditions which correspond to its merits in order to cover the drawbacks of the other voltage control method. As mentioned, the main drawbacks of OLTC are that it cannot be used in voltage regulation of long radial distribution systems and its response is not fast. These problems can be solved by the fast response of D-STATCOM. On the other hand, the drawback of reactive compensation is that it can increase system losses. In order to solve this problem, based on the proposed method, at first voltage violations are removed by D-STATCOM instantly and then OLTC starts to work in order to lessen the D-STATCOM output. In other words, the functionality of D-STATCOM is limited to the fast control of voltage or when the OLTC action reaches its limits. Generally, safe operation of the network in terms of voltage management is defined by a permitted range. In this paper, the permitted range of voltage is equal to 0.03 pu of the reference voltage (1 pu) [11]. Therefore, if the voltage of the regulated point is within these limits (1.03 and 0.97 pu for the maximum and minimum permitted values, respectively), no corrective action of the controllers is needed. As the OLTC action changes the sending point voltage of the feeder to keep the voltage along the feeder within the permitted limits, the action of OLTC can only manage the voltage violations lower than 0.03 pu (from either upper or lower limit). Clearly, if the action of OLTC is allowed to manage the voltage variations of more than 0.03 pu, the sending point voltage will be violated from the tolerated range. In this paper, the OLTC action is proposed as the main voltage control method in the steady state condition. However, when the voltage violations are more than 0.03 pu (from either upper or lower limit), OLTC cannot manage the system voltages effectively. Thus, the OLTC action
If VT > 1.03 If 0.97 VT 1.03 If VT < 0.97
then then then
VREF = 1.03 VREF= VT VREF = 0.97
OLTC action starts when VT 1.03 or VT 0.97 and it stops when either VT rises to 0.972 or VT falls to 1.028 pu. As it can be understood, when voltage exceeds the permitted range, both D-STATCOM and OLTC are triggered to act at the same time but with different response times. After finishing fast response of D-STATCOM, the working conditions of OLTC are still remained as its starting condition is when VT is equal or greater (or smaller) than the permitted limits. Thus, OLTC can continue to work in order to reduce D-STATCOM output until the time that its stop conditions are triggered. The forced limitation of changing sending point voltage (between the permitted voltage values) is considered as another stop condition of OLTC action. V. INVESTIGATED SYSTEM
In order to validate the proposed method, a simple radial distribution system is considered which is shown in Fig. 2. The system under study consists of a DG unit which is located at the end of the feeder where D-STATCOM is also installed. DG unit generates its power with unity power factor and it assumes that there is no exchange of reactive power between DG and the grid. Parameters of the investigated systems are as follow: Impedance of each section of the feeder = 0.5 + j 0.5 Nominal value of each load = 1 MW with lagging power factor equals to 0.9 (loads are voltage independent in the studied range). Rated power of DG = 5.1 MW Nominal system voltage = 11 kV
Figure 2. The investigated system
VI.
SIMULATION RESULTS
Figure 4. System voltages
Simulations are carried out with NEPLAN and the control systems of OLTC and D-STATCOM are implemented in NEPLAN control circuit environment. In the radial distribution feeder with DG unit at the end of the line, the biggest voltage violation happens at the end of the line. Thus, the DG connected bus is chosen as the regulated point (VT =V5) to be sure that voltage of all other buses are within the limits. DSTATCOM is modelled as a pure reactive current source based on its inherent behaviour. Its control system consists of a PI controller with a zero droop characteristic that receives the voltage (bus 5) as an input signal and based on the different explained reference voltages (VREF), regulates the voltage. The OLTC mechanism is installed on the HV/MV transformer. The tap changer action is adjusted by an automatic voltage control relay which continuously monitors the regulated point voltage (V5) as input signal and controls the OLTC action based on the explained start and stop conditions. The tap changing operation is done with the time delay of 30 s and each tap corresponds to a 0.01 pu step change of the sending point voltage (V1) which affects the voltage of all busses along the feeder. In order to show the dynamic response of the proposed voltage control method, a time period with the variations of loads and DG active power is considered. In the first half of this period, the loads are maximal and DG power is minimal and in the second part of the simulation time, loads are minimal and DG power is maximal as it can be seen in Fig. 3. In the proposed time period, by the variations of loads and DG power both voltage rise and voltage drop are occurred. Fig. 4 shows the system voltages at bus 1 and bus 5.
As it can be seen in Fig. 4, during the first half of the simulation time (before 250 s), without any controller, the voltage at bus 5 reaches 0.95 pu (0.02 pu voltage drop from the lower limit). Thus, simple action of OLTC is capable of returning voltage inside the permitted range in steady state conditions. However, in the second part of the simulation (after 250 s), voltage rise from upper limit is more than 0.03 pu (blue dotted line after 350 s) so the single action of OLTC is not enough to return voltage at bus 5 inside the permitted range (without violation of lower limit at bus 1). Fig. 5 shows the system voltage at bus 1 and bus 5 considering the corrective actions of the proposed coordinated controller. Fig. 6 illustrates the D-STATCOM output. D-STATCOM works in inductive mode when its output has positive sign and its negative values correspond to the capacitive behaviour.
Figure 5. System voltages with the coordinated controller
Figure 3. DG power and loads variations Figure 6. Reactive power output of D-STATCOM
As it can be understood from Fig. 5, thanks to the coordinated control of OLTC and D-STATCOM, the voltage was maintained within the limits in both transient and steady state conditions. In the first half of the simulation time, the DSTATCOM firstly kept voltage at 0.97 and with the OLTC action, the output of D-STATCOM was reduced (at 65 s) and finally reached zero at 95 s (see Fig. 6). In the second part of the simulation, D-STATCOM firstly kept voltage at 1.03 pu and OLTC reduced D-STATCOM output (at 293, 323, 353, 383 and 413 s). However, as the voltage rise at bus 5 was about 0.04 pu, 0.03 pu of this voltage rise was managed by OLTC and the rest of that was removed by D-STATCOM. Therefore, the D-STATCOM output did not reach zero by OLTC action. Fig. 7 shows the amount of active power losses with different voltage control systems.
delay of OLTC action or in the extreme voltage violation states (when OLTC reaches its limits). In this way as D-STATCOM is finally unloaded by OLTC action, it does not considerably increase network losses and it would be ready to react with a higher reactive power capability margin. In future research, the voltage control loop of DG units will be added to the proposed method to study the impacts of DG participation in reactive power compensation of the system. Also, the practical aspects of the proposed method including the costs of implementation must be investigated. REFERENCES
[1] T. Xu, and P. C. Taylor, Voltage control techniques for electrical distribution networks including distributed generation, Proceedings of the 17th World Congress, the International Federation of Automatic Control, pp. 11967-11971, July 2008. [2] F. A. Viawan, A. Sannino, and J. Daalder, Voltage control with on-load tap changers in medium voltage feeders in presence of distributed generation, Journal of Electric Power Systems Research, vol. 77, pp. 1314-1322, 2007. [3] F. Bignucolo, R. Caldon, and V. Prandoni, Radial MV networks voltage regulation with distribution management system coordinated controller, Journal of Electric Power Systems Research, vol. 78, pp. 634-645, 2008. [4] J. Mutale, Benefits of active management of distribution networks with distributed generation, Proceedings of Power Systems Conference and Exposition (PSCE), 2006. [5] D. Pudjianto, D. M. Cao, S. Grenard, and G. Strbac, Method for monetarisation of cost and benefits of DG options, Research Project Supported by the European Commission, 2006. [6] G. Won Kim, and Kwang Y. Lee, Coordination control of ULTC transformer and STATCOM based on an artificial neural network, IEEE Transactions on Power Systems, vol. 20, no. 2, pp. 580-586, May 2005. [7] M. S. El Moursi, B. Bak-Jensen, and M. H. Abdel-Rahman, Coordinated voltage control scheme for SEIG-based wind park utilizing substation STATCOM and ULTC transformer, IEEE Transactions on Sustainable Energy, vol. 2, no. 3, pp. 246-255, July 2011. [8] P. N. Vovos, A. E. Kiprakis, A. R. Wallace and G. P. Harrison, Centralized and distributed voltage control: impact on distributed generation penetration, IEEE Transactions on Power Systems, vol. 22, no. 1, pp. 476-483, February 2007. [9] E. Chabod, L. Karsenti, J. Witkowski, and G. Malarange, Local voltage regulation influence on DG and distribution network, CIRED 2012. [10] A. E. Kiprakis, and A. R. Wallace, Hybrid control of distributed generators connected to weak rural networks to mitigate voltage variations, CIRED 2003. [11] C. L. Masters, Voltage rise the big issue when connecting embedded generation to long 11 kV overhead lines, Power Engineering Journal, vol. 16, no. 1, pp. 5-12, February 2002.
Figure 7. Active power losess of the system
Evaluation of system losses in different cases confirms that the proposed control method slightly increases system losses comparing to the case of single OLTC action but single OLTC action is not sufficient for voltage violations of more than 0.03 pu as depicted in Fig. 4. Also, system losses noticeably increase when D-STATCOM is used lonely. To wrap it up, it can be concluded that the proposed voltage control method solves the conventional problem of OLTC in voltage regulation of long radial distribution systems. As the transient control of voltage is supported by D-STATCOM, a longer time delay for OLTC can be set that leads to reduction of its maintenance costs. On the other hand, since DSTATCOM is only used for transient control of system or when OLTC reaches its limits, it does not considerably influence the system losses. VII. CONCLUSION In this paper, a new idea for the voltage regulation of long radial distribution systems with DG unit at the end of the line was presented. The proposed idea was based on the combination of two different control methods which are OLTC action and reactive power compensation by D-STATCOM. The idea was to use each control method in the predefined range corresponding to its merits in the way that the drawbacks of each method are covered by the other voltage control method. Simulation results revealed that the proposed method enables us to efficiently manage the voltage control problem of a radial MV distribution system in both transient and steady state conditions. OLTC action was chosen as the main voltage control method in steady state conditions and D-STATCOM was only used in the fast control of voltage to cover the time
An Improved Lossless Passive Snubber Cell for PFC Boost Converter

Haci Bodur
Department of Electrical Engineering Yildiz Technical University Istanbul, Turkey bodur@yildiz.edu.tr
Huseyin Yesilyurt
Eczacibasi - Lincoln Electric Kaynak Teknigi - Askaynak Kocaeli, Turkey huseyin.yesilyurt@askaynak.com.tr
Hatice Ozel
Eczacibasi - Lincoln Electric Kaynak Teknigi - Askaynak Kocaeli, Turkey hatice.ozel@askaynak.com.tr
Abstract An improved lossless passive snubber cell for power factor correction (PFC) boost converter is proposed. All of the semiconductor devices in the boost converter are operated under soft switching condition. The switch operates at zero current switching (ZCS) turn on and zero voltage switching (ZVS) turn off. The boost diode is turned off under both ZCS and ZVS conditions and is turned on under ZVS condition. The converter has larger overall efficiency and wider soft switching range than that of the most active and passive snubber cells proposed before. The proposed snubber cell in this study has simple structure. The detailed steady-state analysis and the simulation of the proposed circuit are carried out, and this theoretical analysis is verified by implementing a 50 kHz 1 kW converter. The overall efficiency of the proposed passive snubber cell is measured about %97.5 at full load. Keywords Power factor corrector, passive lossless snubber, boost converter
power dissipations and EMI [1]. High current level on auxiliary switch and snubber inductor occurs in most of active snubber cells. This fact also complicates design of snubber inductance. Furthermore, most of active snubber cells have low load range and high circulating energy. Passive snubber cells proposed previously have too much drawbacks. Additional current stress on the switch and/or voltage stress on the boost diode occur in most of snubber cells as proposed in [3]-[11]. The switch is hard switched during turn off process in some of passive snubber cells as proposed in [3], [6], [9], [12]. Some of them cause additional voltage stresses on auxiliary diodes as proposed in [4], [12]. Some of proposed passive snubber cells cause high current ripple because of magnetic coupling with input inductance as proposed in [7-8] and [10-11]. Some of passive snubbers use more than one snubber inductance and additionally saturable inductor. This situation causes high cost and additional voltage stress on the main switch as proposed in [13]. Some of the passive snubber cells cant maintain soft switching for low input current values as proposed in [2], [5], [8]. In this study, an improved lossless passive snubber cell for PFC boost converter is proposed. The circuit scheme of the converter with the proposed snubber cell is shown in Fig. 1.
I.
INTRODUCTION
The increase in the use of grid connected nonlinear loads has caused lots of harmonic currents which cause distortion, dissipation and inefficient use of electrical grid. Power factor correction (PFC) boost converter operating in continuous conduction mode (CCM) is generally used in industry to ensure desired power factor and harmonic current limits. However, the conventional boost converter operating in CCM causes serious switching losses and electromagnetic interference (EMI) during turn on process of the switch because of the reverse recovery current of the boost diode. Furthermore, fast increase of the switch voltage during turn off process also causes EMI and switching losses [1]. High power density and fast response can only be achieved by increasing switching frequency. However, switching losses and EMI also increase by increasing switching frequency. Soft switching techniques are applied to suppress switching losses and EMI caused by increasing switching frequency [2]. In literature there are many studies about active and passive snubbers to provide zero voltage switching (ZVS) or zero current switching (ZCS) for power devices have performed. Active snubbers have complicated structure and high cost because they require additional switch and control. In some of the active snubber cells, the auxiliary switch is operated under hard switching condition. This situation causes substantial
Fig. 1. Circuit scheme of the converter with the proposed passive snubber cell
The proposed lossless passive snubber cell provides ZCS and ZVS together at turn off and ZVS at turn on for the boost diode. The switch operates at ZCS turn on and ZVS turn off. Furthermore, all the semiconductor devices operate in soft switching condition. It provides wider soft switching range and larger overall efficiency than that of the most active and
passive snubber cells proposed previously. A detailed steadystate analysis of the proposed lossless passive snubber cell is carried out. This theoretical analysis is verified by implementing a 50 kHz1 kW converter. II. OPERATING PRINCIPLES
cos sin cos sin / 1/ Time intervals are given as follows: 2

.
(4)
A. Deffinitions and Assumptions The circuit scheme of the proposed passive snubber is shown in Fig. 1. It has simple structure and consists of the , the snubber capacitor , the buffer snubber inductance and four auxiliary diodes , , , . capacitor During one switching cycle, the following assumptions are made in order to simply analysis of the circuit. 1. 2. 3. 4. Input voltage is constant. Output capacitor is large enough to assume that the output voltage is constant. Main inductor inductor is much greater than snubber
(5) (6) (7) (8) (9)
All semiconductor devices are ideal except the boost diode .
(10)
B. Operating Stages Nine different stages occur in steady state over one switching cycle. An additional interval also occurs for low values of the instantaneous input voltage. Equivalent circuit schemes and key waveform for operation of the proposed snubber cell are given in Fig. 2. and Fig. 3. respectively. : . 2 ]: At the beginning of this Stage 1[ stage the boost diode is conducting input current and 0, 0, , and 0 are valid. At a turn on signal is applied to gate of the the moment switch S. S turned on under near ZCS. For this stage the following equations are obtained: (1) (2) In this interval the switch current is increasing and is decrasing simultaneously and linearly. limits current switch current rise rate and boost diode reverse recovery reaches input current and falls to current. At continues to fall. As a zero. Then continues to rise and drops to and reaches to result at is turned off under ZVS and ZCS and this stage thus is the reverse recovery current of . finishes. Here : . 2 ]:At the moment , 0, , and 0 are valid. After turning off of , a parallel resonant with constant input current starts. via For this resonant following equations are derived: , Stage 2[ cos sin (3)
(11)
, current reaches its maximum At the moment value . At the moment , the snubber capacitor falls to zero, falls to and this stage voltage finishes. : . 2 ]: At the moment t t , Stage 3[ , 0, , 0 and are falls to zero and valid. The snubber capacitor voltage turns on under ZVS condition and this stage starts. For this stage the following equations are derived: cos sin cos sin (13) (14) (12)
, current falls to , buffer At the moment reaches its maximum value and capacitor voltage turns off under ZCS condition and this stage finishes. The : following equation is derived for the maximum value of (15) Stage 4[ : . 2 ]: At the moment , and turn off and this stage starts. This stage is the same as conventional on state duration of PWM boost converter. For this stage the following equations are valid:
Fig. 2. Equivalent circuit schemes of the operation modes in the converter with the proposed passive snubber cell
(16) (17) 0 (18)
Stage 5[ : . 3 ]: At the moment t t , , 0, , 0 and are valid. at t t the gate signal of S is removed and S turned is turned under near ZVS condition and this stage starts. charges with on with ZVS. The snubber capacitor voltage constant input current. For this stage following equations are obtained: (19)
is turned on and this stage starts. A parallel resonant with constant input current shown as Fig. 2(f) occurs. During this is charging and buffer capacitor stage snubber capacitor has initial current is transferring its energy to load. Since and its capacity is smaller than , charging amount of is . For this stage the larger than discharging amount of following equations are derived: 1 cos 1 1 sin (23) / (24) , sin (22) (21)
reaches to ,the At the moment diode is turned on with ZVS and ZCS and this stage finishes. Time interval of this stage is given as follows: (20) Stage 6[ : .2 the snubber capacitor voltage ]: At the moment reaches to , ,
Consequently, at the moment reaches to is turned on with ZVS and this stage finishes.
Fig. 3. Key waveforms for operation of the proposed soft switching converter
Stage 7[ : . 2 ]: at the moment snubber capacitor voltage reaches to , is turned on 0, 0, with ZVS and this stage starts. at , and are valid. A parallel resonant with constant input current via occurs. For this resonant the following equations are derived: 1 cos sin cos cos (25)
Stage 8[ 0, 0,
: 0,
.2
]: at the moment and are valid.
During this stage is discharging with constant input current. For this stage the following equations are obtained: (28) (29) Energy transferring process of buffer capacitor exactly completed at . is
sin
(26)
and are turned off with ZCS At the moment and this stage finishes. Energy transferring process of snubber . inductance is exactly completed at For this stage the time invertal is given as belows: = arcsin (27)
Stage 9[ : . 2 ]: During this stage, is conducting the input current and the the boost diode snubber circuit is not active. For this stage the following equation is valid: (30) the gate signal Consequently at the moment of S is given and the other switching period starts. Stage 7 doesnt occurs for low values of the instantaneous input voltage. For this case, the additional stage is also given in Fig. 2(j). occurs after Stage 8. For this stage following equations are derived:
(31) (32) (33)
4. 5. 6.
The proposed snubber cell has simple structure, low cost and easy control. The proposed converter operates as a conventional PWM converter during most of the switching period. The proposed converter increase input current quality. V. EXPERIMENTAL RESULTS
(34)
III. DESIGN PROCEDURE
A. Snubber Inductance is selected to provide ZCS turn off of the boost diode and ZCS turn on of the switch. This case can be defined as (35) 3 (36)
A prototype of a 1-kW 50-kHz PFC boost converter with proposed passive snubber cell has been built to verify the principle operation and the theoretical analysis. IKP20N60T and STTH1210D are used as switch and boost diode respectively. MUR460-E3/54 diodes are used as auxiliary diodes. The converter has 220-V AC input and 400-V DC output. The efficiency measured as 97.5% at full load. Fig. 4 shows the IGBT is turned off under ZVS condition for 90% period of input voltage. Fig 5 shows the boost diode is turned off under ZVS and ZCS condition and turned on under ZVS condition. Fig. 6 shows the IGBT is turned off under ZVS and turned on under ZCS condition. Fig. 7 shows voltage and current. Fig. 8 shows the snubber inductance the input voltage and current. After Stage 3 and Stage 7 high frequency resonances occur between parasitic capacitors and snubber inductance. Therefore a very light passive RC snubber applied on snubber inductance. This very light RC snubber increases efficiency by assisting ZVS turn off of the IGBT.
Here is rise time of switch and is reverse recovery time, is amplitude of the input current. (35) and (36) provide near ZCS turn off for the boost diode and near ZCS turn on for the switch respectively. is calculated as 7.74 and chosen as 10 H. Since, big snubber inductance value prevents resetting the energy stored in , it is not recommended to choose too big in this snubber cell. B. Snubber capacitor Snubber capacitor is selected to provide ZVS turn off for the switch. This case can be defined as : (37) Where, 4.7 . is fall time of the switch. From (36) is chosen as
C. Buffer capacitor Buffer capacitor is principally selected to guarantee resetting the energy stored in the snubber inductance . In addition, big value of decreases additional voltage stress on the boost diode . It is recommended to choose big in order to reset the energy stored in at the same time snubber inductance for this snubber cell. Simulation and ratio as 3 is experimental results show that choosing and decrase additional voltage stress on advisable to reset is chosen as 15 nF. the boost diode . Therefore, IV. CONVERTER FEATURES The features of the proposed soft switched converter are summarized as follows: 1. All of the semiconductor devices in the converter are operated under soft switching conditions. 2. 3. The proposed converter provides soft switching for the switch at both turning on and turning off transient. There is no additional voltage stress on the switch or auxiliary diodes.
Fig. 5. Voltage (upper) and current (lower) waveforms for the boost diode Fig. 4. Snubber capacitor voltage waveform with scale 50V/div 2ms/div
theoretical analysis is verified by implementing a 50 kHz 1 kW converter. Experimental results show that the switch operates at ZVS turn off and ZCS turn on. The boost diode is turned off under both ZCS and ZVS condition and is turned on under ZVS condition. Furthermore all of the semiconductor devices in the boost converter are operated under soft switching conditions. There is no additional voltage stresses on the switch and on the auxiliary diodes. The proposed converter has larger overall efficiency and wider soft switching range than that of the most of the active and passive snubber cells proposed before. Soft switching is maintained approximately full period of the input voltage. The overall efficiency of the passive snubber cell is measured about %97.5 at full load. In addition, the proposed snubber cell increases input current quality by decreasing zero transition distortions and input inductance parasitic capacitor current.
Fig. 6. Voltage (uper) and current (lower) waveforms for the IGBT [1] [2]
REFERENCES
Ching-Jung Tseng and Chern-Lin Chen, Member, IEEE, Novel ZVTPWM Converters with Active Snubbers, IEEE Transitions on Power Electronics, vol. 13, no. 5, September 1998 H. Bodur, A.F. Bakan, M. Baysal (2003), A Detailed Analytical Analysis of a Passive Resonant Snubber Cell Perfectly Constructed for a Pulse Width Modulated D.C.D.C. Buck Converter, Springer Verlag Electrical Engineering Brian T. Irving and Milan M. Jovanovi, Analysis, Design, and Performance Evaluation of Flying-Capacitor Passive Lossless Snubber Applied to PFC Boost Converter, Applied Power Electronics Conference and Exposition, 2002. APEC 2002. Seventeenth Annual IEEE vol. 1, 2002 Ran Zhao, Jian Pan and Jing Hui, A Novel Soft-Switching Boost PFC with a Passive Snubber, Industrial Electronics and Applications, 2007. ICIEA 2007. 2nd IEEE Conference, 2007 Ching-Jung Tseng and Chern-Lin Chen (1998), A Passive Lossless Snubber Cell for Nonisolated PWM DC/DC Converters, IEEE Trans Industrial Electronics 45: 593 601 Qinglin Zhao, Jianyong Zhang, Chuang Zhao (2011), Passive Lossless Snubber for CCM PFC Based on Magnetic Coupling, Electrical Machines and Systems (ICEMS), International Conference on, Beijing, pp 1-6 Ho-Sung Kim, Ju-Won Baek, Jee-Hoon Jung, Jong-Hyun Kim, MyungHyo Ryu, and Hee-Je Kim, A boost PFC rectifier with a passive lossless snubber circuit using coupled inductors methods Applied Power Electronics Conference and Exposition (APEC), 2012 TwentySeventh Annual IEEE pp 1148- 1152 (2012) Wei Dong, Qun Zhao, Jinjun Liu and Fred C. Lee (2002), A Boost Converter with Lossless Snubber under Minimum Voltage Stress,. Applied Power Electronics Conference and Exposition, Seventeenth Annual IEEE, Dallas, TX, vol 1, pp 509 515 Qun Zhao, Fengfeng Tao, Peng Xu, Jia Wei, and Fred. C. Lee, Improving Performance of Continuous Current Mode Boost Converters for Power Factor Correction, Power Electronics Specialists Conference, 2001. PESC. 2001 IEEE 32nd Annual vol. 2, pp 642 647 Jose Antonio Lambert, Joao Batista Vieira, Jr., Luiz Carlos de Freitas, Lucio dos Reis Barbosa, and Valdeir Jose Farias (1998), A Boost PWM Soft-Single-Switched Converter with Low Voltage and Current Stresses, IEEE Trans Power Electronics 13: 26 35 J. A. Lambert J, B. Vieira Jr. L. C. de Freitas M. S. Vilela V. J. Farias, Applied Power Electronics Conference and Exposition, 1996. APEC '96. Conference Proceedings 1996., Eleventh Annual vol 1, pp 469 474 (1996) Jae-Jung Yun, Hyung-Jin Choe, Young-Ho Hwang, Yong-Kyu Park, and Bongkoo Kang (2012), Improvement of Power-Conversion Efficiency of a DCDC Boost Converter Using a Passive Snubber Circuit,. IEEE Trans Industrial Electronics 59: 1808 1814 River T. H. Li, Henry Shu-Hung Chung, and Anson K. T. Sung (2010), Passive Lossless Snubber for Boost PFC With Minimum Voltage and Current Stress. IEEE Trans Power Electronics 25: 602 613
[3]
[4] [5] [6]
Fig. 7. Voltage (upper) and current (lower) waveforms for the snubber inductance
[7]
[8]
[9]
[10]
[11]
Fig. 8. The input voltage and current waveforms.
[12]
VI.
CONCLUSION
[13]
An improved lossless passive snubber cell for PFC boost converter is proposed. The detailed steady-state analysis and the simulation of the proposed circuit are carried out, and this
Breakdown Voltage Analysis of Insulating Oils Under Different Conditions

Celal Kocatepe*, Oktay Arkan, Eyp Taslak, Celal Fadl Kumru
Department of Electrical Engineering Yildiz Technical University Istanbul, Turkey kocatepe@yildiz.edu.tr, oarikan@yildiz.edu.tr, etaslak@yildiz.edu.tr, cfkumru@yildiz.edu.tr
*
Abstract The dielectric performance of insulating oils used in high voltage equipment is very important for the power system continuity and reliability. In this study, breakdown voltage measurements of mineral insulating oil were carried at different temperatures and water quantity rates were researched. Additionally, the effect of water content inside the oil sample, approximately 2.5 percent, on the breakdown voltage was analyzed for particular temperatures. In the experimental study, the breakdown voltage behavior of mineral oil which is widely used in power transformers was acquired according to the ASTM D877 standard for different service conditions and the obtained results were presented. Keywords-breakdown voltage; insulating oils; water content mineral oil; temperature;
I. INTRODUCTION The need of power transformers are increasing as a result of increasing electrical energy demand in the world day by day. It is known that some of the electrical faults occurring in power transformers arise according to the weak insulating performance of used oils. While choosing insulating oil, characteristic features such as thermal cooling, endurance, price, efficiency and being environment-friendly should be considered. The insulating oils used in power transformer serve as a coolant by absorbing the temperature occurring in transformer winding and core besides provides insulation. Therefore, various insulating oils are used in transformers for thermal cooling and isolation. Insulation oils have generally two types which are mineral oils procured by refining crude oil and synthetic liquids procured chemically. Mineral oils have been preferred for a long time because of being low-cost and having a good insulation performance and thermal cooling capability. Therefore, some of the factors, which could deteriorate the insulation quality of the mineral oil, were analyzed in this study. According to the research studies in literature, breakdown voltages of mineral and natural ester oils were compared after exposed to thermal aging at 140C. The results showed that the decrease in breakdown voltage of natural ester was less than the other [1]. In the study[2], the effects of low temperature rating on the breakdown voltage of insulating oil were
researched according to the IEC 156 standard by the breakdown measurements of insulating oils between -20C and 10C temperature. The heat transfer capability of the oil is also important for a suitable insulation in transformers. An effective thermal cooling in transformer is related to the viscosity of the oil. While the temperature of the oil decreases, its viscosity value increases and so insulating performance is affected adversely. For this purpose, the viscosity of different kinds of insulation liquids used in transformer at different temperatures was measured and its electrical impacts were presented in [3]. Also in [4], the temperature-based breakdown voltages of ester and silicon oils used in electrical power equipment were compared. At each measured temperature, the breakdown voltage of ester oil was more than the silicon oil. The effects of water content quantity on insulating oil breakdown characteristics were examined in [5, 6]. Electrode geometries have different effects on the breakdown voltage of the oil. In [7], breakdown voltages were analyzed by using sphere and plane electrode by applying impulse voltage on insulating oils at different temperatures. It was seen that the breakdown voltage of natural ester oil was more than the other tested oils. In this study, the breakdown voltages of mineral oil, which has been used as insulating oil in transformers for a long time, were researched according to the standards of ASTM D877. In experimental study, the breakdown voltages of the mineral oil, depending on temperature changes and water content were measured and analyzed. II. TEST SAMPLE Severely hydrogenated light naphthenic oil has low pour point besides low viscosity and a good cooling. The inhibited oils are commonly used in power transformers. Because of its convenient properties inhibited naphthenic oil is investigated in this paper. Some of the electrical and physical features of the insulation oil used in the experimental study are given in Table 1.
TABLE I.
PROPERTIES OF USED I NSULATING O IL Severely Hydrotreated Light Naphthenic Petroleum Oil 0.89 10 >145 < -45 <0.005 >30
Properties Density (15C) g/ml Viscosity (40C) mm2/s Flash Point (C) Pour Point (C) Dissipation Factor (90C) Breakdown voltage (kV) (2.5mm gap)
voltage variation of the mineral oil was measured according to particular temperatures after the experimental setup had been established. The breakdown voltage of the oil was measured by gradually heating the insulating oil to 30C, 50C, 70C, 90C and 110C (1C) temperatures with a resistance. The warmth of the insulating oil was provided at desired temperature with the resistance placed into the test cell. Before the test, 30 minutes waiting period was implemented for each temperature. After that, for researching the effect of water content, which could be found at varying rates in the insulating oil, on the breakdown voltage, water having 72.88S/cm conductivity with the amounts of 5ml, 10ml, 15ml, 20ml, and 25 ml was added respectively into the insulating oil. For analyses, the measurements were realized in the oil test cell filled with 1000ml mineral oil. Additionally, the effects of temperature changes were researched by adding 25ml water having 72.8S/cm conductivity into the test cell. In the case of having 2.5% water content in the insulating oil, the temperature-dependent breakdown voltage was determined by performing measurements 30C, 50C, 70C, 90C and 110C (1C) temperatures such as the first experimental application. IV. EXPERIMENTAL RESULTS In the experimental study, the tests researching the changes of breakdown voltages according to the temperature changes on the insulating oil and the water content inside it, was carried out. At the first experimental study, breakdown voltages changes of the new oil sample dependent on temperature were researched. Breakdown voltage of the mineral oil was measured by heating the oil at 20C raises from 30C to 110C. In Table 2, measured Vaverage,Vmin and Vmax values of the mineral oils breakdown voltages at the particular temperatures with a fudge factor of 1C are given.
TABLE II. BREAKDOWN VOLTAGE OF MINERAL OIL AT DIFFERENT
TEMPERATURE
III.
EXPERIMENTAL METHOD
The breakdown voltage measurements on the insulation liquid were performed according to the standard of ASTM D877. After the oil was slowly shaken for a few times, driblet oil was used for cleaning of the test cell. Then, the cleaning oil was drained and test cell was filled with test sample without producing air bubbles. For purging the insulating oil from air, test cell was closed and it was waited for 2.5 minutes before applying test voltage. Five breakdown voltage values are measured, by one minute breaking intervals between them, depending on ASTM D877 standard. The electrode used in the measurement was a polished brass disc electrode with 25.4 mm diameter. The gap between the electrodes was arranged to 2.54 mm and the voltage was applied at the rate of 3 kV/s in [8]. Hipotronics OC90D testing device seen in Figure 1 is used for measurement. The device gives an opportunity of testing with three different voltage raising rate. The specified electrode geometry in ASTM D877 was given in Fig. 2. These electrodes were used in experimental application.
Temperature (C) Figure 1. Oil Tester (Hipotronics OC90D) 30 50 70 90 110
Dielectric breakdown voltage (kV)

Vmin. Vaverage Vmax.
53 49.8 58.5 61.4 78.4
59.2 60.8 65.0 80.4 81.6
71.2 75.5 69 89.5 85.4
Figure 2. Disc electrode
Firstly, the effect of temperature changes on the breakdown voltage of insulating oil was researched. The breakdown
90
Breakdown voltage (kV)
80 70 60 50
breakdown between electrodes begins. As seen in Figure 4, the increasing water content inside the insulating oil significantly decreases the breakdown voltage. While the 25ml water content added into the insulating oil, the breakdown voltage value of the oil is affected negatively and it decreases about %50. For determining the electrical characteristic of the mineral oil dependent on temperature and water content, the breakdown voltages at changing temperatures were measured by adding 25ml water into the insulating oil. In Table 4, for a possible situation that could be seen in a transformer mineral oil in a service (contamination of oil), Vaverage, Vmin and Vmax breakdown voltages dependent on the temperature of the insulating oil are given.
TABLE IV. RELATIONSHIP BETWEEN BREAKDOWN VOLTAGE AND TEMPERATURE (25 ML WATER ADDED) Dielectric breakdown voltage (kV) Vmin 25.0 42.8 64.0 61.5 15.5 Vaverage 30.1 46.0 59.0 69.8 37.0 Vmax 35.6 51.1 55.3 75.1 50.5
30
50
70 90 Temperature (C)
110
Figure 3. Breakdown voltages dependent on temperature of the mineral oil
As seen in Figure 3, the breakdown voltage of the mineral oil is about 60kV at 30C and it slowly increases up to 70C. But it is seen that the increase in the breakdown voltage between 70C and 90C is much more. When the breakdown voltage at 110C and 30C compared, it could be easily seen that the increase is about %30. It is thought that the reason of the increase derives from the fact that the water and moisture content inside the oil decreases while temperature increasing [9]. Also, while the oil temperature increases, the gas pressure in the mineral oil increases too and so it retards the beginning of the breakdown [10]. For researching the effect of water contents at different rates on breakdown voltage of the insulating oil, different amounts of water was injected into the oil. Vaverage,Vmin and Vmax breakdown voltages of mineral oil depending on water contents are indicated in Table 3.
TABLE III. RELATIONSHIP BETWEEN BREKDOWN VOLTAGE AND WATER CONTENT Dielectric breakdown voltage (kV)
Vmin Vaverage Vmax
Temperature (C) 30 50 70 90 110
80
60
40
Water (ml) New oil 5 10 15 20 25
20
30
50
70 Temperature (C)
90
110
44,6 41.0 35,2 32.0 30,2 25.0
51,5 46,9 40,2 38,1 33,9 27,8
56,7 51,5 47.0 42.0 39,2 32,5
Figure 5. Breakdown voltages dependent on temperature of the mineral oil (25ml water added)
60 50 40 30 20
New oil
5ml
10ml 15ml Water content (ml)
20ml
25ml
In Fig. 4, the attenuation of the breakdown with the effect of water is indicated. Whereas in Fig. 5, breakdown voltage changes dependent on temperature of the insulating oil containing blobs are seen. At the measured temperatures except 110C, increases in the breakdown voltages of the insulating oil are observed. When the temperature of the insulating oil is about 100C, the water inside the oil is began to vaporize. Thus the air bubbles, which are resulted from vaporization process shown Fig. 6, decrease the breakdown voltage value of insulating oil. As known, these air bubbles facilitate the beginning of breakdown in the oil by moving to between the electrodes.
Figure 4. Breakdown voltages dependent on water content of the mineral oil
The water being as dissolved and free inside the insulating oil decreases the voltage of the insulating oil. If the applied magnitude of electrical stress (E) is equal or greater than the breakdown strength (Eb) of the insulating material (EEb), the
As the total water content increased, electrical resistance of the insulation system attenuated. Therefore, each increase of the water content in the oil is one of the factors of decreasing the breakdown voltage. REFERENCES
[1] M-S. Shim, Comparative evaluation of aging of insulating, material in natural ester and mineral oil, 2010 International Conference on High Voltage Engineering and Application, pp. 393396, Oct. 2010. [2] M. Gao, G. Li, J. Li, and Z. Zhao, The Temperature Dependence of Insulation Characteristics of Transformer Oil at Low Temperature, in Power Engineering and Automation Conference (PEAM), 2011 IEEE, 2011, pp. 2730. [3] C. Perrier and A. Beroual, Experimental Investigations on Insulating Liquids for Power Transformers: Mineral, Ester , and Silicone Oils, IEEE Electrical Insulation Magazine, vol. 25, pp. 613, 2009. [4] K. Yasuda, S. Arazoe, T. Igarashi, S. Yanabu, G. Ueta, and S. Okabe, Comparison of the insulation characteristics of environmentallyfriendly oils, IEEE Transactions on Dielectrics and Electrical Insulation, vol. 17, no. 3, pp. 791798, Jun. 2010. [5] M. Kohtoh, S. Kaneko, S. Okabe, and T. Amimoto, Aging effect on electrical characteristics of insulating oil in field transformer, IEEE Transactions on Dielectrics and Electrical Insulation, vol. 16, no. 6, pp. 16981706, Dec. 2009. [6] X. Li, J. Li, and C. Sun, Properties of Transgenic Rapeseed Oil based Dielectric Liquid, Proceedings of the IEEE SoutheastCon 2006, pp. 81 84, 2006. [7] S. Arazoe, D. Saruhashi, Y. Sato, S. Yanabu, G. Ueta, and S. Okabe, Electrical characteristics of natural and synthetic insulating fluids, IEEE Transactions on Dielectrics and Electrical Insulation, vol. 18, no. 2, pp. 506512, Apr. 2011. [8] Standard Test Method for Dielectric Breakdown Voltage of Insulating Liquids Using, ASTM D877-02, 2007. [9] R. Suwarno, M. Ilyas, Effects of temperature on Dielectric properties of Rhicinnus Oils as insulating liquid, in Condition Monitoring and Diagnosis, 2008. CMD 2008. International Conference on, 2008, pp. 286 289. [10] T. Suziki, H. Hiramoto, and M. Umeda Dependence of brekdown voltage of silicone liquid on temperature, IEEE Transactions on Dielectrics and Electrical Insulation, vol. EI-18, no. 4, pp. 462464. 1983
Figure 6. Contaminated mineral oil (25 ml water added)
V. CONCLUSION In this study, the breakdown voltage changes of the mineral oil frequently used in service were researched experimentally. In the experimental studies, the effect of temperature by considering different temperatures and the effect of the foreign materials on breakdown voltage by adding water at different contents to simulate undesired foreign materials that could be found inside the insulating liquid were analyzed. It was seen that the breakdown voltage of the mineral oil was increased slowly with the temperature increase and there was a %30 increase in the breakdown voltage at 110C. It was also revealed that, when different contents of water was added into the insulating oil, this was affected the breakdown voltage negatively by attenuating the breakdown voltage of the insulating oil. When the added water content was increased, the breakdown voltage decreased significantly. When 25ml water was added into the insulating oil, the breakdown voltage dependent on temperature increased so as to the vaporization temperature of water. The increases in the breakdown voltage were lower in proportion to the values dependent on temperature measured for the new oil. The air bubbles derived from the vaporization of the water inside the oil cause to the decrease of the breakdown voltage of the insulating oil.
A Novel Fuzzy Cause-and-Effect-Networks Based Methodology for a Distribution Systems Fault Diagnosis
Muhammad Mustafa Arab academy for science and technology
Cairo, Egypt eng.mmi@gmail.com
Walid El-Khattam Ain Shams University
Yasser Galal Arab academy for science and technology

Cairo, Egypt eng.yassergalal@gmail.com
Cairo, Egypt walid_el_khattam@hotmail.com
Abstract Distribution systems gained importance due to their proximity to huge numbers of consumers. Power distribution system operation is essential for consumers and equipment safety. In case of abnormal events, accurate and fast fault diagnosis (identifying faults location and type) is a vital issue to retrieve a sound distribution systems operation. Thus, in this paper, a novel on-line distribution systems fault diagnosis methodology using fuzzy-cause-and-effect-networks (FCE-Nets) is presented. Both, a faults location and type are identified using simple matrix operations, Fuzzy logic data base, and if-thenrules. Two case studies are carried out to evaluate the proposed methodology. The obtained results are compared with two different methods: Expert System (ES) with Artificial Neural Networks (ANNs) and cause-and-effect-network (CE-NETs) to evaluate the proposed method. The comparison outcomes are discussed and conclusions are reported. IndexTermsFault diagnosis, Fault location, Fault type, Fuzzy-cause-and-effect-networks, Fuzzy logic.
I. INTRODUCTION The main objective of power systems is to provide a reliable and continuous supply for their consumers. In the event of fault occurrence, the operator/dispatcher aims to minimize the damages to equipment in faulted areas, isolate faulted equipment and components, and restore the system as soon as possible. However, the operators/dispatchers may find it difficult and time consuming to provide the right decisions. This mission becomes more complex if multiple faults have occurred or some equipment have malfunctioned. Therefore, a successful decision will be achieved after recognizing and identifying the faults location, characteristics, and type, which are the basic tasks of the fault diagnosis in power systems. The accuracy of fault diagnosis depends on the completeness of the events messages, i.e. how much information can be obtained from them about the fault case. Researchers investigating fault diagnosis are concerned with the data to be used for fault diagnosis and the implemented identification methodologies to improve their accuracy and speed. With respect to data, the most widely used data are the status of Circuit Breakers (CBs) and protective relays. Besides, most power systems are equipped with Supervisory Control And Data Acquisition (SCADA) systems to improve and facilitate monitoring and operating processes. Furthermore, Phase Measurement Units (PMUs) show promising results to identify the faults location and type
utilizing data of power angles, voltage, and current waveforms recorded in detecting devices [1]. Unfortunately, these devices are costly, thus, not available at all substations. On the other hand, the most widely used techniques were based on the Expert System approach (ES) as in [2-3]. These ES rule-based techniques suffer from slow response time due to conventional knowledge representation and inference mechanisms. In [4-5], the Artificial Neural Networks (ANNs) were presented. They do not require the defining of the system schemes and configurations and have quick response and short diagnosis process time. However, ANN needs additional training for every modification in the system besides the initial training which takes a long time to derive the required network weights. These drawbacks are enlarged as the network becomes more complicated. The Fuzzy logic technique was implemented in [6-7]. Results showed that the fuzzy logic technique is able to express incomplete or uncertain information. In [8-9], cause-and-effect-network (CE-NETs) and Boolean matrix were used for fault diagnosis methodology to overcome the common drawbacks in the ES and ANN systems. However, CE-NETs methodology still suffers when dealing with any inaccurate situation to provide accurate results. In this paper, a novel fault diagnosis methodology is investigated based on a fuzzy-cause-and-effect-networks (FCE-Nets) technique to assist the operator/dispatcher when he faces alarm attacks due to faults in distribution systems substations. It is based on mixing the cause-and-effect networks, Boolean matrix operations, and fuzzy logic techniques. This methodology is concerned with fault diagnosis using the status of protective devices as it is commonly available in existing SCADA systems. Two case studies were carried out on two systems to evaluate the speed and accuracy of the proposed methodology. The assessment of the proposed methodology to identify both a faults location and type is identified even when dealing with inaccurate situations. The simplicity in construction and ease in modification of the developed methodology, with practical considerations to suite a wide range of distribution systems configurations, are illustrated. In Section II, the proposed methodology of FCE-Nets and fuzzy logic are presented. Two case studies are carried out and the obtained results are presented and discussed in Section III. Finally, discussion and conclusions are provided in Section V.
II. PROPOSED METHODOLOGY Fault diagnosis is mainly concerned with identifying a faults location and type. In the proposed methodology: Fault location identification is carried out using FCE-Nets. Various conditions are evaluated such as occurrence of single/multiple faults and with/without failure equipment. Fault type identification is carried out by implementing the Fuzzy-Data-Base and If-Then rules. Four fault types are applied; Single-Line-to-Ground (SLG), Double-Line (DL), Double-Line-to-Ground (DLG) and Symmetric-ThreePhase (3 . The following subsections explain mathematically in details the proposed methodology: A. Data Base and Rule Base of the Fuzzy Logic Fuzzy sets provide a mathematical way based on the concept of possibility to define an element in the universe X giving it various degrees of belonging between 1 and 0. 1) Membership Function: The membership function describes the degree that the element belongs to the fuzzy set A. A higher value of ( ) means a greater degree of membership. An element of a fuzzy set is an ordered pair containing a set element and the degree of membership in the fuzzy set. : 0,1 (1) A fuzzy set A can be defined as follows: A . A (2)
B. Fuzzy Cause-Effect Networks (FCE-Nets) The Cause-Effect Networks (CE-Nets) and FCE-Nets are graphical tools for database representation which make it easier for operators/dispatchers to understand and be able to maintain the database for any system. However, the proposed FCE-Nets are preferably used as they have the ability to express the degree of certainty of any information as a real number between 0 and 1 which will not be achieved by the classical CE-Nets. Furthermore, FCE-Nets can easily represent causality between faults, actions of protective relays and CBs by three kinds of nodes and arcs as shown in Figure 2.
Fig.2 Basic node-arc relations in FCE-Nets
The following matrix, vectors and operations are defined to develop the inference procedures: 1) Rule Matrix of FCE-Nets: This matrix ( ) describes the relations between the nodes and arcs of the FCE-Nets. Its construction is n by n, where n is number of conditions. , , 0, 2) Reasoning with Fuzzy Rule Matrices: In [10], some vectors were defined to transform string based conditions into numerical form for reasoning and computation as following: a) Truth State Vector (T): It represents the fault symptom with the status of protective devices. b) Fault Node Vector (F): It represents the fault section nodes of the FCE-Net. 1 , if C fault section node Fi 0, otherwise c) Backup Node Vector (B): It represents the backup relay nodes of the FCE Net. 1 , if C backup relay node Bi 0, otherwise d) Transformation vector (TV): It is responsible for inferring from the truth state of given fault symptoms to the fault cause. TV RT T (4) Where " " is the fuzzy multiplication operator.
The trapezoidal fuzzy Membership M was used in the proposed method. It is characterized by four points (a, b, c, d), where (bc) denotes the core in which the membership value is equal to 1 and (ab and cd) indicate the left and right boundaries of the trapezoidal distribution. Figure 1 shows a trapezoidal fuzzy membership function M parameterized by (a, b, c, d).
Fig.1 Trapezoidal fuzzy membership function
2) If-Then Rules: A rule consists of prior parts describing causes and the resultant parts describing effects. The general formulation of a fuzzy implication rule can be denoted as: (3) This infers that the truth of condition implies the truth of with a certainty factor . So the larger the condition value of a certainty factor is, the more reliable the rule is.
e) The updated transformation vector : The updated transformation vector is used to remove the status of backup relays from fault section candidates. T TV RT B T Where " "is the fuzzy min-operator. (5)
f) Estimated Fault Section (EFS): By the following equation the faulted section can be identified: EFS T F (6)
C. Inference of the Proposed Fault Diagnosis s Method 1. Fault section estimation by FCE-Nets The step-by-step procedure to identify the fault f location is as follows: a) From SCADA data, identify the relations r between protection devices to set the nodes and arcs, Figure F 2. They are used to construct the relations diagraph netw work of the system , , the fault and create the rule matrix of the system vector F i , and the backup vector B i . His storical records of relays and CBs behaviors when dealing wi ith fault situations were used to identify the limits of the sets of o the membership functions of the systems rule matrix. b) Get the truth state vector entries by b detecting fault symptoms from SCADA systems. c) Calculate entries of the TV vector. d) Compare the vectors T and TV. If f they equal, then will equal TV. But if they are not equal, eq quation (5) will be used to get the value of . e) Determine the fault section from the EFS E vector. Figure 3 shows the flow chart of fault locatio on estimation.
while, more accurate results can used in practice. In the meanw be obtained if the system was a special equipped with PMUs that are based on phase angle beside the current and voltage values. However, implementin ng the proposed method is more economic than using PMUs. The step-by-step procedure to o identify the fault type is as follows: a) SCADA system provid des measurements of the feeder currents and bus voltages. b) These values are conver rted into fuzzy terms then these terms are specified by the mem mbership functions of the fuzzy sets. All measurements, in root mean square value, are described as three fuzzy sets: Low L (L), Normal (N) and High (H). The related membership functions are defined in fuzzy data base. c) These sets are then used d in the evaluation of the fuzzy rules. In [11], Table I show ws the summary of rules for identifying fault types suitable e for any earthing or grounding system. d) The maximum mem mbership grade (degree of belonging) stands for the dom minant rule and is selected to be the final result. This step sho ows the main advantage of the fuzzy logic by which identifyin ng of the fault type process will cover several cases even if there are common characteristics between any type of fault and d another. Figure 4 shows the flow chart of fault location estim mation. Table I. Summary of rule es for identifying fault types
Rule # R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 H N N H N H H N H H N H N H H N H H N H N N H N H H N H H H H H H N N N H H H N L N N L N L L N L L N L N L L N L L N L N N L N L L N L L L Type SLG A SLG B SLG C DL AB DL BC DL AC DLG AB DLG BC DLG AC 3 fAULT
Fig.3 The flow chart of fault location esti imation
2. Fault Type identification In this paper the membership function ns and the fuzzy rules used in the proposed methodology are based on the values of the voltage and the current during the fault implemented using default equipped system ms that are widely
Fig.4 The flow ch hart of fault type identification
III. CASES UNDER STUDY To verify the proposed methodology, various case studies are carried out. The obtained results were compared with those reported in literatures. The proposed methodology was modeled mathematically using Mat-lab fuzzy tool-box and carried out on two actual distribution systems. A. CASE 1 In [12], a method based on Expert System (ES) with connection with ANN was used for online power substations fault diagnosis. An actual system, Figure 5, was used assuming no transmission errors. The scenario under study is as follows; a single LG fault occurs at phase A of feeder 1. CO relay "C03A" operates and trips breaker "CB3", however, the LCO relay "LC03" fails. The steps of the implementation of the proposed methodology are as follows:
Table II. Set of node conditions

Node C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 C11 C12 C13 C14 Description A fault occurs at feeder F1 Relay CO1 operates CB CB1 tripped CO1 operates but CB1 fails Relay COM1 operates CB CBM1 tripped A fault occurs at feeder F2 Relay CO2 operates CB CB2 tripped CO2 operates but CB2 fails A fault occurs at feeder F3 Relay CO3 operates CB CB3 tripped CO3 operates but CB3 fails R[i, j] R[5,18] R[5,22] R[5,26] R[5,27] R[6,5] R[6,6] R[7,7] R[8,7] R[8,8] R[9,8] Node C15 C16 C17 C18 C19 C20 C21 C22 C23 C24 C25 C26 C27 Description A fault occurs at feeder F4 Relay CO4 operates CB CB4 tripped CO4 operates but CB4 fails A fault occurs at feeder F5 Relay CO5 operates CB CB5 tripped CO5 operates but CB5 fails A fault occurs at feeder C1 Relay COC1 operates CB CBC1 tripped COC1 operates but CBC1 fails A fault occurs at section BUS1
Table III. Nonzero Entries of Rule Matrix

R[i, j] R [1,1] R[2,1] R[2,2] R[3,2] R[3,3] R[4,4] R[5,4] R[5,5] R[5,10] R[5,14] R[i, j] R[9,9] R[10,10] R[11,11] R[12,11] R[13,12] R[13,13] R[14,14] R[15,15] R[16,15] R[16,16] R[i, j] R[17,16] R[17,17] R[18,18] R[19,19] R[20,19] R[20,20] R[21,20] R[21,21] R[22,22] R[23,23] R[i, j] R[24,23] R[24,24] R[25,24] R[25,25] R[26,26]
Table IV. Nonzero entries of fault vector (F)

F[i] F[1] F[7] F[11] F[15] F[19] F[23] F[27] Value 1 1 1 1 1 1 1 Description F1 F2 F3 F4 F5 C1 BUS1
Fig.5 Distribution system of Case 1
a) Fault section estimation process Step 1: The relation diagraph of this system is shown in Figure 6 to create the node conditions, the rule matrix of the system and the fault vector in Tables II-IV. Step2, 3: The non-zero entries of T vector and TV vector are as shown in Table V. Step4: Although the T vector and TV vector arent equal, the is equal to TV as there is no backup protection. Step 5: The fault section determined from EFS vector is as shown in Table V.
Table V. Non zero entries of the operating vectors and the results of Case 1
Conditions hold C2,C3 T 2 ,3 TV 1,2,3 EFS 1 Fault sections F1
b) Fault type estimation process Step1: The membership function ranges were defined. Step2: The data of current and voltage measurement in this case were considered to be received from SCADA. Step3: Although the degrees of belonging were very close in the case of fault types SLG-A (0.5942) and DLG-AB (0.430), the heights degree of the belonging of the rules was selected to be the estimated type as shown in Table VI. Table VI. Fault type estimation results of Case 1
Rule #
R1 506 282 293 H N N 25 H 8.9 L 9.4 N 9.8 N
Fault Type
SLG-A
Degree of belonging
0.5942
Fig.6 FCE-Net of Case1
It was found that the obtained results from the proposed methodology were exactly the same as those reported in [12]. Therefore, the accuracy, the simplicity of the knowledge base, and the ability to modify this base were approved. On the other hand, the proposed method has the advantage of no need
to be trained for every modification as in the ES-ANN-based method. Thus, it is very applicable for online diagnosis. B. Case 2 In Case 2, the accuracy of the fault section estimation process of the proposed method when dealing with uncertainty or missing data is evaluated by comparing the results with the crisp CE-Nets based method proposed in [9]. The author defined a vector called Elimination vector (E) to identify the cause of the fault to get the faulted section as shown in the following equation: (7) Also, the ability of the proposed method when dealing with more than failure devices is presented. Furthermore, the effectiveness of the proposed method in setting an FCE-Net diagraph and rule matrix for more than backup buses is presented. An existing substation, Figure 7, is employed to illustrate the reasoning process of the proposed FCE-Net technique.
The substation is composed of one sub-transmission line, one main transformer and one auxiliary transformer feeding two 6.6kV bus bars. BUS1 feeds three radial feeders each one of them protected by three relays CO and one CB. BUS2 feeds five radial feeders each one of them protected by 3 relays CO and one CB. The protective relays of the 6.6kV buses serve as the backup protection for their connected feeders. The scenario under study is described as follows; a three phase fault hit feeder (F1) and protective relay CO1 operates and gives a signal to CB1 to trip. Although CO1 operates, CB1 failed to trip. So the protective Relay of the BUS1 operates as a backup and CBM1 tripped. Also, a single line to ground fault hit feeder (F5) and protective relay CO5 operates and gives a signal to CB5 to trip. Although CO5 operates, CB5 failed to trip. So the protective Relay of BUS2 operates as a backup and CBM2 tripped. A missing signal was considered to make the studied case more challengeable. The steps of implementing the proposed methodology are as follows: a) Fault section estimation process Following the same steps mentioned earlier, the FCE-Net diagraph of this system is shown in Figure 8. The node conditions, the rule matrix, the vectors F and B are shown in Tables VII-X. The non-zero entries of the vectors T, TV, T , EFS, E and results are as shown in Table XI.
Fig.8 FCE-Net of Case2
Table VII. Set of node conditions of Case 2

Node Description Node Description C1 A fault occurs at feeder F1 C20 Relay COM2 operates C2 Relay CO1 operates C21 CB CBM2 tripped C3 CB CB1 tripped C22 A fault occurs at feeder F4 C4 CO1 operates - CB1 fails C23 Relay CO4 operates C5 Relay COM1 operates C24 CB CB4 tripped C6 CB CBM1 tripped C25 CO4 operates - CB4 fails C7 A fault occurs at feeder F2 C26 A fault occurs at feeder F5 C8 Relay CO2 operates C27 Relay CO5 operates C9 CB CB2 tripped C28 CB CB5 tripped C10 CO2 operates - CB2 fails C29 CO5 operates - CB5 fails C11 A fault occurs at feeder MTR1 C30 A fault occurs at feeder F6 C12 Relay COMTR1 operates C31 Relay CO6 operates C13 CB CBMTR1 tripped C32 CB CB6 tripped C14 COMTR1 operates-CBMTR1 fails C33 CO6 operates - CB6 fails C15 A fault occurs at section BUS1 C34 A fault occurs at feeder MTR2 C16 A fault occurs at feeder F3 C35 Relay COMTR2 operates C17 Relay CO3 operates C36 CB CBMTR21 tripped C18 CB CB3 tripped C37 COMTR1 operates-CBMTR2 fails C19 CO3 operates - CB3 fails C38 A fault occurs at section BUS2
Fig.7 Distribution system of Case 2
Table VIII. Nonzero Entries of Rule Matrix R[i, j] R[i, j] R[i, j] R[i, j] R[i, j] R[i, j] R [1,1] R[7,7] R[12,11] R[18,18] R[24,23] R[29,29] R[2,1] R[8,7] R[12,12] R[19,19] R[24,24] R[30,30] R[2,2] R[8,8] R[13,12] R[20,19] R[25,25] R[31,30] R[3,2] R[9,8] R[13,13] R[20,20] R[26,26] R[31,31] R[3,3] R[9,9] R[14,14] R[20,25] R[27,26] R[32,31] R[4,4] R[5,10] R[15,15] R[21,20] R[27,27] R[32,32] R[5,4] R[5,14] R[16,16] R[21,21] R[28,27] R[33,33] R[5,5] R[5,15] R[17,16] R[22,22] R[28,28] R[34,34] R[6,5] R[10,10] R[17,17] R[23,22] R[20,29] R[35,34] R[6,6] R[11,11] R[18,17] R[23,23] R[20,33] R[35,35] Table IX. Nonzero entries of fault vector (F)
F[i] F[1] F[7] F[11] F[15] F[16] F[22] F[26] F[30] F[34] F[38] Value 1 1 1 1 1 1 1 1 1 1
R[i, j] R[36,35] R[36,36] R[20,37] R[20,38] R[37,37] R[38,38]
Moreover, it has the ability to define two backup buses in the same FCE-Net diagraph and in the system rule matrix reflect the capability of the proposed method in dealing with any distribution system construction. V. CONCLUSIONS This paper proposes a novel fault diagnosis methodology to identify and estimate fault location and fault type. This method is based on utilizing FCE-Nets as graphical tools and matrixbased operations to estimate the fault location. In the meanwhile, Fuzzy Logic database and rule-base are implemented to identify the fault type. Comparing the results obtained from the proposed methodology with ES-ANN based and CE-Nets techniques, shows the high accuracy, ability to deal with uncertain situations, fast response, suitability for online operation, no need for training, and the simplicity for establishing the model of the proposed methodology. Furthermore, the proposed methodology can be integrated to existing SCADA systems like a tool box with no need for extra devices. References
[1]
Biswarup Das,Fuzzy Logic-Based Fault-Type Identification in Unbalanced Radial Power Distribution System, IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 21, NO. 1, JANUARY 2006
Description F1 F2 MTR1 BUS1 F3 F4 F5 F6 MTR2 BUS2
Table X. The nonzero entries of Backup vector (B) with Unity Values
B[i] B[4] B[10] B[14] B[19] B[25] B[29] B[33] B[37]
Table XI. Nonzero entries of the operating vectors and results of Case2
Cond. C2,C4,C5,C6,C20,C21, C27,C28 T 2,4,5,6,20,21, 27,28 1,15,26,38 TV 1,2,4,5,6,10,14,15,19,20,21, 25,26,27,28,29,33,37,38 EFS E F1, BUS1 F5 BUS2 4 F1
1,2,5,6, 10,14,15,19,20,21,25, 26,27,28,29,33,37,38
b) Fault type estimation process Performing the proposed method, the estimated fault types are as shown in Table XII.
Table XII. Fault type estimation results of Case 2
Rule # 1400 1360 R10 H H 500 236 R1 H N 1300 H 240 N 20 N 80 H 5.6 L 5 L 6 L 6.2 N Fault Type 6 3 fAULT L 6.4 SLG-A N Degree of belonging 0.670 0.638
[2] C. Fukui and J. Kawakami, An expert system for fault section estimation using information from protective relays and circuit breakers, IEEE Trans. Power Del., vol. 1, no. 4, pp. 8390, Oct. 1986. [3] Y. L. Zhu, Y. H. Yang, B. W. Hogg, W. Q. Zhang, and S. Gao, An expert system for power systems fault analysis, IEEE Trans. Power Syst., vol. 9, no. 1, pp. 503509, Feb. 1994. [4] G. Cardoso, J. G. Rolim, and H. H. Zurn, Application of neural-network modules to electric power system fault section estimation, IEEE Trans. Power Del., vol. 19, no. 3, pp. 10341041, Jul. 2004. [5] T. Bi, Z. Yan, F. Wen, Y. Ni, C. M. Shen, F. F. Wu, and Q. Yang, Online fault section estimation in power systems with radial basis function neural network, Int. J. Elect. Power. [6] H. J. Cho and J. K. Park, An expert system for fault section diagnosis of power systems using fuzzy relations, IEEE Trans. Power Syst., vol. 12, no. 1, pp. 342348, Feb. 1997. [7] S. W. Min, J. M. Sohn, J. K. Park, and K. H. Kim, Adaptive fault section estimation using matrix representation with fuzzy relations, IEEE TRANSACTIONS ON POWER System, vol. 19, no. 2, pp. 842 848, May 2004. [8] Wen-Hui Chen, Chih-Wen Liu, and Men-Shen Tsai, Fast Fault Section Estimation in Distribution Substations Using Matrix-Based Cause Effect,,NetworksIEEE TRANSACTIONS ON POWER DELIVERY,
VOL. 16, NO. 4, OCTOBER 2001.
[9] Wen-HuiChen,Chih-Wen Liu, Men-Shen Tsai, Fault Diagnosis in Distribution Substations Using CE-Nets via Boolean Rule Matrix Transformations,IEEE,2000. [10] Wen-Hui Chen, Fault Section Estimation Using Fuzzy Matrix-Based Reasoning Methods, IEEE TRANSACTIONS ON POWER DELIVERY,
VOL. 26, NO. 1, JANUARY 2011
It is worth mentioning that, when the information is complete, FCE-Nets and CE-Nets inference can obtain the correct results. However, if a status signal is missing, CE-Nets cant find the fault sections. Therefore, the proposed method has the ability to identify multiple fault sections even when a failure device and incomplete or missing information exist.
[11] Onojo Ondoma James, Ononiwu Gordon Chiagozie, FAULT DETECTION

ON RADIAL POWER DISTRIBUTION SYSTEMS USING FUZZY LOGIC, ASIAN JOURNAL OF NATURAL & APPLIED SCIENCES, Vol.1, No.2, June 2012.
[12 Hong-TzerYang,Wen-YeauChang,Ching-Lien Huang, ON-LINE FAULT

DIAGNOSIS OF POWER SUBSTATION USING CONNECTIONIST EXPERT SYSTEM, IEEE Transactions on Power Systems, Vol. 10, No. 1, February
1995.
Impact of Distributed Generation in Transmission Network Expansion Planning Problem

Chandrakant Rathore Electrical Engineering Department S V National Institute of Technology Surat, India chandrakant.r@rediffmail.com
Abstract Expansion of the electricity sector and liberalized electricity market leads to more jobs challenging the designer of the transmission network. In order to match with the load growth and generation patterns schemes optimal transmission network expansion planning (TNEP) has to be done. It will find out where, when and how many new transmission lines should be added in the network. From the recent past, renewable energy resources are becoming major attraction part to the power sector. Large scale integration of distributed generation (DG) affects the electrical industry and as well as the strategy of the transmission network expansion. In this paper the impact of distributed generation (DG) in transmission network expansion planning (TNEP) is presented. The objective of TNEP is to minimize the transmission line investment cost required to meet the growing load and the added constraints. Static TNEP problem is modeled by using DC power flow model and to solve this problem a population search based algorithm named, Gbest Artificial Bee Colony (GABC) optimization algorithm is proposed. To show the effectiveness of the proposed method, it is evaluated on Garvers 6-bus network, and IEEE 24-bus test system. Results obtained are compared with the previous published literature. Keywords DC power flow, distributed generation, gbest artificail bee colony algorithm, rescheduling, transmission expansion planning .
Ranjit Roy Electrical Engineering Department S V National Institute of Technology Surat, India rr@eed.svnit.ac.in receives an important role in power system planning. Also as the network size increases more computational effort is required to solve the TNEP problem. Garver who has first applied linear programming approach to solve the TEP problem [2]. Later on various methods are applied to solve the TEP problem which includes dynamic programming [3], interactive method [4], hierarchical decomposition [5], simulated annealing for long term TNEP [6], constructive heuristic algorithm (CHA) applied for solving AC TNEP problem in [7]. Application of CHA in a branchand- bound to solve DC model static TEP problem is presented in [8], Branch-and-Bound algorithm [9], Artificial techniques such as, ANN [10], Fuzzy systems [11], discrete particle swarm optimization (DPSO) algorithm [12], and Projection-Adapted Cross Entropy method [13], novel differential evolution algorithm (DEA) [14] are reported in the literatures . Many works related with the distributed generation optimal location and size in both radial and meshed connected network system can be found from [15- 19]. But in TEP problem few papers are reported in literatures [20-21]. Therefore, in this study impact of large scale integration of DG in TNEP problem is presented. In this work, Gbest Artificial Bee Colony (GABC) algorithm [22-23] which is a population-based search procedure is adopted to solve the TEP problem. The proposed algorithm is implemented to solve the DC- Static transmission network expansion planning (STNEP) problem for three different cases. The rest of the paper is organized as follow: Section II describes the mathematical model for transmission expansion planning. Section III presents overview of Artificial Bee Colony Algorithm and its implementation on TNEP problem. Section IV presents the results obtained and discussion and finally conclusion in Section V. II. MATHEMATICAL MODEL FOR TEP Generally, for the TNEP problem DC power flow model is used because of its robustness. Many other models are reported in [24] as an alternative to DC model. A. Static TNEP model The objective function for the TNEP problem is to minimize the transmission line investment cost of the system. In this paper, the static transmission expansion planning
I.
INTRODUCTION
As per present scenario demand of electric power generation is increasing. It is difficult for utility to provide economical and technical solution to their customers. This imputes to use new source of energy such as renewable energy as decentralized power. In recent years, distribution generation (DG) has been one of the most attractive research areas in the field of power generations. In general, DG can be defined as electric power generation within distribution networks or to the customer side of the network. Hence, it is essential to study the impacts of DG in the transmission network expansion planning (TNEP). TNEP problems are generally, classified as static and dynamic. Static transmission network expansion planning (STNEP) method determines new transmission facilities needed to meet the system requirement for a specific planning horizon. Dynamic TNEP is timing based method. Transmission investment cost is more when generating station is far away from load centers [1]. Therefore, TNEP problem
problem can be formulated based on lossless DC power flow model as follows [8-9, 24], Minimise C Subjected to f f n |f | g dg 0 n g n n g dg n dg n g d f dg 0 (2) (3) (4) (5) (6) (7) ,
CL
known as onlooker bees and a bee carrying out random search around the swarm to find food source is called the scout bee. ABC generates arbitrarily distributed initial population P of Ns vectors of candidate solutions as in (8), P X , X , . . , XN
T
(1)
(8)
Each candidate solution X is a D-dimensional vector, containing as many integer-valued parameters in (9) as D is the number of optimized parameters. X x , , , x , , . . , xD, , i = 1, .........., Ns (9)
where C , CL , n and represents, the total transmission investment cost in US $, construction cost of each line in branch i-k in US $ and the number of circuits added in each right-way-of i-k. represents set of all right-of-way paths for candidates network expansion. f is power flow in branch ik, f is the thermal limit of the circuit g and d are the real power generation and the corresponding demand at each node i, and are the voltage angle at i and k bus, n , n , n are the number of existing circuits, maximum number of added circuits in branch, the total integer number of circuits added to the branch i-k. is the susceptance between buses i and k, g , g ,g represents the real power generation, the maximum, lower limit of generation capacity at bus i, and , dg and dg represents the maximum and dg minimum capacity of distributed generation and distributed generation at bus i respectively. Equation (1) represents the total investment cost of new lines added, (2) and (3) represents power balance constraint at each node and voltage balance constraint in each loop both are based on kirchhoffs law, (4) represents the power flow limit for each branch, (5) represents power generation limit at each bus, (6) represents distribution generation limit at each bus and (7) represents link expansion limit for each branch. The objective is to minimize the total investment cost of the new transmission lines to be build satisfying the constraints mentioned in above equations. . III. OVERVIEW OF ABC ALGORITHM Karaboga and Bastruk [25] have introduced the Artificial Bee Colony (ABC) algorithm which is meta-heuristic algorithm based on intelligent behaviour of honey bee swarm. The position of a food source signifies a possible solution of the optimization problem and the fitness solution of the problem a corresponding nectar amount of a food source. It consists of three groups of bees namely employed bees, onlooker bees, and scout bee. Bees going to food source randomly. They carry information and share it with other bees waiting in the hive regarding location and the profitability of that particular food source are called employee bees. The bees waiting in the dance area for making decision to choose a food source based upon information given by the employed bees
In order to update the new candidate food position, employee bee uses (10) to search for new position from the old in memory v x x x (10)
where is a random number between [-1, 1], k {1, 2..., Ns} and j {1, 2..., D} are randomly chosen indexes. Onlooker bee selects a food source according to the probability calculated by using (11) associated with that food source pp . pp
f N f
(11)
where fitness is the fitness value of the solution i, Ns is the number of food source. In ABC, if a position cannot be enhanced during set number of cycles, then it is assumed to be abandoned. This control parameter is named limit. Assume that the abandoned source is x and j {1, 2..., D}, then the new food source found out by the scout replaced the abandoned one by using (12) x x rand 0,1 x x (12)
For each candidate source position v is produced and estimated by the artificial bee, its quality is compared with its old one. If its found better than old one, it is replaced by old one and if not old one is retained in memory. In ABC algorithm the solution search equation described as in (10) has good at exploration but poor at exploitation. In order to achieve good optimization performance the exploration and exploitation abilities should be equally balanced. Therefore to achieve this, Eqs (10) is modified to improve the exploitation as follows [23]
vij = x ij + fij * (x ij - x ki ) + ij (yi - x ij )
(13)
Where the term added in (10) is gbest term is the jth element of the global best solution, and is a uniform random number in [0, C] where C is a non-negative constant. The modified ABC was named as Gbest-guided ABC (GABC) algorithm. A) Implementation of GABC to TEP problem This section provides application of GABC algorithm to solve STNEP problem as follows:
1) Input parameter and initialization step: After receiving required input parameters such as population size, number of food source, number of employed bees, onlooker bees and limit value. Lower and upper bounds of candidate solution are specified. Algorithm generates randomly distributed initial population of Ns size by (8). In this TEP problem formulation, each candidate solution X is n , , , n , , . . , nD, , i {1, 2..., Ns}, D an integer value. X is the number of possible and nji represents the number of possible lines between each branch j and i. 2) Employee bees serach phase: In this algorithm, to optimize food source employed bees select the position by using (13). Penalty factor method is used to handle the constraints. After calculating the position, the value of objective functions will be calculated and it keeps the best solution obtained in memory. 3) Onlooker bees serach phase: Based on the probabilities, find out from (11), onlooker bees search for food source in swarm, and if the newly obtained food source dominates the old one by using (13) that will updates its position. 4) Scout bee serach phase: If a solution found in onlooker search phase cannot be improved further for certain numbers of cycles, then scout bee finds a new source randomly and replace the abandoned source using (12). If the maximum number of cycles is not reached, jump to step 3. The algorithm is terminating when it reaches to its number of cycles. The control parameters for GABC algorithm to obtain optimal solution for two test systems are as follows: For Garvers 6-bus, and IEEE 24-bus system 50 employee bees, 500 onlooker bees, limit is 4 and C is 1.5 and maximum number of cycles is taken as 500. IV. RESULTS AND DISCUSSION Static transmission network expansion planning problem is solved for two test systems by applying proposed algorithm and is implemented in Matlab 7.9. To validate the performance of GABC algorithm the results obtained are compared with other optimization methods. The static TEP problem was analyzed for three different cases. In this, research work the value for g is set to 0 MW and g is set to rated capacity for all test cases. The best results for minimum investment cost, Cinv in US $ is obtained by proposed algorithm after 30 trial runs. GABC control parameters of the best solution were found through several numbers of runs. A. Garvers 6 bus system Garvers system [1] is used as a first test system in this paper, which comprises of 6 buses, 15 new candidate links, and 760 MW of demand all the necessary system data can be found in [24]. The upper bound of new line in each right-of ways is 4.
In this system, bus 6 is a new generation bus that needs to be connected with existing network. 1) Case study 1 In this case, the optimal solution found by GABC is C = US $ 200,000 with the following system configuration: n2-6 = 4, n3-5 = 1& n4-6 = 2. The optimal solution reported for this case was previously found in by branch and bound algorithm [9], hybrid methods of ANN, GA and TS in [10], DEA [14] and CBGA [26] is same as result obtained. 2) Case study 2 In this case, the optimal solution found for static TEP problem with generation resizing consideration by GABC is C =US$ 110,000 with the following system configuration: n3-5 = 1& n4-6 = 3, which is same as result found in [9] by branch and bound algorithm and CBGA [26]. The cost convergence curve characteristic for both the cases is shown in Fig. 1. The simulation results for cases 1 and 2 giving details about optimal, worst, average value, and standard deviation and number of lines added for optimal solution are shown in Table I.
TABLE I. RESULTS OF GARVERS 6 BUS SYSTEM FOR CASE 1 AND CASE 2
Results of static TEP Best, US $ Average, US $ Worst, US $ Standard deviation Loss, MW Line additions for the best result
Case 1 200000 214400 272000 30357.87 55.848 n2-6 = 4, n3-5 = 1& n4-6 = 2
Case 2 110000 145200 242000 35329.56 52.495 n3-5 = 1& n4-6 = 3
3) Case study 3 In this case, the static TNEP problem is solved with distributed generation incorporation. To show the more impact in this case different DG penetration level is selected and it is assumed all load buses are probable location of DG placement. a) With DG penetration 20% of the total load demand:The optimal solution for static TEP problem is found when DG is placed at bus 5 and there results are as follows: =US$70,000 with the following system configuration: n23 = 2 & n4-6 = 1 and line loss is 23.921 MW. The simulation results are shown in Table II. b) With DG penetration 15% of the total load demand:The optimal solution for static TEP problem is found when DG is placed at bus 2 and the results are as follows: =US$80,0000 with the following system configuration: n3-5 = 1 & n4-6 = 2 and line loss is 27.862 MW. The simulation results are shown in Table II.
than the reported in [27] and the investment cost is found out US$ 224,000,000. The cost convergence curve characteristic for both the cases 1 and 2 is shown in Fig. 2. The simulation results for cases 1 and 2 giving details about optimal, worst, standard deviation and etc. optimal solution are shown in Table III.
Fig. 1. Cost convergence curve for Garver 6-bus system TABLE II. Results of static TEP Bus No Best, US $ Average, US $ Worst, US $ Standard deviation Loss, MW Number of lines added 2 70000 78000 80000 4216.3 7 24.575 3 RESULTS OBTAINED WITH DG PENETRATION With different DG penetration level 20% 4 70000 95400 182000 40901.51 29.075 3 5 70000 78000 80000 6324.5 55 23.921 3 2 80000 92200 182000 32172. 45 27.862 3 15% 4 80000 102200 182000 29581.5 3 29.377 3 5 8000 0 9200 0 1200 00 1316 5.61 28.17 1 3
Fig. 2. Cost convergence curve for IEEE 24-bus system TABLE III. RESULTS OF IEEE 24 BUS SYSTEM FOR CASES 1 AND 2 Case 1 390000000 400300000 493000000 32571459.9 240.551 n1-5 = 1, n3-24 = 1, n6-10 = 1, n7-8 = 2, n14-16 = 1, n15-24 = 1, n16-17 = 2, n16-19 = 1 & n17-18 = 2 Case 2 152000000 167400000 194000000 16971872 211.551 n6-10 = 1, n7-8 = 2, n10-12 = 1& n14-16 = 1
Results of static TEP Best, US $ Average, US $ Worst, US $
B. IEEE 24-bus system The second test system is IEEE 24-bus system consists of 24 buses, 34 existing branches and seven new candidate lines, 41 possible right-way paths and 8550 MW of load demand. The maximum number of three new parallel lines can be installed. The system data is available in [27]. 1) Case study 1 In this study, static TENP problem without generation rescheduling is considered with one form of five possible generation plans. Result obtained for this case by proposed algorithm is investment cost C =US$ 390,000,000 with the following topology: n1-5 = 1, n3-24 = 1, n6-10 = 1, n7-8 = 2, n14-16 = 1, n15-24 = 1, n16-17 = 2, n16-19 = 1 & n17-18 = 2 and line loss is 240.551 MW which is better than the result found by CHA [28]. 2) Case study 2 In this case, the optimal solution found for static TEP problem with generation rescheduling consideration by GABC is C =US$152,000,000 with the following system configuration: n6-10 = 1, n7-8 = 2, n10-12 = 1& n14-16 = 1and line loss is 211.551 MW. It is the best solution and found better
Standard deviation Loss, MW Line additions for the best result
3) Case study 3 In this case, the static TENP problem is solved with distributed generation incorporation. a) With DG penetration 20% of the total load demand:The optimal solution found by GABC when DG is placed at bus 10 and the results are as follows: =US$48,000,000 with the following system configuration: n6-10 = 1 & n7-8 = 2 and line loss is 191.045 MW. The simulation results are shown in Table IV. b) With DG penetration 15% of the total load demand:The optimal solution for static TEP problem is found when DG is placed at bus 10 and the results are as follows: =US$48,000,000 with the following system configuration: n6-10 = 2 & n7-8 = 1 and line loss is 138.674 MW. The simulation results are shown in Table V.
TABLE IV. Results of static TEP Bus No Best, US $ Loss, MW Line additions Number of lines added 3 101000000 191.604 n1-5 = 1, n3-9 = 1, n6-10 = 1 & n7-8 = 2 5 4 97000000 248.092 n1-5 = 1, n4-9 = 1, n6-10 = 1 & n7-8 = 2 5 5 101000000 241.639 n5-10 = 1, n6-10 = 1, n7-8 = 2 & n20-23 = 1 5 TABLE V. Results of static TEP Bus No Best, US $ Loss, MW Line additions Number of lines added 3 124000000 170.374 n1-5 = 1, n610 = 1, n7-8 = 2 & n14-16 = 1 5 4 97000000 185.018 n1-5 = 1, n4-9 = 1, n6-10 = 1 & n7-8 = 2 5 5 98000000 236.184 n6-10 = 1, n7-8= 1, & n11-13 = 1 3
RESULTS OBTAINED WITH 20% DG PENETRATION DG penetration level 20% of total load 6 64000000 175.599 n6-10 = 2 & n7-8 = 2 4 8 78000000 239.808 n6-10 = 1 & n13-14 = 1 2 9 70000000 155.614 n1-5 = 1, n6-10 = 1, & n7-8 = 2 4 10 48000000 191.045 n6-10 = 1 & n7-8 = 2 3 14 98000000 139.401 n6-10 = 1, n7-8 = 2, & n10-11 = 2 4 19 152000000 195.889 n6-10 = 1, n7-8 = 2, & n10-12 = 1 5 20 152000000 216.048 n6-10 = 1, n7-8 = 2, n10-11 = 1 & n14-16 = 1 5
TABLE IV RESULTS OBTAINED WITH 15% DG PENETRATION DG penetration level 15% of total load 6 48000000 199.755 n6-10 = 2 & n7-8 = 2 3 8 78000000 181.084 n6-10 = 1 & n13-14 = 1 2 9 70000000 165.392 n1-5 = 1, n6-10 = 1, & n7-8 = 2 4 10 48000000 138.674 n6-10 = 2 & n7-8 = 1 3 14 98000000 146.448 n6-10 = 1, n7-8 = 2, & n10-11 = 2 4 19 124000000 178.272 n1-5 = 1, n610 = 1, n7-8 = 2 & n14-16 = 1 5 20 152000000 223.146 n6-10 = 1, n7-8 = 2, n10-11 = 1 & n14-16 = 1 5
Form the above analysis it is found that when generation values are fixed the total investment cost is more than when generation rescheduling is considered. It is observed that as distributed generation is incorporated in the systems the total investment cost, new lines added and line losses gets reduced. Table VI provide the detail comparison results for static TEP problem by proposed method and other methods for case 1 and case 2.
TABLE VI. COMPARSION OF RESULTS FOR STNEP PROBLEM FOR CASE 1 AND CASE 2 Optimal cost (US $) Methods Garvers 6-bus system Case 1 B&B [9] ANN, GA & TS [10] DEA [14] CGA [14] CBGA [26] CHA [28] GABC 200000 200000 200000 200000 200000 ____ 200000 Case 2 110000 ____ 110000 110000 110000 ____ 110000 IEEE 24 bus system Case 1 ____ ____ ____ ____ ____ 438000000 390000000 Case 2 ____ ____ ____ ____ 152000000 ____ 152000000
Results obtained by the proposed algorithm for Garvers 6bus system with case 1 and case 2 are same as found by other methods. For IEEE 24-bus system, GABC yields least cost compared with CHA for case 1 and with CBGA in case 2. V. CONCLUSION In this paper, Gbest Artificial Bee Colony algorithm has been applied for static TEP problem. Garvers 6 bus and IEEE 24 bus test systems are used to solve the proposed problem and results obtained are compared with other optimization methods in existing literatures. Results indicate that without and with generation rescheduling, the proposed algorithm yields the optimum results for Garver 6-bus, and IEEE 24-bus system compared with published results. Moreover, results obtained with distributed generation placement indicate that as the penetration level of distributed generation increases, the total investment cost and network line losses decreases. REFERENCES
[1] F.F Wu, F.L. Zheng, F.S. Wen, Transmission investment and expansion planning in a restructured electricity market, Energy, vol. 31, no. 67, pp. 954-966, May-June 2006 L.L. Garver, Transmission network estimation using linear programming, IEEE Trans Power Appl. Syst. vol. PAS-89, no. 7, pp. 16881697, Sept./ Oct. 1970.
[2]
[3] Y.P .Dusonchet, A.H. El-Abiad, Transmission planning using discrete dynamic optimisation, IEEE Trans. Power Appl. Syst., vol. PAS-92, no. 4, pp.1358-1371, April 1973 A.O. Ekwue, B.J. Cory, Transmission system expansion planning by interactive methods, IEEE Trans. Power Appl. Syst., vol. PAS-103, no. 7, pp. 1583-1591, July 1984 R. Romero, A. Monticelli, A hierarchical decomposition approach for transmission network expansion planning, IEEE Trans. Power Appl. Syst. vol. 9, no. 1, pp. 373-380, Feb. 1994 R. Romero, R.A. Gallego and A. Monticelli, Transmission expansion planning by simulated annealing, IEEE Trans. Power Appl. Syst. vol. 11, no. 1, pp. 364-369, Feb. 1996 M.J. Rider, A.V. Gracia, R. Romero, Power system transmission network expansion planning using AC model, IET Gener. Transm. Distrib., vol.1, no. 5, pp.731-742, Sept. 2007 R. Romero, E.N. Asada, E. Carreno and C. Rocha, Constructive heuristic algorithm in a branch-and-bound applied to transmission network expansion planning, IET Gener. Transm. Distrib., vol. 1,no. 2, pp. 318-323, March 2007 M.J. Rider, A.V. Garica and R. Romero, Transmission system expansion planning by a branch-and-bound algorithm, IET Gener. Transm. Distrib., vol. 2,no. 1, pp. 90-99, Jan. 2008 Al-Saba Tawfiq, El-Amin Ibrahim, The application of artificial intelligent tools to the transmission expansion problem, Elect. Power Syst. Res., vol 62, no. 2, pp 117-126, June 2002 A.S. Sousa and E.N. Asada, Combined heuristic with fuzzy systems to transmission system expansion planning, Elect. Power Syst. Res, vol. 81, no. 1, pp 123-128, Jan. 2011 Y. X. Jin, H Z. Cheng, J Y. Yan and L. Zhang, New discrete method for particle swarm optimization and its application in transmission network expansion planning, Elect. Power Syst. Res., vol. 77, no. 34, pp. 227-233, March 2007 A. Eshragh, J. Feliar, A. Nazar, A projection-adapted cross entropy (PACE) method for transmission network planning, Int. J. Elect. Power Energy Syst., vol. 2, no.2 pp. 189-208, June 2011 T.S. lm, G.A. Taylor, M.R. Lrving and Y.H. Song, Differential evolution algorithm for static and multistage transmission expansion planning, IET Gener. Transm. Distrib., vol.3 no. 4, pp. 365-384, 2009 C.Wang, M. H Nehrir, "Analytical approaches for optimal placement of distributed generation sources in power systems". IEEE Trans. Power Syst, vol. 19, no. 4, pp. 2068-2076, Nov 2004 Q.Kang, T.Lan,Y.Yan ,L.Wang ,Q.Wu, "Group search optimizer based optimal location and capacity of distributed generations". Neurocomputing, vol. 78, no. 1, pp. 55-63, 2012 S. H.Lee, J. W. Park, "Selection of optimal location and size of multiple distributed generations by using kalman filter algorithm". IEEE Trans. Power Syst, vol. 24, no. 3, pp. 1393-1400, Aug 2009 M. F.Akorede, H.Hizam, I.Aris, M. Z. A. Ab Kadir, "Effective method for optimal allocation of distributed generation units in meshed electric power systems". IEE Proc. Gener. Transm. Distrib, vol. 5, no. 2, pp. 276-287, 2011 A.Kumar, W.Gao, "Optimal distributed generation location using mixed integer non-linear programming in hybrid electricity markets". IEE Proc. Gener. Transm. Distrib, vol. 4 no. 2, pp. 281-298, 2010 J. H.Zhao, J.Foster, Z. Y.Dong, K. P.Wong, "Flexible transmission network planning considering distributed generation impacts". Power Systems, IEEE Transactions on, vol. 26, no. 3, pp. 1434-1443, Aug. 2011 P.Vasquez, F.Olsina."Valuing flexibility of DG investments in transmission expansion planning". In Proc. IEEE Lausann on Power Tech, pp. 695-700, July 2007 W.Gao, S. Liu, L.Huang, "A global best artificial bee colony algorithm for global optimization". J. Comput. Appl. Math. vol. 236, no. 11, pp. 2741-2753, 2012 G.Zhu, S.Kwong, "Gbest-guided artificial bee colony algorithm for numerical function optimization". Appl. Math. Comput. vol. 217, no. 7, pp. 3166-3173, 2010 [24] R. Romero, A. Monticelli, A. Garcia and S. Haffner, Test systems and mathematical models for transmission network expansion planning, IET Proc-Gener. Transm. Distrib, vol. 149, no. 1, pp. 27-36, Jan. 2002 [25] B. Basturk, D. Karaboga, An artificial bee colony (ABC) algorithm for numeric function optimization, In IEEE swarm intelligence symposium, pp. 12-14, May 2006 [26] I.J.Silva, M.J.Rider, R.Romero, A.V.Garcia,C.A. Murari,Transmission network expansion planning with security constraints, IEE Proc. Gener. Transm. Distrib., vol. 152, no. 6, pp. 828-836, 2005 [27] R. Fang, D. J. Hill, A new strategy for transmission expansion in competitive electricity markets, IEEE Trans. Power Syst. vol. 18, no. 1, pp. 374-380, Feb. 2003 [28] R. Romero, C. Rocha, J.R.S. Mantovani, I.G. Sanchez, Constructive heuristic algorithm for the DC model in network transmission expansion planning, IEE Proc- Gener. Transm. Distrib, vol. 152, no. 2, pp. 277-282, March 2005.
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
Effect of the Tap Winding Configurations on the Electromagnetic Forces Acting on the Concentric Transformer Coils
Guven KOMURGOZ 1
1, 2
Tayfun GUNDOGDU 2
Istanbul Technical University, Faculty of Electrical-Electronics Engineering Department of Electrical Engineering, 34469-Maslak/Istanbul, TURKEY 1 2 komurgoz@itu.edu.tr tgundogdu@itu.edu.tr
Abstract Knowledge of the electromagnetic forces at different winding topology provides a good basis for designers and researchers with which to deal with some of the demands that exist for the next generation transformers. Location of the regulating coil in transformer determines to effect of the electrodynamic forces on the operational behavior of the transformer. This paper presents design principles of the regulating coil in transformers and shows the electrodynamics forces by comparing the analytical results with Finite Element Method (FEM) analysis results. In this paper, electrodynamics and leakage field of magnetic force of the 25 MVA power transformers were analyzed under short-circuit condition of the high voltage windings. Keywords Transformers, electromagnetic forces, tap windings, approximate calculation method, FEM.
studied domain reluctivity
I.
INTRODUCTION
NOMENCLATURE
a A, , A , , , Bz , 1 , 2 dl , Js , 1 , 2 , , S1, S2 U, V v , 0 , isolation thickness vector potential and its components, respectively flux density and its components, respectively radial and axial magnetic field density, respectively winding thickness of primer and seconder, respectively unit length of wire cooling gap (oil-gap) force radial and axial force, respectively axial height of the winding coercive magnetic field intensity tangential component of magnetic intensity instantaneous value of the current maximum current current source density directional short-circuit current density Rogowski correction factor number of turns in the low and high voltage winding, respectively unit vectors in cylindrical coordinate parallel and vertical boundary condition, respectively rated voltage and scalar potential, respectively velocity short-circuit impedance of the system and transformer, respectively phase angle magnetic permeability of the air-gap conductivity and conductance, respectively
Power transformers are very critical, costly, and capitalintensive assets for utilities and industry. On the other hand, power transformer is the most important unit in an electrical distribution network. Their reliability not only affects the electric energy availability of the supplied area, but also affects the economical operation of a utility. Under the deregulation policy of electric systems, each utility is trying to cut its cost and provide to the prevention of accidental loss. The capital loss of an accidental power transformer outage is often counted in million dollars for output loss only. Therefore, design of a power transformer taking into account to fault analysis is of great importance in terms of reliability and also national wealth. Electromagnetic force is a very important reliability parameter for a transformer. The analysis of electromagnetic forces is essential for mechanical considerations. By the interaction of leakage field and the short-circuit current, the huge short-circuit force is generated in the windings. One of the important problems in the design of transformers is radial and axial forces, being proportional to the square of the short-circuit current. Forces arising during short-circuit may be as high as ten thousand to million N. By the effect of so large forces and thermal expansion of wires, the insulation of transformer windings can be distorted, even collapsed, shortcircuit error occurs or damage to the clamping structures. The leakage flux and short-circuit current not only cause the additional losses and forces, but also creates heating to the internal components. Furthermore, the location of the tapings has the predominant effect on the axial forces since it controls the residual ampere turn. Failure of transformers due to shortcircuit is major concern for power utilities and manufactures. These hazards can be avoided by proper design of windings structure against thermal and mechanical strains to prevent permanent deformations and movement of windings if forces can be calculated correctly. The forces, which acting on the transformer windings, have to be evaluated before the manufacture process. Evaluation of the magnetic field distribution of the transformer is essential for the calculation of electromagnetic forces.
During the last decades, several authors have studied on the equations for calculation the electromagnetic forces acting on the windings in transformers [1-11]. Furthermore, classical methods were used to compute the short-circuit forces in windings [12-15] and transient formulations are described in [16-18]. By simplifying the configurations with some assumptions, electromagnetic force computation methods have also been proposed [19, 20]. Furthermore these methods are simple, fast and easy, but not accurate and not suitable for predicting the performance of special types of transformers, especially the axial length of windings is not equal [21-24]. It is, however, obvious that by using modern computerized methods, sophisticated methods, it is possible to calculate forces acting on the elements of winding, the effect of any arrangements of parts and asymmetries. If magnetic field is calculated accurately, it is possible to define electromagnetic forces in the detailed transformer model by using numerical methods such as; Finite Element Methods (FEM), Finite Difference Methods (FDM), Boundary Elements Methods (BEM) and etc. In recent years, a significant development of FEM software has enabled the force calculation to be accomplished easily in where the winding and tapping arrangement is complex [7-11]. This paper concentrates on the use of FEM for analyzing of the models. This method provides a comprehensive view of the overall transformer mechanic and electromagnetic behavior under normal and disturbance conditions. The effect of tap winding configurations, which are on +15% top, -15% middle and -15% upper, which creates asymmetry on the windings are also analyzed. The results obtained from FEM analysis of transformers using MAXWELL are validated by the analytical solution of the radial and axial electromagnetic forces acting on the windings of the transformer. The analysis study carried out to provide guidelines for optimal design in terms of forces acting on the transformer windings. II. FORCES ACTING ON THE TRANSFORMER WINDINGS
are separated into two groups and each group has balanced ampere-turns. The radial ampere-turns produce radial flux which causes axial force in the windings as it seen in the Fig.1. This assumption allows calculation of the axial forces.
C1
C2
D1avg
D2avg
Figure 1. Winding dimensions.
The algebraic sum of the ampere-turns of low voltage (LV) and high voltage (HV) windings at any point and at end of the windings gives the radial ampere-turns at that point in the winding. A curve is plotted for every points called residual or unbalanced ampere-turn diagram which the method derives its name [24]. It is clear that windings without axial displacement and windings have the same length have no residual ampereturns or forces between windings. However, there are some internal compressive forces and forces on the end coils, although there is no axial thrust between windings. Tapings location on the winding has a great effect on the axial forces since it controls the residual ampere-turn diagram. The radial forces develop due to interaction of coil currents with the axial component of its own magnetic flux. In a transformer with concentric windings, radial forces considered insignificant because, the radial strength of the winding is high. Most of the problems occur because of axial forces and axial movement cause more damage to the winding and insulation than radial movements. The inner coil is subjected a pressure tends to collapse to the core. At the same time, the outer coil is under a pressure to extend the diameter of the coil which produces a stress. Preferable choice in a transformer is circular coils, because they are the strongest shape to withstand the radial pressure mechanically [26]. III. CALCULATION OF ELECTROMAGNETIC FORCES
When the electromagnetic force becomes greater than the strength of the windings, the windings will fail. The types of failure, electromagnetic forces, acting on transformer can be classified as radial forces which develop in the x direction and axial forces develop in the y direction. Axial forces creates slipping or breakdown of windings as a whole standingup of part of windings, tilting and deformation of coils. Radial forces create buckling phenomena of inner windings, excessive elongation of outer windings. The windings must be designed to withstand such forces. But this is not validated by test mostly. In this case, accurate calculations are essential. Manufactures have to design the transformers to withstand short-circuit, based on computer programs that they both are familiar with and have confidence in. For the calculation of these forces, both analytical and numerical methods are presented such as residual ampere-turn method, Robins solution, Smythes solution, calculation using Fourier series, two dimensional methods of images, FEM, image method with discrete conductors, and etc. [25]. Residual ampere-turn method, one of the elementary and simplest methods, gives closer approximations and reliable results for the calculation of axial forces. Concentric windings
A. Electromagnetic Forces Transient analysis allows calculating electromagnetic forces for every time step by calculating the leakage flux and full field in winding region. Fully coupled dynamic physical solution is given in (1).
A J s
A V H c v A t
(1)
The differential equation and the boundary conditions of transient axial symmetric electromagnetic field can be expressed in the cylindrical coordinate as in (2), respectively.
rA rA rA v ' v ' J s ' Z Z r r t (2) S1 : rA rA 0 S2 : v ' rA n Ht
The short-circuit current has two components as steady state and exponentially unidirectional. In Fig. 2, applied steady-state and short-circuit currents on the windings of the power transformer in Maxwell software is shown. The exponentially unidirectional component is ignored to make calculations simpler. As seen in the figure, the LV windings have been shortcircuited after 100 ms. C. Approximated Calculation Method In this method, quantities of the formulation, which gives the radial and axial forces, are indicated in the Fig. 1. Therefore, the radial force acting on the low voltage winding is estimated by using (8) neglecting the no load current.
For 2D analysis, the radial and axial components of the magnetic flux density can be expressed as in (3), respectively.
Br A z , B 0, Bz 1 rA r r
(3)
The magnetic flux density can be calculated as in (4) when it is decomposed into its radial and axial components.
Fx 0
IN 2
2h
Davg K
(8)
B r Fz z d Fr r F J r Bz z
(4)
where K is the Rogowski correction factor which is calculated as in (9).

h s K 1 1 e , h
In brief, the force on the power transformer is expressed by the Lorentz force as given in (5).
d F idl B
(5)
s c1 c2
(9)
In similarly, the radial and axial forces of unit length can be calculated as given in (6).
The axial forces acting on the low and high voltage windings are calculated by using (10), respectively.
Fx By I max dl ,
Fy Bx I max dl
(6)
B. Short-Circuit Current Short-circuit currents on the windings have a significant effect on calculation of electromagnetic forces. Generally, the short-circuit current is calculated for different situations by considering [27];
Tapping arrangement Fault position Short-circuit power combination (network and transformer) Short-circuit type (e.g. three phase symmetrical)
Fy
IN
4h
2
Davg 2 K 1 '
(10)
where ' is equal to (11). If the lengths are in [m] and the currents are in [A], then the forces are in [N]. In order to calculate the Fx or Fy on the HV or LV side, related Davg shown in Fig. 1 should be taken into account.
'
IV.
c1 c2 3
(11)
In order to see the effects of the short-circuit current on power transformers, three phase short-circuit scenario is investigated. The amplitude of the first peak of the shortcircuit current is calculated as given in (7) [12, 28].
RESULTS & DISCUSSIONS
I max
1200 900 600
R 2 X 1 e
A B C
U 2 sin 3 Zt Z s
A 25MVA, three phase, 120/11kV transformer was analyzed under short-circuit and various tap winding condition which are: +15% top, -15% middle and -15% upper tapping position of HV windings. A. Model Detailed knowledge of the flux distribution of a transformer plays a very important role in a safe estimation of the forces of the transformer. Complex computer programs are required to obtain a reasonable representation of the field in different parts of the windings. Using the above models for determination of the forces, a numerical application (FEM) has been implemented for a 25MVA power transformer. To reduce computing time and avoid excessive use of RAM, the insulating materials and supporting structure are neglected, besides analyses were done in 2-D structure. The characteristics of the studied transformer are presented in Table I and geometry details of the analyzed transformer are shown in Fig. 3.
(7)
LV Current [A]
300 0 -300 -600 -900 -1200 Time [ms] Fig. 2. Input currents of LV windings. 0 25 50 75 100 125 150
TABLE I. TRANSFORMER DATA Rated Power 25 [MVA] Rated Frequency 50 [Hz] Rated Voltages 120/11 [kV] Rated Currents 120/1310 [A] Turns Ratio 1000 / 159 Connection Yd11 Tap setting 15 % Transformer short-circuit voltage (%) 9
B. Electromagnetic Results Transformers require an accurate mathematical model for system simulation and performance evaluation. In this study, magnetic analysis of the designed machines according to different tab conditions has been investigated using Maxwell 2D program. Different tap conditions are given in the Fig. 3.
shows axial flux distribution with respect to height of the winding for at +15% top, -15% middle and -15% upper tapping position of HV windings under the short-circuit condition. To determine the axial forces, it is necessary to find the radial flux produced by the radial ampere-turns. As seen from Fig. 4, axial flux density of +15% top is approximately constant along the winding due to symmetrical windings (with fully balanced ampere-turns) and it depends on the position of tapping winding due to -15% middle and -15% upper position of the HV winding. All the magnetic flux distributions and densities under the different tapped position of HV windings are obtained. However, just the -15% upper tapping position of HV is given as seen in Fig. 5, because it has the highest flux density. According to (4) and (5), as the leakage flux increases, electromagnetic forces are occurring rapidly. Flux density, which is given in Fig. 5 for -15% upper tapping position condition, plays very important role in the determination of the area and magnitude of the forces according to (6). On the other hand, Fig. 5 gives insight about the total deformation of the transformer. Using (6) and Fig. 5, the total deformation can be estimated.
Fig. 3. 2-D model of analyzed transformers with different tap conditions.
The simulations were completed using the following steps;

1) 2) 3) 4) 5) 6) 6 5 4
Geometric model creation, The appointment of the materials that make up the structure of the machine, Boundary conditions and mesh process, The appointment of currents in windings, Analyze, Examination of the results.
(-) 15% Middle (-)15% Upper (+)15% Top
Fig. 5. Flux density at -15% upper tapping position of HV windings.
3 2 1
After the determination of the electromagnetic results, radial and axial forces acting to HV and LV coils are calculated by MAXWELL. The analysis results are also validated by using approximate calculation method. And a comparison table is given in Table II where X, Y and M indicate; radial, axial and magnitude forces, respectively. The results are nearly equal to each other as seen in the table. Furthermore, it is realized that the insertion of tap section on the upper side of the coil tends to cause more increase according to middle of the coil.
TABLE II. RADIAL & AXIAL FORCE COMPARISONS [MN] +15%
Force / Coil Analytical FEM
H (2) L (1) H (2) L (1) H (2) L (1)
B [T]
-15% upper
Analytical FEM
-15% middle
Analytical FEM
0 0 0.25 0.5 0.75 1 Distance [m] 1.25 1.5 1.75 X Y M
Fig. 4. Axial magnetic Flux densities created by the HV windings.
After completion of the analysis, axial magnetic flux densities created by the HV windings vs. core length are obtained as in Fig. 4 under the different tap conditions. Fig. 4
13,2 -9,92 -0,117 -0,1 13,2 9,92
12,05 8,678 -0,088 -0,103 12,05 8,678
12,52 -9,47 -6,435 1,28 14,077 9,556
11,11 12,32 -7,5 -9,438 -7,94 3,46 0,194 -6,2E-05 13,65 12,8 7,5 9,438
12,21 -7,14 -2,4 -5E-05 12,44 7,14
The electromagnetic forces in the winding of the power transformer are calculated with the leakage flux and transient currents. The radial and axial forces at different tapping position of each conductor coil in the HV and LV windings are given in Fig. 6 - 9. Calculated and sketched axial and radial forces in the windings under short-circuit condition when the windings are axially non-symmetrical (-15% middle and -15% upper) are higher. Due to the symmetry of winding and regular distribution of flux, force values are lower than asymmetrical winding arrangement. If there is an asymmetry in the winding heights due to the tap position (or for some other reasons), the ampere-turn
unbalance increases and gives rise to forces, and result of this, tending to break the winding. Furthermore, the forces acting on the LV coils are not only lower and also in the negative direction according to LV windings as seen in Fig. 12 and 13. Total deformation estimation in windings when the windings are axially or radially symmetrical or non-symmetrical can be obtained by using Table II and Fig. 5. Total deformation depends on the tap position and at +15% tapping position of HV windings, they will bigger than which at -15% tapping positions of HV windings. The location of forces shifts to the upper side of the winding.
Fig. 6. Radial Forces at different tapping positions of HV windings.

5 4 3 2 1 0 -1 -2 -3 -4 -5 -15% Middle
Fig. 8. Radial Forces at different tapping position of LV windings.

0.03 0.025 0.02 0.015 0.01 0.005 0 -0.005 -0.01 -0.015 -0.02 -0.025 -0.03 0.04 0.03 0.02 0.01 0 -0.01 -0.02 -0.03 -0.04 0.035 (+)15% Top 0.025 0.015 0.005 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 -15% Middle
10
11
12
13
14
15
10 11 12 13 14 15
Axial Force on LV Coils [MN]
4 3.5 3 2.5 2 1.5 1 0.5 0 -0.5 1 -1 -1.5 -2 -2.5 -3 -3.5 -4 0.045 0.035 0.025 0.015 0.005 -0.005 -0.015 -0.025 -0.035 -0.045 1
Axial Force on HV Coils [MN]
-15% Upper
-15% Upper
10
11
12
13
14
15
(+)15% Top
10 11 12 13 14 15
-0.005 -0.015 -0.025 -0.035
10 11 12 13 14 15
Coil Number
Coil Number
Fig. 7: Axial Forces at different tapping position of HV windings.
Fig. 9. Axial Forces at different tapping position of LV windings.
[11] Kladas, M.G., Papadopoulos, M.P., Tegopoulos, J.A., Leakage flux and force calculation on power transformer windings under short circuit: 2D and 3D models on the theory of images and the finite element method compared to measurements, IEEE Trans. Magn., vol. 30(5), pp. 3487 3490, 1994 [12] Waters, M., The Short Circuit Strength of Power Transformers. London, U.K. Macdonald & Co., 1966 [13] Yun-Qiu, T., Jing-Qiu, Q., Zi-Hong, X., "Numerical Calculation of Short Circuit Electromagnetic Forces on the Transformer Winding", EEE Transaction on Magnetic vol. 26(2), pp. 1039-1041, 1990 [14] Satish, L., Jain, A., "Structure of transfer function of transformers with special reference to interleaved windings", IEEE Trans. Power Deliv. vol.17 (3), pp. 75460, 2002 [15] Rahimpour, E., et al., "Transfer function method to diagnose axial displacement and radial deformation of transformer windings", IEEE Trans. Power Deliv. vol.18 (2), pp. 493505, 2003 [16] Renyan, T., et al., Numerical calculation of 3D transient eddy current field and short ciruit electromagnetic force in large transformers, IEEE Trans. Magn., vol. 28(2), pp. 14181421, 1992 [17] Arturi, C.M., Electromagnetic force calculation on a 3-phase autotransformer under time-varying fault by 3-D non-linear finite element code, IEEE Trans. Magn., vol. 29(2), pp. 20102013, 1993 [18] Ho, S.L., et al., Numerical simulation of transient force and eddy current loss in a 720-MVA power transformer, IEEE Trans. Magn., vol. 40(2), pp. 687690, 2004 [19] Kumbhar, G.B., Kulkarni, S.V., "Analysis of short-circuit performance of split-winding transformer using coupled field-circuit approach,", IEEE Transactions on Power Delivery, vol. 22(2), pp. 936-943, 2007 [20] Waters, M., "The Measurement and Calculation of Axial Electromagnetic Forces in Concentric Transformer Windings," Proceedings of the IEE Part II: Power Engineering, vol. 101, pp. 35-46, 1954 [21] Najdenkoski, K., Manov, D., Electromagnetic forces calculation on power transformer winding under short circuit, Int. J. Comput. Math. Elect. Electron. Eng., vol. 17,(1/2/3), pp. 374377, 1998 [22] Salon, S., Lamartine, B., Sivasubramaniam, K., Comaparison of short ciruit forces in transformers, IEEE Trans. Magn., vol. 36(5), pp. 3521 3523, 2000. [23] Arturi, C.M., 3-D FE analysis of the axial forces on the step-up transformer winding with helical LV, IEEE Trans. Magn., vol. 31(3), pp. 20322035, 1995 [24] Beavers, M.F. Adams, C.M., "The Calculations and Measurement of Axial Electromagnetic Forces on Concentric Coils in Transformers," Power Apparatus and Systems, Part III. Transactions of the American Institute of Electrical Engineers, vol. 78, pp. 467-477, 1959 [25] Minhas, M.S.A., "Dynamic Behaviour of Transformer Winding under Short-Circuits," Ph.D. Thesis, University of the Witwatersrand, Johannesburg, 2007 [26] Say, M.G., "The Performance and Design of Alternating Current Machines," Sir Issak Pitman & Sons Ltd, London, 3rd Edition, 1958. [27] Mahomed, N., "Electromagnetic Forces in Transformers under ShortCircuit Conditions," Energize Online, pp. 36-40, 2011 [28] IEC Standard 60076-5: Power Transformers-Part 5: "Ability to withstand short circuit, 2006
V.
CONCLUSION
This paper presents comprehensive forces analysis of 25_MVA power transformer with tap windings under shortcircuit condition. By changing the location of the regulating coil three different conditions which are: +15% top, -15% middle and -15% upper tapping position of HV windings are composed. And all tapping conditions are analyzed by using MAXWELL based on the FEM which is a powerful tool on the estimation of the forces with in transformers. The analysis results are also validated by using approximate calculation method. The insertion of tap sections in the windings, which produces asymmetries between LV and HV windings, causes to increase in the radial and axial forces that damage to transformers. Total deformation depends on tap position. These forces shifts the windings towards to axial and radial side of the cooling gap (). Furthermore, it is realized that, to insert a tap section on the upper side of the coil causes more increase in forces acting on the windings according to middle of the coil. The method of calculation offers a reference to the design stage of the transformer. REFERENCES
[1] [2] Karsai, K., Kerenyi, D., Kiss, L., Large Power Transformers, Elsevier Science, 1987 Jiao, L., Bai, B., Li, H., "The calculation of ampere force on electric power transformer under the short circuit situation," International Conference on Electrical Machines and Systems (ICEMS'08), pp. 4423 - 4426, 2008 Behjat, V., et al. "Identification of the most sensitive frequency response measurement technique for diagnosis of interturn faults in power transformers," Measurement Science and Technology, vol. 21(7), pp. 1-14, 2010 Azevedo, A.C., et al., "Transformer mechanical stress caused by external short-circuit: a time domain approach", International Conference on Power Systems Transients (IPST07), pp. 1-6, 2007 Azevedo, A.C., et al., "Investigation of transformer electromagnetic forces caused by external faults using FEM," Transmission & distribution conference and exposition: Latin America, pp. 1-6, 2006 Heathcote, M., Franklin, D.P., "The J&P Transformer Book" Reed Educational and Professional Publishing Ltd, England, 12th edition, 1998 Tang, R., et al., Transient simulation of power transformers using 3D finite element model coupled to electric circuit equations, IEEE Trans. Magn., vol. 36(4), pp. 14171420, 2000 Wang, J.S., A nodal analysis approach for 2D and 3D magnetic -circuit coupled problems, IEEE Trans. Magn., vol. 32,(3), pp. 10741077, 1996. Tsukerman, I.A., Konrad, A., Lavers, J.D., A method for circuit connections in time-dependent eddy current problems, IEEE Trans. Magn., vol. 28,(2), pp. 12991302, 1992.
[3]
[4]
[5]
[6] [7]
[8] [9]
[10] Arturi, C.M., Electromagnetic force calculation on a 3-phase autotransformer under time-varying fault by 3-D non-linear finite element code, IEEE Trans. Magn., vol. 29(2), pp. 20102013, 1993
Transmission Expansion Planning Via a Constructive Heuristic Algorithm in Restructured Electricity Industry
H. Khorasani, M. Pourakbari-Kasmaei and R. Romero Dept. of Electrical and Computer Engineering Universidade Estadual Paulista Ilha Solteira-SP, Brazil khorasani.hamid@gmail.com, {mahdi@aluno , ruben@dee}.feis.unesp.br
Abstract The transmission expansion planning problem in modern power systems is a large-scale, mixed-integer, nonlinear and non-convex problem. this paper presents a new mathematical model and a constructive heuristic algorithm (CHA) for solving transmission expansion planning problem under new environment of electricity restructuring. CHA finds an acceptable solution in an iterative process, where in each step a circuit is chosen using a sensitivity index and added to the system. The proposed model consider multiple generation scenarios therefore the methodology finds high quality solution in which it allows the power system operate adequacy in an environment with multiple generators scenarios. Case studies and simulation results using test systems show possibility of using Constructive heuristic algorithm in an open access system. Keywords-Transmission expansion planning; Constructive heuristic algorithm; Open access; Multiple generating scenarios; hybrid model.
I.
INTRODUCTION
Transmission system expansion problem consists of finding the optimal expansion plan of the electrical system in terms of number and location of transmission lines and/or transformers in order to support secure and economical operation in a specified planning horizon. The available data are: system's base topology, candidate circuits, generation and demand forecast in the planning horizon, investment constraints, etc. In the static planning, there is only one planning horizon and a generalization is the multi-stage planning, where the horizon is split up into various stages. In this paper, only the static planning problem is analyzed, however, the methodology can be extended to a multistage planning as well. Transmission expansion planning has been introduced in 1970 by Garver [1] while several different techniques like Branch and Bound [2], Sensitivity Analysis [3], Benders Decomposition [4], Simulated Annealing [5], Genetic Algorithm [6], Tabu Search[7] and other heuristics algorithm [8] are used to studying such a challenging problem. Commonly used models are in a centralized and vertically integrated power system. It
This work was supported by FAPESP under processes 2010/19032-1 and 2011/13995-5.
can be said that these methods might not be suitable for competitive electricity markets environment. In recent years transmission expansion planning in deregulated power systems is much of interest [9]. Deregulation has changed the structure of power systems incorporating market issues in operation, planning and management [10]. One of the most important characteristic of restructuring is facilitating a competitive environment for power markets but today's transmission networks may not sufficiently support electricity transaction, causing congestion in transmission lines. Therefore, in restructured power markets, consumers are paying increasing congestion costs. From social welfare perspective, if the total costs of congestion that might be relieved by an investment in transmission network is higher than its investment costs, the economic transmission investment is justified. However, since congestion costs as an operational expense may occur at any generating dispatch scheme, and transmission investment costs as a capital expense that is allocated at the beginning of the economic life of the project, it is difficult to compare these two types of costs. In literatures two measures for congestion costs such as: redispatch costs and congestion rent are commonly used. Redispatch costs refers to the systems' costs due to congestion, namely the difference between the total generation costs without transmission constraints and the total generation costs with transmission constrains. In some studies, the term redispatch is also referred to as out-of-merit generation costs, costs of constraints, or congestion costs. On the other hand, the difference between the total payment that a load requires and the total payment that the generators receive is defined as congestion rent. For today's networks, it is not fair to analyze the topology of transmission network without considering operation within a competitive market. In new environment of deregulated power markets, the price that is determined by the least-costs dispatch is called the user prices that may affect the connectivity of the load and generation indirectly affected by the capacity of transmission lines. Hence, the transmission investment problem should try to find minimum investment costs that guarantee least-costs dispatch for the entire system. Therefore system objectives should look for the minimum of both investment costs as well as system re-dispatch costs [11, 12]. Ideally modern transmission networks expansion planning should omit the congestion for all feasible and future
generation scenarios to get an efficient market condition as well as the least-costs dispatch. On the other side, since the future dispatches are unknown; consequently an exhaustive analysis requires considering that the generators can assume producing any value between their lower and upper limits. For that reason, from the viewpoint of the complete elimination of future congestion, the planning process should have a look at all the feasible and future generation scenarios to assure leastcosts in the future dispatch patterns. This can lead to the excessive investment costs and it is necessary to know those costs, while this paper presents a new methodology to determine such costs. In most addressed researches for TEP problem, the open access issue as a requirement of electricity restructuring is not considered, where the optimal expansion plans are determined for only one or a few generation scenarios [13,14]. In this research a new transmission network expansion planning considering multiple generation scenarios is proposed, in which the injected power at each generation bus is not an exact amount and the generation is represented by a set of feasible scenarios. The mathematical models in deregulated environment is more complicated than regulated environment they are usually multi objectives with various constraints they have many integer variables and usually meta-heuristics algorithm are employed to solve this problem . In this paper we try to solve this problem using a constructive heuristic algorithm (CHA). Until now all the CHA algorithms are applied in regulated environment and there is no report about using CHA algorithms in deregulated environment. In this paper the problem is solves in two stage, first stage; define all feasible generator scenarios using extreme bound of generators to model all the possible future scenarios that power market might be encounter in an open access system, second stage employing constructive heuristic algorithm and considering all generator scenario identify the best lines for installation in an iterative fashion, and finally removing the unnecessary lines from the solution. In this work a solution algorithm using constructive heuristic algorithm is proposed, while the Garver and IEEE 24bus test systems are used to validate the proposed methodology. II. GENERATION SCENARIOS To satisfy the conditions that the expanded transmission network does not cause congestion for any feasible generation scenario, the following conjecture is made. If a system can operate adequately for all extreme and feasible plans, then it will be able to operate for any feasible generation plan since the constraints for the feasible plans are less restrictive than those for extreme and feasible plans. This conjecture has an important implication, and it is that the problem can be mathematically formulated since the feasible and extreme scenarios are one measurable and reduced subset of scenarios, while the group of feasible scenarios is infinite. Initially the concept of feasible and extreme generation scenario is defined. An extreme and feasible generation scenario is a plan in which some generators will be functioning at their upper limit ( g ),
Start
k=1
for i = 1 to ng and i k g i = g i if i 1 g i = g i if i 2
g k = Pt ( g i +
i 1
i 2
g)
i
gk gk gk
k=k+1
No
Yes
Save Scenario
No
k ng Yes
End
Fig.1: Creating Generation Scenarios
while others will remain at the lower limit ( g ); the k th single free generator will generate at: (1) g k = dt ( gi + gi )
i1 i 2
Where
d t is the total demand, gi and g i are the upper and
lower limits of the ith generator, respectively. 1 is the set of generators operating in the upper limit and 2 is the set of generators operating in the lower limit. An extreme and feasible generation scenario should satisfy the following constraint: (2) gk gk gk Where: g k is upper limit of the k th free generator and g is k its lower limit. Therefore, in a power system with ng generators, the number of extreme and feasible plans will be ng 2 (ng 1) which is generated using the following procedure: Step 1: Separate ( ng 1) generators in two subsets. In the first subset the generators are in their upper limits. In the second subset the generators are in their lower limits. The remaining generation of k th single free generator will be derived using Eq. (1). This step is repeated for all possible combinations of generators.
Step 2: The extreme and feasible generation scenarios are selected from the previous combinations which satisfy the Eq. (2). Figure 1 shows the flowchart of creating generation's scenarios via proposed method. III. MATHEMATICAL MODEL Two types of mathematical model for a static transmission network expansion planning considering p generation scenarios are presented in this paper; a DC model and a hybrid model, which are briefly outlined in the following section. A. DC Model The DC model for static transmission network expansion planning considering p generation scenarios presents the following Format:
Min v =
constraints. The constraints (7), (8) and (9) should be defined for each extreme and feasible scenario of generation. The rest of the constraints are related to the operational limits of transmission devices. The biggest difference between this formulation and the formulation of basic planning, where only one scenarios is considered, is that, now, the generation is fixed and associate with p extreme scenarios of generation, and the p generation scenarios should be solved simultaneously. The number of operation variables ( fij , i , gi , ri ) increases p times, and the group of operation variables associated to one generation scenario ( fijq , iq , g iq , ri q ) are related with the group of operation variables of the other scenarios through investment variables. The number of investment variables doesn't change in relation to the basic model.
( i , j )0
0 ij
cij nij + i ri q
q =1 i
(3) (4)
s.t. Sf q + g q + r q = d
f ij (n + nij )( ) = 0
q ij q i q j
0 f ijq ( nij + nij ) f ij
(5) (6) (7) (8) (9) (10) (11)
B. Hybrid Model DC Model is a mixed-integer nonlinear programming problem and is very difficult to solve. If we assume that constraint in Eq. 4, KCL, is satisfied only by existing circuits (and not necessarily by the added circuits) hybrid model is obtained. In this context, the hybrid linear model employed in CHA assumes the following form:
Min v =
( i , j ) 0
g k g kq g k
q
q q j
cij nij + i ri q
q =1 i
(12)
giq = g i i 1q
g = g j j
0 ri q d i 0 nij nij
q 2
s.t. Sf q + S 0 f 0 q + g q + r q = d
(13) (14) (15) (16) (17) (18) (19) (20) (21)
0q ij
0q ij
n ( ) = 0 (i, j ) 0
0 ij ij q i q j
0 nij f ij (i, j ) 0
nij integer and (i, j ) 0

Where cij, Yij, nij, nijo represent, respectively, the cost of a circuit that can be added to the i-j right-of-way, the susceptance of that circuit, the number of circuits added to the i-j right-of-way, the number of circuits in the base case, v is the investment, S is the branch-node incidence transposed matrix of the power system , p is the extreme and feasible scenarios of generation f ijq , iq , g iq and ri q represent the operation variables for the generation scenario q which are respectively the total power flow, the phase angle, the generator value and the amount of load shedding at k th bus. q f q , g q and r are the vectors with elements f ijq , g iq and ri q and d is the demand vector with elements d i . nij is the maximum number of circuits that can be added to the i-j rightof-way . fij is the maximum power flow by circuit in the i-j
q is the set of generators in the upper limit for right-of-way . 1 q the q scenario; 2 is the set of generators in the lower limit for the q scenario. and 0 are the set of load buses and all buses and branches respectively . Constraint (4) represents the conservation of power in each node. This constraint models Kirchhoff's Current Law (KCL) in the equivalent DC network. Constraint (5) is an expression of Ohm's Law for the equivalent DC network and so Kirchhoff's Voltage Law (KVL) is implicitly taken into account which are non-linear
f ijq nij f ij (i, j )
g k g kq g k
q
giq = g i i 1q
q gq j = g j j 2 q
0 ri q d i
0 nij nij
nij integer
In which S 0 is the transpose incidence node-branch matrix formed by circuits and buses of the base topology; f 0q is the vector of power flow through the circuits of the base topology with elements fij0 q for the scenario q, S is the transpose incidence matrix of the entire system and f q is the vector of power flows through added circuits with elements f ijq for scenario q . 0 represents the base case circuit indices and the set with indices of all circuits. In the hybrid model, power flows through circuits which belong to the base case are represented separately from flows of the new added circuits. Power flows in the base circuit are represented by fij0 q and in the new circuits by f ijq , values can be different among them. 0 and represents the base case circuit indices and the set
with indices of all circuits. In the proposed liner model, only circuits of the base topology must follow KVL and this requirement is represented by constraint (14). In this paper the hybrid linear model (HLM) is employed for calculation of sensitivity index used to determine the circuit to be added in the electrical system at each step of CHA. IV. APPLICATION OF CHA TO TEP In this section some fundamental components and main characteristic of CHA are presented. In fact, CHA may find a good quality solution in an iterative process. The fastness and robustness are the main characteristics of CHA. In order to obtain a feasible and high quality solution, in each iteration a circuit is added to the network, where the aforementioned circuit is selected based on a sensitivity index. In order to achieve an optimum expansion plan, the load shedding is not acceptable. General CHA process is explained through different steps as follows: Step 1: Assuming a base topology as the current topology. Step 2: Choose a mathematical model for TEP. Step 3: Solve LP/ NLP to determine parameters used in the sensitivity index defined in CHA algorithm that considers operational conditions. If LP or NLP solution indicates that the system is adequately operate in new additions, it means that a feasible solution is in hand. Then go to step 5. Step 4: Use a sensitivity index to identify the most attractive circuit. Update the current topology with the chosen circuit, then go to step 3. Step 5: Sort the added circuits costs in decreasing order. Using an LP, it verifies that whether the removal of a circuit keeps the system in an adequate operational condition or not. If yes, remove the circuit, otherwise keep it. Repeat circuit removal process until all the circuits have been tested. All added circuits that werent removed, represent CHAs final solution. Many CHAs in literature are of two following categorizes. i) Algorithms that use electrical system performance to make sensitivity index ii) Algorithms that use the relaxed version mathematical model. Algorithms similar to least-load-shedding [15] and least effort [16] belong to group (i), and Garvers [1], Villasana GarverSalon (VGS) [17] and the algorithm proposed in this paper also belong to group (ii). In least-load shedding algorithm, sensitivity index tries to identify the circuit that would provide the most significant reduction in load shedding. For this case, in step 3 CHA solves an LP while operation constraint is load shedding. It can be mentioned that sensitivity index is an approximation due to the fact that the selected circuit may not guarantee the least load shedding. Although the selected circuit may provide a reduction in load shedding, it may not facilitate the optimal topology. All these problems may partially arise when the sensitivity index considers the circuits' costs. A major advantage in using either the least-effort algorithm or the least-load-shedding algorithm is that both of them employ DC model directly. The model that is used in Graver's algorithm is transportation model (TM). The TM is a relaxed version of
DC model provided from elimination of third constraint in DC model. In fact, TM is a mixed-integer linear optimization problem. Graver's algorithm relaxes the integrality of the investment variable and solves TM, i.e. making and solving the problem as an LP. The LP solution might not be feasible for TEP problem, therefore this solution is deployed as a sensitivity index for CHA. The sensitivity index can be defined by Eq. (22). (22) SI = max{SI ij = nij f ij ; nij 0} Where: nij is the solution given by LP after relaxing integrality of nij. In Garver's algorithm at each step an LP with the current topology is solved while the number of new circuits might not be an integer that may facilitate a minimum investment. In this regard, Garver algorithms will face with two crucial difficulties. On the other hand, VGS algorithm can find a good solution for DC model than the best CHA ever proposed in the literature. By relaxing the third constraint in DC model, a hybrid model will be produced where an LP solver can be applied to solve such a hybrid model identifying the most important circuit at each step of algorithm. It has worth to be mentioned that in hybrid model, the relaxed constraint will only be considered to those circuits of the current topology. V. CHA FOR EXTENDED HYBRID MODEL Unlike the CHAs in literature that solves only a simple model without considering security constraints, the proposed CHA, works with an extended hybrid model that considers security constraints. The CHA solves hybrid model after relaxing the integrality of investment variables, i.e. the integer is changed to another feature presented in this algorithm is that every circuit added in the process must comply with both KCL and KVL which means compatibility between current solution and the DC model solution. The major drawback of this method is that at each CHA step a very large LP must be solved where it gets considerable for large scale power systems. The proposed CHA employed in this work is as follows: Step 1: Assume the base topology as current topology. Step 2: Solve LP to determine those parameters used in the sensitivity index Eq. (22). If LP solution indicates that the system is adequately operate in new additions, it means that a new solution for DC model has been obtained, then go to step 4 otherwise go to the next step. Step 3: Use sensitivity index of Eq. (22) to identify the most attractive circuit. Update the current topology with the selected circuit, then go to step 2 otherwise go to the next step. Step 4: Sort the added circuits in a descending order of costs. Using an LP, it verifies that whether the removal circuit keeps the system in adequate operational conditions or not. If yes, remove the circuit, otherwise keep it. Continue circuit removal until all circuits have been examined. All added circuits that werent removed represent the CHAs solution. It can be notified that although this CHA uses a hybrid linear model to identify the best circuit to add in an iterative process, it complies with both Kirchhoffs Laws after adding a new circuit, thus the final solution is also a feasible solution of the DC model.
VI. CASE STUDIES AND SIMULATIONS The proposed algorithm was implemented within MATLAB and CPLEX is used as a LP subroutine. It might be noted that since the number of variables and constraints are extremely large for solving this problem the issue of sparse matrix should be employed to avoid possible errors due to the lack of memory.
A. Garver Six-Bus System Garver system has 6 buses, 15 candidate branches, a total demand of 760 MW, and a maximum possible number of added lines per branch are equal to 5. The Garver system data are given in [18]. Maximum generation capacities for this system in buses 1, 3, and 6, are 150MW, 360MW, and 600MW respectively. To obtain extreme and feasible generation scenarios, all combination sets are organized such that 2 of 3 of generators produce at max or min of their generation capacity and the other generator should generate its power which is the difference between total demand and total power generated by other two units. There are 12 generators' scenarios which shown in Table I. In this table the slack generator is shown with G and other generators which have fix value are shown with lower case g. Some of these extreme generators scenarios are feasible and infeasible combinations (those violate generation constraint of slack generator) should be eliminated. The only 4 combinations out of the above 12 combinations are feasible. Consequently the number of restrictions, variables and equality constraints will be 184, 115 and 24 espectively. Therefore the feasible are as following: {(150, 360, 250), (150, 10, 600), (0,360,400), (0,160,600)} Which are highlighted in the Table I.
Table I: Generating Scenarios for Garver System
B. IEEE 24-Bus System This system has 24 buses, 41 branches and a total demand of 8550MW. The data of IEEE 24_Bus system is available in [19]. This system has ten generators, in consequence 102(10-1) = 5120 possible generation scenarios appear. Of these 5120 scenarios, only 168 are feasible generation scenarios, because only 168 scenarios satisfy the constraint Eq.(2). Consequently the number of restrictions, variables and equality constraints will be 24900, 13487 and 3984, respectively. The proposed CHA converges after solving 34 LPs and removes seven circuit in step 4.
Table II Garver solution iteration by iteration Iter.# Index sensitivity Selected line n1-5=0.2918, n2-6=2.3551 1 n3-5=1.198, n3-6=0.44082
n4-6=3.2367
n4-6
n1-5=0.2918, n2-6=2.3551 2 n3-5=1.198, n3-6=0.4082 n4-6=2.2367 n1-5=0.5388, n2-6=2.8918 3 n3-5=1.4147, n3-6=0.1612 n4-6=0.9579 n1-5=0.5399, n2-6=1.8902 4 n3-5=1.4143, n3-6=0.1601 n4-6=0.9628 n1-5=0.5403, n2-6=0.8895 5
n3-5=1.4143, n3-6=0.1597
n2-6
n2-6
n2-6
n3-5
n4-6=0.9649 n1-5=0.3397, n2-6=0.7310 6 n3-5=0.4147, n3-6=0.3603

n4-6=0.9589
G1
760 160 400 200
g2
0 0 360 360
g3
0 600 0 600
g1
0 0 150 150
G2
760 160 610 10
g3
0 600 0 600
g1
0 0 150 150
g2
360 0 360 0
G3
400 760 250 610
n4-6
7 8 9 10 11
n1-5=0.3384, n2-6=0.7332 n3-5=0.4135, n3-6=0.3616 n1-5=0.3516, n2-3=0.2570

n3-5=0.4606, n3-6=0.0914
n2-6 n3-5 n3-6 n2-6 -
After finding extreme and feasible generation scenarios, Garver system's solution will be obtained by solving ten LP. Procedure of line addition to the network is according to Table II. In this table the first column represents the number of iteration and the second column represents the candidate lines sensitivity index, while the third columns shows the selected lines. Note that the sensitivity index for candidate lines that are not shown in the second column is zero. Chosen lines for adding to the network in each iteration, is represented in third column. As shown in Table II. After solving ten iterations, Garver system's solution is found, while for each iteration one LP is solved, means total LPs are ten. Finally, total investment costs for open access model without eliminating any line in step 4, is equal to US$ 270,000,000 with the following topology: n2-3=1 ; n2-6=5; n3-5=2; n2-6=2.
n1-5=0.1003, n2-3=0.4424
n3-6=0.1537
n1-5=0.0056, n2-6=0.1821 -
In IEEE 24-bus system, the optimum solution that will not produce congestion in any 168 feasible generation scenarios is with an investment equals to US$ 1,477,000,000 associated with adding the following lines: n01-02=1; n01-03=1; n03-24=2; n04-09=1; n05-10=1; n06-10=3; n07-08=2 n08-09=1; n09-11=1; n10-11=1; n10-12=2; n11-13=1; n12-23=1; n14-16=2 n15-21=1; n15-24=1; n16-17=2; n16-19=1; n17-18=1; n20-23=1; n21-22=1 n01-08=3; n02-08=1; n06-07=1; n13-14=1 The lines that are removed from the network in step 4 are the following lines: n6-7; n6-7; n1-8; n1-8; n1-8; n15-16; n6-10
VII. RESULT ANALYSIS AND DISCUSSION The point which is more noticeable in this paper is the large difference between the costs of centralized model with the planning that is called open access model. For example in Graver system the costs of centralized model is 110 M$ and the costs of this system is 270 M$ in open access model. On the other hand, for IEEE 24-bus system, in centralized model the cost is about 152 M$ while the costs of open access model is about 1477 M$. The question that may arise: "is the open access model economically significant or not"? In the other word, such increasing in the costs may satisfy the other goals of transmission owners, where it may not create any line congestion. An important point that should be considered is the obtained solution for the centralized model is just for one generation scenario. In fact, if the generation output of generators change, the transmission grid may not capable of supporting such produced power and the lines might be congested. But in open access model it is able to support several generating scenarios without any congestion and as these scenarios are marginal generation scenarios, the transmission grid is able to transfer electric power for any other scenarios. Therefore, it can be said that the open access model has the maximum compatible manner. Now for finding the answer for the above question, the owner of transmission lines should analyze if the profit of its achievements is more than the difference between two planning models plus the profit in the centralized model or not .If the outcome is positive, the open access model is beneficial otherwise not. For clearing the problem, consider Garver system. The difference between two planning model is 158 million dollar, where one reason of this additional cost can be considering N1 security constraint which makes it a proper system under contingency condition. If the profit of completion for the special time period; which is commonly 20 years; is more than 160 M$ plus the profit of centralized planning and also considering the high social welfare of this method which nowadays plays an important role in power markets, the open access model is considered a significant model. In some papers being nearer in to the transmission line is defined as completion metric and the ideal transmission grid is defined, the grid that no obstacle for competition isn't in that. So the grid which no transmission constraint is considered in that is supposed. Now obtain this transmission grid .i.e. the lines are specified. This metric for being nearer into the transmission line can be used more suitable. VIII. CONCLUSIONS TEP needs be revised for its suitability in competitive electricity markets because existing methodologies may not necessarily support competition. In restructured power markets, consumers are paying incurred congestion costs. To have an efficient market environment, ideally, modem TEP should eliminate the congestion for all feasible and future generation scenarios to obtain always the least-costs dispatch. In this paper a mathematical model for TEP problem that can consider multiple generation scenarios in a competitive electricity market is proposed. Case studies considering proposed CHA are also presented. Simulation results show that the algorithm developed for traditional planning can also be employed for planning considering multiple generation
scenarios. The results indicate a direct relation between system flexibility and investment costs. The main contribution of this study is thus the clarification of basic mechanisms for the representation of the possible generation scenarios which should be considered in the modern TEP problem. IX.
[1] [2]
REFERENCES
[3]
[4] [5] [6] [7] [8] [9] [10]
[11] [12]
[13]
[14] [15]
[16]
[17] [18] [19]
L.L. Garver.: Transmission network estimation using linear programming, IEEE Trans.Power App. Syst., 89, pp. 16881697, 1970 S. Haffner, A. Monticelli, A. Garcia and R. Romero.: Specialised branch-and-bound algorithm for transmission network expansion planning, IEE Proc. Gener. Transm. Distrib. , 148, (5), pp. 482488, 2001 M. V. F. Pereira and L. M. V. G. Pinto, Application of sensitivity analysis of load supplying capacity to interactivetransmission expansion planning, IEEE Trans. Power App. Syst., vol. PAS-104,\ pp. 381389, 1985. S. Binato, M. V. Pereira, and S. Granville,A new benders decomposition approach to solve power ransmission network design problems, IEEE Trans. PowerSyst., vol. 16, pp. 235240, 2001. R. Romero, R. A. Gallego, and A.Monticelli, Transmission system expansion planning by simulated annealing, IEEE Trans. Power Syst., vol.11, pp. 364 369. 1996. E. L. Silva, H. A. Gil, and J. M. Areiza, Transmission network expansion planning under an improved genetic algorithm, IEEE Trans. Power Syst., vol.15, pp. 1168 1175, 2000. R. A. Gallego, R. Romero, and A. J. Monticelli, Tabu search algorithm for network synthesis, IEEE Trans. Power Syst., vol. 15, no. 2, pp.490 495, 2000. S. Binato, G. C. de Oliveira, and J. L. de Arajo, A greedy randomized adaptive search procedure for transmission expansion planning, IEEE Trans. Power Syst., vol. 6, no. 2, pp. 247253, 2001. T. Sebastin, A. J. Conejo and J. Contreras.:Transmission Expansion Planning in Electricity Markets , IEEE Trans Power, vol. 23, no. 1, pp. 238 248 , 2008. M. Pourakbari-Kasmaei and M. Rashidi-Nejad, An effortless hybrid method to solve economic load dispatch problem in power systems, Energy Conversion and Management, vol. 59, no. 89, pp. 28542860, 2011. H. Gil, E. da Silva and F. Galiana.: Modeling Competition in transmission expansion IEEE Transactions on Power Systems, vol. 17,pp. 1043-1049,2002 A. Braga and J. T. Saraiva.: A multiyear dynamic approach for transmission expansion planning and long-term marginal costs computation IEEE Transactions on Power Systems, vol. 20, no. 3 pp. 1631-1639,2005. R. Romero, A. Monticelli, A. Garcia, and S. Haffner.: Test systems and mathematical models for transmission network expansion planning lEE Proceedings - Generation, Transmission and Distribution, vol. 149, no.1, pp. 27-36,2002. R. Fang and DJ. Hill.: A new strategy for transmission expansion in competitive electricity markets IEEE Trans. Power Systems, vol. 18,no. 1, pp. 374-380, 2003. M.V.F. Pereira and L.M.V.G. Pinto.: Application of sensitivity analysis of load supplying capability to interactive transmission expansion planning IEEE Transactions on Power, vol. 104, no 2, pp. 381-389, 1985. A. Monticelli, A.J. Santos, M.V.F. Pereira, S.H. Cunha, B. J. Parker and J.C.J. Praca.: Interactive transmission network planning using a least effort criterion IEEE Transactions on Power ,vol. 101, no 10, pp. 39193925, 1982., R. Villasana, L. L. Garver, and S. J. Salon.: Transmission network planning using linear programming, IEEE Trans. Power App. Syst., vol. 104, no. 2, pp. 349356, 1985. R. Fang, D.J. Hill.: A new strategy for transmission expansion in competitive electricity markets IEEE Transactions on Power, vol. 18, no. 1, pp. 374380, 2003. R. Romero, C. Rocha, J. R. S. Mantovani and I. G. Sanches.: Constructive heuristic algorithm for the DC model in network transmission expansion planning, IEE Proc. Gener. Transm. Distrib. vol. 152, no. 2, pp. 277282, 2005.
Optimal FACTS Location to Enhance Voltage Stability Using Multi-objective Harmony Search
Abdelaziz Lafa
Dpartement de gnie lectrique Universit 20 Aot 1955 - Skikda Skikda, Algeria e-mail: azlaifa@yahoo.fr
AbstractIn this paper, a new multi-objective planning framework, namely Non-dominated Sorting Improved Harmony Search (NSIHS), is presented, which is able to evaluate the impact of FACTS placement for an enhancement of voltage stability. It is based on the Improved HS algorithm which has been extended to the multi-objective optimization problem by non-dominated sorting and ranking with crowding distance strategy. Simulations are performed on IEEE 14 test system with FACTS controllers modeled for steady state studies. The approach is compared with Multi-Objective Particle Swarm Optimization (MOPSO) method. From the simulation results, it is clear that the proposed method is able to give better distributed Pareto-optimal solution than MOPSO method. It can then be a robust tool towards multi-objective optimization in power system management and planning. Keywords- Voltage Stability, SVC, TCSC, FACTS location, Harmonic Search algorithm, Multi-objective Optimization.
an actual and important subject to appropriately select the suitable place for the FACTS device installation at the viewpoint of voltage security enhancement. This problem has retained the interest of worldwide researchers in power systems. Then, various methods and criteria were proposed and used to optimal allocation of FACTS devices in power systems [7-11]. In this paper, an optimization approach is used for SVC and TCSC installation. The problem is formulated as a nonlinear constrained multi-objective optimization problem where voltage deviations, voltage collapse impact and power losses are treated as competitive objectives in order to determine the optimal locations of SVC and TCSC. The multi-objective optimization approach has been the subject of many power system problems, and approached with different methods [12]. However, the objectives are often conflicting, leading to the problem of finding a set of optimal solutions, called Paretooptimal front. A good approximation of the Pareto-optimal front helps the Decision Maker in choosing the best compromise solution among all solutions [13, 14]. The use and development of heuristics-based multi-objective optimization techniques have significantly grown. As a consequence, various heuristic approaches have been adopted by researches including genetic algorithm, evolutionary programming and particle swarm optimization to evaluate and enhance network operation [12]. In [15], the authors used an elitist MOPSO algorithm for finding optimal FACTS locations and sizes based on Pareto-optimal front. Harmony search (HS) algorithm has been recently developed [16] in an analogy with an improvisation process where musicians always try to polish their pitches to obtain a better harmony. Music improvisation process is similar to the optimum design process which seeks to find optimum solution. The pitch of each musical instrument determines the certain quality of harmony, just like the objective function assigned to the set of variables. In this paper, HS algorithm is extended using fast non dominated sorting and ranking performed in well known NSGA-II algorithm [17]. A fuzzy based mechanism can be used to find a compromise solution from the obtained Pareto-optimal front. This proposed multi-objective harmony search algorithm has been tested on the standard IEEE 14 bus test system. The Pareto front solution was compared with the one obtained by the MOPSO method [15]. Simulation results clearly show the robustness of the multiobjective HS method to obtain well distributed Pareto-optimal solutions than MOPSO method.
I.
INTRODUCTION
Flexible AC Transmission System (FACTS) devices, which can provide direct and flexible control of power transfer, can be very helpful in the operation of power networks. Both the power system performance and the power system stability can be enhanced by utilizing FACTS devices. They provide the opportunity to influence power flows and voltages in such a way that thermal limits are not violated, losses minimized, stability margin increased, and contractual requirement fulfilled etc, without the need of new lines [1, 2, 3, 4]. However, our concern in this paper is to enhance the voltage stability limit with the help of FACTS devices. Two main types of devices are considered here, namely, Static Var Compensators (SVC) and Thyristor Controlled Series Compensators (TCSC). TCSC as an effective series compensation device can be used for adjusting the limitation created by generation and transmission systems. SVC as a shunt compensation component is designed for voltage maintenance in power systems. It makes it possible the functioning of the system by increasing its loading margin. Thus, they are increasingly used in nowadays stressed transmission systems [5, 6]. For practical power systems, different buses are differently sensitive to the overall power system voltage stability. Some buses are more, and some are less. In order to profit from the benefits of FACTS devices, proper allocation and appropriate set values for the controllers of these devices have to be determined. To a large extent, proper allocation of SVC can make great enhancement to voltage stability [3]. Therefore, it is
The remaining of this paper is organized as follows: next, the models for the FACTS devices applied in this paper are given. The optimization problem is formulated in Section III. To make a proper background, the basic concept of the HS and the improved version of the HS algorithm IHS reported in the literature are briefly explained in Section IV. Also, in Section IV, the proposed multi-objective HS (NSIHS) is explained. Results obtained are given in Section V and some conclusions are drawn in Section VI. II. MODELING OF FACTS DEVICES As the intention is to improve the static voltage stability margin, the power system as well as the FACTS devices is modelled using static equations. The FACTS devices considered in this paper are SVC and TCSC, since these FACTS devices, besides Phase Angle Regulators, are used most frequent in power systems [2]. An SVC is shunt-connected to a bus and influences the voltage at the bus to which it is connected by injecting or absorbing reactive power. It combines a series capacitor bank shunted by thyristor controlled reactor. Then, the SVC can be considered as a synchronous compensator modeled as PV bus, . This with limits designed by its rated size characteristic can be modelled by a shunt-connected variable susceptance (Fig. 1). As the size of the SVC is limited, a and , exist for the lower and an upper bound, . These values are determined by the effective susceptance dimensions of the reactors and capacitors integrated in the elements of the SVC. Thus: (1) The reactive power injected into the system is then given by: . (2) In the other hand, the TCSC is a series compensation component which consists of a series capacitor bank shunted by thyristor controlled reactor. A TCSC is connected in series with a transmission line and is able to influence the active power flow through the line by adapting the reactance of the line. Hence, the device is modelled as a variable reactance connected in series with the line, as shown in Fig. 2. This reduces the transfer reactance between the buses to which the line is connected. This leads to an increase in the maximum power that can be transferred on that line in addition to a reduction in the effective reactive power losses. The series capacitors also contribute to an improvement in the voltage profiles. The total reactance of the line including the TCSC is therefore: + (3) The effective reactance of the TCSC is limited to a and determined by the size range between of the device but also by the allowed line compensation, set usually to 80% capacitive and 40% inductive [18], resulting in (4)
QSVC
VSVC
j BSVC
Fig. 1. Model of SVC
R+j X
j XTCSC
jB/2
jB/2
Fig. 2. Model of TCSC
III.
PROBLEM FORMULATION
As we already mentioned, this paper focuses on the optimal location and design of two kinds of FACTS, namely the SVC and the TCSC. The goal of the optimization is to obtain the best utilization of the existing power network. Only the technical benefits of the FACTS controllers are taken into account (no equipments cost considered). In this respect, the FACTS devices are located so to maximize the loading level, minimize power system total loss and flatten buses voltage. The optimal location and design of SVC and TCSC is then formulated as a mixed continues-discrete multi-objective optimization problem. The optimization parameters are the FACTS locations and sizes. A. Objective Functions The goal is to determine the optimal locations and parameters settings of the SVC and TCSC in the power network to eliminate or minimize the risk of voltage violations and the increase of voltage stability margin. The objectives selected for this study are presented below. Voltage Stability Enhancement Voltage stability enhancement is achieved through maximizing the Voltage Stability Margin (VSM) or Loading Margin, which is the most widely accepted index for proximity of voltage collapse. It is defined as the largest load change that the power system may sustain at a bus or collective of buses from a well defined operating point (base case). The maximization of VSM can be presented as follows: (5) where is the value of loading factor at the critical point or voltage stability margin. The method used for this margin calculation is the Continuation Power Flow (CPF) [19]. The loads are increased gradually and the power sources will
compensate for this increase by increasing its generation. At every load level, the system state is calculated until the maximum or critical condition is reached. Voltage Deviation To have a good voltage performance, the voltage deviation at each load bus must be made as small as possible. The voltage deviation to be minimized is as follows:
(6)
copied from the memory; pitch adjustment, which disturbs the chosen component and random selection that randomly generates a new component [21]. The parameters are: the Harmony Memory Consideration Rate (HMCR), which varies between 0 and 1, is the rate of choosing one value from the historical values stored in the HM, Pitch Adjustment Rate (PAR) and the Bandwidth (BW). The pitch adjustment operation is as follows: + . (10) The detailed description of the HS algorithm can be found in [2224]. The improved version of harmony search (IHS), proposed in [25], provides the fine-tuning of the parameters PAR and BW, making them change dynamically with generation number or iteration (Iter) and the total number of improvisations (NI ), as shown in equations (11) and (12). +
( )
Real Power Losses As economic aspect, the minimization of active power losses is taken into account. can be expressed as:
+ 2 cos (7)
B. System Constraints The equality constraints represent the typical load flow equations as follows:
cos sin 1, , + sin cos 0 0 (8) (9)
(11) (12)
The inequality constraints represent the system operating limits like limits on reactive generation and bounds on tap settings of transformers. IV. MULTI-OBJECTIVE HARMONY SEARCH
A. Harmony Search Optimization The harmony search (HS) algorithm, proposed by Geem et al. [20], is a nature inspired algorithm, mimicking the improvisation of music players. The harmony in music is analogous to the optimization solution vector, and the musicians improvisations are analogous to the local and global search schemes in optimization techniques. The HS algorithm uses a stochastic random search, instead of a gradient search. This algorithm uses harmony memory considering rate and pitch adjustment rate for finding the solution vector in the search space. The optimization procedure of the HS algorithm is as follows: 1. Initialize the optimization problem and algorithm parameters. 2. Initialize the harmony memory HM. 3. Improvise a new harmony (or solution). 4. Update the harmony memory: if the new solution is better than the worst one in HM, the worst is replaced. 5. Check for stopping criteria (number of improvisations). Otherwise, repeat step 34. The improvisation can be divided in three phases: memory consideration, where an existing harmony component is
B. Non-dominated Sorting Improved Harmony Search A Multi-objective optimization problem always has a set of optimal solutions, for which there is no way to improve one objective value without deterioration of at least one of the other objective values. Pareto dominance concept classifies solutions as dominated or non-dominated solutions and the best solutions are selected from the non-dominated solutions. In this work, fast non-dominated sorting and ranking scheme, proposed by Deb et al. [17] in their NSGA-II algorithm is used to find Pareto optimal solutions. To sort nondominated solutions, the first front of the non-dominated solution is assigned the highest rank and the last one is assigned the lowest rank. When comparing solutions that belong to a same front, another parameter called crowding distance is calculated for each solution. The crowding distance is a measure of how close an individual is to its neighbors. Large average crowding distance will result in better diversity in the population. In order to investigate the multi-objective version of the improved HS algorithm using the NSGA-II framework, named Non-dominated Sorting Improved Harmony Search (NSIHS) some modifications in the HS algorithm were made. The details of NSIHS are as follows: 1. Define objective functions and decision variables, and input the system parameters and the boundaries of the decision variables. 2. Initialize the harmony memory (HM). Initial population is produced randomly within the range of the boundaries of the decision variables in the HM so that none of them are repeatable. 3. Start the improvisation. Set iteration counter Iter:=0. 4. Set another new memory (HMnew) empty.
5. Generate new harmonies using HM and the improved improvisation process. Fill the HMnew with these new harmonies. 6. The new harmony memory, generated by improvisation process, is combined with the existing harmony memory to form 2HMS solution vectors. 7. The non-dominated sorting and ranking procedure is performed on the combined harmony memory. 8. At the end of each iteration, the harmonies in the combined HM are sorted according to the definition of domination. Update the HM by taking the first HMS best solutions. 9. Check for stopping conditions. If the number of improvisations has been reached to the maximum, go to the next step. Otherwise, return to step 4. 10. The non-dominated harmonies in the combined HM are Pareto front (or non-dominated solution vectors). C. Best compromise solution Having obtained the Pareto optimal set, choosing a best compromise solution is important in decision making process. The best compromise solution is obtained by using fuzzy membership approach. Due to imprecise nature of the decision makers judgment the ith objective function Fi of individual k is represented by a membership function defined as [26] 1 (13) 0 where and are the minimum and maximum value of the ith objective function among all non-dominated solutions, respectively. For each non-dominated solution k, the normalized membership function uk is calculated as

According to this initial CPF runs and based on the maximum value of the tangent associated to voltage variation at the collapse point, bus 14 is the first most sensitive bus and seems needing Q support (tangent= 0.1284 for the base case). It can be selected as a suitable placement for the SVC to the power system voltage stability enhancement. Before NSIHS is applied to location problem, the algorithm parameters need to be defined. They were tuned based on many trials and previously works. The selected tuned parameters are: Harmony Memory Size: HMS = 100 Harmony Memory Considering Rate: HMCR = 0.95 Pitch Adjusting Rate: PARmin = 0.2 and PARmax = 0.6 Bandwidth: BWmin = 0.01 and BWmax = 0.9 Number of Improvisations: NI = 1000 A. Optimal Location of SVC The goal is to find the best location of SVC which is needed to enhance voltage stability. The optimization is made on two parameters: location and size. The SVC size limits are fixed at the beginning. In doing so, the SVC is considered as a synchronous compensator with a reactive power changing continuously between 0 and 2pu. The optimal location of SVC is considered as a discrete decision variable, where all load buses are candidates to be the optimal location of SVC. The problem is formulated as bi-objective optimization considering the minimization of real power losses and the maximization of VSM. Here, the VSM improvement and power loss minimization are done at the same time. Then, the power losses are calculated at the critical point. The critical point corresponding to the voltage collapse condition can be determined using the CPF method. This multi-objective optimization problem was solved by the proposed approach and MOPSO method for comparison purpose. The Pareto optimal solution obtained using the proposed NSIHS algorithm and MOPSO are shown in Fig. 3. From the Pareto optimal solution, it is clear that the proposed method is able to give better distributed solutions than MOPSO. The compromise solution, shown in Table I, was found using the fuzzy membership approach. B. Optimal Location of SVC and TCSC The decision variables considered here are the location and size of both TCSC and SVC. The number of FACTS to be installed is chosen one for each type; also the limits are fixed at the beginning. The SVC is considered as previously. However, the reactance of TCSC is considered varying continuously between -20% and 80% of the line reactance. The placement of TCSC is considered as a discrete variable, where all lines of the system are selected to be the candidates location of TCSC. The problem is again formulated as biobjective optimization considering the minimization of real power losses and the maximization of VSM. The Pareto optimal solution obtained by the proposed method and MOPSO are shown in Fig. 4. It can once again be proved that the proposed method is giving well distributed Pareto optimal front. Compared with the case of the SVC
(14)
where M is the total number of non-dominated solutions. The best compromise solution is that having maximum value of uk. V. SIMULATION RESULTS AND DISCUSSION
The NSIHS algorithm and all calculation programs used in this study were written in Matlab7.1 and used to perform the optimization routines. The application is carried out on the IEEE 14-bus test system, which consists of two generators, located at bus 1 and 2; three synchronous compensators used only for reactive power support at buses 3, 6 and 8. The generators are modeled as PV buses with Q limits; the loads are typically represented as constant PQ loads with constant power factor, and increased gradually via the load parameter . With the increasing of the system loading condition, the power system is stressed gradually to the critical point where the power system will lose its voltage stability. This condition, determined first for the base case, is reached at = 1.7401.
allocation, the Pareto front of the non-dominated solutions is somewhat larger and enhances the power losses while obtaining the same loading margin in comparison with the case of SVC alone (Fig. 5). We give in Table I compromise solutions for this case with the corresponding losses and VSM.
2.5
N SIH S MOP SO
TABLE I. case SVC location (bus #) QSVC (pu) TCSC location (line #) XTCSC
BEST COMPROMISE SOLUTIONS
SVC 4 1.5961 2.4118 1.067
SVC+TCSC 5 1.9353 10-11 0.7269 2.6396 1.2524
Three Objectives 1 0.0760 1.3502 0.1747
Ploss (pu)
losses (pu)
1.5
0.5
1.2
1.4
1.6
1.8
2.2
2.4
2.6
2.8
load factor
C. Three Objective Case In this case, all the three competing objectives were taken: VS margin, power losses and voltage deviations. These three objectives were optimized simultaneously by the proposed NSIHS. Fig. 6 shows the Pareto optimal set for the proposed method. The best compromise solution is given in the same Table as previously. From the results, it is clear that the proposed method is giving well distributed solution.
Fig. 3. SVC allocation
2.5
N SIH S MOP SO
0.4
v oltage deviation
0.3
0.2
los ses (pu)
1.5
0.1
0 2 1.5 3 1 0.5 1.5 0 1 2.5 2
0.5
los ses (pu)

0 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
load factor
load factor
Fig. 6. Three objectives SVC allocation
Fig. 4. SVC and TCSC allocation
VI.
CONCLUSION
2.5
S VC+TC SC S VC
los ses (pu)
1.5
0.5
1.2
1.4
1.6
1.8
2.2
2.4
2.6
2.8
load factor
Fig. 5. SVC and TCSC allocation using NSIHS
In this paper, a new multi-objective harmony search algorithm, namely NSIHS, has been presented and successfully applied to FACTS allocation problem for the enhancement of voltage stability. Non-dominated sorting and ranking with crowding distance strategy were used to find and maintain the Pareto optimal solutions. The proposed algorithm has been validated on the standard IEEE 14 bus system and compared with MOPSO method. A fuzzy membership approach has been used to identify the best compromise solution. From the simulation results, the proposed NSIHS algorithm is able to give better distributed Pareto optimal solutions than MOPSO method for FACTS location problem in order to improve voltage stability margin and considering the minimization of power losses and voltage deviations. Its effectiveness and implementation on power system management and planning can be a robust tool towards multiobjective optimization.
REFERENCES
N. Yorino, E. E. El-Araby, H. Sasaki, S. Harada, A New Formulation for FACTS Allocation for Security Enhancement Against Voltage Collapse, IEEE Trans. on Power Systems, Vol. 18, No. 1, Feb. 2003, pp. 3-10. [2] G. Hug-Glanzmann, G. Andersson, Coordinated Control of FACTS Devices in Power Systems for Security Enhancement, Bulk Power System Dynamics and Control - VII, August 19-24, 2007, Charleston, South Carolina, USA. [3] J. Zhang, J. Y. Wen, S. J. Cheng, J. Ma, A Novel SVC Allocation Method for Power System Voltage Stability Enhancement by Normal Forms of Diffeomorphism, IEEE Trans. on Power Systems, Vol. 22, No. 4, Nov. 2007, pp. 1819-1825. [4] S. Rahimzadeh, M.T. Bina, Planning Required for FACTS Devices to Improve the Steady State Efficiency of Restructured Power Systems: A Review, International Review of Electrical Engineering (IREE), Vol. 4, n. 3, June 2009, pp. 417-424. [5] S. Mollazei and al., Multi-objective Optimization of Power System Performance with TCSC Using the MOPSO Algorithm, IEEE Trans. on Power Systems, Nov. 2007. [6] R. Mnguez, F. Milano, R. Zrate-Miano, A. J. Conejo, Optimal Network Placement of SVC Devices, IEEE Trans. On Power Systems, Vol.22, No.4, Nov. 2007. [7] Y. Mansour et al., SVC Placement Using Critical Modes of Voltage Instability, IEEE Trans. on Power Systems, Vol. 9, No. 2, May 1994, pp. 757-763. [8] L. A.S. Pilotto, Determination of Needed Facts Controllers That Increase Asset Utilization of Power Systems, IEEE Trans. on Power Delivery, Vol. 12, No. 1, Jan. 1997, pp. 364-371. [9] M. K. Verma, Optimal Placement of SVC for Static and Dynamic Voltage Security Enhancement, International Journal of Emerging Electric Power Systems, Vol. 2, Issue 2, Article 1050, 2005. [10] A.K. Sharma, Optimal Number and Location of TCSC and Loadability Enhancement in Deregulated Electricity Markets Using MINLP, International Journal of Emerging Electric Power Systems, Vol. 5, Issue 1, Article 6, 2006. [11] Malihe M. Farsangi, H. Nezamabadi-Pour, K. Y. Lee, Multi-objective VAR Planning with SVC for a Large Power System Using PSO and GA, 2006 Power System Conference & Exposition, pp. 274-279. [12] M. A. Abido, Multi-objective Evolutionary Algorithms for Electric Power Dispatch Problem, IEEE Trans. On Evolutionary Computation, Vol. 10, No. 3, June 2006, pp. 315-329. [13] K. Deb, Multi-objective genetic algorithms: Problem difficulties and construction of test problems, Evolutionary Computation, vol. 7, pp. 205230, 1999. [1]
[14] A. Zhou, B.-Y. Qu, H. Li, S.-Z. Zhao, P. N. Suganthan, and Q. Zhang, Multiobjective evolutionary algorithms: A survey of the state of the art, Swarm and Evolutionary Computation, vol. 1, no. 1, pp. 32 49, 2011. [15] A. Laifa, M. Boudour, Multi-Objective Particle Swarm Optimization for FACTS Allocation to Enhance Voltage Security, International Review of Electrical Engineering (IREE), Vol. 4, n. 5, October 2009, pp. 994-1004. [16] E. Zitzler, J. Knowles, and L. Thiele, Quality assessment of pareto set approximations, pp. 373404, 2008.J. Williams, Narrow-band analyzer (Thesis or Dissertation style), Ph.D. dissertation, Dept. Elect. Eng., Harvard Univ., Cambridge, MA, 1993. [17] K. Deb, A. Pratap, S. Agarwal, T. Meyarivan, A fast and elitist multiobjective genetic algorithm: NSGA-II, IEEE Transaction on Evolutionary Computation 6, 2002, pp. 182197. [18] A. Sode-Yome, N. Mithulananthan, Static Voltage stability Margin Enhancement using STATCOM, TCSC and SSSC, IEEE/PES Transmission and Distribution Conference & Exhibition: Asia and Pacific Dalian, China, 2005. [19] C. A. Canizares, Applications of Optimization to Voltage Collapse Analysis, IEEE/PES Summer Meeting, San Diego, July 14, 1998. [20] Z. Geem, J. Kim, et al., A new heuristic optimization algorithm: harmony search, Simulation 76 (2) (2001) 6068. [21] Z. W. Geem, State-of-the-art in the structure of harmony search algorithm, in Recent Advances In Harmony Search Algorithm, ser. Studies in Computational Intelligence, Z. Geem, Ed. Springer Berlin / Heidelberg, 2010, vol. 270, pp. 110. [22] M. Fesanghary, M. Mahdavi, M. Minary-Jolandan, Y. Alizadeh, Hybridizing harmony search algorithm with sequential quadratic programming for engineering optimization problems, Comput. Meth. Appl. Mech. Eng. 197 (3340) (2008) 30803091. [23] L. Coelho, V. Mariani, An improved harmony search algorithm for power economic load dispatch, Energy Convers. Manage. 50 (10) (2009) 25222526. [24] M. Fesanghary, M. Ardehali, A novel meta-heuristic optimization methodology for solving various types of economic dispatch problem, Energy 34 (6) (2009) 757766. [25] M. Mahdavi, M. Fesanghary, and E. Damangir, An improved harmony search algorithm for solving optimization problems, Applied Mathematics and Computation, vol. 188, no. 2, pp.15671579, May 2007. [26] T. Niimura, T. Nakashima, Multiobjective tradeoff analysis of deregulated electricity transactions, International Journal of Electrical Power & Energy Systems, 25(3) (2003) 179-185.
A Fuzzy Logic Controller for Reactive Power and PQ Bus Voltage Control
F. Benhamida*(1), S. Souag, Y. Salhi,
Irecom laboratory, dept. of electrotechnics UDL university of Sidi Bel Abbes Sidi Bel Abbes, Algeria (1) dr.benhamida@yahoo.fr
Abstract Unlike the active power control problem, the reactive power control problem is generally difficult because of the large scale and non-linear characteristics in power systems. Based on the power system status, the reactive power and voltage control is required to make the correct decision in order to get the system back to an acceptable operating state. For this purpose we have developed in this paper an algorithm based on Fuzzy Logic in order to control the VAR resources, transformer tap changer and PV pus voltage in a power system under abnormal and/or contingency operation to maintain the voltage at all buses within acceptable limits, when minimizing the number of control actions. The algorithm is iterative and designed to take advantage decoupling the load flow Jacobian matrix to decrease the computation time. The effectiveness of the developed algorithm is identified through its application to the IEEE 14 test system. The calculation results show excellent performance of the proposed method, in regard to computation time and quality of results. Keywords-component; Reactive power control, voltage control, fuzzy logic.
A. Graa, A. Bendaoud(2)
Apelec laboratory, dept. of electrotechnics UDL university of Sidi Bel Abbes Sidi Bel Abbes, Algeria (2) babdelber22@gmail.com Many deregulated electricity markets have established financial compensation mechanisms for reactive power services [3], and consider the concept of voltage stability margin to avoid voltage collapse occurred in heavily stressed power system [4]. Nowadays in the operation of power systems, the reactive power controls the reactive power sources in the system in a synchronized approach to recover the system voltage profile in nominal margins and to minimize a suitable objective function. Based on the power system status, the reactive power control is required to make the correct decision in order to get the system back to an acceptable operating state. For this purpose we have developed an algorithm based on fuzzy logic in order to control the VAR resources, transformer tap changer and PV pus voltage in a power system under abnormal and/or contingency operation to maintain the voltage at all buses within acceptable limits. Several papers [4] have been published on reactive power reserve management with the perspective of ensuring voltage stability by providing an adequate amount of reactive power reserves. Menezes in [5] propose a strategy to improve the voltage stability by dynamic VAR source scheduling. In the same context in [6], the authors established a method for reactive power injection rescheduling from generators and synchronous condensers based on participations factors and optimal power flow solution. The authors in [7] introduce a reactive power optimization scheme by optimizing the corrective controls. Evolutionary algorithms like differential evolution, genetic algorithm, and evolutionary programming [8,9] have been extensively used during the last 2 decades in the field of optimization. Such intelligent approaches were used for optimal reactive power dispatch in recent works [10][12]. The objective considered when controlling the reactive power in this paper, is the minimization of the number of control actions needed and keep the PQ bus voltage deviation within an accepted deviation from the nominal voltage. The algorithm is iterative and designed to take advantage of a method of decoupling the load flow Jacobian matrix to decrease the computation time. The method use sensitivity information derived from the load flow Jacobian and augmented with equations relating the desired control and dependent variables.
I.
INTRODUCTION
Today, the power systems are often interconnected in large power pools. Due to the continuous and rapid growth of loads without a corresponding increase in transmission capability as well as economical and environmental constraints, operation of such power systems becomes increasingly complicated. On the other hand and as a result of deregulation, the power transfers increase, transmission margins reduce and, at the same time, voltage security margins reduce, so the electric power systems utilities around the word have to make a better use of their respective available transmission systems [1]. The controls of real and reactive power are two major Energy Management Systems (EMS) functions. These controls are required to maintain operation and quality of supply (low cost, constant voltage magnitude and constant frequency) [2]. Unlike the active power control problem, the reactive power control problem is generally difficult because of the large scale and non-linear characteristics in power systems. In the past two decades there has been growing concern about reactive power operation [2] which is a significant factor to support the system security and reliability. This control is achieved through coordination between generator voltage setting, switchable shunt compensators, and transformer taps.
Identify applicable sponsor/s here. If no sponsors, delete this text box. (sponsors)
II. OBJECTIVE FUNCTION The method for optimizing reactive power control used in this paper is the minimizing the number of controllers that may be changed in order to improve voltage profile (minimize the control actions, i.e.,) under the constraints of: (1) reactive power limits of the generators, (2) voltage level limits at all PQ buses, and (3) the operating limits of the control variables. The control variables used are: PV voltages, transformer tap changer setting, and switchable reactive power resources. The constraints are: (1) limits on the control variables and (2) limits on the dependent variables. III. OPTIMIZATION PROBLEM FORMULATION The total number of buses in a power system (n) can be divided into 3 categories: (1) generator buses (g); (2) switchable VAR compensator buses (p) and (3) PQ buses without any source of VAR compensation (l), where: n = g + p + l. The general form of this optimization problem is: Min(Fobj) = c.u Subject to dependent variable constraints : ximin xi ximax for : i = 1,2,, n and control variable constraints : ummin um ummax for : m = 1, 2,, g + t + p
x = S .u
u = [ a jk
V jmin V j V jmax
(9)
j = g+1, g+2,,n where Qimax= Qimax-Qi; Qimin= Vimin= Vimin-Vi . Qimin-Qi; Vimax= Vimax- Vi and
B. Constraints on the Control Variables These constraints represent the limits for the generator terminal voltages, the limits for the transformer taps, and those for the switchable shunt capacitors, these limits in linearized which depend on the current values of (ajk, Vi, Qk) and their upper and lower limits:
a min a jk a max jk jk Vi min Vi Vi max
(10) (11) (12)
pour : i = 1, 2,,g
(1) (2) (3) (4)
Q kmin Q k Q kmax
k = g+1, g+2,,g+p where p is the number of switchable shunt capacitors; t is the number of tap changing transformers; akmax= akmax-ak; ajkmin= akmin-ajk; Qkmax= Qkmax-Qk; Qkmin= Qkmin-Qk; Vimax= Vimax- Vi and Vimin= Vimin-Vi . IV. TAP CHANGING TRANSFORMER MODEL The tap ratio of a transformer can be changed by changing number of turns of transformers windings using a tap changer. The representation of an equivalent circuit of transformer with tap changer under load is shown in Fig. 1.
Ij + Vj Ijk Y/aik Ikj + Ik + Vk -
( ) , (a )
1 jk
, , a jk
( ) , V , V , ,
t 1 2
.... V g , Q1 , Q 2 , , Q p ]T
(4) (5)
x= Q1 , Q 2 , , Q g , V g + 1 , V g + 2 , V n
where c is a row vector of the linearized objective function sensitivity coefficients; u is a column vector of linearized control variables; umin,umax are the lower and upper limits on the control variables ; x is a column vector of linearized dependent variables; xmin, xmax are are the lower and upper limits on the dependent variables and S is a linearized sensitivity matrix relating the dependent and control variables. A. Constraints on the Dependent Variables These constraints represent the limits for the reactive power generation of the generators, and the limits for the voltage magnitudes of all PQ buses, these limits are as follow:
Qimin Qi Qimax
1 a Y 2 jk a jk
a 1 Y jk a jk
Figure 1. Equivalent circuit of a transformer with a tap changer
where I is the current, V is the voltage and Y admittance, respectively; ajk, is turn ratio and j is the tap side of the transformer. The derivatives of active and reactive power injected at bus j and k of circuit of Fig. 1 can be found as:
Pj a jk = 2G V j2 a
3 jk
for : i = 1, 2,,g
(6) (7)
V jVk + 2 a jk
G cos jk + B sin kj
( )
( )
(13)
V jmin V j V jmax
j = g+1, g+2,,n The inequalities (6) and (7) can be rewritten in a linearized form which depends on the current values of Qi, Vi, and their upper and lower limits (Qimax, Vjmax, Qimin, Vjmin) as:
Qimin Qi Qimax
V jVk Pk = 2 a jk a jk
G cos jk B sin jk
( )
( )
(14)
Q j a jk
= 2B
V j2 a
3 jk
V jVk + 2 a jk
g sin jk B cos jk G
( )
( )
(15)
pour : i = 1, 2,,g
(8)
V jVk Qk = 2 a jk a jk
G sin jk + B cos jk
( )
( )
(16)
Equation (22) can be divided into three equations. Assuming Q / 0 . The third one is:
Q1 = Qn a jk V1 V n
with y=G+jB; i and k are the phase angle of Vi and Vk, respectively and ik= i -k. The derivatives of active and reactive power flow through branch jk in Fig. 1 can be found as:
Pjk a jk
Pkj a jk
Q a jk
Q V
(23)
= G
V j2 a2 jk
V jVk + 2 a jk
G cos jk + B sin jk
( )
( )
(17)
= G
Vk2 V j Vk + 2 a jk a2 jk
G cos jk B sin jk
( )
( )
(18)
where the solution are the form:

x = S .u
(24)
Q jk a jk
Q kj a jk
=B
V j2 a2 jk
2 k 2 jk
V jVk + 2 a jk
V jV k + 2 a jk
G sin jk B cos jk
( )
( )
(19)
where variables u and x are defined by (4) and (5), and the matrix S:
S1 ( S2 )( S4 )1 ( S3 ) [S ] = 1 ( S4 ) ( S3 )
=B
V a
G sin jk + B cos jk
( )
( )
(20)
( S2 )( S 4 )
V.
SENSITIVITY MATRIX
Unity( p p) Zeros(l p) Unity p p ( ) 1 ( S4 ) Zeros ( l p)

1
(25)
The Newton Raphson load flow equations in matrix form are as follows:
P2 P2 2 Pn 2 Pn = Q Q2 2 2 Qn Qn 2 P2 P2 n V2 Pn P2 n V2 Q2 Q2 n V2 Qn Qn n V2 P2 Vn Pn Vn Q2 V n Qn Vn
where I is a unity matrix and 0 is a zero matrix.

Agt S1 = , S2 = Agp Agg Agl
(26) (27)
2 n V 2 Vn
(21)
Apt S3 = Alt
Apg J pp J pl , S4 = Alg Jlp Jll
Agt Agg Agp Agl Q Apt Apg App Apl = a jk A A A A ll lt lg lp
Q V
(28)
VI.
NUMBER OF CONTROL VARIABLES MINIMIZATION
Let augment the power of the slack bus and the power flow equations are augmented to the load flow equations in (21):
P1 P Pn P = Pjk jk Q1 Q Qn 1 n a jk V1 Vn
P a jk Pjk a jk Q a jk
P V Pjk V Q V
Consider a power system with n buses, bus 1 to g are generator buses, bus g+1 to n are PQ buses, and bus 1 is the slack bus, the system has t tap changing transformers, and a p buses with a switchable shunt capacitor. By adjusting the controlling device at PQ bus j by acting on uj, the change in voltage profile at bus i is:
Vi = Sij .u j
(22)
(39)
for i=g+1, g+2, .,n and j=1,2,.t+g+p where Vi is the voltage change at PQ bus i; uj is the regulation of the controlling device j and Sij is the sensitivity matrix element of the controlling device j on voltage at PQ bus i. The regulation of the controlling devices is constrained by the upper and lower limits as:
u min u u max
where P / , P / V , Q / , Q / V are (nn) submatrices ; P / a jk , Q / a jk are (nt) sub-matrices ; Pjk / , Pjk / V are (tn) sub-matrices and Pjk / a jk is a (tt) submatrix.
(30)
The objective in this paper is to keep the PQ bus voltage Vi variation within 5% of the nominal voltage for i = g+1,
g+2,, n. Based on the upper and lower limits for every controller, the controlling capacity for each controller can be divided as follows:
c+ = S.m+ + ci = ci ci c = S.m
direct measure of the severity of the problem, voltage violation. On the other hand, controlling ability cij is selected as a second input because it represents the ability of the available controller to fix the problem. B. PQ Bus Voltage Violation The membership functions that represent the voltage deviations are shown in Fig. 3. In this paper, we need to control the voltage deviation at any PQ bus to be within 5% of the rated voltage magnitude. The minimum and maximum voltage deviation at a PQ bus can be calculated from:
Vi min = Vi min Vi nom nom max max Vi = Vi Vi
(31)
S: is the sensitivity matrix of dimension ((n-g)(t+g+p)) represents derivatives of voltage magnitude at all PQ buses with respect each control variable. This matrix shows the effect of any change in one control variable on all the dependent variables. m+ is a diagonal matrix of dimension (t+g+p)((t+g+p)). The diagonal elements represent positive margins of the controlling devices based on the current operating point and equal to (umax u). m- is a diagonal matrix of dimension ((t+g+p)((t+g+p)). The diagonal elements represent negative margins of the controlling devices based on the current operating point and equal to (umin - u). VII. FUZZY LOGIC SOLUTION The term Fuzzy Logic was introduced with the 1965 proposal of fuzzy set theory by Lotfi A. Zadeh [13][14] Fuzzy logic has been applied to many fields, from control theory to artificial intelligence [15]. FL mimics how an expert person would make decisions, but at very high rate depending on the knowledge base. FL provides an easy way to arrive at a definite conclusion regarding a problem when there is no mathematical model for the problem, or there is a mathematical model but difficult to understand, or there is a mathematical model but it is complex to be used in real time. Fig. 2 shows the fuzzy logic block diagram.
Fuzzification Rules base Inference System Figure 2. Fuzzy system block diagram Defuzzification
(32)
where Vimin is the minimum accepted voltage magnitude at a PQ bus. In this pper it is 0.95 pu; Vimax is the maximum accepted voltage magnitude at a PQ bus. In this paper it is1.05pu and Vinom is the rated voltage magnitude in per unit. In this work it is 1.0 pu
Figure 5. Membership function for voltage violation
Input
Output
A. Fuzzy Logic Modeling The main problem of the conventional optimization algorithms, is the strict modeling of network constraints, as a result of such strict modeling, the resulting solutions of this modeling may be incapable of representing practical cases. Consequently, a more reasonable, and reliable model for the voltage magnitude control is needed. In the proposed paper, fuzzy sets are used to model the objective function and network constraints. Two different variables are observed and used as two inputs to the proposed model, these inputs are: Figure 3. PQ bus that has the largest voltage violation, Vi; Figure 4. Controlling capacity cij of the controlling device j that has maximum control ability on bus i, that has the largest voltage violation. The main problem is to improve the voltage profile of a power system. The largest voltage violation Vi at a PQ bus is selected as one input to the proposed fuzzy logic because it is a
C. Controlling Ability of the Controller The membership functions that represent the controlling capacity of the controller are shown in Fig. 4. In the case of an over-loaded system, the bus voltage magnitude is lower than the minimum acceptable voltage limit, the controller increase the voltage magnitude to return to tolerable limits. The maximum controlling ability of all the controllers on bus i (bus with minimum voltage), is the maximum of (31). If the system is under-loaded, the bus voltage magnitude is larger than the maximum tolerable voltage limit. The controller decreases the voltage magnitude. The minimum of (31) represents the maximum controlling ability to decrease the voltage magnitude at bus i (bus with maximum voltage).
Figure 6. Membership function for control ability
D. Controller Output The PQ bus with the largest voltage violation is fuzzified according to the membership functions in Fig. 3, in the same manner the control device that has the largest control capacity on that bus is fuzzified using the membership functions shown in Fig. 4. The decision taken by the controller is based on the rule base designed according to the experience with the power system. The membership functions of the controller output are depicted on Fig. 5. The design of the membership functions completely depends on the power system under study, and may be changed from one to other according to the acceptable voltage violations. The controller uses Mamdani FIS structure method, and the defuzzification method adopted is the center of gravity method.
The rule base is designed based on the operators experience using:

if ( X) and ( Y) then ( Z)
TABLE I. RULE MAP OF FUZZY CONTROLLER
(33)
cij
VN N P VP VN Z Z N2 N3 N Z Z N1 N2 P P2 P1 Z Z VP P3 P2 Z Z
V i
Figure 7. The output of the fuzzy controller
VIII. SOLUTION PROCEDURE The details of the solution process for fuzzy logic controller algorithm are summarized in the following steps: 1. Execute a power flow solution. 2. Check voltage magnitudes at PQ buses. If voltage enhancement is needed proceed to Step 3, otherwise, stop. 3. Calculate the sensitivity matrix (S), m+, m-. 4. Locate (Vmin) and (Vmax) within the system voltages. 5. Locate the PQ bus with the most voltage violation. 6. In the case of over loaded system, locate the maximum value of matrix, [c+ c-], corresponding to the bus i with the largest violation Vi., otherwise calculate the minimum of [c+ c-]. 7. Fuzzify the voltage violation Vi. and the control ability of controller j on bus i cij. As inputs to the fuzzy logic controller. 8. Apply the fuzzy controller to get the new setting of the controller j. 9. Update the setting of the controller j and execute a power flow solution and go back to Step 2. IX. RESULTS AND DISCUSSION The proposed algorithms have been tested the IEEE 14 bus system. The complete description and data of the modified IEEE 14 system can be found in [16]. The result of dull AC load flow is presented in Fig. 7. The voltage profile is presented in Fig. 8. From Fig. 8 we can remark that bus 7, 9, 10, 11, 12 and 13 violate their maximum voltage levels. We apply the proposed FL controller programmed using LabVIEW platform to correct this violation. Table II present the parameters of controllable devices and their limits. The last column of Table II presents the correction action solved by the FL controller where the sufficient and optimum actions are increasing in the tap changing a47 and a79 and decrease a78. The computational time is about 0.1 sec. The voltage profile after correction is presented in Fig. 9.
Figure 8. The input/output of the fuzzy controller
E. Input/Output Fuzzy System We use 4 membership functions for both of the two inputs and 7 membership functions for the output which are sufficient to give a fast control solution. The 4 membership functions for the voltage violation input are: very negative (VN), negative (N), positive (P), and very positive (VP). These membership functions are shown in Fig. 3. On the other hand, the 4 membership functions of tranformer tap changers input are: very negative (VN), negative (N), positive (P), and very positive (VP) as shown in Fig. 4. While the 7 membership functions for the output of the controller are: negative 3 steps (N3), negative 2 steps (N2), negative 1 step (N1), no control action (Z), positive 1 step (P1), positive 2 steps (P2) and positive 3 steps (P3) as shown in Fig. 5. The complete rules of the fuzzy controller are given in Table 1.
Figure 9. Full AC load Flow of IEEE 14 bus system

1,20 1,15 1,10
X. CONCLUSION The reactive power control problem is a complex problem because due to the large scale and non-linear characteristics of power systems. To make the correct decision in order to get the system back to an acceptable operating state, we must get a control of reactive power and voltage. We have developed in this context Fuzzy Logic controller for VAR resources, transformer tap changer and PV pus voltage control in a power system under abnormal and/or contingency operation to maintain the voltage at all buses within acceptable limits, when minimizing the number of control actions. Two different variables are observed and used as two inputs to the proposed model; these inputs are (1) PQ bus that has the largest voltage violation and (2) controlling capacity of the controlling devices. The results show excellent performance of the proposed method, in regard to computation time and quality of results. REFERENCES
V Vmin Vmax
[1] [2]
g (p )
1,05 1,00 0,95 0,90 0,85 0,80 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
[3] [4]
Bus N
Figure 1.
TABLE II.
Voltage profile of IEEE 14 bus system
[5] [6] [7]
CONTROL ACTION OF FUZZY CONTROLLER OF IEEE 14 BUS SYSTEM
Control device
umin 0.8 0.8 0.8 0.8 0.8 0.9 0.9 0.9 0.9 0.9 0
u 0.932 0.978 0.969 1 1 1.06 1.045 1.01 1.07 1.09 19
umax 1.2 1.2 1.2 1.2 1.2 1.1 1.1 1.1 1.1 1.1 40
FL control action (u) 0 0.044 0 -0.1 0.06 0 0 0 0 0 0 [8]
a56 a47 a49 a78 a79 V1 V2 V3 V6 V8 Qc9 (Mvar)

1,20 1,15 1,10
[9] [10] [11] [12]
V Vmin Vmax
g (p )
1,05 1,00 0,95 0,90 0,85 0,80 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
[13] [14] [15] [16]
Bus N
Figure 2.
Voltage profile of IEEE 14 bus system after correction
G.Verbic and F. Gubina, A new concept of voltage collapse protection based on local phasors, IEEE Trans. on Power Delivery, vol. 19, no. 2, pp. 576-581, 2004. S. Maheshwarapu, New static- coefficients for reactive power planning in power systems, in 1998 Proc. TENCON Conf. IEEE Region 10 International Conference on Global Connectivity in Energy, Computer, Communication and Control, pp. 518521, 1998. N. Sinha, L. L. Lai and P. K. Ghosh, GA based algorithm for optimum allocation of reactive power under deregulated environment, in 2008 Proc. DRPT Conf , Nanjing China, pp, 926932, 2008. F. Capitanescu, Assessing reactive power reserves with respect to operating constraints and voltage stability, IEEE Trans. on Power Systems, vol. 26, pp. 22242234, 2011. T. Menezes, L.C. da Silva, V.F. da Costa, Dynamic VAR sources scheduling for improving voltage stability margin, IEEE Trans. on Power Systems, vol. 18, pp. 969971, 2003. S. Granville, Optimal reactive dispatch through interior point methods, IEEE Trans. on Power Systems, vol. 9, pp. 136146, 1994. E. Vaahedi, Y. Mansour, C. Fuchs, S. Granville, M. D. L. Latore, H. Hamadanizadeh, Dynamic security constrained optimal power flow/Var planning, IEEE Trans. on Power Systems, vol. 16, no. 1, pp. 3843, Feb. 2001. R. Storn, K. Price, Differential evolution - a simple and efficient adaptive scheme for global optimization over continuous spaces, Technical Report, International Computer Science Institute, Berkley, pp. 112, Mar. 1995. D. B. Fogel, Evolutionary Computation: Toward a new Philosophy in Machine Intelligence, IEEE Press, 1995. P. Subbaraj, P.N. Rajnarayanan, Hybrid particle swarm optimization based optimal reactive power dispatch, International Journal of Computer Applications, vol. 1, pp. 6570, 2010. K. Mahadevan, P.S. Kannan, Comprehensive learning particle swarm optimization for reactive power dispatch, Electric Power Systems Research, vol. 10, pp. 641652, 2010. P. Aruna Jeyanthy, D. Devaraj, Optimal reactive power dispatch for voltage stability enhancement using real coded genetic algorithm, International Journal of Computer and Electrical Engineering, vol. 2, pp. 734740, 2010. B. C. Arabacioglu, Using fuzzy inference system for architectural space analysis, Applied Soft Computing, vol. 10, no. 3, pp. 926937, 2010. L. Biacino, G. Gerla, Fuzzy logic, continuity and effectiveness, Archive for Mathematical Logic, vol. 41, no. 7, pp. 643667, 2002. M. E. El-hawary, Electric Power Applications of Fuzzy Systems, IEEE press, Jun 1998. Power Systems Test Case Archive, the department of Electrical Engineering, University of Washington, www.ee.washington.edu/research/pstca/
Extended Pruning-Grafting Operators for Efcient Distribution Network Reconguration

Saeed RamezanJamaat, Youhei Akimoto, Hernan Aguirre, and Kiyoshi Tanaka
Department of Electrical Engineering, Faculty of Engineering, Shinshu University, Nagano, Japan s.ramezanjamaat@iplab.shinshu-u.ac.jp, y akimoto@shinshu-u.ac.jp, ahernan@shinshu-u.ac.jp, ktanaka@shinshu-u.ac.jp
AbstractNetwork reconguration is a complicated combinatorial, non-differentiable, constrained optimization problem due to the incorporation of many candidate switching combinations. Introduction of evolutionary algorithms (EAs) for application to distribution network operation has opened many new opportunities. However, these methods suffer from high computational burden. Performance of the EAs is signicantly affected by modeling of the problem and employed operators. This paper employs a branch based modeling of distribution network and proposes two new EA operators which are an extension and redenition of PAO and CAO operators to be utilized for a more efcient application of EAs to the network reconguration problem. Performance of the new operators has been compared to original operators as well as an enhanced version of conventional crossover/mutation operators. Simulation results have proved efciency of the proposed method.
I. I NTRODUCTION Electrical distribution networks are always subject to continuous extension due to the development of society. This has led to complex spider-net urban distribution networks with many possibilities to be recongured. Reconguration means to alter the status of open or closed switches of the network. Switches could be located inside a feeder or between two different feeders. Inside-feeder reconguration is mainly performed in order to reduce loss or to balance loading inside the feeders. On the other hand, inter-feeder load transfer is commonly utilized in order to restore service in contingencies following a fault or to balance loading amongst the feeders. Thus, solving distribution systems operational problems such as loss reduction, load balancing, and service restoration highly depends on the performance of optimization plans. Several evolutionary algorithms (EAs) have been developed in order to deal with the reconguration problem [1]-[3]. Although they have presented encouraging results, the majority of them still demand high running time. Two essential factors affecting the solutions speed are modeling of the problem and employed operators. In order to improve the performance of EAs in the reconguration problem, Santos et al. [4] employ the concept of node-depth encoding and propose two operators called Preserve Ancestor Operator (PAO) and Change Ancestor Operator (CAO) which produce only radial congurations with low computational burden. However, their method still requires extra numerical analysis such as processing intermediate representations. In addition, the operators are limited to inter-feeder load transfer. In order to provide an integrated solution to distribution network reconguration, an
approach which could address both inside-feeder reconguration and inter-feeder load transfer simultaneously is required. In addition, it has to mitigate the computational burden problem, which is a discouraging feature for the application of conventional EAs and provide the solution in reasonable time. The main contribution of this paper is to introduce a branchbased tree modeling of distribution network and to propose an extension to PAO/CAO operators in order to enable them for handling inside-feeder reconguration of complex urban distribution networks in addition to inter-feeder load transfer, while retaining PAO/CAO operators merits. The new set is called PG (Pruning-Grafting) operators and includes EPAO (Extended PAO) and ECAO (Extended CAO). Loss reduction has been considered as the objective of optimization. In order to verify the effectiveness of this extension, original PAO/CAO operators are implemented and their performance is compared to PG operators. In addition, crossover/mutation operators are modied in order to enhance the conventional EA and their performance has been compared to PG and original PAO/CAO operators considering convergence speed, running time and the quality of results. This paper is organized as follows: Section II provides formulation of the problem, Section III presents the proposed method, Section IV discusses on implementing EAs using different operators, Section V presents simulation results, and Section VI concludes this work. II. P ROBLEM FORMULATION A mathematical formulation of the reconguration problem is introduced in this section. Theory of graphs with some adaptations to electricity network concepts has been employed for the modeling of distribution network. A. Graph modeling concepts A graph G is a pair (N (G), E (G)), where N (G) is a nite set of elements called vertices and E (G) is a nite set of elements called edges. A graph without loops is a tree. One of the tree vertices is usually named the root which is the node where the tree initiates from. More details on graph theory fundamentals could be found in [4]. Some adaptations of graph concepts to electricity network terms are: i) using term node instead of vertex, ii) using term branch instead of edge, iii) using term source instead of root, and iv) feeder concept which is a set formed by a branch directly connected to the source and all of its downstream branches.
B. Network modeling Single line diagram has been used for representation of the network. It has an acceptable accuracy for modeling of almost balanced MV distribution networks which is the scope of this work. The proposed method could be easily extended to a three-phase unbalanced network model. Two types of switches have been considered on branches: i) sectionalizing switch that is a normally closed switch, and ii) tie switch that is a normally open switch which has both end nodes as energized. Normally open switches with one or two de-energized end nodes isolate a part of the network and are out of the scope of the proposed method. Vicinity set dened for each sectionalizing switch is the set of tie switches that are directly connected to the receiving node of that sectionalizing switch. Two types of loops have been considered in this work: i) simple loop which is formed by branches of a single tree, and ii) extended loop which is formed by branches of two different trees assuming sources to be connected to an innite bus. A fundamental loop, that could be a simple or extended loop, is assigned to each tie switch and is the set of branches that connect both ends of the tie switch together or to the source(s). Therefore, each tie switch opens a fundamental loop. An advanced modeling of the loads has been employed [5]. It considers each load as a composition of three contributions: a) constant power, b) constant current, and c) constant impedance, using following formulation: P = P0 (a0 + a1 V + a2 V ) Q = Q0 (b0 + b1 V + b2 V 2 ) and a0 + a1 + a2 = 1 b0 + b1 + b2 = 1 where, (a0 , b0 ), (a1 , b1 ), and (a2 , b2 ) are coefcients related to constant power, constant current, and constant impedance contributions of the loads, respectively. Feasibility concept which would be frequently referred in this work is claried in order to avoid ambiguity. A feasible conguration is a tree (radial) structure which serves all nodes. Lavorato et al. [6] prove that a feasible conguration has to fulll two conditions: i) the number of open switches has to be equal to the number of fundamental loops, and ii) all nodes have to be served. According to [6], satisfaction of only rst condition is necessary but not enough for feasibility and both should meet, simultaneously. The number of fundamental loops could be calculated using the following equation: Nf l = Nbr Nno + Nsrc (2)
2
and voltage margins could be easily included in advanced applications of the proposed method. C. Network evaluation Each candidate conguration has to be evaluated using a load ow analysis. An efcient load ow method proposed in [7] has been employed which is a fast and robust branchbased sweep method capable of compensation for loops and PV nodes. Based on experiments, this method converges very fast, commonly in less than 3 iterations when no loops and PV nodes are involved. Santos el al. [4] claim that for EA application, merely a single iteration of sweep method is adequate which makes this load ow method even faster. In this work, sweep method runs until convergence. In addition, this method has potential of parallel processing such as calculation of nodal current injections for all nodes or performing the sweeps on all branches of the same layer, in parallel [7]. After running the load ow, parameters of the network such as currents of all branches and voltages of all nodes are identied. These data are used in order to compute the sending and receiving active powers of each branch. The difference between these powers is the active power loss in that branch. Total active power loss of the network is the summation of active power loss in all branches, as calculated using the following equation:
nb
Ptotal =
i=1
(PiS PiR )
(3)
(1)
where PiS and PiR are the sending and the receiving active powers in branch i, and nb is the total number of branches. This formula assumes lumped load modeling in which loads are compact energy sinks at nodes. This modeling inates the calculated power loss compared to a distributed load model. However, due to the focus of this research on MV distribution networks which commonly have lumped loads, this modeling still remains accurate. III. T HE PROPOSED METHOD A branch-based tree modeling has been employed which offers a natural representation of the actual power network and allows for the use of graph concepts in the optimization of its conguration. In this modeling, an individual is represented as a set of one or more trees and each tree is composed of some feeders sharing a same source. Status of switches and impedances and connection data of branches are stored as attributes of branches in the tree. In addition, information about transformers, loads, generators, and compensators are stored as attributes of sending or receiving nodes of branches. Some benets of this modeling are: i) an easy access through each branch to all its upstream and downstream branches which facilitates the application of this modeling to more sophisticated analyses of the network, and ii) providing an adaptive model with employed branch-based load ow method which is the evaluation core of the proposed method. PG operators which manipulate this representation are an extension and redenition of PAO/CAO operators proposed
where, Nf l , Nbr , Nno , and Nsrc are the number of fundamental loops, branches, nodes, and sources, respectively. Note that only engineering constraint is checked for feasibility in this step of research in that the main contribution has been to propose an efcient modeling and a set of operators which are fast and keep the search in the space of radial congurations. Operational constraints such as loading limits
by Santos et al. [4]. The extension has been performed by adopting the concept of fundamental loops. PAO/CAO operators are applicable only to inter-feeder load transfer while PG operators perform inside-feeder reconguration as well as inter-feeder load transfer which enables them for a wider application in the network operation. Therefore, the optimization process could nd an integrated solution to reconguration problem in a single run. In addition, PG operators maintain the merits of PAO/CAO for keeping the optimization process in the feasible search space and low computational burden. PG operators maintain the search in the feasible space by selecting closing/opening switches from same fundamental loops. Thus, both criteria for feasibility mentioned in subsection II-B would be fullled. In the reconguration problem, unfeasible search space could have a large extent as it would be demonstrated in the results. Consequently, checking the feasibility of new individuals and possible correctional actions are not necessary which accelerates the whole process. Details on the implementation of PG operators are presented here. 1) EPAO operator: this operator prunes a part of a feeder (thus creates a sub-feeder) and grafts it to a feeder. The pruned node remains the root of sub-feeder. In addition, the destination feeder could be the same as the source feeder, which is a new feature compared to PAO/CAO. EPAO needs two switches for implementation: pruning switch (p) and grafting switch (g ). By new denition, a p candidate is a sectionalizing switch which has at least one vicinity in the same fundamental loop. For each p, g candidates are its vicinities which are in the same fundamental loops with p. EPAOs sample p and g candidates are illustrated in Fig. 1(a). In this gure, bold and dashed lines represent sectionalizing and tie switches, respectively. Application of EPAO results in a minor change in the networks conguration because: i) by denition, p and g candidates are directly connected to each other, thus the sub-feeder is moved to a nearby location, and ii) since the root of sub-feeder does not change, direction of energy does not alter in any branch. 2) ECAO operator: this operator prunes a part of a feeder (thus creates a sub-feeder) and grafts it to a feeder, similar to the previous one. However for ECAO, the root of subfeeder changes before being grafted. To imagine, it is similar to spinning the sub-feeder before grafting. Again, the destination feeder could be the same as the source feeder which is a new feature compared to PAO/CAO. For ECAO, a p candidate is a sectionalizing switch which has a sectionalizing switch in its downstream that has at least one vicinity in the same fundamental loop with p. For each p, g candidates are those vicinities which are in the same fundamental loops with p. ECAOs sample p and g candidate selection is illustrated in Fig. 1(b). Application of ECAO results in a more substantial change in the networks conguration because: i) by denition, p and g candidates are located far from each other, thus the sub-feeder is moved to a distant location, and ii) since the root of sub-feeder changes, direction of energy alters in some branches that follows new requirements for network operation such as re-setting of directional protection relays.
(a) For EPAO Fig. 1.
(b) For ECAO
Selection of p and g candidates
In order to execute an EPAO or ECAO operation, a p is selected randomly from the list of networks p candidates and one of its g candidates is selected, randomly. Then, p is opened and g is closed. Enforcing the selection of vicinities from different feeders in [4] has limited the application of PAO/CAO operators to inter-feeder load transfer. However, inclusion of fundamental loops concept in this work has extended their application to include inside-feeder reconguration as well. In order to identify p and g candidates, a subroutine has been added to the network setup function which simply detects them due to its broad access to upstream and downstream of all branches. This knowledge cannot be easily obtained by the indirect representation used in [4], which is another advantage of the proposed modeling. IV. EA IMPLEMENTATIONS In order to evaluate performance of the proposed method, three EAs have been implemented using: i) PG operators, ii) enhanced conventional operators, and iii) original PAO/CAO operators. Details of implementations are as follows. A. EA using PG operators In this EA, one of PG operators is randomly selected and applied to an individual in order to make an offspring in each iteration. The population size of 5 individuals has been selected. Based on experiments, size of 1 allows faster convergence for PG operators converting them into climbers. However, climbers are prone to be trapped into local optima. In order to diminish the possibility of a premature convergence, population size has been increased to 5. Total active power loss in (3) has been considered as the objective function which has to be minimized through the network reconguration. Steps of the EA are briey discussed here. 1) Initialization: the current feasible conguration of the network is added to the population as the rst individual. 2) Parent selection: in each iteration, an individual is selected as parent using tournament selection method, where two randomly selected individuals compete considering their objective function values. Obviously, this selection method is applied only when population has more than one individual. Otherwise, the only individual would be the parent. 3) Variation: one of PG operators is selected randomly and applied to the parent in order to generate an offspring. Fixed and adaptive probability adjustment strategies could be used
in order to determine the probability of selecting one of PG operators. In the xed strategy, the probability is tuned and remains constant during the EA. While in the adaptive strategy, evolution process starts with an equal probability (of 50%) for both operators. Then, if one of them generates an offspring that survives, its probability increases and others decreases by a step (for example 1%) [4]. Both strategies have been examined in a tuning process of the parameters. 4) Survival selection: the created offspring is evaluated using the method discussed in subsection II-C. If the population is not full, it is added to the population without any comparison. However when it is full, if the new individual is better than the worst individual of the population, it replaces the worst. Otherwise, the new individual is discarded. Steps 2-4 are repeated until a predened number of evaluations are performed. Then, the best individual of the last iteration is introduced as the nal solution. B. EA using enhanced conventional operators In this EA, a binary representation is used in order to model the status of switches in which each 1 represents a closed and each 0 represents an open switch. Two conventional operators of crossover and mutation are enhanced in order to t the requirements of reconguration problem. The modication aims to maintain the number of open switches in the generated offsprings constant and equal to the number of fundamental loops in (2). Although this does not guarantee the feasibility of individuals, it keeps the search close to the feasible spaces borders since new individuals will have at least one feature in common with feasible ones: the number of open switches. Actually, the size of search space would be signicantly limited from 2n to Cm,n where, n is the total number of switches and m is the number of open switches. Experiments have proved the better performance of enhanced operators compared to standard one-point crossover and bit-ipping mutation which is not presented here due to the lack of space. A masked crossover has been employed which recombines open switches (zeros) of two parents and generate two offsprings by a probability Pc . Then, mutation is applied to all bits of each offspring by a probability Pm . In order to enhance these operators, a compensation routine is applied that maintains a constant number of zeros. However, compensation of crossover operation might lead to a more explorative crossover. It means that some offsprings could have genes which do not exist in any of the parents. To avoid this, rst, all bits of each offspring are set to 1. Then, bits of two parents are examined, equal bits are directly copied to both offsprings, and a mask is closed at these locations. Next, for each offspring, the rst parent is used in order to ll the open-mask locations of the offspring with a permutation and predened probability Pf pc . Finally, if the number of zeros in the offspring is not still enough, the second parent is used and lls the open-mask non-zero locations with a permutation but without considering any probability this time. For mutation operator, once each bit changes, compensation toggles a random different bit.
In practice, unfeasible congurations commonly have more attractive objective values compared to feasibles in that they cut a portion of the network or have loops which both entail less active power loss. However, the nal solution of evolution has to be feasible. In order to enforce the process to prefer feasible solutions as well as keeping a population of unfeasibles for avoiding a premature convergence, two subpopulations (SPs) have been used for feasible and unfeasible individuals with the size of 5 individuals for each SP. Newly generated offsprings try to enter the proper SPs and the nal solution is selected from the feasible SP. In addition, in order to favor less violating individuals in the unfeasible SP (less number of loops or unserved nodes), a penalty term is required. The objective function for individuals of the feasible SP is the total active power loss in (3). This is multiplied by a penalty term in (4) for those in the unfeasible SP. P enalty = Nun Nloops (4)
where, Nun is the number of unserved nodes and Nloops is the number of loops. Main steps of the EA are as follows. 1) Initialization: uses the current feasible conguration as the rst individual and applies mutation to this individual in order to generate a second individual. These individuals enter proper SPs. 2) Parent selection: in order to select each of two required parents, a SP is selected randomly and one of its individuals is selected by tournament method. 3) Recombination: the enhanced crossover operator is applied to parents in order to generate two offsprings. Then, the enhanced mutation is applied to each offspring. 4) Survival selection: one of the offsprings is selected randomly and if outperforms the worst individual of the related SP, replaces it. Otherwise, the new individual is discarded. The evaluation of both offsprings is avoided in order to maintain an equal number of evaluations compared to the implementation of other operators and to have a fair comparison. Steps 2-4 are repeated until a predened number of evaluations are performed and the best individual of the last iteration in the feasible SP is introduced as the nal solution. C. EA using PAO/CAO operators In order to illustrate the benets of the extension applied to original PAO/CAO operators, they are also implemented and applied to the same problem with the same population size, objective function, steps, and the number of evaluations as the algorithm in subsection IV-A. V. S IMULATION RESULTS The proposed method has been implemented using Visual C++ programming language and has run on an Intel Pentium IV 3.4GHz desktop PC. It has been applied to a Taiwan Power Companys network [8] with 94 buses, 96 branches and automatic switches on all branches, being operated in 11.4kV. It has 13 tie switches on branches number 84 to 96 including 12 inter-feeder and 1 inside-feeder tie switches (Case I). In
TABLE I T UNED SETTINGS OF PG, ENHANCED CONVENTIONAL , AND PAO/CAO
OPERATORS
540 PG Enhanced Conventional PAO/CAO
530
Operator PG (proposed)
Probability EPAO (xed) ECAO (xed) Crossover (Pc ) First parent crossover (Pf pc ) Mutation (Pm ) PAO (xed) CAO (xed)
Setting 80% 20% 40% 50% 1.04% 80% 20%

Active power loss (kW)
520
510
500
Enhanced conventional
490
PAO/CAO
480
470
TABLE II R ESULTS AT FINAL EVALUATION IN AVERAGE OF 30 RUNS Method PG (prop.) Enh. conv. PAO/CAO Case I Aver. SD 468.07 468.16 468.07 3.4e-13 0.166 3.4e-13 Case II Aver. SD 391.03 395.78 397.50 1.662 3.299 2.769 Case III Aver. SD 334.66 335.61 395.49 2.794 3.837 2.127
460 1 10 100 1000 10000
Number of evaluations
(a) Case I
PG Enhanced conventional PAO/CAO
540
520
500
480
addition, two variations of this test case have been developed by adding new tie lines in order to evaluate performance of the methods under different inside-feeder reconguration possibilities. For Case II, four tie lines are added between nodes 4-9, 19-23, 65-70, 87-92 called branches number 97 to 100, respectively. For Case III, seven tie lines are added between nodes 4-9, 19-26, 19-23, 55-59, 65-70, 75-80, 87-92 called branches number 97 to 103, respectively. Base power has been 5MVA. Proposed methods performance has been compared to enhanced conventional and PAO/CAO operators. Before commencing the experiments, a tuning process has been performed in order to optimize the settings of operators in three implementations. Tuned settings used for generating the results are presented in Table I. Mutation rate is adjusted to around 1/n where, n is the total number of switches. This gives chance to one bit in average to mutate. Tests have been performed using 10000 evaluations and 30 trials for different seeds of the random number generator. The average active power loss of the best individuals in each iteration are presented in Fig. 2 using logarithmic x-axis. The average and standard deviation of results at nal evaluation are summarized in Table II. In addition, lists of the nal open switches for three cases related to the best results of 30 runs, objective values, and frequency of their production are presented in Table III. For all cases in Fig. 2, slow convergence speed of enhanced conventional operators compared to PG and PAO/CAO operators could be clearly observed. Enhanced conventional operators do not converge even after 10000 evaluations while PG operators require 503, 805, and 497 evaluations and PAO/CAO operators require 328, 443, and 571 evaluations for convergence in Cases I, II, and III, respectively. As a reason for slow convergence rate of enhanced conventional operators, based on performed experiments, more than 79% of individuals produced by these operators have been unfeasible even though these operators include compensation. For Case III,
460
440
420
400
380 1 10 100 1000 10000
(b) Case II
550 PG Enhanced conventional PAO/CAO 500
450
400
350
300 1 10 100 1000 10000
(c) Case III Fig. 2. Performance of PG, enhanced conventional, and original PAO/CAO operators for network reconguration
it is interesting to note that results produced by enhanced conventional operators at nal evaluation are statistically better than PAO/CAO operators as it can be observed in Fig. 2(c) and Table II. This is because by denition, enhanced conventional operators have access to wider switching options including
TABLE III B EST RESULTS OF 30 RUNS Method PG (prop.) Enh. conv. PAO/CAO Case I Open switch Obj. 7, 13, 34, 39, 42, 55, 62, 72, 83, 86, 89, 90, 92 Case II Open switch 7, 8, 13, 19, 34, 39, 42, 61, 64, 72, 79, 84, 86, 89, 90, 91, 92 7, 8, 13, 19, 34, 39, 42, 63, 72, 79, 84, 86, 89, 90, 91, 92, 99 Case III Open switch 7, 8, 19, 21, 34, 39, 42, 52, 61, 69, 79, 84, 86, 87, 88, 89, 90, 91, 92, 101 7, 8, 13, 20, 21, 34, 39, 42, 61, 64, 72, 79, 84, 86, 89, 90, 92, 99, 100, 102
Fr. 30 20 30
Obj. 390.45 392.35
Fr. 23 1 1
Obj. 331.56 393.20
Fr. 8 2 2
468.07
inside-feeder reconguration as well as inter-feeder load transfer while PAO/CAO operators are limited to inter-feeder load transfer. On the other hand, PG operators always produce equal or better results compared to enhanced conventional operators. Note from Table III that the frequency of producing the best results in 30 runs is: i) 30, 23, and 8 times for PG operators, ii) 20, 1, and 2 times for enhanced conventional operators, and iii) 30, 1, and 2 times for PAO/CAO operators, in Cases I, II, and III, respectively. Furthermore, the average time taken to generate and evaluate the individuals produced by PG and enhanced conventional operators is analyzed and compared. Based on experiments, in average of 10000 evaluations, it takes 2.83 milliseconds for PG operators and 6.50 milliseconds for enhanced conventional operators that is enhanced conventional operators take about 2.3 times longer than PG operators. This is because every single individual produced by enhanced conventional operators needs to be checked for feasibility and possibly compensated. In addition, some individuals produced by these operators entail loops and consequently their evaluation by load ow method takes more time. On the other hand, PG operators always produce feasible individuals which could be evaluated faster. Thus, even for an equal number of evaluations, PG operators are computationally faster than enhanced conventional operators. Therefore, PG operators outperform enhanced conventional operators when it comes to convergence speed and computational time per evaluation which has been a discouraging feature of EAs in many applications including network reconguration. If the attention is drawn again to the quality of results, PG and PAO/CAO operators exhibit a very close behavior in Case I and both converge to a similar solution. This is expected since Case I has only one possibility (tie switch on branch number 95) for inside-feeder reconguration which could be accessible by inter-feeder load transfer operation after some alterations of the network conguration. In Case II with 4 new possibilities for inside-feeder reconguration, PG operators outperform PAO/CAO operators by producing better results. This is intensied in Case III with 7 new inside-feeder tie lines where PG operators provide signicantly better results than PAO/CAO operators. Therefore, considering the quality of results, PG operators provide better performance compared to PAO/CAO operators when applied to more densely tied urban distribution networks with many inside-feeder tie lines because of their broader access to networks switching operations. In addition, network reconguration has improved the nodal
voltage prole in all cases. The minimum voltage has increased from initial 0.93 pu to 0.95 pu, 0.95 pu, and 0.96 pu for Cases I, II, and III, respectively. VI. C ONCLUSION This paper proposed a new contribution to distribution network reconguration problem. A branch-based modeling scheme was employed and two extended operators called PG operators were developed in order to efciently obtain an integrated solution to switching operations problem in a single run of the optimization program. In order to evaluate the efciency of the proposed method, three EAs using PG, original PAO/CAO and an enhanced version of conventional crossover/mutation operators were implemented. Simulation results showed that PG operators having capability of both inside-feeder reconguration and inter-feeder load transfer outperform PAO/CAO operators when the network is more densely tied with many inside-feeder tie lines commonly seen in urban distribution networks. In addition, they outperformed enhanced conventional operators when it comes to convergence speed and computational time. As future work, the proposed modeling and operators would be applied to more extent optimization requirements of the power distribution network in order to integrate various aspects for smarter control of power system into a single framework. R EFERENCES
[1] K. Nara, A. Shiose, M. Kitagawa, and T. Ishihara, Implementation of genetic algorithm for distribution systems loss minimum reconguration, IEEE Trans. Power Systems, vol. 7, no. 3, pp. 1044-1051, Aug. 1992. [2] J. Z. Zhu, Optimal reconguration of electrical distribution network using the rened genetic algorithm, Electric Power Systems Research, vol. 2, pp. 37-42, 2002. [3] Y. Y. Hong and S. Y. Ho, Determination of network conguration considering multiobjective in distribution systems using genetic algorithms, IEEE Trans. Power Systems, vol. 20, no. 2, pp. 1062-1069, May 2005. [4] A. C. Santos, A. C. B. Delbem, J. B. A. London, and N. G. Bretas, Node-depth encoding and multiobjective evolutionary algorithm applied to large-scale distribution system reconguration, IEEE Trans. Power Systems, vol. 25, no. 3, pp. 1254-1265, Aug. 2010. [5] J. Liu, M. M. A. Salama, and R. R. Mansour, An efcient power ow algorithm for distribution systems with polynomial load, Inter. Journal of Electrical Engineering Education, vol. 39, no. 4, pp. 371-386, 2002. [6] M. Lavorato, J. F. Franco, M. J. Rider, and R. Romero, Imposing radiality constraints in distribution system optimization problems, IEEE Trans. Power Systems, vol. 27, no. 1, pp. 172-180, Feb. 2012. [7] C. S. Cheng and D. Shirmohammadi, A three-phase power ow method for real-time distribution system analysis, IEEE Trans. Power Systems, vol. 10, no. 2, pp. 671-679, May 1995. [8] C. T. Su and C. S. Lee, Network reconguration of distribution systems using improved mixed-integer hybrid differential evolution, IEEE Trans. Power Delivery, vol. 18, no. 3, pp. 1022-1027, Jul. 2003.
A Complementary Hybrid System for Electricity Generation Based on Solar and Wind Energy Taking into Account Local Consumption - Case Study
Ajla Merzic, Mustafa Music, Elma Redzic
Department for Strategic Development Public Enterprise Electric Utility of Bosnia and Herzegovina Sarajevo, Bosnia and Herzegovina a.lukac@elektroprivreda.ba, m.music@elektroprivreda.ba, e.turkovic@elektroprivreda.ba
AbstractIn this paper, the complementary nature of wind and solar energy has been considered, especially by analyzing output power variations from a photovoltaic power plant (PVPP), a wind power plant (WPP) and their combination in a hybrid system. For these purposes, an own model has been made, with calculations based on real wind and solar energy potential data. Two hybrid system configurations have been observed; the first where both generating facilities have equal installed capacities (i.e. 2 MW each) and the second where the installed capacity of the WPP is five times larger. Effects of implementing these hybrid power systems have been verified by carrying out a simulation study using a practical load demand profile, with a recorded maximum hourly value of 3.5 MW. The complementary nature is more expressed in the first hybrid system configuration, where a decrease in the output power variations from the hybrid system considered as a whole in comparison with variations when considering generating facilities individually, has been observed. But, this hybrid system configuration manages to meet hourly consumption in only 7.06% of time. The complementary effect is reduced in the second hybrid system configuration, and output power variations have been observed in a very wide range. However, this system manages to meet hourly consumer needs in 41% of the time. Neither one of the two cases can independently satisfy observed demand on hourly basis. Acknowledging the fact that in the second case there are periods of time when the generated electricity is up to 3.5 times higher than the hourly consumption, as well as the positive difference between generation and consumption on annual basis, an off grid system would be feasible with proper storage facilities. Keywords-hybrid system; solar energy; wind energy; complementary nature; power output variability; load demand
in that way provide higher quality and more reliable power to customers than a system based on a single resource. Because of this feature, hybrid energy systems have caught worldwide research attention [2], [3], [4], [5], [6]. In this paper, special attention is given to the utilization of solar and wind energy, since advances in wind and photovoltaic (PV) energy technologies have increased their use in hybrid wind/PV configurations. For the purpose of the study, which results are presented in this paper, an own model has been developed for converting the raw energy resource, namely solar and wind energy, into output power, i.e. generated electricity in considered hybrid energy systems. Analyzes are based on real, measured, one year data1, taken from an actual wind and solar data acquisition and monitoring system located on the territory of Bosnia and Herzegovina (B&H). The model took into account currently available technologies, all restrictions in conversion of wind and solar energy into electricity in wind power plants (WPP) and PV power plants (PVPP), as well as the available area at the location of interest. The model has been developed for two cases, namely: Case I: total installed capacity of 4 MW: 2 MW in a PVPP and 2 MW in a WPP Case II: total installed capacity of 12 MW: 2 MW in a PVPP and 10 MW in a WPP.
I.
INTRODUCTION
Global environmental concerns and the ever increasing need for energy, coupled with steady progress in renewable energy technologies, are opening new opportunities for utilization of renewable energy sources (RES) across the world [1]. There is need to utilize these resources through commercially viable technologies to generate power, since they offer sustainable energy alternatives with zero environmental pollution. Also, they can complement each other to some extent in certain configurations of hybrid energy systems and
Main objectives of the study were to examine the complementary nature of the two RES and to elaborate hourly output power variations from each of the generating facilities, as well as from the hybrid power system as a whole. Analyzes were extended by considering real load data from a consumption centre near the location of the potential hybrid system. A rough evaluation of effects of constructing a hybrid system has been provided in this paper. II. PROBLEM FORMULATION The use of renewable energy technologies to meet energy demand has been steadily increasing for the past few years. However, important drawbacks associated with renewable
1
Unfiltered data has been used in order to avoid subjectivity.
energy systems are their inability to guarantee reliability and their lean nature. In this paper attention is devoted to PVPP and WPP, as two of the fastest developing renewable technologies. Their complementary nature has been discussed. Possibilities of electrifying consumption centers remote from the power network, characterized by significant wind and solar energy potential and suitable for the implementation of appropriate configurations of hybrid systems, have been elaborated, in order to reduce high transmission and distribution costs of supplying electricity to remote areas. Analyzes were made on the basis of real measurement data of wind potential and solar radiation at a selected location in B&H for the time period 01.10.2011 - 30.09.2013 on site Medvedjak. The site has been chosen for this study, since it has been characterized by good solar and wind energy potential. This site is also located near a small village with a recorded maximum hourly load 3.5 MW. One of the main disadvantages of generating facilities using intermittent renewable energy is that the availability of these resources has variations on minutes and hourly time basis, as well as daily and seasonal fluctuations, which results in difficulties in regulating the output power to cope with the varying load demand [7], [8]. This is specially emphasized for wind energy and can be seen from Figure 1, which shows wind power density (WPD) [W/m2] and solar irradiation values [W/m2] in one year time period, based on hourly data. These generating facilities are also characterized by low conversion efficiency of raw potential into useful energy, i.e. electricity; special requirements for optimal utilization of these resources (available space, slope, orientation); high investment costs of available technologies; relatively short lifetime compared to conventional power facilities, etc.
Wind power density and solar irradiation hourly values
4500 4000 3500 3000
in accordance with IEC 61400-12 [9] and MEASNET recommendations [10]. The average annual 10 minute wind speed at 30 m height is 5.2 m/s. Insolation is a measure of solar radiation energy received on a given surface area and recorded during a given time, and it is calculated as integral of solar irradiance over the considered time period, equation (1): (1) and present initial and final time period of where calculation, respectively and presents measured value of solar irradiance. The average annual insolation measured at this site equals 1,740 kWh/m2. This location has been selected after a previously performed evaluation and analysis of measured data from ten different locations spread throughout B&H. Analysis and modelling, as well as the choice of an appropriate location for this type of study, were based on following criteria: available wind potential available solar potential available space consumption centre vicinity power network distance.
Since the measured wind speeds relate to 30 m and 10 m heights, an extrapolation to a height of 78 m has been done2. For these purposes, following logarithmic function (2) was used: (2) where z and z0 present heights above the ground at 78 m and 30 m, respectively; v(z) presents the calculated 10 minute average wind speed at 78 m height, v(z0) denotes the known 10 minute average wind speed at 30 m height and is the roughness length. Wind power calculated for the wind speed as given in (2) is defined by (3): (3) denotes air density, A the surface and c the degree of aerodynamic conversion. The theoretical maximum of c is cited as Betzs limit and equals 16/27 = 0.593. Thereby, the most efficient technologies in the application can achieve approx. up to 0.45. Considering the efficiency of converting solar radiation into electricity using polycrystalline PV panels, this value is lower, amounting up to 15%
[W/m2]
2500 2000 1500 1000 500 0
WPD
Time [h] Total
Solar irradiance
Figure 1. WPD and solar irradiation based on hourly values - one year data
Data used for the performed analysis and modelling are overtaken from an active 30 m high wind and solar data acquisition system, Medvedjak. The system is equipped with two first class anemometers, a wind vane, an air pressure, humidity and temperature sensor, as well as a pyranometer, all
Analyzes were extended by considering real load data from a consumption center near the location of the potential hybrid system. It is approx. 70 km away from a big consumption center, with which it is linked through a 110 kV transmission line. It is difficult to match variations of consumption with the volatile generation of facilities based on intermittent RES, thus
2 This height was selected as the height of the chosen wind turbine type, which was later on used for modeling of output power simulation of the WPP.
this problem has been given a special evalua ation further in this work. cy of output power For an insight into the range and frequenc variations from generating units based on n solar and wind energy, a model of the hybrid system connected to the grid was ere considered and developed. Following two configurations we their results were discussed: P Case I: 2 MW PVPP + 2 MW WPP Case II: 2 MW PVPP + 10 MW WP PP.
ystem, operating capabilities and the transmission/distribution sy power system service insuranc ce as regulation of active power and frequency as well as reac ctive power and voltage, power systems are not able to accept t unlimited amounts of installed power from PVPP and especially not from WPP.
Frequency of output power p variations based on hourly data 2 MW W PVPP + 2 MW WPP
45 40 35 30 25 20 15 10 5 0
Selection of these sizes is not usual fo or hybrid systems. This choice was driven by exploring possibil lities for electricity supply to consumers located near the potent tial hybrid facility, in order to achieve known positive effects on n distributive losses decrease, security and reliability of supply in ncrease, as well as exploitation otherwise unused area. r with the Visual The software tool WindPRO, together Basic platform have been used for the model l development and output power and energy yield calculations s. Simulations are done with respect to currently available te echnologies, space availability and other prevailing conditions at the considered ces air mass factor location. For the PVPP near shadings, indic (IAM), PV conversion factor, PV loss due to t irradiance level and temperature, array soiling loss, module quality q and module array mismatch loss, ohm wiring loss and in nverter losses were considered [11]. All this has been employed in i the Visual Basic model. The considered type of solar cells is polycrystalline, e optimal distance where special attention has been paid to the and configuration of the PV panels at the site. eld from the WPP Simulations of output power/energy yie were done for the wind turbine type Vesta as V 80.2.0, using WindPRO in combination with a Visual Ba asic program code. This type was chosen because of its widespre ead presence in the world market. The hybrid system configuration consi idered in Case II arose as a result of comparative analysis of o the total annual consumption and the possible annual prod duction of hybrid systems of different sizes. A. Simulation Results for Case I Simulation results of hourly output po ower variations of each of the generating facilities under th his hybrid system configuration are presented in Figure 2. cluded that hourly Analyzing gained results, it can be conc variations in output power of the assumed WPP W appear in the range from -90% of the nominal installed capacity (Pn) up to 80% of Pn. Only in 17.9% of time there are no variations between two adjacent values. In the case of the PVPP, the wer is narrower, i.e. range of hourly variations in the output pow 60% of Pn. In 44.9% of time there are no variations v between two adjacent values, in this case. In 90.2% % of time, hourly output power variations are in the range of f 20% Pn for the WPP and 95.2% for the assumed PVPP. ficant problems in This kind of variations, may pose signif managing and operating of a power system, especially in cases g facilities. Thus, with high penetration of such generating because of technical limitations, primary th he development of
[%]
[% %Pn] WPP output po ower variations [%] PVPP output power p variations [%] Total output power variations [%] Figure 2. Frequency of output powe er variations based on hourly values for Case I - on ne year data
B. Simulation Results for Case e II In Case II the installed cap pacity in WPP is increased five times, resulting in 10 MW, wh hile the installed capacity in the PVPP remained the same. Sim mulation results, which provide insight into annual hourly output power variations of the d on hourly values, are presented considered hybrid system based in Figure 3. The installed ca apacity ratio of the considered hybrid system is more common n in real conditions, but currently installed hybrid systems have much lower installed capacity (few hundred Watts).
Frequency of output pow wer generation variations based on hourly data 2 MW M PVPP + 10 MW WPP
45 40 35 30 25 20 15 10 5 0
[%]
[% %Pn] WPP output pow wer variations [%] PVPP output pow wer variations [%] Total output pow wer variations [%] Figure 3. Frequency of output powe er variations based on hourly values for Case II - on ne year data
Analyzing gained results, it can be conc cluded that hourly variations in output power of the assumed WPP W appear in the same range as in Case I, since the frequency of output power tential data. variations depend on the considered wind pot III. PROBLEM ANALYSIS, DISCUSSION AND SOLUTION Configurations of the two hybrid power systems s considered were chosen on the basis of maximal hourl ly load (3.5 MW), for the same time period, of the consumptio on center near the potential hybrid system site. A. Analysing Possibilities of Satisfying Loca al Consumption Needs Throughout Case I Although the configuration considered in Case I is not common, this system is reviewed in order to investigate the ources in a hybrid complementarity nature of the two energy so system with equal installed capacities, and to t examine output power variations reduction of the system as a whole, compared to variations of individual generating facilitie es. From Figure 2, significant reduction in the range of hou urly output power variations can be observed. They appear in the t range of 50% of Pn, whereby in 96.8% of the time, variations are in the range e are no variations of 20% Pn. Only in 8.8% of the time there between two adjacent values. Reduction of output power nd management of variations positively affects the operation an the power system, given that for the ter rtiary (as well as secondary) regulation lower amounts of reser rves are necessary. For further analyzes, consumption of a center near the t into account. location of the potential hybrid system was taken Electricity generation on hourly bases was w simulated and calculated from each of the individual gene erating facility, i.e. PVPP and WPP, as well as from the hybrid system s as a whole. Calculation results are given in Table I. % more energy than On annual basis, the WPP produces 65% the same sized PVPP. The annual difference between generation and consumption equals -12.1 GWh, which means that such a hybrid system cannot meet consum mer needs. In only 7.06% of the time, the hybrid system manages m to meet consumer needs on hourly basis. Difference between b electricity generation and the appropriate consumption n on monthly basis is always negative, which means that not even e in one of the considered months the consumption can be b satisfied. Data showing the percentage rate of time in which h the hybrid system can meet the hourly needs of the treated con nsumption are also very low. In June the highest generation n of the PVPP is n and consumption observed; the difference between generation is the lowest compared to other months, but b the generation meets the hourly consumption in only 10 0.6% of the time. During December the highest generation of the WPP is g matches observed but only in 7.2% of the time the generation the hourly consumption in the observed time period.
TABLE I. INDICATORS OF CONSUMPTION SAT TISFACTION - CASE I
Hybrid system [MWh] 507.48 273.79 Consumption [MWh] 1,672.37 1,651.09 Difference [MWh] -1,166.93 -1,379.23 Consumption unsatisfied [% of time] 95,17 99,86
Dec 11 Jan 12 Feb 12 Mar 12 Apr 12 May 12 Jun 12 Jul 12 Aug 12 Sep 12 TOTAL
79.96 93.90 132.22 270.30 250.88 342.24 447.59 431.74 426.83 251.47 3,022.1
569.56 397.33 350.53 415.45 515.63 343.58 393.78 373.88 337.40 437.06 4,620.5
649 9.51 491 1.23 477 7.11 691 1.39 766 6.50 685 5.82 841 1.37 805 5.62 764 4.23 688 8.54 7,64 42.6
1,724.18 1,521.14 1,578.58 1,712.05 1,628.12 1,645.81 1,656.93 1,657.89 1,724.20 1,618.05 19,790.4
-1,074.78 -1,032.03 -1,055.97 -1,064.72 -862.07 -953.72 -814.98 -853.01 -959.83 -924.16 -12141,4
92,75 93,02 95,07 91,28 91,67 93,02 89,44 88,84 93,55 91,11 92,94
p difference between Figure 4 indicates the percentual electricity generation and con nsumption in one year period, based on hourly data. It can be e seen that for the hybrid system in Case I for 29.9% of the tim me the electricity generation can satisfy only 10% of the hourly y consumption. Also, in 0.1% of the time, hourly generation exceeds the consumption by 60%. Hybrid system gen neration relative to consum mption
-90% -70% -50% -30% -10% 0 20% 40% 60% 80% 200% 350% 0 1 10 20 tage of time [%] Percent Case II Case I 30
Figure 4. Hybrid system electricity generation g relative to consumption - one year r data
At this point, it is important t to emphasize that this study did not consider storing of energy y excess, which would partially improve effects of the implem mentation of this hybrid system, but would not be able to meet lo ocal consumer needs. B. Analyzing Possibilities of Satisfying S Local Consumption throughout Case II p insight into frequency of Simulation results, which provide output power variations based on hourly values of the hybrid ions of individual generating system, compared to variati facilities are presented in Figur re 3. A reduction in the range of hourly output power variations s considering the hybrid system as a whole, instead of each gen nerating unit separate, is evident. Comparing results shown in Fig gure 3 with the ones in Figure 2, a reduction in the complem mentarity effect of these two generating units can be noticed d. This feature is a consequence of unequal power, as well as differences in the efficiency of the d availability of the considered two applied technologies and resources. In this case, output t power variations of the entire hybrid system range from -90 0% to 70% of Pn, dominantly caused by the WPP. For Ca ase II in 85.6% of the time, variations are in the range of 20% Pn. Only in 8.8% of the ween two adjacent values. time there are no variations betw
Month
PVPP [MWh] 167.23 127.74
WPP [MWh] 340.25 146.05
Oct 11 Nov 11
Generation/load [%]
In Table II calculation results on the difference between the hybrid system generation and the local consumption for Case II are presented. The electricity generation from the PVPP remained same, whereas the WPP has been increased five times, due to the increased installed capacity. Also, time when hourly consumption was not satisfied by this hybrid system configuration has been expressed for each month in percentage.
TABLE II.
Month Oct 11 Nov 11 Dec 11 Jan 12 Feb 12 Mar 12 Apr 12 May 12 Jun 12 Jul 12 Aug 12 Sep 12 TOTAL
INDICATORS OF CONSUMPTION SATISFACTION - CASE II

Consumption [MWh] 1,672.37 1,651.09 1,724.18 1,521.14 1,578.58 1,712.05 1,628.12 1,645.81 1,656.93 1,657.89 1,724.20 1,618.05 19,790.4 Diffe-rence [MWh] 196.12 -794.02 1,204.46 559.07 432.47 501.05 1,208.93 414.30 760.17 639.83 390.10 821.67 6,334.2 Consumption unsatisfied [% of time] 64.43 86.69 53.42 61.88 57.39 58.39 44.38 59.33 54.31 54.30 58.60 54.17 58.97
Hybrid system [MWh] 1,868.48 857.07 2,928.64 2,080.21 2,011.05 2,213.11 2,837.05 2,060.12 2,417.10 2,297.72 2,114.30 2,439.72 26,124.5
the performed analysis point out the complementary nature of solar and wind energy, as two intermittent renewable energy sources used for electricity generation. Due to equal amounts of installed capacities, the hybrid system configuration considered in Case I has a more positive effect on reducing the output power variation range, compared to the ones gained in Case II. The monthly difference between electricity generation and consumption is somewhat more favorable during summer months, especially in the period April 2012 - September 2012. This feature is attributed to favorable operating conditions for the PVPP and higher values of solar irradiation. The hybrid system configuration considered in Case II is characterized by the usual ratio of installed capacities of the two generating facilities, but in practice it usually refers to a micro facility with installed capacity of few hundreds of W, connected to a low-voltage distribution network. This configuration meets the hourly consumption in 41% of the time. The complementary nature of the two energy resources considered is less expressed, due to a 5 times larger installed capacity in the WPP, but, due to dominant electricity generation from this power plant, better results in satisfying local consumer needs are achieved. Independent electricity generation of any of the two hybrid systems considered could not meet local consumer needs. In respect to this, an alternative supply is necessary, either through the grid or by providing appropriate storage of a useful form of energy that could be converted into electricity, when needed. Energy storage would especially make sense in Case II, given that the annual difference between electricity generation and consumption is positive. However, the generated electricity is up to 3.5 times higher than the consumption for 8.4% of the time and storing of such amounts would require large overall dimensions, which, as with currently available technologies is not profitable. ACKNOWLEDGMENT Authors would like to thank Elektroprivreda B&H for the outsourced data. Appreciation to the anonymous reviewers for useful comments and suggestions. REFERENCES
[1] T.F. El-Shatter, M.N. Eskander, M.T. El-Hagry, Energy flow and management of a hybrid wind/PV/fuel cell generation system, Energy Conversion and Management, vol.47, no.9-10, ELSEVIER, pp. 1264 1280, 2006. C.Wang, M.H. Nehrir, Power management of a stand-alone wind/photovoltaic/fuel cell energy system, IEEE Transactions on Energy Conversion, vol.23, no.3, pp. 957-967, September 2008. P.Dalwadi, C.Mehta, Feasibility study of solar-wind hybrid power system, International Journal of Emerging Technology and Advanced Engineering, vol.2, no.3, pp. 125-128, 2012. H.Yang, L.Lu, W.Zhou, A novel optimization sizing model for hybrid solar-wind power generation system, ELSEVIER Solar Energy, vol.81, pp. 76-84, 2007. P.Nema, R.K.Nema, S.Rangnekar, A current and future state of art development of hybrid energy system using wind and PV-solar: A review, ELSEVIER Renewable and Sustainable Energy Reviews, vol.13, issue 8, pp. 2096-2103, 2009. Y.M.Atwa et al. Adequacy evaluation of distribution system including wind/solar DG during different modes of operation, IEEE Transactions on Power Systems, vol.26, issue 4, pp. 1945 - 1952, November 2012.
The absolute difference between generation and consumption equals 6.3 GWh, which means that the electricity generated in this hybrid system configuration would be sufficient to meet consumer needs on annual basis. In 41% of the time, this hybrid system configuration manages to meet consumer needs on hourly basis. Difference between electricity generation and the appropriate consumption on monthly basis is only negative in November, where the lowest electricity generation from the WPP is observed. The considered hybrid system meets the highest amount of hourly consumption in April (55.6% of the time), especially due to favorable wind potential at the site, but also good conditions for PVPP operation. From Figure 4 it can be seen that, for the hybrid system in Case II, for 16.6% of the time the electricity generation can satisfy only 10% of the hourly consumption. Also, there are occurrences when the hourly generation exceeds the consumption by 350%. This happens even in 8.4% of the time. In cases with this amount of energy excess, when the off-grid system is considered, it would be good to have some kind of storage. Energy storage would solve the problem of satisfying local power consumption. A detailed optimization with appropriate operation strategies has not been considered for the generating facilities in this paper. This would be a further step. Although, additional analyzes have shown that, in case of energy excess storage, a hybrid system consisting of a 2 MW PVPP and a 8 MW WPP could meet local consumer needs. In this case the capacity of energy storage would be lower. IV. CONCLUSION An important characteristic of generating facilities based on intermittent renewable resources considered through this research is their output power variability. In this paper, hourly variations are considered and a positive effect on reducing the range of output power variation in the case of a hybrid system, in comparison with variations when considering generating facilities individually, has been observed. This approach and
[2]
[3]
[4]
[5]
[6]
[7] A.Lukac, M.Music, S.Avdakovic, M.Rascic, Flexible generating portfolio as basis for high wind power plants penetration - Bosnia and Herzegovina case study, IEEE Xplore, 2011. A.Merzic, M.Music, M.Rascic, First aspect of conventional power system assessment for high wind power plants penetration, International Journal of Renewable Energy Development, vol.1, no.3, pp. 107-113, 2012. IEC Standard, IEC 61400-12-1, Wind turbines Part 12-1: Power performance measurements of electricity producing wind turbines, 2005. [10] MEASNET - Measuring Network of Wind Energy Institutes, Evaluation of site-specific wind conditions, Version 1, 2009. [11] . Salam, Meteorological parameters effects on solar energy power generation, WSEAS Transactions on Circuits and Systems, vol.9, iIssue 10, pp. 637-649, 2010.
[8]
[9]
Experimental Verications of Offshore Marine Substation for Grid-Connection of Wave Energy Farm
Rickard Ekstr om Senad Apelfr ojd Mats Leijon
Division of Electricity Division of Electricity Division of Electricity Department for Engineering Sciences Department for Engineering Sciences Department for Engineering Sciences Swedish Centre for Renewable Swedish Centre for Renewable Swedish Centre for Renewable Electric Energy Conversion Electric Energy Conversion Electric Energy Conversion Uppsala University Uppsala University Uppsala University Box 534, SE-751 21 Uppsala, Sweden E-mail: Rickard.Ekstrom@angstrom.uu.se
AbstractAn offshore marine substation has been designed and constructed for grid-connection of a wave energy farm. The substation will be deployed 2 km off the Swedish West coast, on the seabed at a depth of 25 m. Before the deployment, the substation electrical circuit has been tested in the laboratory. The functionality and total substation efciency have been evaluated at different voltage levels. A synchronous generator with variable speed control has been programmed to generate a similar voltage output as a single wave energy converter. The generator output has been connected to the substation and power transferred to the local grid at unity power factor. The impact on the total efciency by adding a DC/DC boost converter is also discussed. The conducted experiments verify the functionality of the marine substation before deployment.
Wave power farm
Power cable
Measurement cabin
I. I NTRODUCTION The worldwide energy demand is continuously increasing. To meet this in environmentally friendly ways, renewable and sustainable energy sources have to be used [1]. One of the potential energy resources is wave energy, with the global power potential estimated to be between 1-15 TW [2], and in Sweden the yearly wave energy potential is 515 TWh [3]. Many different wave energy converters (WECs) have been developed during the past decades, and may be divided into three different groups depending on their power absorption technique. These are the oscillating water column, the overtopping device and the point absorber. A WEC concept of the point absorber type has been developed at the Swedish Centre for Renewable Electric Energy Conversion, Uppsala University. The WEC consists of a oating buoy, connected to a linear generator put on the seabed. The buoy is directly connected to the translator of the generator with a steel wire. Potential energy in the waves is absorbed by the buoy and converted into electrical energy in the linear generator. The induced stator winding voltage varies in both magnitude and frequency, which requires an electrical conversion step to grid-connect the WEC. The concept is mechanically very robust with only one moving part and no
Fig. 1. Site location of the Lysekil wave energy farm, located outside the Swedish West coast. The power cable connects the farm to the onshore measurement cabin, and further to the electric grid.
gear box. There is only electrical damping in the system, which is controlled by the electrical conversion stage. Earlier studies and results from this research project include e.g. [4][6]. A wave energy farm of seven WECs has been constructed and deployed in an area 2 km off the Swedish West coast. A map of the site location is shown in Fig. 1. The WECs are of varying mechanical and electrical properties, further discussed in [7]. To transfer the converted wave power into the local electric grid, an offshore marine substation has been designed. In Fig. 2, an overview of the experimental research site is shown, with the generator and substation put on the seabed at a depth of 25 m. A power cable connects the output of the substation to the local 1 kV-grid onshore. There is also an observation tower for monitoring of buoy motion, as well as dummy buoys for biological studies of the impact of the wave energy farm on the environment. In this paper, the electrical circuit of the marine substation is described, and its functionality evaluated in the laboratory. The total efciency of the substation is measured for different voltage and power levels. A synchronous round-rotor
150m
12m Wave Wave
2km
Control cabin & grid connection
200m
2.9km
(a)
25m
(b)
Fig. 3. (a) Computer rendering of the substation hull. (b) The substation under construction in the workshop.
Fig. 2. Overview of the Lysekil research site. The offshore marine substation is put on the seabed within the wave energy farm. There is also an observation tower for monitoring of buoy motion.
at higher voltages are connected to the second. To connect the system to the same Point of Common Coupling (PCC), different transformer voltage ratios are used. The transformers also have different primary taps to further increase system exibility.
WECs Tap transformer
permanent magnet (PM) generator, with variable speed drive, has been controlled to behave like a linear generator in the WEC. The varying generator output is fed to the substation and transferred to the grid at unity power factor, while keeping the DC-level in the substation constant. II. E LECTRICAL DAMPING OF THE WEC S To optimize the delivered power from the WECs, the electrical damping strategy is most vital. A popular strategy used for point-absorbing WECs is the articial latching [8], where the aim is to align the buoy velocity with the wave excitation force. Correct active damping control may increase the power absorption by multiples. Several examples of active damping implementations may be found in e.g. [9][11]. Despite an increase in the power absorption, many of the proposed active control strategies result in more complex circuits and control systems with reduced reliability. As a simpler alternative, the strategy of constant DC-level is proposed in [12], and discussed in more detail in [13]. In this case, the WEC is passively rectied onto a DC-bus maintaining constant DC-level. The optimal value of the DC-level, with respect to power delivery, is a function of the buoy and generator characteristics as well as the sea state [13]. A great advantage of this is that WECs of similar properties may be connected in parallel to the same DC-link, simplifying the electrical circuit and enhancing the power smoothing. III. E LECTRICAL C IRCUIT L AYOUT The CAD rendering of the substation hull is shown in Fig. 3a, and the experimental set-up in Fig. 3b. The electrical devices are mounted with direct contact to the inside of the hull wall, and will be passively cooled by the sea water. In Fig. 4, the one-line diagram of the substation circuit layout is presented. Two parallel systems with the same topology are designed and mounted within the same hull. The systems are rated for different voltage levels. WECs rated at lower voltages are connected to the rst system while WECs rated
DC/DC/AC
LCL-filter
DC/DC/AC
Tap transformer
LCL-filter
1kV
Marine substation
Fig. 4.
One-line diagram of the electrical layout of the marine substation.
All WECs are passively rectied onto the constant DCbuses. The DC/DC-converter may be used to decrease the DCvoltage seen by the WECs further. The DC/DC-circuits can also be bypassed for better system efciency. There is also the possibility of actively varying the DC-level of an individual WEC using the DC/DC. The DC/DC-converter basic circuit diagram is shown in Fig. 5a, and the used ad-hoc inductor (L) in Fig. 5b. The ratio between the primary DC-voltage VDC 1 and the secondary DC-voltage VDC 2 during continuous conducting mode is ideally given by: VDC 2 1 = (1) VDC 1 1d where d is the duty cycle. A two-level voltage-source inverter (2L-VSI) is used to generate the sinusoidal voltage output, using sinusoidal pulsewidth modulation (SPWM). The inverter is directly connected to a step-up tap-transformer. The selection of tap sets the limitations on the DC-bus, and will be chosen to optimize the WEC output as well as the substation efciency. A lowpass LCL-lter is put on the transformer secondary side, to reduce the current ripple from the inverter. This is not the standard topology, and results in more stress on the transformer windings as well as higher transformer magnetization losses [14].
L Vdc1
Diode Vdc2
IGBT
(a)
(b)
Time [s]
(a)
Fig. 5. (a) The DC/DC boost converter circuit layout. (b) The ad-hoc DC/DCinductor.
Voltage![V]
However, this is considered acceptable at lower voltage levels, and will reduce both the losses and the ampacity requirements of the lter inductors. The two systems are synchronized with the grid voltage at the PCC, and power is injected to the grid at unity power factor. IV. S OFTWARE I MPLEMENTATIONS In order to handle all the system measurements in realtime, an extensive measurement system has been designed for the substation, thoroughly discussed in [15]. The software is developed using the compactRIO platform from National Instruments, containing both Field-Programmable Gate Array (FPGA) chips for more time-critical assignments, and realtime processors with heavier computational capacity. The main control functions include:

100 50 0 50 100 0 5 10 15 Time![s] 20 25 30
(b) Fig. 7. (a) Measured no-load output voltages for a WEC during sea trials. (b) The no-load voltage for the round-rotor generator with variable speed control.
Damping control of the WECs DC/DC control Inverter grid synchronization Power ow control and secondary DC-level control System monitoring and protection
V. G ENERATOR S PEED C ONTROL The synchronous round-rotor PM generator, shown in Fig. 6a, is described in [16]. It has been programmed with variable speed control to behave in a similar way as the WEC depicted in Fig. 6b. The incoming waves and excitation forces on the buoy are for simplicity assumed sinusoidal, and the complex buoy-generator dynamics are not accounted for. The sinusoidal behaviour is an adequate estimation for heave motion of the wave, while the downward motion is set predominantly by the gravitational force on the translator. Thus, a plateau is introduced in the downward motion to limit the speed. In Fig. 7a, the typical output voltages from a deployed WEC at noload are shown. Implementation of the above described speed prole during laboratory testing resulted in the no-load output voltage shown in Fig. 7b. Here, the wave time period was set to TE = 12s. The round-rotor generator has a nominal main voltage at 270 V and a nominal electrical frequency at 33.9 Hz. VI. E XPERIMENTAL S ET- UP The WECs deployed at the experimental site are rated for peak powers of 20 kW each. The power limitation of the substation electrical circuit, shown in Fig. 4, is rated for this with signicant overcapacity, and will be limited by the passive cooling capabilities of the hull. The transformers are rated for 80 kVA and 60 kVA respectively. The passive rectication bridges were built using SKKD 162/80 diodes. If
(a)
(b)
Fig. 6. (a) The PM round-rotor synchronous generator with variable speeddrive. (b) The WEC with a toroidally shaped buoy.
the DC/DC-converter is used, the primary DC-bus is 34 mF and the secondary 136 mF, whereas if the DC/DC-converter is bypassed these will be connected in parallel. The DC/DCconverter is made from three devices, a 3 mH ad-hoc inductor, a 600GA126D IGBT and a SKKE 301F diode. The three-phase inverter consists of six 400GB126D IGBTs with 2SC0108T2Ax-17 Concept driver boards. The characteristics for tap 2 on the 80 kVA transformer are displayed in Table. I. The LCL-lters are designed with two threephase 1.5 mH inductors and 40 F lter capacitors in Yconguration.
TABLE I T RANSFORMER CHARACTERISTICS Power rating Voltage ratio Rp Rs Lp Ls Lm (50 Hz, 1kV) Rm (50 Hz, 1kV) 80kVA 345V Y / 1kV Yn 0.0030535 pu 0.0096 pu 0.19858 pu 0.57303 pu 500 pu 89 pu
The DC/DC-converter efciency is measured and shown in Fig. 9. The duty cycle is set to 0.5, doubling the DCvoltage from 140 V to 280 V according to Eq. 1. As expected, the current-dependent losses increase faster than the input power, resulting in an efciency drop at higher power levels. The overall efciency improves for higher voltage levels. The utilization of a DC/DC-converter in the power circuit has to be evaluated together with the other losses of the system. Also, an increase in WEC power output may motivate the added losses of the DC/DC-converter.
0.96
0.94 DC/DC!efficiency
0.92
0.9
0.88
0.86 0
VII. R ESULTS AND D ISCUSSION Initially, the total system efciency was evaluated while bypassing the DC/DC-converter. The efciency depends on power and voltage levels as well as the selected tap on the transformer. In Fig. 8, three different cases are shown. In general, the overall efciency improves at higher power levels, as the no-load losses reduce relatively. At the lowest tap, two different DC-voltages, 150 V and 250 V respectively, were kept constant while increasing the power ow. It is evident that the lower DC-voltage produces a much better efciency. This is due to that the amplitude modulation of the inverter may be set at a higher value, reducing the harmonic content of the inverter current and thus increased transformer losses.
1 0.8 System!efficiency 0.6 0.4 0.2 0 0
8 PIN![kW]
10
12
14
16
Fig. 9.
DC/DC efciency for d=0.5.
Tap!1,!VDC=150V Tap!1,!VDC=250V Tap!2,!VDC=250V 2 4 6 PIN![kW] 8 10
Fig. 8. Substation efciency of the entire system from WEC to grid, when bypassing the DC/DC-converter. The efciency depends on selected DC-level as well as transformer tap ratio.
In Fig. 10a the phase voltage and phase current for the loaded generator is displayed, and the total three-phase power in Fig. 10b. In this case, the DC-level is too high for the generator to deliver any power in its downward motion. The generator was connected to the substation, and VDC kept constant. In Fig. 11a, VDC = 130V and TE = 6s, and in Fig. 11b VDC = 120V and TE = 12s. The system is able to handle both types of waves with small uctuations in VDC . In fact, this is a kind of worst case scenario, as the power uctuations will decrease when more units are connected in parallel. The difference in VDC between the two cases results in different damping of the generator, and thus different power outputs. This will be a very important research aspect of the deployed substation. Unity power factor is maintained with some uctuations in reactive power for the fast changes in active power. This is due to the LCL-lter, which enforces the coupling between active and reactive power. To improve the reactive power control further, cross-coupling terms must be included in the control system . However, for the current set-up it is considered satisfactory. In Fig. 12a, the grid current is shown during one wave cycle. Fig. 12b displays voltage and current at peak power for the wave cycle. VIII. C ONCLUSIONS A marine substation has been designed for grid-connection of a wave energy farm, and its electrical circuit veried in the lab. The system efciency is evaluated, and depends a lot on the selected DC-level and transformer ratio. The efciency
Selecting tap 2 with a VDC = 250V improves the system efciency further compared to tap 1. This is partially explained by the increased amplitude modulation, but also because the transformer magnetizing currents decrease for higher voltage levels.
150 120 90
VLN I
80 60 40
12 10 Input!power![kW] 8 6 4 2
60 VLN![V] 30 0 30 60 40 90 120 150 0 2 4 6 Time![s] 8 10 12 60 80 20 0 20 I![A]
0 0
6 Time![s]
10
12
(a) Fig. 10.
(b)
(a) Input phase voltage and current for the synchronous generator. (b) Input power uctuations.
12 10 8 Grid!power![kVA] 6
Pgrid 140 Q grid 120 V

DC
12 10 8
Pgrid 140 Q grid 120 V

DC
100 Vdc![V] 80 4 60 2 40 0 20 2 0 5 10 15 20 Time![s] 25 30 35 0 40
Grid!power![kVA]
100 80 60 40 20 10 20 30 40 Time![s] 50 60 70 0 80 Vdc![V] 6 4 2 0 2 0
(a) TE =6s Fig. 11.
(b) TE =12s
In (a), a shorter wave period is set. The DC-level is set to 130V. In (b), the DC-level is set to 120V, resulting in higher input peak powers.
20 15 10 Grid!current![A] 5
300 200 100 Vgrid![V] 0 100 200 300
20
grid
Igrid 15 10 5 0 5 10 15
1 2 3 4 5 6 7
0 5 10 15 20 0
Time![s]
2470
2480
2490
2500 Time![ms]
2510
2520
20 2530
(a) Grid current. Fig. 12.
(b) Grid current and grid voltage at peak power.
The uctuations in the grid current during one wave period while maintaining unity power factor.
of a DC/DC-converter is shown, and will be important when evaluating optimal WEC damping with respect to system conversion losses. A round-rotor generator with variable speed drive has been run to simulate the WEC behaviour. The power is successfully transferred to the local electric grid with maintained constant DC-level. There are some variations in the reactive power for large changes in the active power. This may be reduced by introducing cross-coupling terms in the control system between the active and reactive power. ACKNOWLEDGMENT This project is supported by Statkraft AS, KIC InnoEnergyCIPOWER, Fortum oy, The Swedish Energy Agency, Draka Cable AB, The Gothenburg Energy Research Foundation,
Falkenberg Energy AB, The Wallenius Foundation, Helukabel, ProEnviro, Seabased AB, The Olle Engkvist Foundation, The J. Gust. Richert Foundation, Angpannefreningens Foundation for Research and Development, CF Environmental Fund. The G oran Gustavsson Research Foundation, Varg ons Research Foundation. This support is gratefully acknowledged. R EFERENCES
[1] P. Droege, Urban Energy Transition - from Fossil Fuels to Renewable Power. Elsevier Science, 2008. [2] T. Thorpe, A brief review of wave energy. etsu report r-120 for the uk department of trade and industry, 1999. [3] A. Clement, P. McCullen, A. Falco, A. Fiorention, F. Gardner, K. Hammarlund, G. Lemonis, T. Lewis, K. Nielsen, S. Petroncini, M. Pontes, P. Schild, B. Sjstrm, H. Sorensen, and T. Thorpe, Wave energy in europe: current status and perspectives, Renewable and Sustainable Energy Reviews, vol. 6, pp. 405431, 2002.
Igrid![A]
[4] M. Leijon, H. Bernhoff, O. Agren, J. Isberg, J. Sundberg, M. Berg, K.-E. Karlsson, and A. Wolfbrandt, Multiphysics simulation of wave energy to electric energy conversion by permanent magnet linear generator, IEEE Transactions on Energy Conversion, vol. 20, no. 1, pp. 219224, March 2005. [5] O. Danielsson, M. Leijon, K. Thorburn, M. Eriksson, and H. Bernhoff, A direct drive wave energy converter simulations and experiments, in Proceedings of the 24th International Conference on Offshore Mechanics and Arctic Engineering, OMAE 2005, 2005. [6] M. Eriksson, R. Waters, O. Svensson, J. Isberg, and M. Leijon, Wave power absorption: Experiments in open sea and simulation, Journal of Applied Physics, vol. 102, p. 084910 (5 pages), 2007. [7] E. Lejerskog, H. Gravr akmo, A. Savin, E. Str omstedt, K. Haikonen, S. Tyrberg, R. Krishna, C. Bostr om, R. Ekstr om, M. Rahm, O. Svensson, J. Engstr om, B. Ekerg ard, A. Baudoin, V. Kurupath, L. Hai, W. Li, J. Sundberg, R. Waters, and M. Leijon, Lysekil research site, sweden: Status update, in Proceedings of the 9th European Wave and Tidal Energy Conference, EWTEC11, Southhampton, UK, 2011. [8] J. Falnes, Ocean Waves and oscillating systems, 1st ed. Cambridge, United Kingdom: The press syndicate of the University of Cambridge, 2002. [9] P. Ricci, J. Lopez, M. Santos, P. Ruiz-Minguela, J. Villate, F. Salcedo, and A. Falcao, Control strategies for a wave energy converter connected to a hydraulic power take-off, IET Renewable Power Generation, vol. 5, no. 3, pp. 234244, 2011. [10] H. Luan, O. Onar, and A. Khaligh, A dynamic model for control purposes of a wave energy power plant buoyancy system, in IEEE Applied Power Electronics Conference and Exposition, 2009, pp. 739 743. [11] D. Valerio, P. Beirao, and J. S. da Costa, Optimisation of wave energy extraction with the archimedes wave swing, Ocean Engineering, vol. 34, no. 17, pp. 23302344, 2007. [12] R. Ekstr om, V. Kurupath, C. Bostr om, O. Svensson, R. Waters, and M. Leijon, Evaluating constant DC-link operation of a wave-energy converter, in Proceedings of the ASME 2012 31st International Conference on Ocean, Offshore and Arctic Engineering, 2012. [13] V. Kurupath, R. Ekstr om, and M. Leijon, Optimal constant DC link voltage operation of a wave energy converter, Energies, vol. 4, pp. 19932006, 2013. [14] R. Ekstr om, S. Apelfr ojd, and M. Leijon, Transformer magnetization losses using a non-ltered voltage-source inverter, Advances in Power Electronics, vol. 2013, 2013. [15] R. Ekstr om, V. Kurupath, O. Svensson, and M. Leijon, Measurement system design and implementation for grid-connected marine substation, Renewable Energy, vol. 55, pp. 338346, 2013. [16] S. Eriksson, A. Solum, M. Leijon, and H. Bernhoff, Simulations and experiments on a 12kw direct driven PM synchronous generator for wind power, Renewable energy, vol. 33, no. 4, pp. 674681, 2008.
CAD of Electrical Machines Using Coupled FEMMMATLAB Softwares

Tariq Benamimour Ammar Bentounsi and Hind Djeghloud
Laboratory of Electrical Engineering of Constantine ,LGEC University of Constantine 1,Algeria Email:hinddjeghloud@yahoo.fr Laboratory of Electrical Engineering Laboratory of Electrical Engineering of Constantine ,LGEC of Constantine ,LGEC University of Constantine 1,Algeria University of Constantine 1,Algeria Email: tarekbenamimourelt@yahoo.fr Email:amar.trotek@gmail.com
Abstract This paper presents a software platform performed in the laboratory of EE, LGEC, at University Constantine 1, Algeria, dedicated to the CAD of electrical machines and based on an interactive coupling of finite element free software FEMM with MATLAB tool. The first step carried out under MATLAB is to predetermine analytically the geometrical parameters of a given machine from its specifications and to draw the structure by using FEMM software. The second step allows to solve by FEM the equations of the studied system and to determine its electromagnetic characteristics. This methodology was implemented for a 6/4 VRM prototype which characteristics will be used in look-up tables under MATLAB/Simulink for the dynamic study of performance in a non-linear mode. Keywords-CAD; electrical machine; FEMM; MATLAB; VRM.
I.
INTRODUCTION
The problematic of designing an electromechanical system is to find a structure corresponding to a given need and expressed in a specification. A design process often requires several software tools to study all phenomena characterizing the operation of a given system. Reduction of costs and deadlines being a priority for the industrial sector, many approaches were developed in this direction. These include software platforms implementing multiphysics models through the coupling of dedicated software tools. The work described in this paper focuses on developing a tool for the design and analysis of electromechanical converters. It is based on an original approach combining an analytical approach to pre-design according to specifications, performed in MATLAB software and coupled to a numerical finite element approach using FEMM software (Finite Element Modeling Method). Electromagnetic characteristics obtained by FEMM are either directly plotted under MATLAB or used in look-up tables for dynamic simulation of non-linear mode of a variable reluctance machine (VRM) in MATLAB/Simulink. The numerical results of the FEA were then validated against the analytical solution using equivalent magnetic circuits to seven equiflux tubes according to an approach proposed by Krishnan and that we have implemented in a MATLAB program called VRM.m. This CAD tool has been developed by a research team of the Laboratory of Electrical Engineering, LGEC, Faculty of Technology Science, University Constantine1, Algeria.
The idea of linking between several softwares to extract different characteristics is very interesting in order to reduce the time of analyzing and to get more exact results, but most of them are complicated and expensive sometimes because of the license agreement of the FEA used software, such as [1-5]. The originality of this paper shows how to link between MATLAB and open source software called FEMM [6]. As never done before in similar works using this software classically [7-9], here FEMM is controlled remotely with MATLAB because of the large using and good manipulating of this last software by the students. For that, the first step was to find several geometric parameters using a Matlab*.m file, taking in account the desired specifications such as power, speed and the envelope size [10]. After that as a second part, a sub-program always under MATLAB is calling automatically FEMM, which executes the orders given by the Matlab*.m file (draw, assign materials, excitations, boundaries, solve and analyze). When the analyze finished, the same Matlab*.m file recovered all needed data to plot any magnetic characteristic under MATLAB following the flowchart depicted in Fig.1.The geometric parameters of the studied 6/4 VRM are calculated with a sub-program under MATLAB, based on the approach developed in [11]. These parameters are used to model with FEM the considered prototype in order to extract its magnetic characteristics (flux, inductance...). From the obtained curves, the analytical study of the same machine presents results very close to the numerical one.The paper is organized as follows. Section II presents fundamentals of the studied variable reluctance machine. Section III explains the drawing process of the geometry. Section IV describes the interactive control of FEMM software from MATLAB. Section V discusses the obtained results. II. BASIC THEORY OF VRM
For illustrative purposes, we use the basic three-phase machine with Ns=6 stator poles and Nr=4 rotor poles, a 6/4 VRM is shown in Fig. 2. The concentrated coils around each stator pole form the phase windings, while there are no windings or magnets on the rotor. The principle of operating is based on the change of reluctance (and hence the inductance seen by the stator winding) due to a geometrical salience (or magnetic anisotropy in the case of sheeted rotor or axially laminated rotor) during the displacement of the rotor.
Start of the approximate design
Design specifications of the studied machine (Power, speed)
Under MATLAB
Torque
In the unaligned position corresponding to the maximum reluctance (or to the minimum inductance) if we apply a current in the stator winding, the torque will be produced by the tendency of the rotor to be aligned with the stator poles of the given phase according to the well known principle of minimum reluctance or maximum flux, as shown in Fig. 3. If the energizing is stopped, the rotor will continue its movement until it reaches the next unaligned position. We repeat the operation by supplying the next phase of stator and so on. A continuous torque can be generated by the contribution of all phases. For each phase, the electromagnetic torque equation is given by (1) While the average torque can be expressed as (2) where q is the number of phases, Nr is the number of rotor poles, Wa is the magnetic energy in the aligned position and Wu is the magnetic energy in the unaligned position. III. GEOMETRY DRAW PROCESS
Specific electric and magnetic loadings
Bore volume
Material coefficients
Selection of dimensions Number of turns of coils
Calling FEMM and drawing the geometry
Assigning materials, boundaries and excitations
Analyzing (for different rotor positions)
Under FEMM
The draw process is the most important step of our work seen that it contains all geometric parameters of the prototype. In this step we must identify the unit of measurement, the minimum meshing angle, the current frequencyetc. To get best results, the geometry should be drawn with a perfect accuracy under FEMM, so all points are depending on each other and on other geometric parameters with many mathematical expressions. The draw is based on projecting all points on a (x,y) plan and for that, we needed two new angles (angXr) and (angXs) expressed (in degrees) respectively by (3) and (4) according to Fig. 4. (3) (4) where:
Rsh R Ro Yr Ys e Bs Br Shaft radius Rotor bore radius Outer radius Rotor yoke thickness Stator yoke thickness Air gap Stator tooth angle Rotor tooth angle
Plot results
To MATLAB/Simulink lookup table bloc (for dynamic simulation)
Figure 1. Steps of linking between MATLAB and FEMM.
Figure 2. Diagram of the studied 6/4 VRM.
The final design of the machine under FEMM is shown in Fig. 5 and the geometric parameters are recorded in Table I. Because of the structural complexity of VRMs and highly saturated nature of the flux density, it is very difficult to accurately model the different characteristics analytically. Thus, a 2-D non-linear finite element model in FEMM must be used with a B(H) curve of the magnetic material illustrated in Fig. 6.
Such as in [12], the main steps of design for any selected geometry of VRM are the same. For that, the given results in Table 1 are obtained from the basic *.m file program under MATLAB software as shown in Fig.7.
TABLE I. Parameter Number of turns Rotor yoke thickness Stator yoke thickness Air-gap length Outer radius Rotor bore radius Shaft Radius PARAMETERS OF THE STUDIED 6/4 VRM Symbol N Yr Ys e Ro R Rsh Value 359 14.7 mm 14.8 mm 0.25 mm 114.2 mm 57.1 mm 20 mm Figure 6. B(H) curve of the used ferromagnetic material.
IV.
REMOTE CONTROL OF FEMM FROM MATLAB
FEMM offers a scripting property which enables to control the software remotely by using its inner script LUA or by using a MATLAB.m file. Both LUA scripts and the *.m file contain a list of its inner functions nouns used for calling and commanding FEMM. The m file includes commands used to perform the drawing, assigning materials and excitation, running the simulation, updating rotor position, data values recoveringetc. The sample algorithm given below can create a 2D model of a simple magnetic circuit, after the algorithm is written and compiled under MATLAB, FEMM starts automatically.
openfemm;%calling FEMM create(0);%select a magnetic problem L=10;%Depth of the problem mi_probdef(0,'millimeters','planar',1e8,L,30,0);%(problem definition,units..) main_maximize;%open in a window mi_addboundprop('boundry',0,0,0,0,0,0,0,0,0);%creat a new boundary'Dirichlet' mi_addnode(-100,-100); mi_addnode(200,-100); mi_addnode(200,200); mi_addnode(-100,200); mi_addsegment(-100,-100,200,100); mi_addsegment(200,-100,200,200); mi_addsegment(200,200,-100,200); mi_addsegment(-100,200,-100,-100); z=mi_selectsegment(0,-100); mi_setsegmentprop('boundary',1,1,1,1); z=mi_selectsegment(200,0); mi_setsegmentprop('boundary',1,1,1,1); z=mi_selectsegment(0,200); mi_setsegmentprop('boundary',1,1,1,1); z=mi_selectsegment(-100,0); mi_setsegmentprop('boundary',1,1,1,1) %exterior core mi_addnode(0,0); mi_addnode(100,0); mi_addnode(0,100); mi_addnode(100,100); mi_addnode(100,49.5); mi_addnode(100,50.5); mi_addsegment(0,0,100,0); mi_addsegment(0,0,0,100);
Figure 3. Creation of the electromagnetic torque.
Figure 4. Angles (angXr) and (angXs).
Figure 5. The plotted 6/4 VRM under FEMM.
mi_addsegment(100,0,100,49.5); mi_addsegment(100,100,0,100); mi_addsegment(100,50.5,100,100); %interior core mi_addnode(10,10); mi_addnode(90,10); mi_addnode(90,49.5); mi_addnode(90,50.5); mi_addnode(90,90); mi_addnode(10,90); mi_addsegment(10,10,90,10); mi_addsegment(90,10,90,49.5); mi_addsegment(90,50.5,90,90); mi_addsegment(90,90,10,90); mi_addsegment(10,90,10,10); mi_addsegment(90,49.5,100,49.5); mi_addsegment(90,50.5,100,50.5) %coils mi_drawrectangle(10,20,20,80); mi_drawrectangle(-10,20,0,80); i=10; %excitation mi_addcircprop('current',i,1); %materials mi_getmaterial('Copper');
mi_getmaterial('Air'); mi_getmaterial('US Steel Type 2-S 0.018 inch thickness'); mi_addblocklabel(15,50); mi_selectlabel(15,50); mi_setblockprop('Copper',1,30,'current',0,1,24/2); mi_clearselected; mi_addblocklabel(-5,50); mi_selectlabel(5,50); mi_setblockprop('Copper',1,30,'current',0,1,-24/2); mi_clearselected; mi_addblocklabel(50,50); mi_selectlabel(50,50); mi_setblockprop('Air',1,30,0,0,1,0); mi_clearselected; mi_addblocklabel(50,5); mi_selectlabel(50,5); mi_setblockprop('US Steel Type 2-S 0.018 inch thickness',1,30,0,0,1,0); mi_clearselected; %view and analyze mi_zoomnatural(); mi_saveas('c:\\femm42\\sample.fem'); mi_analyze(0);%analze mi_loadsolution();%get results %end
Figure 7. Screenshots of FEMM window after the compilation of the above algorithm.
V. VRM SIMULATIONS UNDER FEMM

The main program of the VRM is more complicated and contains almost six subprograms, each one of them calculates many specialized variables, as an example, the first subprogram initializes several geometric coefficients and parameters. The last subprogram is the most important one, it contains almost 320 lines of command and it calls all of the other subprograms, for that different VRM configurations can be optimized using its plan of calculating not only the 6/4 VRM topology. The rotation step of any iteration had selected 5 as shown in Fig. 8. So when FEMM starts automatically, draws the
geometry, assigns different materials, defines several boundaries, meshes (Fig. 9) and analyzes the problem for the current rotor position, the torque value and other desired variables are saved in a separated matrix in MATLAB. The next stage is refreshing the rotor position with the above selected step (5). The material labels must be separated in any loop, or must be fixed with its bloc. The value of the current can be modified in any loop of simulation and the using of all library materials is possible too; it aims to realize a material optimization as easily as possible. When the FEA simulation is over, FEMM is closed automatically and the simulation results are plotted under MATLAB. The obtained results of torque, inductance and flux
vs. stator excitation at different rotor positions are shown in Fig. 10, Fig. 11 and Fig. 12, respectively. To validate the results, we have represented the analytical flux-current characteristics corresponding to the two extreme rotor positions (aligned and unaligned) performed under MATLAB subprogram depicted in Fig 13 on the same graph of FEMM numerical results, as shown in Fig. 14. We note a small difference between the curves of the two approaches. For example, we needed the two look-up tables (for torque and flux) to simulate one phase of a VRM under MATLAB/Simulink as shown in Fig.15.
flux vs position 1 0.8 10A 8A 6A 4A
flux[Wb]
0.6
0.4
0.2
0 0
10
20
30
40 50 position []
60
70
80
90
Figure 12. Inductance vs. rotor position for different stator excitations.
inductance vs position
0.2 0.18 0.16 0.14 10A 8A 6A 4A
inductance[H]
0.12 0.1 0.08 0.06 0.04 0.02 0
10
20
30
40
50
60
70
80
90
position []
Figure 9. Screenshot for FEMM analysis in the 2nd rotor position (+5)
Figure 13. Flux vs. rotor position for different stator excitations.
Start with Previously parameters
Initial value of stator mmf (F0=ni) or pole flux density (Bs0)
Figure 10. Meshing under FEMM in the 3th rotor position (+10)
torque vs position 20 15 10
torque[N*m]
For path 1 calculate The flux density in the different machine parts (Bk) Corresponding fields (Hk) and permeability (k) from B-H Mean lengths of fields circulation (lk) Reluctance (R1 = Rk) Total mmf : F1=R1*0
10A 8A 6A 4A
yes Bs=BsBs
F0-F1 > ?
no
5 0 -5
From the final Bsk compute the inductance of flux path 1
-10 -15 -20 0 10 20 30 40 50 position [] 60 70 80 90
Likewise compute inductance contributed by the other paths and sum them : Lu = Lk
Stop Figure 14. Flowchart of sub-program for the analytic calculation.
Figure 11. Torque vs. rotor position for different stator excitations.
flux linkages vs excitation 1.4
REFERENCES
[1] B. Bilgin and M. Krishnamurthy, An FEA/MATLAB Based Machine Design Tool for Switched Reluctance Motors, in Proc. IEEE Vehicle Power and Propulsion Conference, 6-9 Sept. 2011, pp. 1-6. [2] Z. Tang, P. Pillay, Y. Chen and A.M. Omekanda, Prediction of Electromagnetic Forces and Vibrations in VRMs Operating at Steady State and Transient Speeds, IEEE Trans. On Industry Applications Conference, Vol.41, No.4, 2005, pp. 927-934. [3] M. Leroy, Etude et mise au point de motoventilateurs hautes performances pour laronautique,(study and control of motoventilators in high performances for the aerospace), doctorate thesis, UMR CNRS N5828, 2006. [4] A. Matveev, P.J. Van Duijsen , Novel Caspoc-Based Software For Multilevel Simulation Of Switched Reluctance Drives, in Proc. of PCIM, Nuremberg, Germany, May 20-22, 2003. [5] Z. Cheng, D. Liang, Simulation of Switched Reluctance Starter/Generator System Based on Simplorer, in Proc. of the 8th Int. Conf. on Electrical Machines and Systems (ICEMS), 2005,Vol.1, pp.564-567. [6] D. Meeker, FEMM42.exe software. http://www.femm.info. [7] M.B Rego, L.C. Gomes and al, Survey of Inductance Curves in Switched Reluctance Machine Using Finite Elements, International Conference on Renewable Energies and Power Quality (ICREPQ12), Spain, March 2012. [8] D. Uygun, C. Akuner, M. Yumutraci, A New Construction of Switched Reluctance Motor Improving Magnetic Field and Torque Production, in Proceding of the 6th WSEAS, Istanbul, Turkey, 2006. [9] M. Kowol, P. Mynarek, D. Mrochen, The Electromagnetic Field Calculation of the Switched Reluctance Motor, 2nd Int. Students Conference on Electrodynamic and Mechatronic (SCE1109), 19-21 May 2001, pp.13-14. [10] R. Krishnan, R. Arumugam and J.F. Lindsay, Design Procedure for Switched Reluctance Motor, IEEE Transactions on Industry Applications, Vol.24, No. 3, pp. 456-461, May-June 1988. [11] R. Krishnan, Switched Reluctance Motor Drives: Modeling, Simulation, Analysis, Design, and Applications, Industrial Electronics Series 2001, CRC Press. [12] M.N. Anwar, I. Husain and V. Radun, A Comprehensive Design Methodology for Switched Reluctance Machine, IEEE Transactions on Industry Application, Vol. 37, No. 6, pp. 1684-1692, Nov./Dec. 2001.
1.2
Flux linkages [Wb]
0.8 0.6 0.4 0.2 0 0 1 2 3 4 5 current [A] 6 7 8 9 10
Converted energy (Wa-Wu)
Analytical results Numerical results
Figure 15. Comparison between half-numerical and numerical results.
Figure 16. Insertion of the look-up tables blocs in MATLAB/Simulink.
III.
CONCLUSIONS AND PROSPECTS
The study developed in this paper focused around linking two softwares for finding best simulation results. A hybrid methodology was made from the design to the FEA, ending with a non-linear simulation of the studied prototype. The proposed work allows to economize time of initializing design process of any structure of electrical machines under FEMM environment, to command FEMM remotely with MATLAB, to design and extract electromagnetic characteristics perfectly by using simple tools. In the future works, we will try to realize an optimization study and add other subprograms which may make it easy to estimate losses, efficiencyetc.
Transient stability constrained optimal power flow solution of ac-dc systems using genetic algorithm
Ula Kl
Assistant Professor, Celal Bayar University Faculty of Hasan Ferdi Turgutlu Technology Manisa, Turkey ulas.kilic@cbu.edu.tr
Abstract Optimal power flow (OPF) is one of the known problems of the power systems. Many numerical and heuristic methods were used to solve this problem so far. As seen from these studies in literature, heuristic methods are more effective and faster than numerical methods. This case is to make more attractive and mandatory the using of heuristic methods in optimal power flow solution of High Voltage Direct Current (HVDC) systems. In this study, transient stability constrained optimal power flow solution of alternating current-direct current (AC-DC) systems is accomplished by using the genetic algorithm (GA) that is one of the heuristic methods for the first time. The proposed approach is tested on modified New England 39-bus test system. Keywords-Optimal power flow; HVDC system; transient stability; genetic algorithm
Krat Ayan
Assistant Professor, Sakarya University Faculty of Computer and Information Sciences Sakarya, Turkey kayan@sakarya.edu.tr process in GA approach. Therefore GA approach is a robust method. GA approach was only applied to OPF and TSCOPF problems of purely AC power systems until today [11-13]. The purpose of this paper is to solve the TSCOPF in ac-dc systems using GA approach. GA is essentially search algorithm that is based on natural genetics. It evaluates the optimal solution with randomized, structured exchanges of information among exist solutions. The constraints on solution space are not implemented during the process in GA approach. Therefore GA approach is a robust method. The power of GA comes from its ability to exploit historical information from previous solution guesses to increase performance of future solutions [14]. OPF solution of AC-DC systems is accomplished by using GA and the proposed approach is tested on two modified ACDC test systems [15]. In this study, the transient stability constrained optimal power flow (TSCOPF) solution of ACDC systems is also accomplished by using GA for the first time. The proposed approach is also tested on modified New England 39-bus test system. In order to show effectiveness and efficiency of the proposed approach on TSCOPF solution of AC-DC systems, the obtained results are given and the relative rotor angles of the generators during the faults are shown. II. THE MODELING OF DC TRANSMISSION LINK
I.
INTRODUCTION
Many studies were performed for the stability and power flow solutions of HVDC systems so far [1-3]. All of them use the numerical methods. There are two basic approaches for solving the power flow equations of AC-DC power systems in literature. The first is the sequential approach [4-5]. In this method, the alternating current (AC) and direct current (DC) equations are solved separately by successive iterations. Although the implementation of the sequential method is easy, it has convergence problems associated with certain situations and are the state vector doesnt contain explicitly the dc variables. The second approach is known as the unified approach [6]. The scientists have used many different methods for solving OPF problem and the transient stability constraint optimal power flow (TSCOPF) of purely AC power systems so far [7-10]. These methods are numerical and heuristic methods. According to the results reported in literature, it can be seen that heuristic methods are superior from the numerical methods [9-10]. The important one of the advantages of heuristic methods is that they convergence to the optimum solution in more short time than others and convergence fewer local minimum. GA is essentially search algorithm that is based on natural genetics. It evaluates the optimal solution with randomized, structured exchanges of information among exist solutions. The constraints on solution space are not implemented during the
Before analyzing DC transmission system, it is necessary to model DC transmission link and the converters. The modeling is based on the accepted assumptions in the literature [6]. Furthermore, a representation of an AC bus having the generators, AC lines, shunt compensators and converters can be represented by Fig. 1 [4]. The active and reactive power equalities at such a bus are given by Eqs. (1) and (2).
Pgk = Plk + Pdck + Pk

Q gk + Q sk = Qlk + Qdck + Q k
(1) (2)
Vk k
Pk + jQk
I dck
Pgk + jQ gk
Qck
Pdck + jQdck
k
Vk k
Pdck
tdck
V dck
Qdck
I k k
Plk , Q lk
Figure 1. General AC bus representation [4]
Figure 2. Equivalent Representation of a Converter Terminal [4]
Xc
I dck
In Fig. 1 P , Q , V and represent active power outputs, reactive powers, bus voltage magnitudes and bus voltage angles, respectively. The subscripts g , l , s and dc also represent generator, load, shunt reactive compensator and DC link, respectively.
3 2
Vk tdck cos( k )
Vdck
Pk = Vk Qk = V k
V (G
j j =1 N
kj
cos kj + Bkj sin kj sin kj B kj cos kj
)
)
(3)
Figure 3. Converter Circuit Model[4]
V (G
j j =1
kj
(4)
where V j and Vk are the voltage magnitudes of j th and k th buses; G kj and B kj are the conductance and susceptance between buses k and j of the bus admittance matrix ( Ybus ); kj is the angular difference between buses k and j , N is a number of all buses of the power system.
A. DC system equations Direct voltage of m-pulse rectifier in a monopolar HVDC system can be expressed as follows [16]: Vdck = nmtdck
The expression of V dcok = 3 2V k t dck / in the first term of the Eq. (6) is defined as the open circuit direct voltage. Thus, for the actual quantities of HVDC system, the direct voltage, the phase shifted, the active power, and the reactive power expressions can be given by equalities, respectively.
Vdck = Vdcok cos k
Vdck = Vdcok cos k
X ck I dck
X ck I dck
(7) (8)
k = cos 1
V dck V dcok
(9) (10) (11)
2Vk
nmX ck sin cos k I dck 2 m
(5)
Pdck = V dck I dck
where n is the converter number in series, m is peak number on load voltage per period, t dck is the tap ratio of the transformer at dc side, Vk is the effective secondary voltage of the transformer, k is delay angle for the rectifier operation (or extinction advance angle for inverter operation), X ck is the commutation reactance, I dck is the dc current. The equivalent representation of a converter terminal and converter circuit model is given in Figs. 2 and 3, respectively. Rearranging the equation (5) for m = 6 and n = 1 ,
V dck = 3 2
Qdck = Pdck tan k

where
represents phase shifted.
B. Per unit transformation After defining the base active power PACbase and the base voltage V ACbase of ac side, the base current I ACbase and base impedance Z ACbase for ac side can be determined as follows: PACbase = PDCbase I ACbase = PACbase 3V ACbase (12) (13)
V k t dck cos k
X ck I dck
(6)
Z ACbase =
V ACbase 3 I ACbase
(14) III.
q dck = p dck tan k

TRANSIENT STABILITY OF AC-DC SYSTEMS
(29)
After defining the base active power PDCbase , the expressions of V DCbase , I DCbase , and Z DCbase for dc side can be
determined by defining C b = 2 nm sin / as follows: m
V DCbase = C bV ACbase
I DCbase 3 I ACbase = Cb
(15) (16) (17)
In this study, the state space model using generator angle as reference is used. Therefore, the generator of the slack bus is selected as the reference frame and the rotor angles of the rest generators are expressed with respect to the reference frame. The relative rotor angles can be expressed as follows:
i = i 0
M i i = 0 Pmi Pgi Dii
i = 1,..., N g
(30)
2 Z DCbase = C b Z ACbase
where i is the rotor angle deviation of i the rotor speed deviation of i rotor speed of i
th th
th
generator; i is
where V DCbase , I DCbase and Z DCbase show the base active power, the base direct voltage, the base direct current, and the base impedance of dc side, respectively. The expressions in per unit by dividing the actual ac and dc expressions to ac and dc base expressions are given by the following equalities:
vk = Vk V ACbase
generator; i and 0 are the
generator and rated rotor speed; M i is the
moment of inertia of i th generator, Di is the damping constant of i th generator; Pmi and Pgi are the mechanical input power and electrical output power of i expressed as follows:
' Pgi = E i2 G ii +
th
(18) (19) (20)
generator. Pgi can be also
x ck =
rck =
X ck Z ACbase
[E E
i j =1 j i
Ng
' j Bij
' sin( i j ) + E i E j G ij cos( i j )
(31)
nmX ck / 2 nmX ck / 2 x ck = 2 = Z DCbase C b Z ACbase 4nm sin 2 ( / m )
where E i and E j are the constant voltages behind a transient reactance of the i th and j th generators; i and j are the rotor angles of i th
' ' and j th generators; Gij and Bij are
v dck = i dck =
Vdck V DCbase I dck I DCbase
(21) (22)
transfer conductance and susceptance between buses i and j

' is self conductance of bus i of the of the reduced Ybus ; Gii reduced Ybus . The reduced Ybus before, during and after a contingency can be obtained by eliminating all buses except for generator buses.
where v k , x ck , rck , v dck , and idck show ac voltage, commutation reactance, commutation resistance, dc voltage and dc current in per unit, respectively. According to these, the others are determined as follows:
V dck V DCbase nmt dck = 2V k
sin cos k m
V DCbase
nmX ck I dck 2 V DCbase
(23)
Vdck VDCbase
nmtdck =
nmX ck sin cos k I dck m 2 Z DCbase I DCbase 2 nm sin VACbase / m 2Vk
(24)
In this study, the state space model using generator angle as reference is used. Therefore, the generator of the slack bus is selected as the reference frame and the rotor angles of the rest generators are expressed with respect to the reference frame. Hence, the rotor angles of the generators except for slack bus are calculated by using modified Euler method and relative rotor angles can be expressed as follows: i = 1,.., N g 1 (32) irelative = i +1 1
) )
Hence the relative rotor angle irelative is restricted by upper (25) (26) (27) (28) limit max as follows;
v dck = v k t dck cos k rck i dck p dck = v dck i dck v dcok = v k t dck
irelative
max
max
i = 1,.., N g 1
(33)
k = cos 1 (v dck / v dcok )
where max is the maximum allowable relative angle and its value is commonly based on experiences. max is determined by trial and error for test system in this study. These values are different for each system in the literature.
IV. ILLUSTRATION OF GA GA is heuristic algorithm based on natural selection. GAs were firstly utilized by Holland in 1975 for solving optimization problems [17]. The base logic of the algorithm is that gens of powerful individuals are based to be carried over next generation and others are based to be detached in next generation. In natural selection, a human born, grows, and dies. These stages of human life correspond to the different operators in the algorithm. GA operators related these stages can be explained as the following.
A. Initial population Random individuals as initial population size within their limits. The fitness values of individuals within population are obtained to be put in objective function the formed individuals. The individual number within initial population is randomly determined as follows:
D. Crossover New individuals same as the determined number are produced to be used the crossing method with the scattered parameter from parents selected via the tournament method explained in selection stage. Gen number of individual are randomly produced within values between 1 and 0. If the value is 1, then gen is taken from mother, the value is 0, then gen is taken from father and thus the child is produced. Mother: a b c d e Cross: 1 0 1 1 0 Father: x y z u w Child: a y c d w
wij = wmin, j + rand (0,1) wmax, j wmin, j
(34)
where the parameters wmin, j and wmax, j show the minimum and maximum of the variable w j .
E. Mutation The number of individual undergo mutation has to be determined in the beginning of the algorithm. The individuals undergo mutation are reproduced to be formed all the gens of the selected individuals within algorithm. Thus, new individuals same as the number determined by Eq. (34) are randomly produced. F. Stopping algorithm In this study, iteration number is preferred as the stopping criterion. More information related to GA operators is available in Ref. [18]. Final population is formed to be included the reproduced individuals in stages above to initial population. After the individuals within final population are classified according to fitness value, the individual same as initial population is carried over the next iteration. G. Calculating fitness value It is significant the determination of the state and the control variables to solve OPF problem of two-terminal HVDC system. Furthermore, these control variables should be the same as those of the problem to be optimized. The control variables of this optimization problem are selected as follows: u = [u AC , u DC ] u AC = p g 2 ,..., p gNg , v g1 ,..., v gNg , t1 ,..., t NT
B.
Fitness scaling The scaling prevents algorithm to get stuck on a local point. In this study, the better individuals than individual having average fitness value are selected and can be formulated as follows:
Nk Fave = Fi / N k i =1
(35)
where Fave , N k , and Fi show the average fitness value within population, the number of individuals within population, and th the fitness value of i individual, respectively. C. Selection In this stage, the parents to be crossed for producing a child are selected. The tournament method is preferred and can be formulated as follows:
u DC = [ p dr , p di , q dr , q di , i d ]
(38) (39) (40)
i = Fi /
F
j =1
Nk
where p gi except the slack bus p gslack , is the generator active

j
(36)
th
power outputs, v gi is the generator voltage magnitudes, N g is the number of generator buses and N T is the number of transformers. The state variables of this optimization problem are selected as follows: x = [x AC , x DC ] x AC = p gslack , q g1 ,..., q gNg , v L1 ,..., v LNl
x DC = t r , t i , , , v dr , v di , irelative
where i shows the weight of i individual within population. Furthermore, the sum of the elective probabilities of all the individuals within population is 1 as given as follows:
(41)
i =1
Nk
=1
(37)
(42)
(43)
max
where p gslack is the slack bus active power output, q gi is the reactive power outputs, v Li is the load bus voltage magnitudes,
N l is the number of load buses, respectively. The fitness value for each individual is obtained as follows:
lim Fi = K 1 Fcos t + K 2 p gslack p gslack
10 F itn e s s v a lu e 9 8 7 6
x 10
Case A
Case B
) + K (q
2 3 i =1 lim 2 r
Ng
gi
lim q gi
) )
+ K 4 v Li v + K7 + K 10
di
Nl
( (v
i =1
lim 2 Li
+ K 5 tr t
+ K 6 ti t
lim 2 i
(44)
lim 2
+ K8
Ng i =1
lim 2
lim 2 di
+ K 11
+ K 9 v dr v
lim 2 dr
relative i max
lim 2
10
20
30
40 50 60 Iteration number
70
80
90
100
lim lim lim lim lim lim , q gi , v Li , tr , t ilim , lim , lim , v dr , v di where p gslack
Figure 4. Variation of the fitness value versus the iteration number for cases A and B The variation of the relative rotor angles of the generators in the test system for cases A and B are shown in Figs. 5 and 6, respectively. It can be seen from Figs. 5 and 6 that all the generators are stable during the fault.
Phase angle difference (fault cleared at 0.2s) 150
and lim show the limits of the related variables, respectively; K1 , K 2 , K 3 , K 4 , K 5 , K 6 , K 7 , K 8 , K 9 , K 10 and K 11 are penalty weights of the related variables. The total fuel cost is also calculated as follows:
Fcos t =
(a
i =1
Ng
2 + bi Pgi + c i Pgi
(45)
100
Delta, degree
where ai , bi and generator.
ci represent the fuel cost coefficients of i th
50
V.
SIMULATION RESULTS
-50
The proposed algorithm is tested on the modified New England 39-bus test system for two fault cases [19]. AC transmission line between buses 14 and 4 in the original New England 39-bus test system is replaced with a two terminal HVDC link. A lot of trial is made for the different iteration numbers and population sizes. Great iteration number (>>1000) and population size (>>200) is not preferred because the great computational time is necessary for reaching an optimum. In the simulations, the time interval t is taken as 0.01 s, the max simulation duration t max is taken as 5 seconds and as a result of trials, the relative rotor angle max is determined as 200 degree. Case A: Three-phase to ground fault occurs at the bus 6 and in the line 6-11. Critical clearing time t c is 0.1960 s in Ref. [19], it is taken as 0.200 s for this study [20]. Case B: Three-phase to ground fault occurs at the bus 3 and in the line 3-4. Critical clearing time t c is 0.2349 s in Ref. [19], it is taken as 0.2500 s for this study [20]. The variation of the fitness value versus the iteration number for cases A and B is shown in Fig. 4. As seen from Fig. 4, it is reached to optimum by 40 iterations and by 75 iterations for cases A and B, respectively. The population sizes of GA for cases A and B are taken as 30 and 50, respectively. For cases A and B, minimum fitness values are obtained as 63868 and 63630 and the computational times are also determined as 360 and 600 seconds, respectively.
-100
-150
0.5
1.5
2.5 t, sec
3.5
4.5
Figure 5. The variation of the relative rotor angles of the generators in the test system for case A
Phase angle difference (fault cleared at 0.25s) 80 60 40 20 Delta, degree 0 -20 -40 -60 -80 -100
0.5
1.5
2.5 t, sec
3.5
4.5
Figure 6. The variation of the relative rotor angles of the generators in the test system for case B
VI.
CONCLUSION
1.
REFERENCES
F. Yaln, and U. Arifolu, A new approach based on genetic algorithm for optimal reactive power flow solution in multiterminal AC-DC systems, Przeglad Elektrotechniczny, vol. 89, no. 3A, 231-235, 2013. M.A. Pai, K.R. Padiya, and C. Radhakrishna, Transient stability analysis of multi-machine AC/DC power systems via energyfunction method, IEEE Transactions on Power Apparatus and Systems, vol. 100, no. 12, pp. 5027-5035, 1981. H. Sato, and J. Arrillaga, Improved load-flow techniques for integrated AC-DC systems, Proceedings of IEE, vol. 116, no. 4, pp. 525-532, 1969. C.N. Lu, S.S. Chen, and C.M. Ong, The incorporation of HVDC equations in optimal power flow methods using sequential quadratic programming techniques, IEEE Trans. Power Syst, vol. 3, no. 3, pp. 1005-1011, 1988. J. Reeve, G. Fahmy, and B. Stott, Versatile load flow method for multi-terminal HVDC system, IEEE Transactions on Power Apparatus and Systems, vol. 96, no. 3, pp. 925-933, 1977. T. Smed, G. Andersson, G.B. Sheble, and L.L. Gigsby, A new approach to AC-DC power flow, IEEE Transactions on Power Systems,vol. 6, no. 3, pp. 1238-1244, 1991. I. Szuvovivski, T.S.P. Fernandes, and A.R. Aoki, Simultaneous allocation of capacitors and voltage regulators at distribution networks using Genetic Algorithms and Optimal Power Flow, International Journal of Electrical Power and Energy Systems, vol. 40, no. 1, pp. 62-69, 2012. J.G. Vlachogiannis, N.D. Hatziargyriou, and K.Y. Lee, Ant colony system-based algorithm for constrained load flow problem, IEEE Transactions on Power Systems, vol. 20, no. 3, pp. 1241-1249, 2005. M.S. Osman, M.A. Abo-Sinna, and A.A. Mousa, A Solution to the Optimal Power Flow using Genetic Algorithm, Applied Mathematics and Computation, vol. 155, no. 2, pp. 391-405, 2004. S. Kumar, and D.K. Chaturvedi, Optimal power flow solution using fuzzy evolutionary and swarm optimization, International Journal of Electrical Power and Energy Systems, vol. 47 no. May, pp. 416-423, 2013. K. Iba, Reactive power optimization by genetic algorithms, IEEE Trans. Power System, vol. 9, no. 2, pp. 685-692, 1994. T. Bouktir, L. Slimani, and M. Belkacemi, A Genetic Algorithm for Solving the Optimal Power Flow Problem, Leonardo Journal of Sciences, vol. 4, pp. 44-58, 2004. I.G. Damousis, A.G. Bakirtzis, and P.S. Dokopouloset, Network Constrained Economic Dispatch Using Real-Coded Genetic Algorithm, IEEE Transactions on Power Systems, vol. 18, no. 1, pp. 198-205, 2003. D. Walters, and G. Sheble, Genetic Algorithm Solution for Economic Dispatch with Valve Point Loading, IEEE Transactions on Power Systems, vol. 8, no. 3, pp. 1325-1332, 1993. U. Kl, and K. Ayan, Optimal Power Flow Solution of TwoTerminal HVDC Systems Using Genetic Algorithm, Electrical Engineering, Accepted. DOI: 10.1007/s00202-013-0277-7. U. Arifolu, Optimal Power Flow Using Sequential Power Flow Approach For an AC-DC Power System, Ph.D. Thesis, Istanbul Teknik University, 1993. J.H. Holland, Adaptation in Natural and Artificial Systems, Ann Arbor: University of Michigan Press 1975. MATLAB Optimization Toolbox 5 Users Guide 2012, The Math Works, Inc Y.Z. Lin, Z.X. Cai, and Q. Mo, Transient Stability Analysis of AC/DC Power System Based On Transient Energy Function, Third International Conference on Electric Utility Deregulation and Restructuring and Power Technologies, pp. 1103-1108, 2009. U. Kl, Transient Stability Constrained Optimal Power Flow In ac-dc Systems Using Artificial Bee Colony Algorithm, Ph.D. Thesis, Sakarya University, 2013.
In this study, TSCOPF solution of AC-DC systems is accomplished by using GA for the first time. The proposed approach is tested on modified New England 39-bus test system. In order to show validity, effectiveness, and efficiency of the proposed approach on TSCOPF solution of AC-DC system, the active power outputs and the bus voltage magnitudes of all the generators and DC system variables obtained for cases A and B are given in Tables 1 and 2, respectively. The variations of the fitness values versus the iteration number are also represented. As seen from Figs. 5 and 6 that all the generators in the test system are also stable for both fault cases. Table 1. The active power outputs and the bus voltage magnitudes of all the generators for cases A and B
Bus number 30 31 32 33 34 35 36 37 38 39 Case A Case B
2.
3.
4.
5.
6.
pg
3.50 4.44 3.99 5.32 6.12 6.87 6.13 6.30 7.40 12.00
v
0.99 0.95 0.99 0.99 1.00 1.02 1.02 0.97 1.01 1.04
pg
3.50 3.71 7.72 6.57 5.86 5.34 4.86 4.68 7.96 11.87
v
1.03 1.03 1.04 0.96 1.00 0.99 1.00 1.02 1.03 0.99
7.
8.
9.
10.
11. 12.
Table 2. DC system variables obtained for cases A and B

Variable Min. 2.6200 2.5717 1.00 1.00 6.00 16.00 0.85 0.85 0.90 0.90 2.3818 Max. 2.6200 2.5877 1.50 1.50 40.00 40.00 1.15 1.15 1.10 1.10 2.9111 Case A 2.62 2.58 1.10 1.17 10.25 18.50 1.07 1.11 0.99 0.97 2.63 Case B 2.62 2.57 1.09 1.25 7.51 20.06 1.10 1.12 0.95 0.93 2.75
18. 19. 16. 14. 13.
p dr p di q dr
15.
q di
(degree) (degree)
tr ti v dr v di id
17.
20.
Steady-State and Dynamic Performance of Asynchronous Back-To-Back VSC HVDC Link

Djehaf M, Zidi S-A, Hadjeri S, Djilani Kobibi Y, Souag sliman
Intelligent Control and Electrical Power System Laboratory Djillali Liabes University of Sidi Bel-Abbes, Algeria med_djehaf@yahoo.com , sbzidi@yahoo.fr, shadjeri2@yahoo.fr, youcef1230@hotmail.com , slimane.souag@gmail.com
Abstract The VSC HVDC back-to-back arrangement is used

when two asynchronous AC systems need to be interconnected for bulk power transmission or for AC system stabilization reasons. Besides controlling the through power flow, it can supply reactive power and provide independent dynamic control at its two terminals. This paper investigates the steady-state and transient performance of high-voltage DC (HVDC) back-to-back transmission systems based on three-level voltage source converters. The study involves analysis of active-reactive power capabilities (P-Q envelope) including active power reversal and provision of voltage support to AC networks. The transient performance is explored by examining the VSC_HVDC response to external AC faults. Finally, the models and results are presented and tested by simulations using Matlab Simulink and its toolbox SimPowerSystems. Keywords- BTB HVDC, Voltage source converter (VSC), IGBT, SPWM, Control design
The objective of this paper is to study the operational performance of back-to-back VSC-HVDC system and its control strategies. The vector control method is studied using 3-level VSC connected to an active AC system at both ends of the HVDC link. Finally, simulations and results are presented by means of Matlab Toolbox Simpower System. This simulation presents the controller performance in the operation range against some steps change in the load and shows the control of active and reactive power. Following that typical operating contingency scenarios are simulated in order to evaluate transient performance. The simulation results confirm that the control strategy has fast response and strong stability. II. VSC-HVDC TRANSMISSION MODEL
a) Basic principle
I.
INTRODUCTION
The VSC applications include but are not limited to HVDC and flexible AC transmission system (FACTS) devices such as STATCOM, SSSC, UPFC, wind generators and active filters. The VSC based HVDC system is a feasible option for high power transmission over long or short distances and the grid integration of renewable energy sources in existing transmission and distribution systems. The main requirement in a power transmission system is the precise control of active and reactive power flow to maintain the system voltage stability. The VSC operating with the specified vector control strategy can perform independent control of reactive power at both ends and at the same time independently control active power flow. This ability of VSC makes it suitable for connection to weak AC networks or even networks without local voltage (i.e. dead) sources. For power reversal, the DC voltage polarity remains same for VSC based transmission system and the power transfer depends only on the direction of the DC current. This is advantageous when compared to the conventional current source converter based HVDC system. Furthermore, in case of VSC, the reactive power flow can be bi-directional depending on the AC network operating conditions [8-10].
Fig 1: Single Line Diagram of VSC-HVDC System
Fig. 1 shows the one-line diagram of VSC-HVDC system implemented in Matlab Toolbox Simpower System. The VSCHVDC system interconnects two AC networks of different frequencies for power transmission and stability improvement during contingencies. The HVDC system is modeled as a conventional bi-polar transmission system. Two series connected DC capacitors of same size are employed across the DC transmission line with grounded midpoint for VSC operation, to reduce the ripples in DC voltage. A smoothing reactor is also connected in series with transmission line for reducing the ripple in DC current.
During failure or scheduled maintenance of one pole of transmission line, a reduced amount of power can still be transmitted by other pole. b) Fundamental Of back to back Vsc-Hvdc Transmission The fundamentals of VSC transmission operation may be explained by considering each terminal as a voltage source connected to the AC transmission network via a three-phase reactor. Changing the fundamental frequency voltage phase angle across the series reactor controls the power; whereas, changing the fundamental frequency voltage magnitude across the series reactor controls the reactive power. If two VSC converters are connected together as shown in Figure 2, an asynchronous transmission link is formed. The converters can be connected in back-to-back configuration or at either end of a transmission line or cable, as schematically shown in Fig. 2 [4-6].
VSC 1 VSC 2 Rv , X V
modulation (PWM). Using the definition of the apparent power and neglecting the resistance of the transformer results in the following equations for the active and reactive power: U L U V (1) P =Ud Id = sin (2) XL (3) XL The active power and reactive power exchanged by VSCHVDC and AC system can be adjusted promptly by change the magnitude and angle of the output AC voltage of the VSCHVDC. This characteristic of VSC-HVDC makes itself more flexible than other FACTS technology, such as SVC, STATCOM, also than traditional HVDC. By means of Phase Width Modulation (PWM) technology, especially Sinusoidal PWM (SPWM), two degrees of freedom, i.e. phase and amplitude can be acquired. Phase and Amplitude Control (PAC) technology is developed for VSC-HVDC applications [6,7]. The VSC can easily interchange active and reactive power with an AC network as well as a synchronous machine.
Q= U L (U L U V (1) cos )
P, Q
Ud
c
Ud
c
AC filters
V2 V1
i i
But the extent of the active power and reactive power which can be adjusted in VSC-HVDC is subject to the rate power limit and the operation condition of the time. The adjusting AC ability of the active power and that of the reactive power system influence each other dynamically. So it is necessarily to analyze the ability in real time. Figure 4 shows a typical P-Q diagram for a VSC based transmission system expressed in per unit [1] [3].
Fig 2: Basic VSC-HVD back-to-back transmission
Fig. 3 shows a phasor diagram for the VSC converter connected to an AC network via a transformer inductance. The fundamental voltage on the valve side of the converter transformer, i.e. UV(1), is proportional to the DC voltage has been expressed in equation (1):
Active Power -
Active Power +
Reactive Power + Region 3 Active Power -
Reactive Power + Region 1 P

Active Power +
U V (1) = k u U d
(1) I P<0 Q<0

Uv(1) XL . Iv(1)
II P<0 Q>0
Reactive Power Region 4
Reactive Power Region 2
III P>0 Q>0
Green = active power delivery and reactive power delivery Orange = active power delivery and reactive power receipt Yellow = active power receipt and reactive power delivery Magenta = active power receipt and reactive power receipt
Uv(1)
IV P>0 Q<0
Fig 4: P-Q characteristics of a VSC-HVDC system
Fig 3: Phasor diagram of VSC and direction of power flows
The quantity k u can be controlled by applying additional number of commutation per cycle, i.e. applying pulse with
U L U V (1) U2 P + Q L = (4) X X L L If the output voltage of the converter UV (1) is reduced, e.i by using PWM, supply of any active and reactive power within the circle is possible.
2
III.
CONTROL STRATEGY B. Outer active and reactive power and voltage loop The active power or the DC voltage is controlled by the control of and the reactive power is controlled by the control of the modulation index (m). The instantaneous real and imaginary power of the inverter on the valve side can be expressed in terms of the dq component of the current and the voltage on the valve side as follows: 3 3 *dq ) = (u fd ivd + u fq ivq ) p = Re(u dq (5) f iv 2 2
Generally the control strategy of a two terminal VSC-HVDC transmission line is to keep one terminal DC voltage constant as operation point, and adjust the other terminal DC current or active power order. The AC voltage or the reactive power of the two terminals can be controlled. Fig. 5 shows an overview diagram of the VSC control system and its interface with the main circuit [6] [11]. The converter 1 and converter 2 controller designs are identical. The two controllers are independent with no communication between them. Each converter has two degrees of freedom. In our case, these are used to control: I.P and Q in station 1 (rectifier) II.Ud and Q in station 2 (inverter).
AC Filters
Iabc UL Ur Iv R, L Uv Pulses SPWM VSC
Id Ud Ud Id
DC Voltage Balance Control V_ref_abc
3 3 Im(u fdq iv*dq ) = (u fd ivq + u fq ivd ) (6) 2 2 If the reference of the dq-frame is selected such that the quadrature component of the voltage is being very small and negligible (uLq 0) then the equations (5) and (6) indicate that the active and the reactive power are proportional to the d and q component of the current respectively. Accordingly, it is possible to control the active power (or the DC voltage or the DC current) and the reactive power (or the AC bus voltage) by control of the current components ivd and ivq respectively. The active and reactive power and voltage loop contains the outer loop regulators that calculate the reference value of the converter current vector (I*dq) which is the input to the inner current loop [2]. q=
C. Inner current loop The AC Current Control block tracks the current reference vector (d and q components) with a feed forward scheme to achieve a fast control of the current at load changes and disturbances (e.g., so short-circuit faults do not exceed the references) [3] [5] [6]. In essence, it consist of knowing the U_dq vector voltages and computing what the converter voltages have to be, by adding the voltage drops due to the currents across the impedance between the U and the PWMVSC voltages. The state equations representing the dynamics of the VSC currents are used (an approximation is made by neglecting the AC filters). The d and q components are decoupled to obtain two independent first-order plant models. A proportional integral (PI) feedback of the converter current is used to reduce the error to zero in steady state. The output of the AC Current Control block is the unlimited reference voltage vector Vref_dq_tmp. D. DC voltage balance control The difference between the DC side voltages (positive and negative) are controlled to keep the DC side of the three level bridge balanced (i.e., equal pole voltages) in steady-state. Small deviations between the pole voltages may occur at changes of active/reactive converter current or due to nonlinearity on lack of precision in the execution of the pulse width modulated bridge voltage. Furthermore, deviations between the pole voltages may be due to inherent unbalance in the circuit components impedance.
PLL
Inner Current Loop I_ref_dq Outer Active/Reactive Power and Voltage Loop
P_ref Q_ref
Ud_ref
Fig 5: Overview diagram of the VSC control system
A. Phase locked loop The phase locked loop (PLL) shown in fig.6 is used to synchronize the converter control with the line voltage and also to compute the transformation angle used in the d-q transformation. The PLL block measures the system frequency and provides the phase synchronous angle for the d-q transformations block. In steady state, sin () is in phase with the fundamental (positive sequence) of the component and phase A of the point of common coupling voltage (Uabc).
Abc dq0
vdq0
Vabc
PLL
Iabc
Abc dq0
Idq0
Fig 6: Phase locked loop block
IV.
MODEL PERFORMANCE ANALYSIS
The dynamic performance of the transmission system is verified by simulating the: A. Active power flow reversal, B. Dynamic response to step changes applied to the principal regulator references, like reactive power regulator, C. VSC_HVDC response to external AC faults.
1. Case 1 In Fig. 7, the initial reference value of the power flow is at 0.5 pu in the direction from VSC2 (rectifier mode) to VSC1 (inverter mode). At t=1.4 s, the reference value is reversed from +0.5 pu to -1 pu. It can be seen that the active power follows the reference rapidly without any oscillation and reaches the new reference of -0.5 pu within 0.2 s. Correspondingly, the measured active power Pmeas1 at VSC 1 changes around from -0.5 p.u. to 1 p.u.. At the same time, the active power flow reverses with the DC current direction change. However, the reactive powers Q1 and Q2 at VSC 1 and 2 dont change. As can be seen, the active power can track the reference of the active power. The transferred active powers at both sides change the direction which causes transients on the DC voltage
1.5 1 0.5 0 -0.5
then returns to the reference value due to the DC voltage controller. The AC voltages at the filter bus (Uf1 and Uf2), can be kept constant except for some transients that occur when both the step changes are applied.
2.Case 2 In order to test the dynamic responses of the back to back VSC-HVDC regulators, two test cases have been studied. At t = 2.3s, the reactive power step change from 0 pu to 0.2 pu of station 1 (VSC1). At station 2 (VSC2), the reactive power step change from -0.1 pu to -0.3 pu at t = 2.5 s (fig.7, with the constant dc voltage of 0,1pu). These two steps change cause transients on the DC voltage, but, as expected, the step change of the active power causes a much higher transient than that with the change in reactive power. We can see that it has a good tracking accuracy. However, variables P1 and P2 are not affected at all. Obviously, the reactive power flow can be independently controlled at each AC network, and the reactive power and the active power control is independent. Notice that, considering the loss in the transmission cables, the sending active power equals the receiving active power plus the loss power in the cable. The active and reactive power responses are decoupled by the control design.
1.5 1
Udc (pu)
Udc (pu)
1.2
1.4
1.6
1.8
2.2
2.4
2.6 P P ref
2.8 0.5
P and P ref (pu)

1 0 -1 0.2 0 -0.2 -0.4 1 0 -1 1 0 -1 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6
1.4
1.6
1.8
1.2
1.4
1.6
1.8
Q and Q ref (pu)
2.2
2.4
2.6 Q Q ref IVd IVd ref
2.8
1 0 -1
P (pu)
2.2
2.4
2.6
2.8
1.4
1.6
1.8
1.2
1.4
1.6
1.8
IVd (pu)
2.2
2.4
2.6
0.5 0 -0.5 2.8 -1 1 0 2.8 -1 1 0

2.8 -1
Q and Q ref (pu)
2.2
2.4
2.6
2.8 Q Q ref 2.8 IVd IVd ref
1.4
1.6
1.8
2IVd (pu)
2.2
2.3
2.4
2.5
2.6
1.2
1.4
1.6
1.8
IVq (pu)
2.2
2.4
2.6 IVq IVq ref
1.4
1.6
1.8
IVq (pu)
2.2
2.4
2.6
2.8 IVq IVq ref
Uf meas (pu)
1.5 1 0.5 0
1.6
1.8
1.5 1
1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8
Uf meas (pu)
2.2
2.3
2.4
2.5
2.8
0.5
1.4
1.6
1.8
2.2
2.4
2.6
2.8
STATION 1 (VSC 1)
STATION 2 (VSC 2)
Fig 7: Steps on the regulators references
3.Case 3 A single phase to ground fault was first applied at t = 2.1s during 0.12s (7 cycles) at station 2 AC bus in order to investigate the behaviour of VSC-HVDC during unbalanced faults. A second perturbation follows. A three-phase to ground fault is applied at station 1 AC bus at t = 2.8 s and is cleared at 6 cycles after the fault, i.e., at t = 2.1s. Fig.8 presents the simulations results. From the simulation, it can be noted that before a single phase to ground fault at station 1, the active power flow is 1.0 pu, transmitted from converter 1 to converter 2, and is kept constant during the fault. The DC voltage drops and it contains an oscillation during the fault. Consequently the transferred DC power contains also the oscillation. During the station 1 side fault the transmitted power can be kept constant except a small oscillation during the fault.
All oscillations in voltages and currents at both systems, means that the phase voltages and currents at both systems are unbalanced. Note that during the three-phase fault, the transmitted DC power is almost zero. At this moment the two VSC stations can be considered as independent STATCOMs. The system recovers well after the fault within 100 ms. During the severe three-phase fault at station 1 at t = 2.8 s, the AC voltage at station 1 side is decreased to 0 pu during the fault and recovers fast and successfully to 1.0 pu voltage after clearing the fault. The transmitted power flow is reduced to very low value during the fault and recovers to 1.0 pu after the fault. It takes about 100 ms to recover the steady state before the next perturbation initiation. The DC voltage, which can be controlled to 1.0 pu during the fault, has some oscillations at the beginning of the fault and at clearing the fault.
2 1 0 -1 2 2 1 0 -1 1 0 -1 3 2 1 0 -1 2 1 0 -1 1 0 -1 2 5 0 -5 2.1 2.5 2 2.1 2.5 2.1 2.5
Udc (pu)
8 P and P ref2.(pu)
3 P P ref 3 Q Q ref
3.5
1.5 1 0.5 0 -0.5 2 0 -2
Udc (pu)
2.1
2.5
P (pu) 2.8
3.5
Q and Q ref2.(pu) 8
3.5
2.1
2.5
2.1
2.5
IVd (pu)2.8
2.1
2.5
IVq (pu)2.8
2.1
2.5
2.8 Uabc (pu)
1 0 -1 3.5 2 2 IVd 0 IVd ref -2 3.5 2 1 IVq 0 IVq ref -1 3.5 2 1 0 -1 2 2 0 -2
8 Q and Q ref2.(pu)
3 Q Q ref
3.5
2.1
2.5
IVd (pu)2.8
3 IVd IVd ref 3 IVq IVq ref
3.5
2.1
2.5
IVq (pu)2.8
3.5
2.1
2.5
2.8 Uabc (pu)
3.5
Iabc (pu)2.8
3.5
2.1
2.5
Iabc (pu)2.8
3.5
2.1
2.5
STATION1 (VSC 1)
2.8
3.5
2.1
2.5
2.8
3.5
STATION2 (VSC 2)
Fig 8: AC side perturbations
V.
CONCLUSION
VIII.
[1]
REFERENCES
In this paper, we have presented the steady-state and dynamic performances of VSC based back-to-back HVDC transmission system during step changes of the active and reactive powers, balanced and unbalanced faults. In all cases the proposed control strategy has been shown to provide fast and satisfactory dynamic responses of the proposed system. From the simulation, it can be obtained that the VSC-HVDC can fulfill fast and bi-directional power transfers. It can be obtained also that during a single-phase fault the transmitted power can be kept constant except a small oscillation during the fault. However, during a three-phase fault; the decreased voltage at the converter terminals strongly reduces the power flow by the DC link. When the fault is cleared, normal operation is recovered fast. The system advantages of deploying a VSC based back-toback asynchronous Tie with standby dynamic voltage control during network contingencies. The controlled power transfer capability allows the exchange of power between the two networks with different frequencies (50/60 Hz). VI. NOMENCLATURE UL = the sinusoidal AC voltage in the AC network Uf = AC voltage in the AC network at the filter-bus UV (1) = the fundamental line to line voltage (valve side) XL = the leakage reactance of the transformer = phase shift between UL and UV (1) IV = source current L, R = phase reactor inductance and resistance C = DC side capacitance = source voltage angular frequency P, Q = AC active, reactive power inputs Ud, Id, Pd = DC side voltage, current, power , = stationary - axis d, q = synchronous d -q axis p, n = positive, negative components *,ref = reference value for controller VII. Station 1(Rectifier side) APPENDIX 230kV(80), 2000 MVA, SCR = 10, L1 = 31.02 mH, R = 13.79 , L2 = 62.23 mH. f = 50 Hz 230kV(80), 1000 MVA, SCR = 5, L1 = 62.04 mH , R = 27.58 , L2 = 124.46 mH, f = 60Hz Yg/, 230kV/100kV, 200 MVA, 15% 70 F 0.1 Km2 (R=0.0139 /km, L = 0.159 mH/km, C = 23.1 F/km 1350 Hz
Ruihua, Song, Chao, Zheng, Ruomei, Li, Xiaoxin, Zhou VSCs based HVDC and its control strategy, IEEE/PES Transmission and Distribution Conference & Exhibition: Asia and Pacific Dalian, China, 2005. [2] D. Madhan Mohan, Bhim Singh and B. K. Panigrahi A New Control Strategy for Active and Reactive Power Control of Three-Level VSC Based HVDC System, Annual IEEE India Conference (INDICON), 2010. [3] G.P. Adam, O. Anaya-Lara and G. Burt Steady-state and Transient Performance of DC Transmission Systems Based on HVDC Technology , AC and DC Power Transmission,. ACDC. 9th IET International Conference ,2010. [4] T. Larsson, A. Edris, D. Kidd, F. Aboytes, Eagle Pass Back-toBack Tie: a Dual Purpose Application of Voltage Source Converter technology Proceedings of the 2001 IEEE PES Summer Power Meeting, Vancouver, BC, July 2001. [5] Chathura Jeevantha Patabandi Maddumage Co-Simulation of Back-to-Back VSC Transmission System, A Master thesis submitted to the Faculty of Graduate Studies at the University of Manitoba, 2011. [6] Weimers, L. HVDC Light: A New Technology for a Better Environment IEEE Power Engineering Review, Vol 18, Issue 8, Aug. 1998. pp: 19 20. [7] Schettler, F.; Huang, H.; Christl, N. HVDC transmission systems using voltage sourced converters design and applications IEEE Power Engineering Society Summer Meeting. Vol 2, 2000 pp: 715 720. [8] Asplund, G. Application of HVDC Light to power system enhancement IEEE Power Engineering Society Winter Meeting. Vol 4, 2000 Page(s):2498 2503 [9] Weimers, L. New markets need new technology International Conference on Power System Technology, PowerCon 2000. Vol 2, 2000 pp: 873 877. [10] Smith D, Cross Sound HVDC Link Extends IGBT Power, Modern Power Systems Transmission & Distribution, October 2000. pp: 31-36. [11] Lindberg, A.; Larsson, T. PWM and Control of Three Level Voltage Source Converters in an Hvdc Back-to-back Station Sixth International Conference on AC and DC Power Transmission, 29 Apr-3 May 1996. pp: 297 302 [12] Lie Xu; Andersen B.R.; Cartwright P. Control of VSC transmission systems under unbalanced network conditions Transmission and Distribution Conference and Exposition, 2003 IEEE PES Vol 2, Issue , 7-12 Sept. 2003 pp: 626 632
Station 2 (Inverter side)
Transformer Main DC capacitor DC Cables Switching frequency
Mohamed DJEHAF: was born in Oran, Algeria. He received the masters. Degree in electrical engineering, from the Djillali Liabes University of Sidi Bel-Abbes (Algeria), in 2011.He is now a PhD Candidate in the Electrical Engineering Department of Djillali Liabes University. His current interest covers HVDC control and protection. Sid-Ahmed ZIDI: was born in SidiBel Abbes, Algeria. He received the diploma of Electro technical Engineering degree from the University of Science and Technology of Oran, Algeria. The Master degree, from the University of Djillali Liabes of Sidi Bel-Abbes, Algeria in 1994. The PhD degrees from the University of Sidi Bel-Abbes, Algeria, in 2005. He is currently interested by the HVDC link, FACTS and transient in power systems.
Performance Analysis and Economic Evaluation of a Solar Power Tower in Algeria

Salah. Boudaoud
Laboratoire de Modlisation et Simulation en Mcanique Universit Mhamed Bougara UMBB, Boumerdes, Algrie sboudaoud@gmail.com
Abdallah Khellaf
Laboratoire de recherche division hydrogne Centre de Dveloppement des nergies Renouvelables CDER, Bouzareah, Algrie khellaf.at@gmail.com
Kamel. Mohammedi
Laboratoire de Modlisation et Simulation en Mcanique Universit Mhamed Bougara UMBB, Boumerdes, Algrie mohammedi.meso.lemi@gmail.com Abstract in the context of the deployment and implementation
of grid connected concentrated solar power plants which are in still progress in MENA regions, the different authority of this country like Algeria gives suitable environment for the promotion and the diversification of energy sources to produce electricity. The energy policy and legal framework are the main incentive trials to support that future environmental friendly assignment, thus in this work, a thermal performance analysis coupled to economic reliability were elaborated using SAM advisor software as a preliminary results in order to investigate whether the installation of high concentrated central receiver solar power plant in different regions (coastal, highland and Sahara) is feasible. We have carried out a method to optimize the Solar Multiple (SM) to get a trade-off between the incremental investment cost related to the heliostat field size and the Thermal Energy Storage (TES) required, yielding a minimum Levelized Electricity Cost (LEC) in DZA/kWhe and allow higher Capacity Factors (CF) for dispatchability of the generated electricity within critical hours in the day. This study presents also the optimization process to build some design parameter like receiver working temperature with molten salt as Heat Transfer Fluid (HTF) medium and thus perform the allowable flux density admitted by the receiver cavity from a given size of heliostat field. Finally, the electricity unit cost, the Net Present Value (NPV) and the internal rate of return variants depending on the region are calculated. The results indicate that for optimal working parameters and a given size of the solar field, highland and Sahara regions are the suitable regions for grid integration and and economic competitiveness of the central receiver power tower plant. Keywords Energy policy, economic reliability, Levelized Electricity Cost, Net Present Value , Solar Multiple, Thermal Energy Storage, Capacity Factors
Introduction The world experiences since more than one century a significant economic development. The industrial development, the increase in the automobile park and the multiplication of the domestic equipment caused a significant growth of the energy demand. In Algeria, this growth in the energy was mainly covered by the construction of new power plant whose source is in fossil energies (natural gas), stimulated by technico-economic considerations. With 2.381.741km of land area, Algeria is by far the largest country of the Mediterranean. Over 70% of its area are South of 20 latitude. According to a study of the German Aerospace Agency, Algeria has with 1787000 km the largest long term land potential for concentrating solar thermal power plants [4]. The Renewable Energy Development Center (REDC) summarized the available insolation measurements in Algeria. According to the irradiation maps presented fig.1, total annual direct normal irradiation ranges from 2100 kWh/myr to over 2700 kWh/myr and is counted among the best insolated areas in the world [4]. Algerias solar potential and land resources are optimal for the implementation of solar thermal power technologies thus: Most desert areas in Algerias offer direct normal insolation above 2100 kWh/myr
The best sites, in the southern part of the country, exceed 2500 kWh/myr It is estimated that within 50 km from required infrastructure (roads, grid) accessible sites have huge potential far in excess of present consumption. Within its policy of climate and environment protection, the Algerian Ministry for Energy and Mines fully supports the objective of the Concentrated Solar Power (CSP) Global Market Initiative (GMI) to facilitate and expedite the building of 5,000 MWe of CSP worldwide over the next ten years. The Government of Algeria has committed itself to develop solar energy as its largest renewable energy source, to cover 5% of the national electricity needs by 2010 with renewables. Opportunities of combining Algerias richest fossil energy source - the natural gas - with Algerias most abundant renewable energy source - the sun - is the actual transition model bteween fossil and renewable source by using Integrated Solar Combined Cycle System (ISCCS). I. SITE SELECTION CRITERIA FOR PRE-FEASABILITY STUDY FOR CSP PLANT II.1 ENERGY RESSOURCES The first parameter considered in site assessment in term of energy resource is the direct normal irradiance (DNI), which is defined as the radiant flux density in the solar spectrum (0.3 m to 3 m) incident at the earth's surface perpendicular to the direction to the sun integrated over a small cone tracing the sun. The available DNI is affected by absorption and scattering of the solar radiation at air molecules, ozone, water vapor and aerosols. As we can see in figure one, the available DNI by day is very high at the south of Algeria while it reduce gradually when getting to the north side.
Figure 1: Daily direct normal irradiation in kWh/m/day. In Algeria, we can distinguish 03 regions in terms of average duration of sun exposure and the amount of energy received by square meter. See table 1 herein after. We can assume that Algeria has the most important solar energy potential in the world and has over 2000 hours of sun exposure by year. Table 1: Average energy resource received in Algeria. Regions Area (%) Average duration of sun exposure (hours/year) Mean daily sunshine duration(h) Average energy density (kWh/m/year) Coastal regions 4 2650 7.26 1750 Highlands 10 3000 8.22 1900 Sahara 86 3500 9.59 2650
II.2 LAND AVAILABILITY CSP plants need a high land area compared to conventional power plants. The specific surface area for a solar tower power plant is about 0.02 km to 0.025 km per MWThe slightly high comparing to parabolic trough power plant which is in the range of about 0.015 to 0.02 km per MWThe. The availability of land to build large CSP collector fields is therefore an important site criterion. In the case of Algeria, we have more wasted land at the south which is very suitable to CSP plant implementation as subsidy, poorly dense in population and low agriculture usage but less economic activities which is not compatible with other
criteria like existence of roads and electrical networks in the north side, the main strain is the land ownership with a dense population and agriculture activities. II.3 WATER AVAILABILITY Water requirements for the solar thermal power station would be similar to a conventional thermal power station of similar output plus additional water that would be used for solar reflector cleaning. For a wet cooled system the total water consumption would be around 276 ML/a, while if dry cooling was introduced this could fall to around 36 ML/a.[06]. In one hand, There is a dense hydraulic network linked to several big dam in the north of Algeria, and in another hand, in the region of highland and south of Algeria, a great water table exist which is important to feed-in water all economical investment such as new power plants far from urban cities. II.4 NATUREL HAZARD POTENTIAL Natural risks comprise phenomena like earthquakes, storms, and others. These risks can affect the operating safety of a CSP plant. In order to resist to the impacts of these phenomena, the design of the solar field and of the power block must be adapted, which may imply higher construction costs. Additionally, insurance costs may rise at sites with higher damage risks. II.5 INFRASTRUCTURE CONVINIENCE CSP plants need certain infrastructure for their operation. Existing infrastructure is, hence, an important site criterion. Missing infrastructure requires higher investment. A power plant needs access to roads or other transportation ways (navigable waterways), to high or medium voltage power grids and to water resources if wet cooling is planned. Additionally, pipelines may be favorable for water transport or fuel transport for hybrid plant operation. In the context of our country, the government has invested a big highway all over the territory such as the named East west high road and another in construction phase which links the north regions to the south ones. II.6 POLITICAL AND ECONOMIC FRAMEWORKS Political and economic conditions in a country represent important site criteria. Promotion measures for renewable energies are especially decisive. There are different promotion strategies. The most important strategies are special feed-in tariffs or premiums for electricity generated on the basis of renewable energy sources, quotas for the renewable energy share and tax incentives. The politically controlled promotion of CSP is still necessary because of the currently higher
levelized electricity cost of CSP plants in comparison to fossil fired power plants and some other competitors. Political promotion has the aim to make CSP plants economically competitive until they get competitive on their own. Incentive premiums for CSP projects are granted within the framework of Algerias new Decree 04-92 of March 25th, 2004 relating to the costs of diversification of the electricity production. The incentive premiums of this decree shall attract private investors to implement integrated solar combined cycle plants in Algeria. According to the current power expansion planning of the ministry for energy and mines, the capacity targets for CSP power implementation in Algeria are 500 MW of new ISCCS plants until 2020. With these CSP targets and the new decree 04-92, Algeria has established the necessary GMI commitment on national solar thermal power market implementation. As the next GMI step to be agreed at the renewables 2004 conference in Bonn, the government of Algeria pledges to develop a framework for solar thermal electricity export from north-Africa to the European Union. II. CENTRAL RECEIVER THERMAL POWER PLANT
DESCRIPTIONS
Central Receiver System with large heliostat elds and solar heat exchanger located on top of a tower are now in a position for deployment of the rst generation of gridconnected commercial plants. The Central Receiver System (CRS) power plant technology can be considered as sufciently mature after the pioneering experience of several 0.5 to 10 MW pilot plants in the early 1980s see fig. 2. Incident sunrays are tracked by large mirrored collectors called heliostats, which concentrate the energy ux to the receiver. The heliostat eld performance is dened in terms of the optical efciency, which is equal to the ratio of the net power intercepted by the receiver to the product of the direct insolation and the total mirror area. The different losses affecting the system are the cosine effect, shadowing, blocking, mirror reectivity, atmospheric attenuation and spilling effect. The typical optical concentration ratio ranges from 200 to 1000 suns. The high solar ux incident on the receiver (averaging between 300 and 1000 kW/m) enable operation at relatively high temperatures of up to 1000C. Different type of thermal fluid system like air, steam, molten nitrate salt and liquid sodium are used to transfer energy from solar receiver to power bloc via heat exchanger system.
: heliostat azimuth angle, 0 in south. l = , the angle between the vector from heliostat to tower and vertical. Hhelio: height of the heliostat mirror, Whelio: width of the heliostat mirror.
Figure 2: Solar central receiver power plant main components. III. CENTRAL RECEIVER SYSTEM DESIGN AND MODELING
The design and modeling of each component composing the solar central receiver power plant involve an interactive process which gives at each level the physical state and governing equation of each part. IV.1 HELIOSTAT FIELD LAYOUT The optimization method used to tradeoff between cost and performance of the heliostat field is established in codes such as DELSOL3 and SOLERGY [9] to provide relationship between tower high, receiver dimensions and the radially staggered field layout see fig.3. Figure 4: Representation of optimized elds with north configuration using SAM Advisor software. IV.2 CAVITY RECEIVER GEOMETRY In our model, we have used a cavity receiver type which consists of a cavity with a small opening (inlet aperture). The concentrated solar irradiation is aimed at the small opening where inside it impinges on tubes carrying the working fluid see fig.5. The idea behind the cavity receiver is to minimize the radiation losses. From the radiation entering the inlet aperture, only small amounts are reflected back into the atmosphere through the inlet aperture.
Figure 3: Radially staggered field layout. In case of north configuration see fig.4, the correlation between the radial (R) and azimuthal (Az) spacing is given by the equation (1) and (2) [16] R = [63.0093- 0.587313.l + 0.0184239. l + cos(2.80873 - 0.14805.l + 0.0014892 .l)].Hhelio .(1) Az= [2.46812- 0.040105.l+ 9.2359 10-4. l+ cos(0.17345- 0.009113.l+ 1.276110-4 .l)].Whelio(2) Where:
Figure 5: Cavity receiver geometry. IV.3 THERMAL STORAGE SYSTEM The direct storage system has been successfully applied in the solar tower technology, the HTF, which is heated by a receiver, is used directly as a storage medium. The HTF/storage medium can be stored either in a single or dual tank. In this simulation, we have used a two tank model (hot and cold tank). IV.4 SIMULATION DATA We have carried out a simulation using SAM advisor which makes performance predictions and cost of energy estimates for grid-connected power projects based on installation and operating costs and system design parameters that we specify as inputs to the model. These are the main inputs:
The future research work focuses on reducing the cost of the heliostat field to be competitive with conventional power plant.
Figure 6: The levelized cost of energy of the main power plant component. For annual thermal energy output of about 16 KWhth as energy resource from solar field, the electrical energy outlet of the turbine does not exceed 1.9 KWhth which result in a global efficiency of about 12% (base case), this is due to different losses (optical losses of the heliostats field, receiver losses (convective, conductive and radiative), the thermal losses in storage in tanks, thermal conversion efficiency of the Rankine cycle, the control and supervision of the plant and the cooling system, see Figure 7.
Table 1: main values introduced in SAM advisor software Parameter definition Direct Normal Irradiance (DNI) Analysis period Inflation rate Debt fraction PPA escalation rate Heliostat area Solar multiple Cavity aperture Turbine gross output Temperature of hot Tank Temperature of cold Tank Capacity storage IV. Input value 1446.4 kWh/m 30 years 2.5 % 40 % 1.4 % 8 m 1.2 22.5 m 2 MW 565C 290C 2h
RESULTS AND DISCUSSION
Figure 7: The annual energy output flow diagram. For net annual energy production of 1 MWe, the levelized energy cost is about 0.128$/kWh which is very high comparing to LCOE of fossil energy (0.04$/kWh) for 10 to 100MWelect, the land area needed is about 285000 m and an annual need of water about 6600 m3. For instance, the solar tower power plant will be competitive for higher power production level.
The financial evaluation of each part of the plant is given by the diagram bar in fig.6, we see that in a solar tower power plant, the solar field is the most critical part in terms of investment with a cost of 33 cents$/kWh and represent about 35% of the total direct cost.
As the energy resource is intermittent within the day and along the year, we have estimated the best solution relative to the capacity of the thermal storage system given a solar multiplier value suggested by economical consideration. The fig. 8, shows that the optimal value for energy storage system is about 2.5 hours regarding to LCOE which remains constant (0.128$/kWh) and a maximum net annual energy output is about 953 kWh.
Figure 8: Optimal hour value for thermal storage system TES. V. CONCLUSIONS
In this study, SAM Advisor software is used to perform a technico-economic examination of the solar tower power plant of 2 MWe. The simulation has been applied to the project ALSOL1 the future central receiver power plant which will be held at Tipaza state near Algiers. We assume that: In this type of concentrating solar plant, the biggest investment is critical in the heliostat field and all associated operating and Maintenance costs during the life time cycle; The levelized cost of electricity LCOE is inversely proportional to the capacity of the plant, more high the capacity of the power plant less the LCOE value; In the central receiver plant at Tipaza which have a capacity of 2MWe electrical output, the LCOE is about 0.128 $/kWh which is relatively high, thus increasing the capacity of the power plant will be more competitive. REFERENCES
[1] A. Boudghene Stambouli , Z. Khiat , S. Flazi , Y. Kitamura. A review on the renewable energy development in Algeria: Current perspective, energy scenario and sustainability issues,
[2] Hannah Cassard, Paul Denholm, Sean Ong. Technical and economic performance of residential solar water heating in the United States [3] Solar Generation Technology Assessment. B&V Project Number 161098. Black & Veatch Corporation, 11401 Lamar. Overland Park, Kansas 66211, Tel: (913) 458-2000 www.bv.com [4] Report on the SolarPaces Start Mission to Algeria September 14-18, 2003. Michael Geyer IEA SolarPACES [5] Hennecke, K., Schwarzbzl, P. Hoffschmidt, B., Gttsche, J. Koll, G., Beuter, M. The Solar Power Tower Julich, a Solar Thermal Power Plant for Test and Demonstration of air Receiver Technology. [6] Solar Power Plant Pre-feasibility Study. ActewAGL and ACT Government. Level 4, Northbank Plaza 69 Ann Street Brisbane, QLD 4000, GPO Box 2907, Brisbane QLD 4001, Australia. [7] Mohamed Abbas, Bousaad Boumeddane, Noureddine Said, Ahmed Chikouche. Dish Stirling technology: A 100 MW solar power plant using hydrogen for Algeria. [8] Patricia Kuntz Falcone. A Handbook For Solar Central Receiver Design. Sandia National Laboratory Livermore. SAND 86-8009, Unlimited Release Printed December 1986. [9] Clifford K. Ho. Software and Codes for Analysis of Concentrating Solar Power Technologies. SAND2008-8053 Unlimited Release Printed December 2008. [10] Walter Short, Daniel J. Packey, and Thomas Holt. A Manual for the Economic Evaluation of Energy Efficiency and Renewable Energy Technologies. National Renewable Energy Laboratory 1617 Cole Boulevard Golden, Colorado 80401-3393. [12] Ernst & Young et Associes. Middle East and North Africa Region Assessment of the Local Manufacturing Potential for Concentrated Solar Power (CSP) Projects. January 2011. [13] Paul Gilman. Solar Advisor Model User Guide for Version 2.0. Technical Report NREL/TP-670-43704 August 2008. [14] Jim Hinkley, Bryan Curtin, Jenny Hayward, Alex Wonhas (CSIRO). Concentrating solar power drivers and opportunities for costcompetitive electricity. March 2011 [15] Michael J. Wagner. Simulation and Predictive Performance Modeling of Utility-Scale Central Receiver System Power Plants. University of Wisconsin Madison 2008. [16] www.mem.dz, guidelines of renewable energy in Algeria. Edition 2007.
Vibration Suppression in Elastic Drive Systems Using the Immersion and Invariance Methodology
Irfan Ullah Khan, Rached Dhaouadi
Mechatronics Engineering Graduate Program American University of Sharjah Sharjah, UAE b00042370@aus.edu, rdhaouadi@aus.edu
AbstractThis paper presents a new robust control methodology to suppress the torsional vibrations in exible drive systems. The position control of a nonlinear Two-Mass-Model system (2MM) is designed using the Immersion and Invariance approach. First, appropriate mapping functions are derived to convert the total nonlinear 2MM system into an equivalent reduced order system with rigid dynamics. This reduced order system is used as a target system to design a position controller, which is based on the standard PI type control technique. Next through immersion and invariance, the reduced order controller is applied to the nonlinear 2MM system to suppress the torsional vibrations and yield an over-damped response similar to the target system. The control law derivation and stability analysis of the target system are described and discussed. Simulation and experiments using a 2MM drive system are used to validate the proposed control methodology. Index Termstorsional vibrations, immersion and invariance, invariant manifold, two mass model, stabilization
I. I NTRODUCTION The design of high-performance motion control system requires accurate knowledge of the electromechanical dynamics, including linear and nonlinear transmission attributes of the systems. Flexibility in the mechanical transmission elements is inevitable in mechanical systems and is found to be undesirable by control system designers because it is often responsible for causing mechanical vibrations, which deteriorates the performance and restricts the closed-loop bandwidth. Combining friction with compliance makes it difcult to achieve highprecision position and speed control in pointing and tracking servomechanisms. Mechanical vibration suppression has been treated from many different points of view, including static and dynamic compensation models with feedback and feed-forward compensation [1][5] . Most of the techniques used rely on a two-massmodel (2MM) system, i.e., a model in which the actuator and load are coupled by a exible shaft and friction depends only on the relative velocity of the motor and load. These models may be sufciently accurate for systems in which relative velocity is almost always extremely small, such as in rolling mill drive systems or multi-axes exible joint manipulators. The parameters of the 2MM system are thus required. Nonlinear and soft computing control techniques also have gained attention in torsional control [6], [7] and speed control [8] of 2MM drive system. Fuzzy control in [6] increases the
robustness to parameter variation. In industry soft computing control techniques are not that much popular. In [9] a survey is made on immersion and invariance and its applicability is discussed via some examples and the method is also extended to feedback linearization where design of observer is typically required. In [10] controller is designed based on immersion and invariance for active suspension system and then results are compared with back stepping control law. Simulation results show that immersion and invariance controller can stabilize the full-order system as well as the back stepping controller in the nominal case, but is more robust to some parameter changes in the system and moreover, when there is an unknown parameter, the adaptive immersion and invariance controller gives closer response to the known parameter case than the adaptive back stepping controller. In [11] immersion and invariance is used to deal with the stabilization problem of magnetic suspension and this methodology has proven to be a very effective tool for the stabilization of nonlinear systems and its main advantage is making the closed system behave asymptotically like a target system. Experimental results illustrate the effectiveness of immersion and invariance in comparison with different additional studied controllers. In [12] immersion and invariance is used to design controller for antagonistic joint with nonlinear stiffness and then immersion and invariance controller was compared with simple PD controller. The immersion and invariance controller performed very well even in the presence of parameters uncertainties. In [13] a more detailed survey is made on immersion and invariance method with more examples from real life systems. In [14] immersion and invariance is used for transient stability and voltage regulation of power system with unknown mechanical power. The authors of the above publications, [9][16] however, do not elaborate on the derivation of the immersion and invariance control methodology, particularly the mapping between the original nonlinear system and the reduced order target system. Some authors show the immersion and invariance control performance using computer simulation analysis, but they do not address the practical aspects of implementation. In this paper the immersion and invariance control approach will be used to design a robust position controller for a singleaxis exible joint manipulator. The controller will suppress the torsional vibrations and yield an over-damped response.
The system is modeled as a two-mass-model system driven by a dc motor with a hysteresis current torque control inner loop. A prototype experimental system is built to validate the developed control algorithm. The detailed procedure of the immersion and invariance method is presented and the main steps are explained. First, a target system is dened, whose order is strictly smaller than the order of the nonlinear system to be controlled. Next, the mapping between the original system and the target system is dened. Then, the invariant manifold is constructed such that the restriction of the system dynamics to this manifold coincides with the target dynamics. Finally, a control law is designed that renders the manifold attractive and ensures that all signals are bounded. II. I MMERSION AND I NVARIANCE M ETHODOLOGY The theorem proposed in [15] to design a controller using immersion and invariance presents a set of sufcient conditions for the construction of globally asymptotically stabilizing, static, state feedback control law for general nonlinear systems. Consider the system described by, x = f (x) + g (x)u (1)
Fig. 1. Two Mass Model
with state x Rn and control, u Rm with an equilibrium point x Rn to be stabilized. Let p < n and assume we can nd a mapping ( ), such that the following holds. (H1) (Target System) The System = ( ) (2)
Fig. 2. Rigid Dynamics of 2MM
with state, Rp has an asymptotically stable equilibrium at Rp and x ( ) (H2) (Immersion Condition) For all Rp f ( ( )) + g ( ( )) c ( ( )) = ( ) (3)
III. S YSTEM U NDER C ONSIDERATION The system under consideration is a single-link exible joint manipulator. It is modeled by a two mass model system as shown in Fig. 1 . The dynamic equations of the two mass model are Jm
d 2 m dt2 =
2
(H3) (Implicit Manifold) The set identity holds {x Rn | (x) = 0} = {x Rn |x = ( ) , Rp } (4)
m l m K ( m l ) C (7)
(H4) (Manifold attractively and trajectory boundedness) All trajectories of the system x = f (x) + g (x) (x, z ) are bounded and satisfy
t
d l m l l Jl dt2 = K ( m l ) + C
(5)
lim z (t) = 0
(6)
If the target system has asymptotically stable equilibrium at the origin then the manifold will be attractive because every asymptotical stable equilibrium point is attractive. The control law that will be derived at the end will have two parts. First part will bring off the manifold dynamics to the manifold and once all the dynamics are on the manifold then the control law that was derived for the target system will keep all the dynamics bounded to the manifold.
where, Jm = Jmotor + JM disk and Jl = JLdisk + Jlink l is the load position, m is the motor position, , Jm is the motor side inertia, Jl is the load side inertia, K is the shaft stiffness, C is the damping constant, m is the motor torque, and l is the load torque. l , x3 = m , x4 = m as the state Dening, x1 = l , x2 = variables of the system and u = m , the system equations can be written as follows x 1 = x2 d 1 x 2 = [K (x3 x1 ) + C (x4 x2 ) M g sin x1 ] Jl 2 x 3 = x4 1 x 4 = [u K (x3 x1 ) C (x4 x2 )] (8) Jm
The system can be represented as (1),

x = x2 1 d Jl [K (x3 x1 ) + C (x4 x2 ) M g 2 sin x1 ] + x4 1 [ K ( x x ) C ( x x )] 3 1 4 2 Jm 0 0 0
1 Jm
( ) =
1 2 3 ( 1 , 2 ) 4 ( 1 , 2 )
(17)
IV. C ONTROLLER D ESIGN The target system is chosen in terms of rigid body dynamics of 2MM with damping term as shown in Fig. 2 . = 1 2 1 = ( ( , )) 2 1 2 Jt
where 3 ( 1 , 2 ) and 4 ( 1 , 2 ) are functions that need to be constructed. The target system is part of the original system i.e. x1 = 1 and x2 = 2 x 2 = 2 1 d 1 [K (x3 x1 ) + C (x4 x2 ) M g sin x1 ] = ( 1 , 2 ) Jl 2 Jt [K ( 3 1 ) + C ( 3 2) M g d 2 sin 1 ] =
Jl Jt ( 1 , 2 )
(9)
where, ( 1 , 2 ) = b1 sin 1 b2 2 cos 1 + u and Jt = Jm + Jl if u = 0, then the rigid body dynamics becomes the dynamics of single pendulum and (9) becomes, = 1 2 = E ( ) R( ) 2 1 2 1 where, E ( 1 ) = b1 sin 1 Jt b2 cos 1 Jt (11)
(18)
The implementation of PI controller on target system is shown below in Fig. 3 . (10)
R( 1 ) =
(12)
E and R( 1 ) is damping function To where, E ( 1 ) = 1 analyze the stability of the target system, a Lyapunov function is dened in terms of the total energy of the system as given below,
Fig. 3. Block diagram for PI Controller Implementation
1 2 + E ( 1 ) (13) 2 2 where E is potential energy of the target system, The target dynamics will have an asymptotically stable equilibrium at the origin if the following conditions are satised, V = V (0, 0) = 0 V ( 1 , 2 ) > 0, in D {0} , ( 1 , 2 ) < 0, in D {0} , V DR
p
u = (Ki u = (Ki
edt + Kv )ev edt + Kv )(ep Kp 2 ) edt Kv 2 Ki 1 (19)
u = Kv Kp e + Ki Kp
where, ev = velocity error and e = ep = position error. Introducing a PI controller as shown in Fig. 3 gives: (14) ( 1 , 2 ) = b1 sin 1 b2 2 cos 1 + K1 e + K2 K3 2 K4 1 where, K1 = Kv Kp , K2 = Ki Kp , K3 = Kv , K4 = Ki The mapping equations become, (16)
gd 3 = M 2K sin 1 + 1 + a1 e + a2
edt (20)
For asymptotically stable equilibrium of target system at origin (13) becomes, V = 2 b1 + [1 cos 1 ] 2 Jt = R( 1 ) 2 2 V
2
(15)
As it can be clearly seen that rst condition in (14) is satised. For second condition in (14) to be satised, b1 should be positive. For third condition in (14), R(0) should be greater than zero and for that, b2 should be positive. (H2) Immersion condition in the theorem will give us the mapping,
edt + a3 2 (21)
+a4 1
gd 4 = 2 M 2K cos 1 + 2 a1 2 + a2 e + a3 2
+a4 2
(22)
Substituting ( 1 , 2 ), 3 , 4 in (18),
gd [K M 2K sin 1 + a1 e + a2
V. E XPERIMENTAL S ETUP Fig. 4 shows the two mass model prototype. The system was designed to have mechanical resonant frequency fr = 18.57 Hz because most of the industrial elastic drive systems have mechanical resonant frequency in the same range. Parameters in Table 1 are selected in a way to keep the load to motor inertia ratio R = Jl /Jm = 1.209 for maximum power transmission. High resolution encoders with 40,000 ppr are mounted on the motor and load to measure the position and speed signals. The dSPACE system is run with a sampling time Ts = 0.5 ms. The current/torque control loop is implementation in hardware using a hysteresis current controller board. Buhler DC motor is used as an actuator, which can give a rated torque of 0.6 Nm and rated current of 12 amperes.
edt + a3 2 + a4 1
gd +C 2 M 2K cos 1 a1 2 + a2 e + a4 2
M g d 2 sin 1 ] =
Jl Jt
Ca3 Jl
[b1 sin 1 (23)
b2 2 cos 1 + K1 e + K2
edt K3 2 K4 1 ]
Solving (23) to nd the mapping coefcients,

Jl K (CK2 KK1 ) a1 = Jt K 3 C 3 K2 + C 2 KK1 CK 2 K3 + C 2 KK4
a2 =
Jl K K2 Jt K 3 C 3 K2 +C 2 KK1 CK 2 K3 + C 2 KK4
Jl (C 2 K2 + K 2 K3 CKK1 CKK4 ) a3 = Jt K 3 C 3 K2 +C 2 KK1 CK 2 K3 + C 2 KK4
a4 = b1 = b2 =
Jl K 2 K4 Jt K 3 C 3 K2 +C 2 KK1 CK 2 K3 + C 2 KK4 M gd(Jt K 3 C 3 K2 +C 2 KK1 CK 2 K3 + C 2 KK4 ) Jl K 3 CM gd(Jt K 3 C 3 K2 +C 2 KK1 CK 2 K3 + C 2 KK4 ) 2J l K 4
Fig. 4. Experimental Setup
In our case b1 and b2 are positive. Hence it is proved that for the above mapping, target system has asymptotically stable equilibrium at the origin. (H3) Manifold can be described by M = {x R4 , (x) = 0} 1 (x) = x3 3 (x1 , x2 ) 2 (x) = x4 4 (x1 , x2 ) (24)
Table 1 show the parameters of the 2MM system.
Table.1. 2MM Parameters
(H4) The off the manifold coordinates are z1 = 1 (x) and z2 = 2 (x) z 1 = z2 z 2 = x 4 4 (x1 , x2 ) v = k2 z2 k1 z1 (25) The controller for off the manifold dynamics is given by v (26)
VI. S IMULATION R ESULTS The implementation of the I&I controller on the 2MM system is rst veried using Matlab/Simulink as shown in the block diagram in Fig. 5 .
v will bring the error dynamics to the manifold. k1 and k2 are chosen in such a way that, s2 + k2 s + k1 is Hurwitz. The controller equation becomes, (x, z ) = J m [k1 z1 k2 z2 M gd M gd 3K a32 +(x2 2K sinx1 a2 aJ Jl cosx1 )x2 l M gd a3 C +( 2K cosx1 + 1 a1 Jl + a4 ) 1 (J (b1 sin x1 x2 b2 cos x1 + u)) t +K (x3 x1 ) + C (x4 x2 )] (27) As it has been proven that target system has asymptotically stable equilibrium at the origin and every asymptotic stable equilibrium point is attractive. Once v brings the error dynamics to the manifold then, the system will behave similar to the target system. The PI controller comes into effect and will keep all the dynamics bounded to the manifold.
Fig. 5. Block Diagram of Simulink Implementation
Fig. 6 and Fig. 7 shows the load position and motor position respectively, of the PI-controlled target system in response to a step reference. This response is compared to the PI-controlled 2MM system with the I&I controller and without I&I control. The results clearly show with the I&I controller the 2MM system achieves an over-damped response similar to the target system.
fast z1 and z2 goes to zero in our case. VII. E XPERIMENTAL R ESULTS The experimental results are shown below for target system, two mass model and two mass model with immersion and invariance. Fig. 10 and Fig. 11 shows the load position and motor position in real time respectively, of the PI-controlled target system in response to a step reference. The real time results are similar to the simulation results.
Fig. 6. Load Position Simulation
Fig. 10. Load Position Experimental
Fig. 7. Motor Position Simulation
Fig. 11. Motor Position Experimental
Fig. 8. Error Dynamics (z1 ) Simulation
Fig. 12 and Fig. 13 shows how fast z1 and z2 goes to zero in real time.
Fig. 9. Error Dynamics (z2 ) Simulation
Fig. 12. Error Dynamics (z1 ) Experimental
How fast error dynamics goes to zero depends on the selection of k1 and k2 in v . Fig. 8 and Fig. 9 shows how
To check the robustness of the controller to disturbances and parameter uncertainty, the load inertia is increase by 30% and an additional 25% of load torque is introduced. The controller
Fig. 13. Error Dynamics (z2 ) Experimental
in the outer loop to control the load position and in the inner loop hysteresis current controller is implemented to control the current/torque. The settling time that is achieved in the experimental results is 0.4 seconds. As it can be seen clearly that both in the simulation and experimental results, the torsional vibrations are suppressed and the response of controller to disturbances and parameter uncertainty is robust. The 2MM without immersion and invariance implementation goes unstable. The selection of the target system in this method is one of the most important task and also a non-trivial one. In the same way different mappings for the same target system can also be achieved. R EFERENCES
was also tested with a 50% load torque increase. The results are shown in Fig. 14 and Fig. 15. Controller shows very robust response to disturbances and parameter uncertainty.
Fig. 14. Comparison of Experimental Load Position with and without disturbances
Fig. 15. Comparison of Experimental Motor Position with and without disturbances
VIII. C ONCLUSION In this paper immersion and invariance methodology is applied to damp the torsional vibration and to design a robust position controller for a single-axis exible joint manipulator. The controller design via immersion and invariance is explained in detail. The rigid body dynamics are chosen as the target system and asymptotically stability is derived using Lyapunov function. The PI controller is implemented
[1] K. Szabat and T. Orlowska-Kowalska, Vibration suppression in a two-mass drive system using pi speed controller and additional feedbackscomparative study, Industrial Electronics, IEEE Transactions on, vol. 54, no. 2, pp. 11931206, 2007. [2] M. A. Valenzuela, J. M. Bentley, and R. D. Lorenz, Evaluation of torsional oscillations in paper machine sections, Industry Applications, IEEE Transactions on, vol. 41, no. 2, pp. 493501, 2005. [3] R. Dhaouadi, Torque control in harmonic drives with nonlinear dynamic friction compensation, Journal of Robotics and Mechatronics, vol. 16, pp. 388396, 2004. [4] J. Pacas, A. John, and T. Eutebach, Automatic identication and damping of torsional vibrations in high-dynamic-drives, in Industrial Electronics, 2000. ISIE 2000. Proceedings of the 2000 IEEE International Symposium on, vol. 1. IEEE, 2000, pp. 201206. [5] S. N. Vukosavic and M. R. Stojic, Suppression of torsional oscillations in a high-performance speed servo drive, Industrial Electronics, IEEE Transactions on, vol. 45, no. 1, pp. 108117, 1998. [6] T. Orlowska-Kowalska and K. Szabat, Damping of torsional vibrations in two-mass system using adaptive sliding neuro-fuzzy approach, Industrial Informatics, IEEE Transactions on, vol. 4, no. 1, pp. 4757, 2008. [7] K. Itoh, M. Iwasaki, and N. Matsui, Optimal design of robust vibration suppression controller using genetic algorithms, in Advanced Motion Control, 2002. 7th International Workshop on. IEEE, 2002, pp. 8691. [8] R. Dhaouadi and K. Nouri, Neural network-based speed control of twomass model system, J. Adv. Computat. Intell. v3 i5, pp. 427430, 1999. [9] D. Karagiannis, A. Astol, and R. Ortega, Nonlinear stabilization via system immersion and manifold invariance: survey and new results, Multiscale Modeling & Simulation, vol. 3, no. 4, pp. 801817, 2005. [10] P. Santhanapipatkul and W. Khovidhungij, Nonlinear controller design for active suspension systems using the immersion and invariance method, in World Congress, vol. 16, no. 1, 2005, pp. 19341934. [11] E. Hristea, H. Siguerdidjane et al., Stabilization of a magnetic suspension by immersion and invariance and experimental robustness study, in Proceedings of 18th IFAC World Congress, 2011. [12] T. Wimbock, C. Ott, and G. Hirzinger, Immersion and invariance control for an antagonistic joint with nonlinear mechanical stiffness, in Decision and Control (CDC), 2010 49th IEEE Conference on. IEEE, 2010, pp. 11281135. [13] A. Astol, D. Karagiannis, and R. Ortega, Towards applied nonlinear adaptive control, Annual Reviews in Control, vol. 32, no. 2, pp. 136 148, 2008. [14] W. Dib, G. Kenn e, and F. Lamnabhi-Lagarrigue, An application of immersion and invariance to transient stability and voltage regulation of power systems with unknown mechanical power, in Decision and Control, 2009 held jointly with the 2009 28th Chinese Control Conference. CDC/CCC 2009. Proceedings of the 48th IEEE Conference on. IEEE, 2009, pp. 78377842. [15] A. Astol and R. Ortega, Immersion and invariance: a new tool for stabilization and adaptive control of nonlinear systems, Automatic Control, IEEE Transactions on, vol. 48, no. 4, pp. 590606, 2003. Acosta, R. Ortega, A. Astol, and I. Sarras, A constructive solution [16] J. A. for stabilization via immersion and invariance: The cart and pendulum system, Automatica, vol. 44, no. 9, pp. 23522357, 2008.
Robust Sliding Mode Observer/Controller of Wind Energy Conversion System Based on DFIG Fed by a Matrix Converter
A DJOUDI Centre de dveloppement des nergies renouvelables. Ecole nationale polytechnique Algiers, Algeria Email : abdelhak.djoudi@yahoo.fr H CHEKIREB and EM BERKOUK Ecole nationale polytechnique Algiers, Algeria Email : chekireb@yahoo.fr emberkouk@yahoo.fr control methods [3, 4 and 5], where some existing interactions are simply ignored. This VC control exhibits low performances and low robustness compared to those of control methods based on the DFIG nonlinear model. Two main methods are proposed. One is based on the optimal switching table of the switches states related to the rotor side converter where, the control errors of the stator powers are minimized using the DFIG electrical states [6, 7]. This method is valid only in the case of the conventional DC/AC converter, and requires complicated online calculations and it displays oscillations when the generator operates in the vicinity of its synchronous speed [8] beside, the balanced rotor voltage is never guaranteed. The second method concerns the sliding mode control based on DFIG coupled model [8]. In this article, in order to simplify the control design, the controller is derived in the stator reference frame. This latter is independent of the angular position of vector related to some electrical quantity which is essential when a synchronous coordinate transformation is used. Meanwhile, the estimated rotor speed is provided by the derivation of the rotor angle which is measured via a sensor encoder. It is known that this procedure can conduct to a false estimation due to noisy measures. Moreover, the classical method [1-8] applied to estimate the stator or rotor flux, incorporating an integral function, displays high sensibility face to stator or rotor resistances variations. So, the estimated flux can deviate greatly from its real value. To overcome these drawbacks and keep the same control objectives, the sliding mode method is applied, in our work, to derive the stator powers control and to estimate the rotor flux. Indeed, the control law and the observer are developed based on the nonlinear and coupled DFIG model. This model is computed in the synchronous reference frame linked to the stator voltage vector. The rotor position (rotor angle) is deduced by integrating of the rotor speed which is measured directly via a speed sensor. Noticing that in our approach, the number of sensors is reduced since the rotor current sensors are unnecessarily contrary to those reported in [1-8] where stator and rotor current sensors are required to compute the control law. The back-to-back converter is widely used for connecting the rotor to the power grid, and generates the rotor voltage
Abstract--We develop in this paper a sliding mode control of the stator powers of the doubly fed induction generator (DFIG), which is used within a wind energy conversion system (WECS). The elements of the state vector, used for control, are assumed available via measurements except the components of the rotor flux, which are estimated by a sliding mode observer. No rotor current sensor is required. The controller and the observer are involved from the DFIG nonlinear and coupled model. The global stability of the sliding mode observer/controller (SMOC) is treated. Our approach is tested in the case of 1.5 MW WECS rating based on DFIG where its rotor voltage control is done through a matrix converter. The stator active power reference is given to extract the maximum of the wind power available. The stator reactive power reference is given to impose the reactive power at the point of the common coupling (PCC). The robustness and the performances analysis were done for the case of an unbalanced grid, within the presence of disturbance and parameter variations. The proposed method keeps the steady-state harmonic spectra at the same level as the vector control (VC) strategy. The satisfying obtained results have allowed validating the proposed method. keywords--DFIG, Matrix Converter, MPPT, Robustness, Sliding Mode Observer/Controller Wind Energy Conversion System (WECS).
I. INTRODUCTION Because of its economical and environmental advantages, many countries around the word, with important power capacity pay more and more attention and interest of the utilization of wind energy [1]. During last years, many wind farms are based on the doubly fed induction generator (DFIG) technology due to its advantages compared with others generators mainly in the reducing of the dimension of the static converter (AC/AC) to about 25%-30% of the generator rating [1,2], this implicates lower converter costs and lower power losses. Several research works present the increase of the exploitation efficiency of wind chain based on DFIG on active and reactive stator power control. We can distinguish a large family of control algorithms based on DFIG decoupled model to simplify its synthesis such as vector
needed for the control of DFIG [1-8]. However, the back-toback converter presents high switching loss and bulky structure because it has 3-step power conversion such as AC-DC-AC. In order to improve this weak point, the matrix converter was proposed in the case of vector control [9, 10]. Our study proves the feasibility of the matrix converter with the nonlinear sliding mode Observer/Controller that we will present. The rest of the paper is organized as follows: in the 2nd section, the considered WECS. The sliding mode control (SMC) of the stator powers and the sliding mode observer (SMO) of the rotor flux are developed in the section 3. The global stability of the proposed SMOC is treated in section 4. The simulation results and their analysis, related to the considered WECS based on DFIG of 1.5MW power rating, are given in section 5. I. THE CONSIDERED WIND ENERGY CONVERSION SYSTEM A. Preliminarily Considerations The scheme of the utilized WECS is given in fig.1, the main generator is a doubly fed induction generator (DFIG) where the stator is directly connected to the grid and the rotor windings are also coupled to the network through a matrix converter. Under the effect of the wind, the turbine produces the torque on its shaft which rotates at the speed . Due to the effect of the multiplier, with coefficient , the speed related to the shaft generator is amplified and the torque on the shaft generator is reduced as that: = ; = / (1)
;
DFIG
'%&
8 =G $%&
AC Grid
, :;<=> ?@= AB?;CDE@ DFD calculation
'() ,
abc/dq
abc/dq
89
MPPT
SMO 6 7 () Sliding Mode Control SMC
Matrix Converte r
$() P dq/abc W M $%&
$() =
AC Grid Fig.1. Scheme of the considered Wind Turbine System
The torque turbine Can be calculated as follows: = = / , = (2) (3)
The components of the DFIG rotor flux *+ ,( , + ,) - are estimated using sliding mode observer. Moreover, the DFIG stator powers ( , are deduced from measurements of stator currents and voltages, and the speed of the generator shaft is also measured. The SMC syntheses the desired rotor voltage source $( , $) corresponding in three phases form $% , $& , $ which are utilized in the PWM bloc to generate the control impulses in view to drive the matrix converter. In the section 3, we propose a SM controller and an observer of the DFIG rotor flux. B. DFIG State Space Model In view to control the DFIG, its state space model is carried out in the !, " reference frame linked to stator vector voltage which is represented by the components ( $ ( , $ ) . The considered state vector gathers the rotor currents ' ( , ' ) , the rotor flux + ( , + ) , and the rotor rotating pulsation . Moreover, the rotor voltage components $ ( , $ ) form the vector control and the wind turbine exerts on the generator shaft a torque and all frictions on this shaft are represented by the ./ torque. 0 = 2 , 2 , 2 , 23 , 24 = *'
(
Where: , and are respectively the air density, the wind turbine radius and the ratio of the blade tip speed to the wind speed . Moreover, the coefficient is related to the characteristic of the wind turbine and it depends on the ratio and the inclination of the blades . An MPPT algorithm is exploited to detect the stator active power reference necessary to maintain the coefficient at its maximal value, in order to extract the maximum power from the wind turbine. The stator reactive power reference is given to get the desired power factor at the point of common coupling (PCC) where the DFIG is connected to the AC grid. We intend to design a sliding mode controller (SMC) in view to force the stator power ( , to tracks their respective references ( , . In our case, the control law is computed using electric variables referred to the synchronous reference frame !, " # We assume that the AC and the DFIG stator current grid voltage $% , $& , $ '% , '& , ' Are measured and projected on the !, " axes to obtain respective $( , $) and '( , ') .
Therefore, the state vector 0 and the control vector 1 are respectively given by: 1 = *$
5
, $ )-
, ' ), +
5
, + ) , - and
5
By using our notation, the dynamic model of the DFIG is as follows: 2L = NM 0 O PQ $R! ; NM = SPM 2M O T 2U O PU 2Q O K M SPQ 2V 2W S PW $T! I 2 L = N 0 O P $ Q R" ; NU = S T 2M S PM 2U O PU 2W O I U U OPQ 2V 2Q S PW $T" [
J 2L Q = N 0 O $R! ; N = SX2M O T 2W S 2V 2W Q Q I 2L = N 0 O $ ; N = SX2 S 2 O 2 2 R" U T Q V Q W W I W H 2L V = NV 0 O YU $'T O Z ; NV = YM 2W 2M S 2Q 2U
(8) Where the positive coefficients of the system (8) are given by: s =
]5^
L K I L I 2L J2L 3 I I H2L 4
= Ss = = =t =t
uv
(v
Ss O s 2 O s 24 23 O s3 $ St 2 O S 24 23 O $ ( S t 23 S S 24 2 O $ ) S 23 OY
./
S s 24 2 O s 23 O s3 $ ) S s $
(
Ss $
(4)
Where the positive coefficients of the differential system (4) are given by: P =\ X=
]5^
]5_
Considering the equilibrium grid and as the " axis is considered along the stator voltage vector so, the stator active and reactive powers of DFIG are expressed by: l = S! 2 O ! 23 [ = S! 2 O ! 2 ! =
ac a_
And they are related to the following machine parameters: , ; rotor and stator resistances. h , h , hi ; rotor and stator and mutual inductances. ; number of pole pairs. j; rotor turbine inertia. k , k ; rotor and stator time constant. ; dispersion coefficient.
, Y =
`, P =
de f
, Y =
]a_5^
, P =
d f
,g = MS
]a_
, P3 =
a^ a_ ac
e
]ac
b]
The sliding surface vector w = w , w 5 is taken as the tracking control error of the output vector: w = l w = S S [ L (9) and and its are bounded and
]5_ b] a_ ]ac e
b]
;s =
$ ); t =
]ac
b]
$);s =
5_.^m ac
;t =
]ac 5_ 5_
b]
$ ) ; s3 =
Using (9), the dynamic of the sliding surface w is then given by: n With: K I L S L wL = L S L wL = L S L S [ (10)
In the sequel, we assume that L temporal derivative and available.
(5)
.^m ac
With:
$ ), ! =
(6)
J L S L I H Or yet:
= Ss =
Ss
O s 2 O s 24 23 O s3 $ Ss $ ( S L
S s 24 2 O s 23 O s3 $ ) O Ss $ ) S L
(
(11)
III. SLIDING MODE CONTROLLER AND OBSERVER A. Sliding Mode Controller This section is devoted to the control of active and reactive stator powers where we intend to develop a sliding mode control (SMC) based on the DFIG nonlinear model (4). Our objective is to force the outputs and tracking their respective references and . From the DFIG state model, we derive the following control model: n
5 oL = p o, 1 ; o = * , , + ( , + ) , 24 - , 1 = *$ 5 q = r o ;q = , (
With:
, $ )- [
5
(7)
We force the dynamic wL , wL follows: n w L = Sy T'pz w w L = Sy T'pz w
x = x o = S Ss S Os3 $ ) S L Ss Jx =x o =S I Os3 $ ( S L H K I
wL = Sx S s $) [ wL = Sx S s $(
(12) S s 24 2 O s 23 O [
O s 2 O s 24 23 O
(13)
of the sliding surfaces as SZ w [ SZ w
From relations (4) and (5), the state model based on the state vector o can be derived as:
(14)
Where y , y , Z , and Z are positive control gains. In the eyes of the relation (12), the condition (14) can be fulfilled if the sliding control law is taken as:
|v } b~^v / v b ^v v be [ |e } b~ê / e b ê e be
(15)
robust sliding mode observer is synthesized in the following paragraph. We intend to construct a rotor flux observer, inspired from [11], based on the measurement of the stator voltage, stator current and the rotor speed. The estimated flux components 2 , ,2 ,3 are used in the computation of the SMC law. Based on the state model given in (8), we propose the sliding mode observer given in (20) where the first sub-system is related to the dynamic of stator powers and the second one concerns the stator flux where the positive coefficients h v , h e h v , h e , h v and h e are observer gain. K I L = Ss S S s 24 2 , Os 2 ,3 O s3 $ ) S s $ Oh T'pz* v
At this level, we consider the Lyapunov function: = w5w It's time derivative is: L becomes: L = wL 5 w (16)
(17)
By using (14),
L = S / y / w/ O Z / w/ L , w/
(18)
Therefore:
As the components of the DFIG rotor flux + ( , + ) , are not easily measured so, in our case we intend, in this following section, to construct a sliding mode observer able to restitute the flux vector. B. Drawback of Classical Flux Estimation Method & Sliding Mode Observer Synthesis Based on stator or rotor current '%& , and stator or voltage $%& , measures, the classical method [1-8] used to estimate the stator or rotor flux is based on: + , %&
,
Which means that w/ converges asymptotically toward zero and track their respective references and the output and
(19.a) K I
We introduce the states observer errors , , 2 and 2 3 as that: = S , =
,L3 = t J2 I H
2 ,L = t
J L = I H
Ss St 2 , O
Os 2 , O s 24 2 ,3 O s3 $ Oh e T'pz* S 24 2 ,3 O $ Oh T'pz* e e
Ss $
) (
O [
St 2 ,3 S
S 24 2 , O$ Oh T'pz* -
( )
Oh
Oh
v v
T'pz* - O
T'pz* - O
(20.a) [
(20.b)
S ,2 =2 S2 , ,2 3 = 23 S 2 ,3 (21)
Where , is the nominal value of stator or rotor resistance. Due to the error modeling and parametric variations during the function of DFIG, the real value of the stator or rotor resistances become ,, % , and the real stator or rotor flux +%& ,, will be: +%&
,
$%&
'%&
(19.b)
And so, the dynamic of observed errors of the subsystem (20.a) and (20.b) are given by: L = Ss 24 2 Os 2 3 S h v T'pz* [ L = s 2 Os 2 2 S h T'pz* 4 3
e
(22.a)
=G $%&
,
% '%&
!
,
(19c)
The estimation stator or rotor flux error is: = +%&

,
S+ , %&
,
= SG
,
S
,
'%&
%
! (19.d)
,
2 L = 2 L3 S 24 St 2 S 2 S S 24 St 3 h v h e T'pz* h v h e T'pz* = O
(22.b)
For first subsystem (22.a), we consider the following Lyapunov function: (23)
Then: = G
'%&
! ;
=S
S , (19.e)
Its time derivative is then given by: L = L O L (24)
Due to the integral function in (19.e), we conclude that a few deviation ,, Of the stator or rotor resistance real value of its nominal value involves a great value of the stator or rotor flux estimation error. The estimator of the stator or rotor flux (19.b) is then very sensible to the stator or rotor resistance variation. To overcome this weakness point, a
Where:
L = Ss 24 2 Os 2 3 S h v [ [ [ L = s 2 O s 24 2 3 S h e [ [
(25)
If the observer gains h v and h e are forced as that: Since the coefficients s ,s ) are positive and the observer errors are assumed bounded as that: P2 [2 + [ and [ [ P2 2 3 + , so the relation (26.a) can take the form: With P2 [24 i% [ and i% is maximal admissible speed of the DFIG Therefore: L , and (27) Which means that the observer errors , and their dynamics respectively ( L , L ) Converges asymptotically to zero, and the system (22.a) is reduced to: n Os 2 3 S h v T'pz* = Ss 24 2 [ =s 2 O s 24 2 3 S h T'pz* e
h v P2 [Ss 24 2 Os 2 3 [ [ [ h e P2 s 2 O s 24 2 3 [
Since the determinant =C ww (26.a)
always negative hence, the matrix ww ) is invertible which means that the gains observer can be always determinate. The convergence of the estimated rotor flux to their real value is investigated based on the Lyapunov function: L = O2 3 - , by exploiting (32) one has = *2
a^v aê
*s
O s - is
h h
v e
s s
i% Os i% Os
+ [ +
(26.b)
S 2 O 2 3 , 2 ,2 3 , . This implicates that the estimation rotor flux errors converges to zero. Thus, the differential system (20) constructs a sliding mode observer of the rotor flux if the conditions (26) and (33) are satisfied.
IV. GLOBAL STABILITY STUDY In this section, we establish the stability of the SMOC even in the presence of observation errors, modeling errors, parametric variations and bounded disturbance. Using the estimated states 2 , ,2 ,3 in place of 2 , 23 so, becomes as: the used estimate state vector o = , ,2 , ,2 ,3 5 , and the control law (15) is computed o as: { $ $
)
(28)
So, when the first-subsystem (22.a) is in sliding mode, T'pz* - and T'pz* - are equivalent to: T'pz* T'pz* )
With,
ww
s 24 h v S s h e
2 = ww ) 2 3
(29)
S s h v S s 24 h e
(30)
Despite the system (31) depend on the speed variable 24 , it can be considered as an invariant time linear system since the flux dynamic is fast compared to the speed one. In this context, the observer gains h v , h e , h v and h e can be set in order to force the dynamic (31) under the form: 2 L S = 2 L3 2 2 3 S (32)
By exploiting (30), the dynamic (22.b) of the flux observed errors is reduced to: 2 L = 2 L3 h h e St S 24 2 S v ww ) h v h e 2 3 S S 24 St (31)
Since the real values of x and x can be rewritten under the form: n x x
% %
Where: - = S s S x *o K I SP3 ! $ ) S - = S Os Jx *o I SP3 ! $ ( S H
b~^v / v b ^v v |v } ( [ b~ê / e b ê e |e } (
(34)
Os 2 , Os 2 ,3 O L
Os 2 , Ss 2 ,3 O L
(35)
Noticing that the variations x and x include the effects of estimated errors, the effects of the parametric variations, modeling errors and bounded disturbances. Based on (12) the sliding surfaces dynamic expressions wL and wL in real conditions are given by: n wL = Sx wL = Sx
% %
- S x = x *o [ - S x = x *o
(36)
O ! $(
O ! $)
(37)
By substituting the control law (34) in the expression (37) of the sliding surface conduct to: n wL = x S y T'pz w S Z w [ wL = x S y T'pz w S Z w (38)
Where and are positive coefficients therefore, these observer gains can be deduced as: h h e St O S 24 v = ww ) b (33) h v h e S 24 St O S
With the dynamic (38), we join the Lyapunov function 5 L5 L = w w its time derivative is = w w and by using L becomes: (38),
If the coefficient y y
= w x S y w O w x S y w S Z w O Z w (39) P2 x ; y and y are imposed as: P2 x
(40) Therefore L , w and w , which means that w and w Converges asymptotically to zero. This means that the sliding mode is maintained. The global stability and robustness of the SMOC synthesized is then proved. V. SIMULATION RESULTS To carry out the simulations of the considered WECS shown in fig.1, the DFIG is rated at M#V- and its parameters are given in appendix I. The AC grid, the matrix converter and the mechanical wind turbine parameters are specified in the appendix 2. The sub-envelope modulation [12] method used as Pulse Width Modulation (PWM) to generate the output voltage desired of the matrix converter. The starting step of the WECS is not considered, moreover the DFIG is assumed properly magnetized. The SMOC gains are given in appendix 3. A. Comparative Study The advantages of SMC compared with VC are reported in [8]. Contrary to [1-8] that uses all DFIG current sensors and sensible flux estimations based on (20), the main benefits of the developed SMO are on the robustness of the rotor flux estimation, and on the suppression of the rotor current sensors. So, we are interested in the present paragraph to testing the performances of the SMOC and verify its feasibility with the matrix converter. Two simulations have been carried. One is based on a VC method that the flux oriented control applied in [9], which the DFIG
2 0 -2 0
KA
flux are assumed available, and any parametric variations or disturbances are considered. The second simulation is carried out with SMOC where the stator and rotor resistances parametric variations are taken with stator voltage unbalanced. Assume that the DFIG is governed by a fixed speed MWRP!/T. The fig.2 illustrates the simulation results of the DFIG in the case the SMOC. Face to different stator and rotor resistances parametric variations percents (fig.2.e), parametric variation with +50% of the all DFIG inductances, and unbalanced voltage grid with U during #QT and 0.42s as shown on fig.2.f, the stator active and reactive powers tracking theirs references with bounds errors acceptable as figured on fig.2.a,b. Moreover the stator and rotor currents figured respectively on fig.2.c and fig.2.d rest admissible. Concerning the feasibility of the proposed SMOC with the matrix converter, the fig.3 gives the rotor and stator currents harmonic spectra in the case of SMOC and vector control method. The rotor and stator current harmonic spectra are found using the Fast Fourier Transform (FFT) along one period of frequency equal to those of rotor voltage and stator voltage respectively. The fundamentals of the rotor current in the case of SMOC and VC apparent in the frequency of the rotor voltage that is V#U and their Total Harmonic Distortion (THD) are respectively #V and #M . The stator current harmonic spectra take the fundamental at V> (stator voltage frequency) with THD of #V in the case of SMOC, and #V in case of VC. In the light of this, we can conclude that the proposed method keeps the steady-state harmonic spectra at the same level as the vector control (VC) strategy. Consequently, the proposed SMOC method presents interested performance, and feasible with the matrix converter.
MW
2 MVAR 0 0.5 1 1.5

S
-2
0
KA
0.5
1.5
(a)
2 0 -2
S
(b)
1 0 -1 0.05 0.1

0.15
0.2
0.25
0.5
1.5
(c)
% V
(d)
100 0 -100
1000 0 -1000 1 1.5

S S
0.5
0.28
0.3
0.32
0.34
0.36
0.38
0.4
0.42
(e)
(f)
Fig.2. Simulation results in the case of SMOC with parametric variation of 50% in the stator, rotor and mutual inductances. (a) Active stator power and its reference. (b) Reactive stator power and its reference. (c) Three phases stator currents. (d) Three phase rotor currents. (e) Variation of stator and rotor resistances in (%). (f) Three phases stator voltages.
stator current (pu)
10
stator current (pu)
(A)
10
-2
10
(B)
THD 7.65% 50Hz
THD 7.85%
50Hz
10
-2
2000
4000
6000
8000
10000
12000
2000
4000
6000
8000
10000
12000
Hz
Hz
THD 7.59% 5.62 Hz 10

-2
rotor curant (pu)
rotor curant (pu)
10
10
THD 7.61% 5.62 Hz

-2
10
10
-4
10
-4
2000
4000
6000
8000
10000
12000
2000
4000
6000
8000
10000
12000
Hz
Hz
Fig.3. rotor and stator currents harmonic spectra. (A) Case of SMOC with parametric variations. (B) Case of VC with flux orientated without parametric variation.
B. Response to MPPT Taking that the turbine submits to the wind of the speed profile given in fig.4.a. The responses of the WECS are shown through the fig.4b-4f. Their analysis shows that the active and reactive power track their respective reference given by an MPPT algorithm with a satisfactory precision. Moreover, robustness tests involve with increasing and decreasing the rotor and stator resistances percentages from their nominal values as fig. 4h-4i. The obtained responses are affected slightly since the power tracking errors remain satisfactory for a wind speed profile showed in (Fig.4.a), even in the presence of resistances parametric variations, observer and disturbance in the case of unbalanced grid voltage during the time interval MT and MVT . Because the wind turbine power coefficient paste its maximum coefficient, we can conclude that the turbine speed stays near of the optimum speed that the turbine must follows to extract the power turbine maximum (Fig.4.b). The active stator power follows its reference given by an MPPT algorithm (Fig.4.d). The reactive stator power varies within weak bound around by its reference stator reactive power (Fig.4.e). Its for note that the reference stator reactive power given for ameliorates the power factor of the grid that the DFIG is connected. The wind speed regime imposed on the input of wind turbine adapts to DFIG functioning beach because the rotor current and the rotor flux (the state vector) varies weakly in the generator acceptable bound (Fig.4.f , Fig.4.g). we can say that the machine is with forced flux. This explicates the absence of the flux control. These good results are achieved due to the robustness of the robust sliding mode observercontroller applied. VI. CONCLUSION During this article, we have focused to establish a sliding mode observer/controller (SMOC) where we have proved its feasibility. The main objective of the SMOC is to achieve the references stator powers even in the presence of parametric variations, modeling errors and distributions (the cases of unbalanced grid), and minimize the sensor number necessary for the DFIG driving. All with respect the bound variables of the DFIG electrical states. Because of its advantages comparing with classical AC-DC-AC converter, we have demonstrated the
feasibility of the matrix converter with DFIG and the SMOC approach developed. The sliding mode observer allows not only the estimation of the rotor flux with robustness, but on minimizing of the number of the sensors used in the conducting of the WECS based on DFIG, where the rotor current sensors are not used. Appendix 1 DFIG parameters are: Pole pairs number: = U ; D =B?=@F = V> ; Stator resistance: = , MU ; Rotor resistance: = ,UM ; Stator inductance: 9 = , MQ ; Rotor inductance: = , MQ ; Mutual inductance: = , MQV ; Viscosity coefficient: = , M Mb # DFIG side inertia: = V# U
Appendix 2 Grid voltage: 690V, V># Pales number = 3 ; Pale diameter: R= 35.5m; Wind turbine characteristic: #V D@ #M /M#U Orientation angle (in our case = 0). Multiplier gain: G=65.
Appendix 3 Sliding mode controller gains: 9 = 9 = M4 and 9 = 9 = M3 Sliding mode observer gains: h v = M4 , h M4 , = M3 , = M3
VI. REFERENCES [1] Lie Xu ; Cartwright, P.Direct active and reactive power control of DFIG forwind energy generation, IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 21, NO. 3, SEPTEMBER 2006. [2] Hui Li ; Zhe Chen; A New Current Control Strategy of Maximizing the Generated Power from a Doubly Fed Induction Generator System, Power Electronics and Motion Control Conference, 2006. EPE-PEMC 2006. 12th International. [3] Mitsutoshi Yamamoto; Osamu Motoyoshi ; Active and Reactive Power Control for Doubly-Fed Wound Rotor Induction Generator, IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL 6. NO 4. OCTOBER 1991 [4] Arantxa Tapia; Gerardo Tapia; J. Xabier Ostolaza; and Jos Ramn Senz; Modeling and Control of a Wind Turbine Driven
Doubly Fed Induction Generator, IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 18, NO. 2, JUNE 2003 [5] A. Susperregui ; M.I. Martinez ; I. Zubia ; G. Tapia; Design and tuning of fixed-switching-frequency second-order sliding-mode controller for doubly fed induction generator power control, IET Electr. Power Appl., 2012, Vol. 6, Iss. 9, pp. 696706. [6] G. Abad; M. A. Rodriguez; and J. Poza, Two-level VSC-based predictive direct power control of the doubly fed induction machine with reduced power ripple at low constant switching frequency, IEEE Trans. Energy Convers., vol. 23, no. 2, pp. 570580, Jun. 2008. [7] G. Abad, M. A. Rodriguez, and J. Poza, Two-level VSC-based predictive torque control of the doubly Fed induction machine with reduced torque and flux ripples at low constant switching frequency, IEEE Trans. Power Electron., vol. 23, no. 3, pp. 1050 1061, May 2008. [8] Jiabing Hu; Heng Nian; Bin Hu, Yikang He; Z. Q. Zhu; Direct Active and Reactive Power Regulation of DFIG Using SlidingMode Control Approach; IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 25, NO. 4, DECEMBER 2010.
[9] Pena, R. ; Cardenas, R. ; Reyes, E. ; Clare, J. ; Wheeler, P.; Control of a Doubly Fed Induction Generator via an Indirect Matrix Converter With Changing DC Voltage; IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 58, NO. 10, OCTOBER 2011. [10] Cardenas, R. ; Pena, R. ; Tobar, G. ; Clare, J. ; Wheeler, P. ; Asher, G. ; Stability Analysis of a Wind Energy Conversion System Based on a Doubly Fed Induction Generator Fed by a Matrix Converter; IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 56, NO. 10, OCTOBER 2009. [11] Tadjine, M. ; Chekireb, H. ; Djemai, M.; On a sliding mode control and observer of induction motor; Systems, Man and Cybernetics, 2003. IEEE International Conference on, 1371 - 1377 vol.2, 2003. [12] Fang Lin Luo; Zhi Yang Pan; Sub-Envelope Modulation MethVol. O Reduce Total Harmonic Distortion of AC/AC Matrix Converters; Industrial Electronics and Applications, 2006 1ST IEEE Conference on, Page (s): 1 7, 2006.
12 10
(m/s) (rad/s)
200 0.5 150 0.48 0.46 0 100 200 300 400
8 6 0
100
100
200
300
400
100
200
300
400
(a) Wind speed
(b) !"# rotor speed
(c) $o%er coe&&icient
0 -0.5 -1
(MVAR)
1 2 0 -1 -2 100 200 300 400

S (KA)
-1.5
(MW)
-2 -2.5 0
100
200
300
400
0 0 100 200 300 400
(d) Acti'e stator po%er and its re&erence
(e) Reacti'e stator po%er and its re&erence
(&) Rotor current amplitude
4 3
(%)
100 50 0
(%)
100 50 0 -50 100 200 300 400

S S
2
(Web)
-50 -100 2 4 6 8 10
S
1 0
100
200
300
400
(g) Rotor &lu( amplitude
()) Rotor 'ariation parametric in (%)
(i) Stator 'ariation parametric in (%)
Fig.4. Wind chain simulation results with parametric variation of +50% in the stator, rotor and mutual inductances.
Accurate Partial Discharge Classification from Acoustic Emission Signals

Mustafa Harbaji, and Ayman El-Hag Department of Electrical Engineering American University of Sharjah Sharjah, United Arab Emirates mharbaji@alumni.aus.edu aelhag@aus.edu Abstract Accurate partial discharge (PD) classification provides significant information to asses power transformers insulation condition. The work presented in this paper aims to improve classification from acoustic emission signals for oilpaper insulated systems. Three different types of PDs are considered; surface discharge, PD from a sharp point to ground electrode, and PD from parallel plates. The PD types are simulated with aged insulation material (oil/paper), large tank size, and high surrounding noise level. The signals collected from each PD type are preprocessed using Continuous Wavelet Transform. The preprocessed signals are compressed using zonal coding applied over Direct Cosine Transform coefficients to create the feature vectors for classification, where a feed-forward with back-propagation trained neural network is utilized. The results indicates high recognition rate for classifying the different PD types using the proposed method. Index Terms Partial Discharge Classification; Acoustic Emission Signals; Continuous Wavelet Transform; Direct Cosine Transform; Zonal Coding. Khaled Shaban Computer Science and Engineering Department Qatar University Doha, State of Qatar khaled.shaban@qu.edu.qa
I. INTRODUCTION Power transformers are the heart and most important assets in any transmission and distribution systems. Transmission and distribution systems in most countries have existed for more than a century [1]. According to a report prepared for the U.S. Department of Energy (DOE) in June 2012, t he average age of installed large power transformers (LPTs) in the United States is approximately 40 years, with 70 percent of LPTs being 25 years or older [2]. Due to their high cost and critical role in delivering uninterrupted power, large power transformers cannot be simply replaced and usually serve for more than half a century. In case of failure, beside businesses interruption and environmental damage, a typical 100 MVA transformer can cost millions of dollars and an interval of 1.52 years to manufacture or substitute [3]. It has been reported that the main cause of power transformer failure in the US from 1997 to 2001 is the failure of its insulation [4]. The aging of transformer insulation during its operational life is a natural phenomenon [5]. However, insulation aging can be accelerated when transformers are subjected to abnormal electrical, mechanical and thermal stresses [6]. An important cause of such stresses on transformer insulation is partial discharge (PD) activities [6-7]. As a result, it is crucial to continuously assess PD activities as an indicator to potential catastrophic transformer failure by a monitoring system.
When PD activities are initiated the resulting energy is translated into different energy forms as mechanical energy represented by acoustic emissions, electrical, thermal and chemical energy [7]. As a result, a wide range of sensors and techniques can be used to detect PD activities as presented in [8]. However, in oil-paper insulated power transformers, not all measurements are applicable due to their complicated internal structure and large outside dimensions [8]. In practice, three detection methods are applicable for monitoring PD in oil-paper insulated power transformers; namely, Ultra High Frequency (UHF), Dissolved Gas Analysis (DGA), and acoustic emission (AE) [7-8]. UHF has high sensitivity for the electromagnetic pulses that accompany PD. However, UHF has very high bandwidth spans from hundreds of MHz and up to the GHz range, which is very expensive to capture and process. In addition, UHF probes can be hard to install and the number of probes are limited by the available dielectric windows or oil valves of the transformer. DGA has a very high sensitivity for PD detection. However, with DGA analysis it is hard to determine the source of PD. Furthermore, DGA analysis is usually performed by taking oil samples periodically, which may not be effective for continuous PD monitoring. An on-site DGA unit is available recently; however, with a high cost associated to it. AE sensors; on the other hand, has many advantages as compared to the other mentioned methods. AE sensors are way cheaper than the sensors used in the previous methods. In addition, this type of sensors is easy to install while the transformer is energized by simply sticking the sensor on the transformer tanks wall using a magnetic holder. Furthermore, AE sensors are insusceptible to external electrical and electromagnetic interferences. Identifying PD activities has been an important research area for PDs direct effect on insulation failure. There are many attempts that addressed achieving high recognition rate for different PD types. However, most of them were not made to simulate PD in transformer tank environment; that is, they did not take into account the effect of the transformer tank dimensions, transformer complex inner structure, or condition of oil such as age and temperature. Other studies presented in [6] and [9] reported high recognition rate for PD inside transformer but the mechanisms used are only applicable when the transformer is off-line. A recent work [10] used AE sensor to study the effect of increasing the tank size, the presence of barriers between the PD source and the AE sensor,
and oil age. The results of this work show a recognition rate in the range of 96-100% for a small tank dimensions, and using different feature extraction techniques. When barriers were placed between the PD source and the AE sensor, the recognition rate stays high; however, it drops significantly when a large tank, 110.5 m, and old oil were used with recognition rates in the range of 50-78% and 60-88% respectively. Moreover, the classification problem in this work was for only two classes presented as PD (from sharp electrode) or no PD. In [11], three types of PD investigated inside a test oil tank of 505080 cm dimensions; namely, PD in pressboard, floating metal in the pressboard and the bubble in the pressboard. Three AE sensors were used to register the paper signals. For feature extraction, time-frequency transformation is proposed from the STFT using seven descriptors as described in [11]. It shows different descriptors values based on averaging 11 training samples for each PD model. There is no testing provided for classification and the training is performed with a low number of samples. Moreover, there is no consideration of other factors that would affect the feature extraction such as the oil condition. In this paper, the proposed system is to classify different common PD types in an on-site measurement environment. The measurement conditions include having aged insulation material (oil/paper), large tank size, and high surrounding noise level. Three types of PDs are considered for the classification problem; surface discharge, PD from a sharp point to ground plane, and PD from parallel plates. II. METHODOLOGY The objective of the experiment set up is to capture AE signals for different simulated PD types. Ultimately, the captured AE signals are preprocessed for classification. A. Experimental Setup Figure 1 shows the overall setup used to generate and register different types of PDs. The tank used has 110.5 m dimensions filled with aged oil. The high voltage source is a 40kV AC vacuum interrupter with a control panel. The high voltage source is connected to an electrode system that is adjustable to generate the desired PD types. The three types of electrode systems are shown in Figure 2.
Figure 2: from left to right: 1- Surface Discharge; 2- PD from sharp point; 3 PD from parallel plates.
The AE sensor used has a bandwidth of 100-450 kHz with a resonance frequency at 150 kHz. The sensor is fixed at the tanks wall by a magnetic holder and kept as far as possible from the PD source to include the effect of the tank size. To capture the produced PD signal, an oscilloscope interfaced with Matlab is used with a sampling frequency set to 10M sample/sec for a window of 2500 samples (250 second). During the PD acquisition, a three phase induction motor is running in the background to simulate high level of surrounding noise. Figure 3 and Figure 4 shows, respectively, AE sample signals acquired for random noise (no PD scenario) and another with PD.
Figure 3: Random Noise (No PD)
Figure 1: Overall system setup
Figure 4: PD signal from Surface Discharge
B- AE Signal Processing, feature extraction and classification After capturing 100 PD signals for each type, frequency features are extracted and compressed from these signals for PD classification as explained in the following subsections. 1) Signal Processing The PD signals are non-stationary signals that appear at random intervals at specific electric field inception. Therefore, traditional FFT analysis can introduce unwanted frequencies when analyzing such signals. A better alternative is to use Continuous Wavelet Transform (CWT) to extract only the related frequencies to the PD signal. The CWT can be visualized using a scalogram. The scalogram has three dimensions; that are, a dimension to represent time (sample space), second to represent scales, and third to represent coefficient value that can be expressed by varying the color or brightness. Figure 5 shows a PD signal analyzed with CWT. The scales are related to frequencies by a relation that is governed by the mother wavelet and the sampling frequency used. Figure 6 shows the relationship between the scales and the frequencies resulting from using a Gaussian mother wavelet and 10MHz sampling frequency.
2) Feature extraction The scalogram obtained by the CWT can be considered a 2D matrix of size where each element represents a CWT coefficient. If we unfold the matrix into a feature vector, the resulting feature vector will have 320000 elements. However, as discussed previously, and as can be seen in Figure 5 the PD signal is a non-stationary signal and the resulting matrix is a sparse matrix. Consequently, the wavelet coefficients have considerable energy percentage at only specific locations depending on the PD type. Therefore, it would be efficient to extract the useful coefficients and reduce the feature vector size. In this paper, zonal coding is applied over Direct Cosine Transform (DCT) coefficients to create the reduced feature vectors. The DCT is a method that translates data into its basic frequency components and can be applied onto multidimensional data [12]. Since the matrices obtained are 2-D, the 2-D DCT is used. To choose the most representative frequency components, zonal coding is utilized, which is based on maximal variance. The resulting feature vector is reduced from 320000 to 200 elements. 3) Classification In this paper, three types of PDs are considered as classes for the classification problem; PD from a sharp point to ground plane, PD from parallel plates and surface discharge. The number of samples or feature vectors obtained for each class as described previously is a hundred. After testing several classifiers, as linear and KNN classifiers, on the obtained feature vectors, a 10 neurons Neural Network was chosen since it gives better results. The network used is a feed-forward network that consists of three layers as shown in Figure 7 with 55% of the overall data (300 feature vectors) used for training, 15% for validation and 30% for testing.
Figure 5: Continuous Wavelet Analysis (CWT) for a Surface Discharge signal
Figure 7: Architecture of utilized Neural Network
III. RESULTS AND DISCUSSION The classes in this paper are numbered as class 1, 2, 3 that corresponds to PD from a surface discharge, sharp point to ground plane, and from parallel plates respectively. Table 1 shows the confusion matrix obtained by passing each feature vector through the trained network.
Table 1: Classification Confusion Matrix for Test Data
Target
Class 1 Class 2 Class 3
Class 1 27 0 1
Output Class 2 2 29 1
Class 3 1 1 28
Figure 6: Relationship between scales and frequencies
It can be seen that the network managed to achieve high recognition rate of 93.333% from the given samples. This implies that each PD type emits a unique AE frequency pattern that can be used for accurate PD classification. Figures 5, 8, and 9 present a sample scalogram for each PD class. Moreover, the results suggest that in an on-site measurement, the PD signal should be measured more than once to have a firm decision on the PD type inside the transformer.
ACKNOWLEDGMENT This work was made possible by NPRP 5 -044 - 2 016 grant from Qatar National Research Fund (a member of Qatar Foundation). The statements made herein are solely the responsibility of the authors. REFERENCES
[1]. Matthew H. Brown and Richard P. Sedano, "Electricity
Figure 8: CWT for a PD signal from a sharp point to ground plane
Figure 9: CWT for a PD signal from parallel plates
IV. CONCLUSION In this work, partial discharge (PD) classification was investigated for three PD types using acoustic emission (AE) signals. The simulated PD considered some of the on-site measurement conditions as aged insulation material (oil/paper), large tank size, and high surrounding noise level. An artificial neural network was successfully able to achieve high recognition rate from a frequency-based features. The proposed method indicates the advantage of using frequency features to achieve high classification rate as each PD type tends to emit AE frequencies in different patterns. A future work can include investigation of more measurements conditions and PD types.
transmission ," National Conference of State Legislatures, A Primer ISBN 1-58024-352-5 , 2004. [2]. Office of Electricity Delivery and Energy Reliability, "Large power transformers and the u.s. electric grid," U.S. Department of Energy, June 2012. [3]. Bart Tichelman, "Transformer asset management - a new paradigm," in Energy Pulse, 2004, April 16. [4]. William H. Bartley P.E., "Analysis of transformer failures," in International Association of Engineering Insurers, Stockholm, 2003. [5]. Cheng-Chein Kuo and Horng-Lin Shieh, "Artifical classification system of aging period based on insulation status of transformer," in Eighth International Conference on Machine Learning and Cybernetics, Baoding, July 2009, pp. 3310-3315. [6]. M. V. Rojas-Moreno, J. Posada, J. M. Martnez-Tarifa,G. Robles and J. A. Garca-Souto J. Rubio-Serrano, "Electroacoustic detection, identification and location of partial discharge sources in oil-paper insulation systems," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 19, no. 5, October 2012. [7]. Wojciech Sikorski and Waldemar Ziomek, "Detection, recognition and location of partial discharge sources using acoustic emission method," in Acoustic Emission., 2012, ch. 3, pp. 49-74. [8]. Shigemitsu Okabe, Hiroshi Murase and Hitoshi Okubo Masayuki Hikita, "Cross-equipment evaluation of partial discharge measurement and diagnosis techniques in electric power apparatus for transmission and distribution," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 15, no. 2, April 2008. [9]. S. R. Hardie and P. S. Bodger, "A new technique for detecting partial discharges within an on-line power transformer subjected to interference," in Electrical Engineers Association (EEA) of NZ Conference, Auckland, New Zealand, 2006. [10]. A. Swedan, A. H. El-Hag, and K. Assaleh, "Acoustic detection of partial discharge using signal processing and pattern recognition techniques," Insight Non-Destructive Testing and Condition Monitoring, vol. 54, no. 12, December 2012. [11]. Yasmin H. Md Thayoob, P.S. Ghosh, Ahmad Zuri Sha'ameri and Mohd Aizam Talib M.L. Chai, "Identification of different types of partial discharge sources from acoustic emissionsignals in the time-frequency representation," in First International Power and Energy Coference PECon, Putrajaya, Malaysia, 2006. [12]. Andrew B. Watson, "Image compression using the discrete cosine transform," Mathematica Journal, vol. 4, no. 1, pp. 81-88, 1994.
Comparison between Reliability Prediction Methods for Power Electronic Circuits

Ayoub Balador, Negar Noroozi, Mohammad Reza Zolghadri, Saeed Ouni, Mahmoud Shahbazi
Department of Electrical Engineering, Sharif University of Technology Tehran, Iran balador.a68@gmail.com, Nenoroozi@gmail.com, Zolghadr@sharif.edu, Saeedouni@sharif.ee.sharif.ir
AbstractSeveral methods are proposed for predicting reliability of electronic circuits. Among them MILHDBK-217F, FIDES and IEC-TR-62380 are investigated in this paper. Predicted failure rate of commonly used power electronic elements using this three methods are presented. The Predicted failure rates are compared with each other and with observed failure rates of elements that could be measured with acceleration tests. The power electronic elements of buck converter are considered. The results showed that MIL-HDBK-217F produces the worst case failure rate and results for IEC-TR-62380 and FIDES methods showed that IEC-TR-62380 produces more conservative failure rate than FIDES and is more optimistic than MIL-HDBK-217F. Keywords reliability; failure rate; MIL-HDBK-217F; FIDES; IEC-TR-62380
Fig. 1. Bathtub Curve
1.
INTODUCTION
Generally, todays environment of ever increasing

force to execute faster, better, and cheaper, needs that designers have the capability to correctly estimate the reliability of the equipment through design process. Correct estimations in the first phases of design not only contribute to a more robust and reliable product, but also push down the redesign related prices. The failure rate is usually the preferred characteristic index for reliability of electronic components. The time dependence of the failure rate for a given population of items of the same type often exhibits a bathtub curve shown in figure1 [1]. Failure rate index for electronics components in generally refer to the phase with constant failure rate. Several reliability prediction methods propose different calculation methods for this parameter in constant period. Because of the importance of reliability in high quality production and better selling in electronic productions, many methods have been emerged to predict failure rate by huge military and industrial companies. While the first method of reliability prediction came in the 1940s by German scientist and mathematician respectively named Von Braun
and Eric Pieruschka [2]. The most important and prevalent of proposed methods are (MIL-HDBK217F, FIDES, IEC-TR-62380). This paper studies and compares these methods in calculating failure rate in power electronic circuits that may be used for a better reliability prediction and element selection for power electronic circuits. To do so, a buck converter which operates under assigned circumstances is considered. The failure rates for important elements are calculated by aforementioned methods and are compared with measured value driven from accelerating tests by device manufacturers. Accelerating tests are done by factories which produce elements and the results may be found in annual reports of these factories. In section 2, the basics, in sections 3 to 5, the procedure for each failure rate calculation method and in section 6, the results, analysis and comparison of methods and pros and cons of each method are represented. Finally, section 7 is assigned to conclusions. 2. BACKGROUND of RELIABILITY EVALUATION METHODS
MIL-HDBK-217 Created by the U.S. military in 1965s, has undergone six major revisions, and was last edited in 1995.It was developed to supply a reference for predicting the reliability of electronic military devices [4]. The largest defect of the handbook is that it assumes most failures are due to components and does not fully address process based factors such as manufacturing or system management.
The FIDES Guide: Reliability Methodology for Electronic Systems has been developed by a consortium of French industrialists from the aeronautics and defense fields at 2004 and the last edition is for 2009. It is based on the physics of failures and affirmed by the experimental information and existing modeling analysis. The main features of the FIDES method are: provides models for wider electronic components, considers technological and physical factors, considers precisely the life cycle, considers mechanical and thermal overstress, and provides the possibility for distinguishing the failure rate of a specific supplier of a component. Also, it covers failures linked to development, manufacturing and maintenance processes. The FIDES method also supplies a reliability process control and audit guide which helps recognize factors that influence to equipment failure. Recognition of these factors allows equipment reliability to be controlled and improved throughout the lifecycle by changing design or producing processes [5]. The UTE-C 80810 reliability handbook is the major overhaul of the RDF93 reliability handbook and was emitted by CNET (France Telecom). The first version of this reliability hand book was published in 2000 under designation RDF200. An update of the document was published in 2004. The IEC defined it as an international standard under the designation IEC-TR-62380 [6]. This reliability predicting method for electronic equipment is an important progress compared to other older methods. Models calculate directly the effect of the environment. The thermal cycling seen by component, function of life cycles undergone by the equipment, replace environment factor which is troublesome to evaluate. These models can handle permanent working, on/off cycling and inactive applications. On the other words, failure rate of component soldering is included in component failure rate [7]. 2.1 MIL-HDBK-217F There are two ways to estimate reliability by MILHDBK-217F: Parts Count Estimation and Parts Stress Estimation. Parts Count Estimation is generally used to estimate the reliability of a component before product design cycle to acquire a tough reliability estimate in accordance with the reliability purpose. A failure rate is determined by literally counting similar components of an item and grouping them into several component types. The amount of components in each group is then multiplied by a base failure rate and quality factor. Finally, the failure rates of all the different part groups are summed together for the final failure rate. Parts Stress Estimation is usually used in the design cycle of product. In Parts Stress contrary to Part
Count the failure rate for every component is exclusively calculated using the specific stress levels the component is exposed (i.e., voltage, temperature, humidity, vibration). An example of a MOSFET transistor part stress model from MIL-HDBK 217 is as follows: Where p= base failure rate, b= part failure rate, Q=quality factor, A=application factor, E= environment factor, T=temperature factor. Values are given to the every stress factor and base failure rate from tables of each element in MILHDBK 217. These values are then assigned in the base part equation and a failure rate is specified for that part under certain environmental conditions. Failure rate calculations for other parts use different stress factors but follow the same [4]. 2.2 FIDES Not like many of the models shown in the earlier reliability Guides for electronic reliability prediction, the models shown in reliability manual FIDES were developed orienting in the physics of the failures and not just from an empirical modeling of the operational feedback. The structure of the models is created on the physics of the failures and reinforced by check information analyses, operational feedback and existing modeling. After perfecting, the models have been calibrated with the operational feedback. At the general level, the FIDES model is composed of three basic factors. = Physical . PM . Process (2) (1)
ProcessThis factor shows the quality and technical control around reliability in the product profile life cycle. It will take into consideration the system products processes. It addresses all processes from specification all the way to operation & maintenance. Some questions are offered in order to survey the product developers processes. The end result is a process grade that matches to the formula 3. The 2 is a correlation factor that specifies the range of the process factor multiplier and has been designed to 2.08 by FIDES designers [5].
(3)
TC e RH RH 70
.
T
PM This factor shows the device quality. This factor takes into consideration the Quality Assurance (QA) level encompassing the part. The rank is depends on the qualification level of the manufacturer (QM). In addition it uses one more ranking of the buyers previous experience with the provider ( ). These numerous rankings are rolled into the Part Grade parameter [5].
.
QM
(10)
.
.
(11) (13)
_
QA RA
(4) (5)
Part_Grade
The variables contained in the equations comprising the MOSFET transistor model come in Table 1. These factors, and especially the acceleration factor, are broken down for every physical stress. A physical stress is any stress commonly applied to the product throughout its working use, consisting of features linked to the design. Physical stresses are arranged in to various families: Electrical: Electrical Thermal: Thermal Temperature cycling: TCy Humidity: RH Chemical: Chemical Mechanical: Mechanical
Physical This is wherever the base failure rate (o) for the element itself in addition for the case type can be used and extended upon. It encompasses parts of another two areas while they relate with the element, and hence, is the most detailed of the three. Things such as for example element family environmental sensitivities, end consumer and system integrator level of education, function of the element plus many other considerations are weaved into this factor [1].
.
P _
The physical factor (Physical) for every element is made up of a base failure rate (o), cceleration and nduced factors. Elements have unique base failure rates and the cceleration factor is composed of various physical limitations for each. nduced for every element may also be different being connected to overstresses and place of elements in the system. For example, the MOSFET transistor model is as follows:
P
The induced factor shows the influence of overstresses not listed as such. It is determined for every phase in the life cycle. It is in these form:

. L C
13
Table1. Physical Stress Parameters Parameter Tannual Tboard_ambient Tcycling Definition Time linked with each activating phase over a year (hours) Average temperature of board during a phase (0C) Variation amplitude linked with a cycling phase (0C) Maximum temperature of board during a cycling phase Number of cycles linked with each cycling phase over a year (cycles) Cycle period (hours) Stress linked with each random vibration phase (GRMS)
TC S
TC S
Tmax-cycling
(7)
Nannual-cy cy
Where T T e 0
.
T
GRMS
(8) (9)
Placement shows the effect of the component location in the equipment or the system. In this case location describes the positioning of the component or the function in which it is incorporated (particularly whether or not it is interfaced). Application shows the effect of the utilization atmosphere for application of the product comprising the item. As an example, contact to a mechanical overstress is really a priori more important in electronics built into a portable system than in a fixed station system. Ruggedising shows the effect of the plan to take consideration of overstresses in the component development. Csensitivity shows the coefficient of sensitivity to overstresses natural to the component technology considered. i is the index of the phase considered.
1. 2. 3.
On/Off working phases Permanent working phases Inactive phases mode.
Parameters necessary to define the life cycle of equipment are defined in handbook such as table1 [7]. The generic mathematical model used in the IECTR-62380 reliability handbook is of additive type. It consists in:
(14)
This addictive standard mathematical approach corresponds to a major change in the modeling of the electronic components reliability. The reliability handbooks more recent than the IEC-TR-62380 also use an additive modeling type. The addictive type modeling enables to differentiate the main independent categories of failure mechanisms and to model their acceleration coefficients [6]. For example, the MOSFET transistor model that IEC-TR62380 has developed is as follows:
2.3 IEC-TR-62380 The data in this guide addresses intrinsic failures and, necessarily, there is a small amount of nonintrinsic failures, since it is difficult to identify all the non-intrinsic failures when they are residual. The other non-intrinsic failures (due to mistakes of design, selection, and usage) are excluded from this guide. The element failure rate depends upon a number of working and environmental factors. For this reason, for every element family, the guide provides a base failure rate value multiplied by a number of influencing factors. This simple, empirical expression takes consideration of the more substantial influencing factors in regards to conditions of use [7]. The key factors used are the following [7]: 1. Factors providing the impact of temperature: Factor t provide the influence of temperature on the predominating failure mechanisms. 2. Factors providing the effect of particular stresses: Usage factor U for Zener and Thyristors. Factor A providing the effect of current pulses for Aluminum liquid electrolyte capacitors. 3. Factors giving the influence of applied voltage : S The influence of applied voltage is taken into account for transistor and optocouplers. Predicted reliability calculation of equipment has to be done according to its circumstances. They are defined by the life cycle. A life cycle must be decomposed in many working phases, on the basis of a normal year of use [7]. These phases can be considered:
2.75
10
(15) Where B: base failure rate of the transistor package. i: influence factor related to the use of the transistor (protection interface or not). EOS: failure rate related to the electrical overstress in the considered application. 0: base failure rate of the die. See table on this S: charge factor. The unit of failure rate is FIT (failure per one billion hours). 3 ANALYSIS AND RESULTS To calculate the reliability of the electronic components a buck converter is considered. Specifications of the buck converter and elements are shown in table 2. To calculate the life cycle, this assumption is same for all methods: Buck converter is considered as an industrial charger operating continuously 10 hours of day. Environmental conditions such as the state like GF in MIL-HDBK-217 with an average temperature around 20 C with an average swing of 10 degrees.
Temperature of elements could be calculated with thermal resistance and dissipated power. Reliability can also be obtained by experiments to determine reliability by examining a large number of the product at their specified temperature. The prediction can be specified sooner by increasing the stress factors on the product such as increasing its operating temperature or humidity. This is known as accelerated life testing. Predictions by these methods take into account the number of units and their operating hours of survival before failure. In references [8-9] completely describes this method. For some component, the manufacturer releases the acceleration test results and failure rate of them since acceleration tests is very expensive and consume time so only do for important elements. Here only the MOSFET failure rate is provided and therefore, is used as reference for verify accuracy of methods. 3.1 Ground Rules The MIL-HDBK-217F predicted failure rates are expressed in failure per one million hours. These values were simply converted to FIT (failure per one billion hours) for an accurate comparison with the other prediction methods. IEC-TR-62380 and MILHDBK-217F predictions are based on operating hours. However, FIDES calculates failure rates in calendar hours. FIDES uses these time factor to be able to take into account for downtime and different profile life cycle:
_
_
-ceramic capacitor model for more similarity likewise, in IE-TR-62380 method used ceramic capacitor model for film capacitor. 3.2 Prediction Comparison Table 3 shows acceleration test results and observed failure rate in these experiments. Two types of acceleration test has been done for transistors and observed failure rate achieved by both tests. It should be considered, if the number of samples and test time are small, the failure rate will be higher than the real value. Thus, the real failure rate is less than the observed value obtained by the tests. Table 4 shows predicted failure rate by using each of three methods. According to the results, it is clear that the predicted failure rate by FIDES method for all elements is the lowest value while MIL_HDBK_217F approach provides the highest failure rate. It can be said that the failure rate predicted by IEC_TR_62380 is between the two other methods. Although predicted failure rate for transistor by IEC_TR_62380 is close to observed failure rate in compare with another methods.
Test 1 Part Number Date Code Voltage [V] Temp.[0C] Time[hours] XTQ110N 10P SS1045 80 125 1000 30 0 30000 Test2 Part Number Date Code Low Temp. [0C] High Temp. [0C] number of cycles Sample Size Failures Device Sampl e [hours] Failure rate 61.27 XTQ110N1 0P SS1045 -55 125 250 30 0 7500
(16)
In FIDES guide calculate the Process is complicate because its need to more information therefore uses the default value as well as default value for ruggedizing is considered. For film capacitor there is not Model in FIDES handbook therefor used of
Table 2: Buck and Component Specification Buck Specification Input voltage Output voltage Output power Output voltage ripple Switching frequency 30 V 15 V 60 W 5% 50K Hertz Component Transistor Diode Capacitor Inductor Mode IXTQ110N10P DSA30I100PA MKT36825155 E35/20/9 CCM
Sample Size Failures Device Sampl e [hours] Failure rate 281
Table 3. Acceleration Tests
Table 4. Predicted Failure Rate (FIT) COMPONENT OF CONVERTER MOSFET DIODE CAPACITOR INDUCTOR PREDICTED FAILURE RATE (PER FIT)
MIL_HDBK_21 7F FIDES IEC_TR_623 80
5460 20 58.4 0.85
3.84 8.05 0.061 0.28
28.75 14.56 0.48 1.19
3.3 Advantage and Disadvantage of Methods The environmental condition is a significant factor in the reliability. Indeed, the reliability of the identical components could widely vary based on the kind of their placement environment. The reliability handbooks present the model of environmental effect based on two opposite methods: Definition of categories of environments (MIL_HDBK_217F) Parameter adjustment of environments (IEC_TR_62380, FIDES)
We could say about available handbooks that FIDES is difficult to handle while IEC_TR_62380 is intermediate and MIL_HDBK_217F is easy. 4 CONCLUSIONS
The first definition has the advantage of simpler use. In contrast to environments based on category, the parametric environments allow their modification and adjustment of the pressure levels to approach to the reality. The table 5 shows the stresses considered within the model of the environments. The parametric environments have the ability to consider the inactive phases; however that is not the case in the MIL_HDBK_217F. Therefore, the inactive phases in many cases are considered as operating phases, decreasing the failure rate. This can be one of the reasons of differences in the results of these methods. FIDES guide model the effect of the process (maintenance, operating, construction, specification and design) on the ultimate reliability of the component. Until now, FIDES is the one method presenting a major detail level in modeling. It enables us to correct calculation starting from the signification number of parameters. Whereas, FIDES needs a noticeable number of parameters and produce an important sensitivity for the assumptions.
Table 5. Stress Take into Account Constraints Thermal stress Environment categories Thermal cycling stress Mechanical stress Thermo-chemical stress Chemical stress Induced stress MILHDBK217F IECTR62380 FIDES
Each method has its own benefits and problems but can be used all of them to compare and improve reliability. But for correctness of reliability prediction, it can be said that FIDES method estimate failure rate optimistically for power electronics component, MIL_HDBK_217F method estimates failure rate pessimistically and the IEC_TR_62380 method is between two other methods and closer to real answer. REFERENCES
[1] Billinton, Roy, Ronald Norman Allan, and Ronald N. Allan. Reliability evaluation of power systems. Vol. 2. New York: Plenum press, 1984. Pecht, M.G., Nash, F.R., Predicting the reliability of electronic equipment, Proceedings of the IEEE, Vol. 82, No. 7, July 1994. Department of Defense, MIL_HDBK_217F_Notice2: Reliability Prediction of Electronic Equipment, 1995. Juan J. Marin, W. Pollard, Raytheon Company, Experience report on the FIDES reliability Prediction Method, Proceedings of the IEEE, jan. 24-27, 2005 Dlgation Gnrale pour lArmement(DGA), UTE_FIDES_Guide_2009_Edition_A: Reliability Methodology for Electronic Systems, April 2010. Institude pour la Mitrise des Risques (IMdR) ,Selection guide for electronic component predictive reliability models, Octobre 2009. International Electrotechnical Commission, IEC-TR-62380: Reliability Data Handbook Universal Model for Reliability Prediction of Electronics components, Agu. 2004. D. Crowe, Design for Reliability, CRC Press LLC 2001. L. A. Escobar and M. Q. William, "A review of accelerated test models", Statistical Science 21.4 (2006): 552-577.
[2]
[3] [4]
[5]
[6]
[7]
[8] [9]
Determination of Worst Case Loading Margin of Droop-Controlled Islanded Microgrids

Morad M. A. Abdelaziz and E. F. El-Saadany
Electrical and Computer Engineering Department University of Waterloo Waterloo, Ontario, Canada m3abdelm@uwaterloo.ca
AbstractThe determination of an islanded microgrid proximity to voltage instability is essential for its operation with an adequate security margin. This paper presents an algorithm for determining the worst case loading margin of droop-controlled islanded microgrids. The problem is formulated as an optimization problem to determine the shortest distance to voltage instability (i.e. the closest saddle node bifurcation point). A detailed microgrid model is adopted to reflect the special features of droop controlled islanded microgrid systems where; 1) the system frequency is a power flow variable, and 2) the power produced by the different DG units is dependent on the system power flow variables and cannot be pre-specified. The optimization problem is subject to different system operational constraints including; the power flow constraints, voltage and frequency regulation constraints and unit capacity constraints. Different numerical case studies have been carried out to test the effectiveness and the robustness of the proposed algorithm. Keywords- Distributed generation (DG); droop-control; islanded microgrid; saddle-node bifurcation; voltage stability
single DG unit serving a single load [4]-[6]. As such, microgrids have recently become an important concept in the integration of the DG units into the distribution system. The recent IEEE Std. 1547.4 presents the microgrid structure as the building block at the core of the smart active distribution system design and planning [7]. Given that the islanded microgrid system is fed from a group of small DG units of limited capacities, its proper planning and operation requires accurate assessment of its proximity to voltage instability. Static voltage stability problems are related to the voltage collapse occurring as the system experience a qualitative change in the structure of its solutions due to small disturbances. The point at which such a qualitative change happens is known as the static bifurcation point of the system. This point is associated with the disappearance of systems equilibrium point (i.e. load flow solution) and the sensitivity of the load flow solution to small disturbances is infinite. A common methodology to assess voltage instability proximity is to determine the load margin (i.e. allowed load increase) from the current operating point to the point of voltage collapse. Generally, two basic interpretations of voltage instability proximity exist, which are: 1) those assuming a prefixed direction of load increase, and 2) those making no assumption about the direction of load increase (aiming for the closest node bifurcation point) [8]-[9]. To the authors best knowledge, the calculation of the droop controlled islanded microgrid proximity to voltage instability was only tackled in [10] and [11]. In [10], a continuation method based on the corrector predictor scheme was adopted to compute the islanded microgrid maximum loading factor. In [11], an optimal power flow problem was formulated to assess the islanded microgrid maximum loading factor. Nonetheless, the work in [10] and [11] assumed a prespecified direction of load increase. As such the work in [10] and [11] did not allow for the consideration of the case where there is no prior knowledge of the expected direction of load increase. Accordingly, in this paper, an approach for the calculation of the worst case loading margin of droop controlled islanded microgrid is presented. In this case the worst case loading margin represents the shortest distance to voltage instability i.e. the closest saddle node bifurcation point. The problem is formulated as an optimization problem for obtaining the closed saddle node bifurcation point. The remainder of this paper is organized as follows; Section II
I.
INTRODUCTION
The electrical power system has been originally based on the generation of electrical power from dispatchable generators centralized in adequately located central power generation stations of ratings ranging from hundreds of MWs to several GWs [1]. However, driven by technical, economic and environmental benefits for different stake holders in the power industry [2], the electric power system is being changed to having a large portion of the growing demand for electrical energy supplied through the use of large numbers of dispersed generation units of ratings ranging from tenth or hundreds of KWs to several MWs [1], which is known as Distributed Generation (DG) [1-3]. These DG units can operate in one of two modes; standalone or grid-connected. In the standalone mode of operation the DG is used to feed a load located in a remote area where its connection to the grid is rather difficult or expensive. In the grid-connected mode, the DG is used to support the grid by feeding a part of the base load, performing peak load shaving, or providing auxiliary power services [1]. Such applications of DG units brought about a new concept of forming autonomous regions in the power system by grouping a cluster of DG units and loads into what is known as microgrids. The idea supporting the formation of microgrids is that a paradigm consisting of multiple DG units and aggregated loads is far more reliable, economical and practical than a
presents the droop controlled islanded microgrid model that reflects its special philosophy of operation. The model presented in section II is then used in formulating the proposed approach for calculating the islanded microgrid closest saddle node bifurcation in section III. Section IV presents the numerical results that verify the effectiveness of the proposed methodology. Finally, Section V concludes the paper and summarizes its findings. II. ISLANDED MICROGRID MODELING Droop chs
PGi,QGi
PCC
| Vi |,
Figure 1. Steady-state model of a DG unit operating in droop mode
In droop control mode, the power sharing is achieved between the different DG units by mimicking the behavior of synchronous generators operating in parallel. As such, the frequency of the DG unit output voltage is drooped as the generated active power by the DG unit increases [12]. Similarly, the magnitude of the DG unit output voltage is drooped as the generated reactive power by the DG unit increases. Accordingly, for a droop-controlled DG unit connected to bus i, the DG output voltage frequency, , and magnitude, Vi , can be given as:
of the systems mismatch equations describing the different system buses. For each droop controlled bus i the mismatch equations can be given as:
PLi (Vi ) = Pi ( Vi , V j , i , j , ) PGi ( ) QLi (Vi ) = Qi ( Vi , V j , i , j , ) QGi ( )
(3) (4)
i {1,2,,nDG }
voltage magnitude at bus i, { j} represents the set of nodes connected to node i. i is the voltage angle at node i. nDG is the number of DG units. From (1) and (2), PGi and QGi can be given as where is the system steady-state frequency, Vi is the
= * i m pi PGi
Vi = Vi nqi QGi
*
(1)
(2)
* where * i and Vi are the output voltage nominal frequency and magnitude set points at no-load, respectively, mpi and nqi are the active and reactive power static droop gains, respectively, and PGi and QGi are the injected active and reactive power by the DG unit, respectively. Fig. 1 shows the steady-state model of droop controlled DG unit. As it can be seen in the figure, the droop-controlled DG unit is represented as an ideal voltage source whose voltage magnitude and frequency are determined using the droop relations given in (1) and (2).
PGi = QGi =
1 * i m pi
(5) (6)
1 * Vi Vi nqi
The steady-state representation of droop controlled islanded microgrids is different than the conventional distribution system representation in three main aspects; 1) unlike conventional distribution system were the DG units are represented as PQ or PV buses, in the islanded mode the DG units representation has to reflect its droop characteristics where the DG generation is not pre-specified and is controlled by the DG droop characteristics. 2) Conventional distribution system representation depends on the modeling of the main substation as a slack bus capable of keeping the system frequency constant by supplying the difference between the time varying system demand and the pre-specified generation units output. However, in the droop-controlled islanded microgrid scenario, the DG units forming the islanded microgrid system are of small and comparable sizes and there is no one DG unit that is capable of performing the slack bus function. 3) In droop controlled islanded microgrid system, the system frequency is not constant and is considered as one of the system power flow variable [12]. The steady state representation of the droop-controlled islanded microgrid system depends on the accurate formulation
With yij ij representing the frequency dependent branch admittance between nodes i and j, the calculated active and reactive power injected to the microgrid at bus i, Pi and Qi respectively, can be given as:
Pi = Vi
j
yij cos ij Vi V j yij cos ij + j i
( )
(
(7)
Qi = Vi V j yij sin ij + j i Vi
j
yij sin ij
( )
(8)
PLi and QLi are the active and reactive load power at node i, respectively and given as PLi = Poi | Vi | QLi = Qoi | Vi |
(9) (10)
where Poi and Qoi are the nominal active and reactive power operating points respectively; and are the active and reactive power voltage exponents. Typical values for and can be found in [13]. as: For each load bus k the mismatch equations can be given
The power flow constraint in (14) ensures that the system fulfills the power balance constraints at the system maximum loading condition.
Voltage and frequency regulation constraints
min max
QLk (Vk ) = Qk ( Vk , Vl , k , l , ) PLk (Vk ) = Pk ( Vk , Vl , k , l , ) (11) (12)
Vi min Vi Vi max
(15) (16)
k {1,2,,nLoad }
where {k} represents the set of nodes connected to node l. III. DETERMINATION OF CLOSEST SADDLE NODE BIFURCATION
To determine the closest saddle node bifurcation point of droop-controlled islanded microgrid systems, no assumption is made on the direction of load increase. Instead, we want to find the set of active and reactive power increment at the different load point whose norm is minimal and which when imposed on the current operating point cause the power flow Jacobian to be singular. The problem of determining the closest saddle node bifurcation point of droop controlled islanded microgrids can be formulated mathematically as:
2 2
where the subscript min and max indicate the lower bound limits and the upper bound limits. The inequality constraints given in (15) and (16) establishes the upper and lower limits on the overall system frequency and the allowable voltage violation at the maximum loading operating point. The fulfillment of the constraints given in (15) and (16) ensures the system voltage and frequency regulation at the system maximum loading condition.
DG units maximum capacity

Typically the DG units are equipped with a current limiter intended to limit the PGi and QGi generation of the DG unit to its rated capability, S Gi ,max [14]. Accordingly, the DG units PGi and QGi generation follows the droop relations given in (5) and (6) up till the DG units maximum active and reactive power generation limits, PGi,max and QGi,max, respectively. Beyond PGi,max the DG unit active power generation is not allowed to follow the droop relation, given by (5), and the DG is transformed to inject a constant amount of active power set at the violated limit. Similarly, beyond QGi,max the DG unit reactive power generation is not allowed to follow the droop relation, given by (8), and the DG is transformed to inject a constant amount of reactive power set at the violated limit. Accordingly, DG units active and reactive power generation capabilities constraints can be given as:
nbus nbus Min. S = Pi + Qi i =1 i =1 Subject to
1 )Load Flow Constra int s 2 )Voltage and frequency regulation Constra int s 3 ) DG Capacity constra int s
(13)
where Pi and Qi represent the change in the active and reactive power load at bus i from the current operating point to the point of voltage collapse, respectively. ||S|| represent the distance from the current operating point to the point of voltage collapse. As it can be seen in (13) the load can be increased in different ways, the objective of proposed optimization problem is to find the direction of load increase such that the distance to voltage collapse is minimal. The optimization problem is subjected to different power balance, voltage and frequency regulation and DG capacity constraints. The detailed explanation of the problem constraints is given here under:
PGi SGi ,max
(17)
QGi
IV.
(SGi ,max )2 (PGi )2

NUMERICAL RESULTS
(18)
Load flow constraints F ( Vi , i , ,Pi , Qi ) = 0 (14)
where F represent the power flow equations of the droop controlled islanded microgrid system, as described in (3)-(12).
The proposed algorithm for determining the closest saddle node bifurcation point of droop-controlled islanded microgrid, presented in Section III, is tested on a 33-bus islanded microgrid system. Fig. 2 shows the single-line diagram of the test system. The feeder parameters and load nominal power are given in [15]. An islanded microgrid is formed by isolating the system from the main grid by the isolation switch. Four DG units have been allocated to feed the system in the islanded microgrid mode of operation. The DG units ratings, locations, and droop settings are given in Table I. The proposed optimization problem for determining the closest saddle node bifurcation point of the droop controlled islanded microgrid is solved by a pattern-search method implemented in Matlab environment. Two case studies have been performed to test the applicability of the proposed method.
Figure 3. Voltage profile at maximum loading point (Case Study #1)
Figure 2. The 33-bus microgrid test system. Data about the line impedances and system loads can be found in [15]
TABLE I.
DG # 1 2 3 4 5
DG UNITS LOCATIONS, STATIC DROOP COEFFICIENTS, NOMINAL SETTING AND RATINGS IN THE 33-BUS TEST SYSTEM
Location mp (p.u.) 5.00e-3 5.00e-3 2.50e-3 2.50e-3 5.00e-3 nq (p.u.) 0.100 0.100 0.050 0.050 0.100 * (p.u.) 1 1 1 1 1 V* (p.u.) 1.02 1.02 1.02 1.02 1.02 Sgmax (p.u.) 1 1 2 2 1
08 09 22 02 25
A. Case Study #1 In this case study no voltage and frequency regulation constraints are imposed at the system maximum loading point. Fig. 3 shows the system voltage profile at the maximum loading condition. as can be seen in Fig. 3 at the point of maximum loading (i.e. the point beyond which any increase in system loading will lead to voltage collapse) the system voltage is not within the voltage regulation bounds. Fig. 4 shows the active and reactive power generation by the different DG units at the point of maximum loading. The results in Fig. 4 show that at the point of maximum loading the DG units are operating at their maximum apparent power rated capacities. Table II show the change in active and reactive power at each load bus which lead to closest saddle node bifurcation in case study #1. B. Case Study #2 In this case study, voltage and frequency regulation constraints are imposed on the maximum loading point. The upper and lower limits on the node voltage variation are taken as 1.1 p.u. and 0.9 p.u., respectively. A maximum frequency variation of 0.5% is considered in the reported case studies. Fig. 5 shows the system voltage profile at the maximum loading condition. As can be seen in the Figure, at the point of maximum loading the system voltage is still within the allowable voltage regulation bounds. Fig. 6 shows the active and reactive power generation by the different DG units at the point of maximum loading. Again, this results show that at the maximum loading point the DG units are generated their maximum apparent power capacity. Table II also show the change in active and reactive power at each load bus which lead to closest saddle node bifurcation in case study #2.
Figure 4. DG units active and reactive power generation at maximum loading condition (Case Study #1)
Figure 5. Voltage profile at maximum loading point (Case Study #2)
Figure 6. DG units active and reactive power generation at maximum loading condition (Case Study #2)
TABLE II. ACTIVE AND REACTIVE POWER INCREMENTS AT DIFFERENT LOAD BUSES WHICH CAUSES VOLTAGE COLLAPSE FOR THE 33-BUS ISLANDED MICROGRID SYSTEM
productions in both cases is due to the difference in the loading margin. V.

CONCLUSION
Case Study #1 Bus # 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 ||S|| P KW 47.28 47.32 48.25 48.82 49.42 51.14 51.28 51.12 51.47 52.89 53.10 53.43 55.02 55.73 56.14 56.52 57.21 57.33 47.25 45.47 44.79 43.40 48.32 48.41 48.09 51.46 51.87 53.51 54.63 55.10 55.93 55.97 56.08 1703.79 2314.22 Q KVAR 44.34 44.33 45.01 45.50 45.95 47.10 46.89 46.75 46.96 47.94 48.06 48.37 49.51 49.93 50.20 50.45 50.88 50.94 44.39 42.79 41.95 40.82 45.18 45.25 45.02 47.35 47.75 49.42 50.64 51.28 51.62 51.76 51.79 1566.12 KVA
Case Study #2 P KW 59.06 59.08 48.52 42.98 38.43 31.21 30.87 29.62 27.76 25.72 25.41 24.88 23.02 22.43 21.95 21.51 20.95 20.80 63.37 165.91 246.49 489.77 48.22 47.89 48.12 30.25 29.04 25.22 23.02 21.96 20.40 20.08 19.91 1873.86 2509.89 Q KVAR 51.31 51.32 46.63 43.71 41.05 33.80 32.36 32.01 30.48 28.76 28.60 28.33 26.52 25.68 25.19 24.78 23.97 23.82 54.40 109.54 159.48 397.78 46.52 46.44 46.67 33.13 32.26 27.94 25.43 24.60 22.82 22.39 22.09 1669.78 KVA
This paper proposes an algorithm for determining the closest saddle node bifurcation point of droop controlled islanded microgrid. The problem is formulated as a nonlinear optimization problem. The problem is subjected to the different system constraints; including DG capabilities, power flow constraints, frequency and voltage regulation. A detailed droop controlled islanded microgrids model that takes into consideration its special philosophy of operation has been adopted in the optimization problem formulation. The proposed methodology has been tested on a 33-bus microgrid system. The numerical results verify the effectiveness of the proposed algorithm in assessing the proximity of the droop-controlled islanded microgrid to voltage instability. The proposed algorithm can be a powerful tool that helps the islanded microgrid operator to perform steady state voltage stability analysis.
REFERENCES [1] W. El-Khattam, and M. M. A. Salama, "Distributed generation technologies, definitions and benefits, Electric Power Systems Research, vol. 71, pp. 119-128, 2004. [2] R. C. Dugan, and T. E. McDermott, "Distributed generation," IEEE Ind. App. Mag., vol. 8, no. 2, pp. 19-25, Mar. 2002. [3] F. Blaabjerg, Z. Chen, and S. B. Kjaer, Power electronics as efficient interface in dispersed power generation systems, IEEE Trans.Power Electron., vol. 19, no. 5, pp. 1184-1194, Sept. 2004. [4] N. Pogaku, M. Prodanovic and T. C. Green, Modeling, analysis and testing of autonomous operation of an inverter-based microgrid, IEEE Trans. on Power Electron., vol. 22, no. 2, pp. 613-625, Mar. 2007. [5] Y. Mohamed and E. F. El-Saadany, Adaptive decentralized droop controller to preserve power sharing stability of paralleled inverters in distributed generation microgrids, IEEE Trans. Power Electron., vol. 23, no. 6, pp. 2806-2816, Nov. 2008. [6] H. E. Farag, M. M. A. Abdelaziz, and E. F. El-Saadany, Voltage and reactive power impacts on successful operation of islanded microgrids, IEEE Trans. Power Syst., to be published, vol. 28, no. 2, pp. 1716-1727, May. 2013. [7] IEEE guide for design, operation, and integration of distributed resource island systems with electric power systems, IEEE standard 1547.4, July 2011. [8] E. Haesen, C. Bastiaensen, J. Driesen, and R. Belmans, A Probabilistic formulation of load margins in power systems with stochastic generation, IEEE Trans. Power Syst., vol. 24, no. 2, pp.951-958, May 2009. [9] I. Dobson and L. Lu, New methods for computing a closest saddle node bifurcation and worst case load power margin for voltage collapse, IEEE Trans. Power syst., vol. 8, no. 3, pp. 905-913, Aug. 1993. [10] G. Diaz, and C. Gonzalez-Moran, Fischer-Burmeister-based method for calculating equilibrium points of droop-regulated microgrids, IEEE Trans. Power Syst., vol. 27, no. 2, pp. 959-967, May 2012. [11] M. M. Abdelaziz, E. F. El-Saadny, and R. Seethapathy, Assessment of droop-controlled islanded microgrid maximum loadability, Proc.PES General meeting 2013, Accepted for publication. [12] M. M. Abdelaziz, H. E. Farag, E. F. El-Saadany, and Y. A.-R. Mohamed, A novel and generalized three-phase power flow algorithm for islanded microgrids using a newton trust region method, IEEE Trans. Power Syst., vol. 28, no. 1, pp. 190-201, Feb. 2013. [13] P. Kundur, Power system loads, in Power System Stability and Control, New York: McGraw-Hill, 1994, ch. 7, pp. 271-314. [14] N.R. Ullah, K. Bhattacharya and T. Thiringer, Wind farms as reactive power ancillary service providers-technical and economic issues, IEEE Trans. Energy Convers., vol. 24, no. 3, pp. 661-672, 2009. [15] M. E. Baran, and F. F. Wu Network reconfiguration in distribution systems for loss reduction and load balancing, IEEE Trans. Power Del., vol. 4, no. 2, pp. 1401-1407, Apr. 1989.
C. Discussion The results obtained in case studies #1 and #2 show the effectiveness of the proposed algorithm in determining the worst case loading margin of droop controlled islanded microgrid systems. As it can be expected, the worst case loading margin in case study #2 is higher than the worst case loading margin in the case study #1 due to imposing the voltage and frequency regulation constraints at the maximum loading point. The active and reactive power sharing between the DG units in case studies #1 and #2, depicted in Figs. 4 and 6, show that the DG unit output power follow the droop characteristics described in section II accurately, which in turn reflect the accuracy of the adopted model in reflecting the droop-controlled islanded microgrid operational characteristics. The small difference between the DG
Morad Mohamed Abdelmageed Abdelaziz (S11) was born in Cairo, Egypt, on September 27, 1984. He received the B.Sc. (with honors) and M.Sc. degrees from Ain-Shams University, Cairo, Egypt, in 2006 and 2009, respectively, both in electrical engineering. He is currently pursuing the Ph.D. degree in the Department of Electrical and Computer Engineering, University of Waterloo, Waterloo, ON, Canada. He was an Electrical Design Engineer with Dar Al-Handasah Consultants (Shair and Partners) from 2006 to 2010. His research interests include dynamics, controls, and analysis of microgrids; distributed and renewable generation modeling, analysis, and controls; and power electronics and their applications in smart grids. Ehab F. El-Saadany (SM05) was born in Cairo, Egypt, in 1964. He received the B.Sc. and M.Sc. degrees in electrical engineering from Ain Shams University, Cairo, Egypt, in 1986 and 1990, respectively, and the Ph.D. degree in electrical engineering from the University of Waterloo, Waterloo, ON, Canada, in 1998. Currently, he is a Professor in the Department of Electrical and Computer Engineering, University of Waterloo. His research interests are smart grids operation and control, power quality, distributed generation, power electronics, digital signal processing applications to power systems, and mechatronics.
Techno-Economic Analysis on Using Battery Storage System with Household PV Panels

Hamed Naderi Rad
Electrical Engineering Department, Faculty of Engineering, Ferdowsi University of Mashhad Mashhad, Iran hamed.naderi.rad@gmail.com
Javad Sadeh
Electrical Engineering Department, Faculty of Engineering, Ferdowsi University of Mashhad Mashhad, Iran sadeh@ferdowsi.um.ac.ir attractive profitability for users and enhance the investments on these systems in Spain. They concluded that economic profitability is related to the power management model and storage capacity. Nottrott et al [7] proposed a routine linear programming to optimize dispatch schedule in order to obtain better financial benefits. Iran, as a developing country, will have a growing need to new sources of energy and also have proper climates in order to use PV household systems. In this paper, a sample house in Mashhad the most populated city in Iran after Tehran- is the case study. Firstly, the components of hybrid PV- battery system and data sources will be introduced. After that, a kind of definite power management algorithm (PMA) will be performed on the hybrid system and the results will be analyzed. Furthermore, this procedure and the result analysis will be done with GA algorithm. And finally, two economic incentives will be applied to improve economic condition of these algorithms. II. HYBRID SYSTEM COMPONENTS PV panels are the main part of this system. Battery storage bank is also working as a buffer and a back-up source beside PV panels. Apart from these, inverter, charge controller, special connection wires, and a two-way meter are other components of this hybrid system. As shown in Fig. 1, power generated from PV is delivered to power manager system (PMS). PMS is a package included an inverter and a charge controller. The PMA is performed in the PMS, so the exchange power with grid (Pgrid) and battery power (Pbat) would be specified. The two-way meter can work with variable pricing and tariffs. In Table I and Table II, technical and economical specifications of hybrid system components are presented. Using 1kWp PV panels is common in household applications. Celik [5] and Colmenar-Santos et al. [6] have used this capacity in their case studies. Capacity of battery was selected based on minimization of capital cost and prohibiting of deep discharge simultaneously. III. DATA USED IN SIMULATIONS Performing PMAs need three types of data. Load data, weather data and pricing data. Each type is introduced and analyzed in following.
Abstract Small-scale PV power generation has been developed in recent years. Power output fluctuations and incapability of energy storing are two main drawbacks of this system. Battery storage system can effectively improves the drawbacks of household PV, but on the other side batteries have high capital cost. In this paper, a techno-economical analysis with two different power management algorithms has been carried out to evaluate the project of using batteries in household PV system. The results indicate that this project can be economically feasible if the government applied incentive policies for household PV systems.
Index Terms- Household PV, Economic evaluation, Battery

storage system
I.
INTRODUCTION
Today, using of renewable energy to provide parts of consumers needs is common in many countries. In smallscale usage like a house, photovoltaic (PV) has many excellences. Household PV generation is also feasible especially in regions with high solar irradiation and too many sunny days. In order to improve some downsides of household grid-connected PV, battery bank storage is a proper choice. It enables the whole system to store the generated power, so this power can be exported to the grid when the energy price is high [1]. On the other side, increasing the penetration level of PV has a negative impact on the distribution network. The main impact is due to output power fluctuations of PV and using battery storage can overcome this problem [2], [3]. In order to use hybrid generation system consists of PV panels and battery bank, an economic analysis should be performed. McKenna et al. [1] presents an economic and environmental impact analysis for the use of lead-acid batteries in PV systems under current feed-in tariff arrangements in the UK and found negative net present value for this usage. Su et al. [4] performs a demand-side technoeconomic analysis with fixed capacity of battery. They concluded that economic feasibility is strongly related to load shape of end user. Celik [5] has carried out a techno-economic evaluation on a household PV- battery system with long time data of Ankara. The results show that the system is unlikely to be applied without government incentives. Colmenar-Santos et al. [6] aimed to find models for the remuneration of energy generated by small photovoltaic systems that provide
A. Load Data In household consumers, it is quit hard to find a static load pattern that repeats every day. So, to find an approximate load pattern, it is necessary to use feeder data and make an estimation based on the large number of consumers. A simple approach is recommended in [2]. This approach is performed on our data and an approximate pattern created. This pattern is shown in Fig. 2. B. Weather Data PV power generation is totally dependent on solar irradiance. Output power is also affected by temperature changes, but it is not as important as irradiance is. Average monthly direct irradiance for Mashhad is adopted form NREL [9] which is shown in Appendix. On the other hand, a normal pattern is used in PMAs. This pattern is created by data sampling operation done on 8th November 2012 in Air and Sun Research Lab. from Ferdowsi University of Mashhad and it is shown in Fig. 3. Sampling period is 10 minute, so in each day 144 samples are available. The average Direct Normal Irradiance (DNI) of this pattern is close to average yearly DNI of NREL, so it can be generalized to whole year because in this study the amount of power generated is more important than the pattern of generation. In order to verify this claim, different patterns are used for each month. Each pattern has similar average DNI to NREL data. In Table III, results are compared to each other. When different patterns are used, the total yearly generated power changes about 3.5% in compare with using single pattern. This amount of difference would be ignorable in long-time study. C. Pricing Data In this paper, it is assumed that the electricity market exists and based on day-ahead market, energy pricing pattern is clear for the following day. On the other hand, Iranian network is a traditional and centralized network which manages by the government. So, there is no spot market and variable energy price for household consumers. According to lack of pricing data in Iran and in order to run PMAs, one pattern is selected randomly (18 Oct. 2012) from Ontario market [10] and it is shown in Fig. 4. IV. PRICE-BASED POWER MANAGEMENT ALGORITHM (PPMA)
Figure 1. Schematic of hybrid renewable system
TABLE I. Photovoltaic module Aria Solar ASM200 200 40.08 4.99 45.7
TECHNICAL SPECIFICATIONS OF HYBRID SYSTEM

COMPONENTS
Battery Faran LAFEI 12V 120 5.76 12 36
Inverter-Charge Controller
Manufacturer
Manufacturer
Manufacturer
Studer XPC 220048 48 20 16 95
Model
Model
Model
Nominal power (W) Voltage (V)
Capacity (Ah) Capacity (kWh) Voltage (V) Max charge current (A) Best charge current (A)
Battery Voltage (V) Charge current (Adc) Maximum current (Aac) Maximum efficiency (%) Output power (kWh)
Current (A) Open circuit voltage (V) Short circuit current (A)
5.42
12
1.6
TABLE II.
Component Photovoltaic Battery Inverter Charge Controller
ECONIMICAL SPECIFICATIONS OF HYBRID SYSTEM

Number of modules Life time (year) Capital Cost ($)
5 4 1
30 6 30
1166 1000 2300
Adding battery bank as a storage system to PV panels makes the owner of system use a power management algorithm. This algorithm has two technical and economic aspects. From technical view, battery bank has some constraints which must be satisfied in order to ensure its health e.g. battery should not charge more than its maximum current or it should not be discharged more than 30% of its capacity. In the economical aspect, make an investment on battery with PV must have more net present value than using the PV alone. This aspect will be discussed in section VI.
TABLE III. COMPARISON OF SINGLE AND DIFFERENT IRRADIANCE PATTERN IN PV POWER GENERATION
Time period Generated PV power (kWh)
Pricing pattern adopted from Ontario market on 18 Oct. 2012 0.06 0.055 0.05 0.045 Price ($/KW h ) 0.04 0.035 0.03 0.025 0.02 0.015 0.01 0 50 Time (10 min.) 100 150
Jan. Feb. Mar. Apr. May Jun Jul. Aug. Sep. Oct. Nov. Dec. Total yearly power generated by different patterns Total yearly power generated by single pattern
124.80 139.03 136.61 162.70 182.85 215.01 218.02 211.26 196.85 169.12 145.76 121.80 2023.81
Figure 4. Energy pricing pattern used in PMAs
generation is not the only goal because the system is gridconnected. Another goal is using PV power generation when it has maximum profit. This profit can be gained by selling PV power generation to grid, or to supply the load or even in storing in battery. The PPMA was performed for one day, using the data which were presented in Figs. 2-4 and the results are shown in Fig. 5. As indicated in Fig. 5(a), battery state of charge is changing between 74% and 89%. By this allowed change, 6 years life time for the batteries can be achieved. Exchange power with grid is also depicted in Fig. 5(b). On the other side, Fig. 3 is also representing the injected power to the grid in PV alone scheme. In PPMA some constraints were also applied to control the amount of exchange power with grid. So, by comparing Fig. 3 and Fig. 5(b), it is clear that the amount of power and the size of fluctuations which are injected to the grid reduced significantly.
1952.4
A Sample Normalized Load data on 19 Jan. 2004 in Mashhad 650 600 550 Load (W) 500 450 400 350 300
V.
0 50 Time (10 min.) 100 150
INTELLIGENT POWER MANAGMENT ALGORITHM (IPMA)
Figure 2. Sample load data used in PMAs

A sample 1kWhp PV power generation on 8 Nov. 2012 1000 900 800 700 PV output power (W) 600 500 400 300 200 100 0
50 Time (10 min.)
100
150
Figure 3. 1kWhp PV panel power output used in PMAs
If it is assumed that an estimation of solar irradiance can be available, then it is possible to use intelligent algorithms. These algorithms search the whole solution space, so they may find better solutions in compare with definite algorithms. On the other hand, forecasting methods of solar irradiance is now under study, so if technical and economical results improve significantly with intelligent algorithm, it will be justifiable to invest on forecasting methods. Genetic algorithm (GA) is the most common and easiest algorithm among intelligent algorithms. GA with elitism is applied in this paper and the results are shown in Fig. 6. As shown in Fig. 6(a), battery charge status is in the allowed range. In Fig. 6(b), the exchange power with grid is shown and like PPMA, the amount of injected power and size of fluctuations have decreased in compare with PV alone scheme. This decrease is due to penalty factors which were applied in the fitness function of GA. VI. ECONOMIC EVALUATION
Furthermore, conventional PMAs generally focus on fulfilling the load with PV power generation. Lack of power is supported by grid or battery and the extra power must sell to grid or store in battery. In PPMA, supplying the load with PV
In order to evaluate the economic efficiency of different projects, it is essential to return all incomes and costs to a
Battery State of Charge (SOC)
100
where is net present income and can be given as [11]: 1 1 1 1
is net present cost. (2)
90
80
70
50 Time (10 min.)
100
150
(a)
800 600 400 Grid's Power 200 0 -200 -400 -600 0 50 Time (10 min.) 100 150
In Eq. (2), is the income of project. This income is obtained by selling renewable electricity to the grid. is also called annual interest rate and can be calculated as [11]: 1 1 1 (3)
where is effective interest rate and is annual increase in electricity price which are assumed 10% and 20%, respectively in this paper. is also given by [11]: & (4)
(b) Figure 5. (a) Battery charge status during one day in PPMA. (b) Exchanged power with grid, negative power means the grid is getting power and positive means it gives power
B attery State of Charge (S OC) 100 90 80
70
50 Time (10 min.)
100
150
(a)
1200 1000 800 600 G rid 's P o w e r 400 200 0 -200 -400 -600 -800
, and are capital costs of PV In Eq. (4), panels, battery bank and inverter-charge controller, respectively. is the replacement cost of batteries during life time of project. It is needed to replace battery bank 4 times because the project life time is 30 years and in 30% depth of discharge each battery has 6 years life time. & is operation and maintenance cost which increases over useful life of the project. In calculation of and & , effective interest rate (ir) is also 10%. By using Eqs. (1)-(5) the project of adding battery to a household PV plant can be comparable economically with the case that PV was used alone. This comparison is made and the results are shown in Table IV. As the results show in Table IV, applying battery bank as a storage system with household PV plant has less NPV in compare with using the PV alone. It mainly happens because of high capital cost and short life time of batteries. VII. INCENTIVE POLICIES FOR THE HOUSEHOLD PV OWNER In order to compensate the NPV decrease and make household PV-battery project profitable, a possible way is providing additional incentives to the household PV owner [2]. These incentives can be in the form of increasing the selling price of PV energy or giving equipment purchase loan to the owner of system. A. Increase PV energy selling price Changing feed-in tariffs can improve NPV for consumers and increase self-consumption simultaneously. In this change, consumers with PV systems can profit more by using their generated power than injecting it into grid [1]. But in Iran, the policy of government is to develop household PV systems, so feed-in tariffs are specified to encourage PV owners export their generated electricity into grid. So, incentive coefficients are defined to differentiate between buying and selling price
50 Time (10 min.)
100
150
(b) Figure 6. (a) Battery charge status during one day in IPMA. (b) Exchanged power with grid, negative power means the grid is getting power and positive means it gives power
reference time and present time is more common. Net present value (NPV) is a good criterion and it can be represented by: (1)
e.g. if 0.05 $/kWh is the price for imported electricity, by applying 30% incentive coefficient the export price would be 0.065 $/kWh. Different incentive coefficients are applied and the results are shown in Table V. The results indicate that NPV and incentive coefficients have an approximate linear relationship in PPMA but in IMPA increasing the incentive coefficient to 30% makes the NPV of PV-battery scheme equal to the PV alone scheme. B. Equipment Purchase loan Economic analysis clarify that the main cost in NPV is capital cost of equipment. The government can improve the NPV significantly by giving equipment purchase loan. This loan can be paid in two cases. In case 1, the loan is paid just at the beginning of the project to buy equipment. In case 2, the loan is paid both for buying the equipment and replacing them during the project life. The interest of loan is assumed 5% and the results are shown in Table VI. As indicated in Table VI, equipment purchase loan is changing the NPV significantly and in case 2 both batterybased schemes have more NPV in compare with the PV alone scheme. VIII. CONCLUSION Economic analysis has been made on the insertion of battery bank to a grid-connected household PV system in Mashhad, Iran. In this paper, the purpose of adding is to store generated power and use it when the energy price is high. Battery bank can also be used to smooth PV power fluctuations. In order to evaluate the operation of batteries in this system, a conventional definite and an intelligent power management algorithm have been applied. The results indicate that the net present value (NPV) of both algorithms is less than the PV alone scheme. This mainly happens due to high capital cost of equipment (especially batteries) and low energy price in Iranian network. In order to improve economic condition of PV-battery scheme, incentive policies have been proposed. These incentives are in form of increasing the selling price of renewable generated energy and giving equipment purchase loan to the owner of household PV system. By applying incentive policies, the NPV of PV-battery scheme improved significantly and in some cases it became economically feasible. ACKNOWLEDGMENT The authors gratefully acknowledge Air and Sun research laboratory in Ferdowsi university of Mashhad for supporting the data sampling operation in this project.
TABLE IV.
Scheme
ECONOMIC ANALYSIS ON DIFFERENT RENEWABLE SCHEMES AND PMA

Power Management Algorithm Net Present Value ($) Payback Time (year)
PV PV + Battery PV + Battery
PPMA IPMA
4211 2663 3499
22.02 26.47 25.59
TABLE V.
Incentive coefficient (%)
EFFECT OF DIFFERENT INCENTIVE COEFFICIENTS ON NPV

NPV for PV alone ($) NPV for PV + Battery ($) Priced-based Intelligent PMA PMA
0 5 10 15 20 25 30
4211 4601 4992 5382 5772 6162 6553
2663 3130 3597 4065 4532 5000 5467
3499 4008 4517 5026 5535 6045 6553
TABLE VI.
EFFECT OF EQUIPMENT PURCHASE LOAN ON NPV

NPV without loan($) NPV with loan($) Case 1 Case 2
PV PV + Battery PV + Battery
PPMA IPMA
4211 2663 3499
5565 4403 5239
5565 5879 6715
APPENDIX
TABLE VII. AVERAGE DNI FOR MASHHAD ADOPTED FROM NREL [9]
Month Avg. DNI (kWh/m2 per day)
Jan. Feb. Mar. Apr. May Jun Jul. Aug. Sep. Oct. Nov. Dec. Avg. Yearly
4.14 4.62 4.52 5.15 6.1 7.12 7.24 7.06 6.56 5.63 4.82 4.08 5.59
REFERENCES
[1] E. McKenna, M. McManus, S. Cooper, and M. Thomson, Economic and environmental impact of lead-acid batteries in grid-connected domestic PV systems Applied Energy, vol. 104, pp. 239-249, April 2013 W. A. Omran, M. Kazerani and M. M. A. Salama, Investigation of methods for reduction of power fluctuations generated from large gridconnected photovoltaic systems, IEEE Transactions on Energy Conversion, vol. 26, no. 1, pp. 318-327, March 2011 R. Passey, T. Spooner, I. McGill, M. Watt, and K. Syngellakis, The potential impacts of grid-connected distributed generation and how to
[2]
[3]
address them: A review of technical and non-technical factors, Energy Policy, vol. 39, no. 10, pp. 6280-6290, July 2011 W-F Su, C-E Lin, and SJ Huang, Economic analysis for demand-side hybrid photovoltaic and battery energy storage system IEEE Transactions on Industry Applications, vol. 37, no. 1, pp. 171-177, Jan./Feb. 2001 A.N. Celik, Present status of photovoltaic energy in Turkey and life cycle techno-economic analysis of a grid-connected photovoltaichouse, Renewable and Sustainable Energy Reviews, vol. 10, no. 4, pp. 370-387, August 2006 A. Colmenar-Santos, S. Campez-Romero, C. Prez-Molina, and M. Castro-Gil, Profitability analysis of grid-connected photovoltaic facilities for household electricity self-sufficiency, Energy Policy, vol. 51, pp. 749-764, December 2012 A. Nottrott, J. Kleissl, and B. Washom, Energy dispatch schedule optimization and cost benefit analysis for grid-connected, photovoltaicbattery storage systems, Renewable Energy, vol. 55, pp. 230-240, July 2013 [8] P. Salah, E. A. Basiri, D. Abedi, and E. A. Ghadimi, Finding behavior and the hourly demand of household consumers using load cumulative measurement and normalized billing 16th Electrical Power Distribution Conference (EPDC), Bandar Abbas, Iran, 18- 20 April, 2011 [9] NREL Solar Radiation Research Laboratory (BMS). [Online] Available: http://www.nrel.gov/midc/srrl_bms/ [10] Hourly Ontario Energy Price (HOEP) Report. [Online] Available: http://www.ieso.ca/imoweb/marketdata/hoep.asp [11] G. J. Thuesen, W. J. Fabrycky, Engineering Economy, 9th ed., PrenticeHall, New York, 2000 [7]
[4]
[5]
[6]
PSCAD Simulation of Grid-Tied Photovoltaic Systems and Total Harmonic Distortion Analysis
Abdulrahman Kalbat Electrical Engineering Department Columbia University in the City of New York New York, NY 10027 Email: ak3369@columbia.edu
AbstractWith the increasing fears of the impacts of the high penetration rates of Photovoltaic (PV) systems, a technical study about their effects on the power quality metrics of the utility grid is required. Since such study requires a complete modeling of the PV system in an electromagnetic transient software environment, PSCAD was chosen. This paper investigates a grid-tied PV system that is prepared in PSCAD. The model consists of PV array, DC link capacitor, DC-DC buck converter, three phase six-pulse inverter, AC inductive lter, transformer and a utility grid equivalent model. The paper starts with investigating the tasks of the different blocks of the grid-tied PV system model. It also investigates the effect of variable atmospheric conditions (irradiation and temperature) on the performance of the different components in the model. DC-DC converter and inverter in this model use PWM and SPWM switching techniques, respectively. Finally, total harmonic distortion (THD) analysis on the inverter output current at PCC will be applied and the obtained THD values will be compared with the limits specied by the regulating standards such as IEEE Std 519-1992. Index TermsEMTDC, photovoltaic systems, power system harmonics, power system simulation, PSCAD, total harmonic distortion.
I. I NTRODUCTION The future plans for smart grid deployment , which mainly depend on the concept of distributed generation, make gridtied photovoltaic systems more attractive solution, especially in regions with the high solar irradiation values. With the continuous advancement in power electronics and the increasing number of consumer appliances which utilize power electronics, both the customers and the utility operators are more aware of the importance of power quality of the utility grid. One of the most important power quality issues is harmonic distortion which is mostly caused by nonlinear power electronic based loads that are connected to the grid. Connecting PV systems to the utility grid, which is already being injected with harmonics from nonlinear loads, will introduce more stress on the power quality of the grid. All previously addressed issues makes it clear that more studies should be conducted on the effects associated with the high penetration rates of photovoltaic system on the power quality of the utility grid. Total Harmonic Distortion analysis could be used to measure the magnitude of the harmonic distortion in any power network. Since such study requires a complete modeling of the PV system in an electromagnetic transient software environment,
PSCAD was chosen to simulate the grid-tied PV system in this paper. PSCAD is a powerful and exible graphical user interface to the world-renowned EMTDC simulation engine but it does not provide power ow solutions. Another feature of PSCAD is its ability to interface with Simulink. This feature enables the researchers to combine the exible power systems simulation of PSCAD with the rich and ready-to-use control systems library of Simulink which helps in minimizing the modeling time. Also, PSCADs interface is designed in a way that is easily used by researchers and developers in the power systems eld [1]. In section II of this paper, a detailed description of the grid-tied photovoltaic system PSCAD model will be given. The description include a detailed discussion of the different components of the PV model along with the operation, modeling and control principles of each component. Also, the effects of variable atmospheric conditions on the output of some components will be presented. The output signals of most important stages of the system will plotted. In section III, the Total Harmonic Distortion (THD) analysis will be applied to the PSCAD model and the obtained results will be discussed by comparing them to the IEEE Std 519-1992. II. PSCAD M ODEL OF G RID -T IED P HOTOVOLTAIC S YSTEM The PSCAD model used in this paper is based on [2] and it mainly consists of PV array model, DC link capacitor, DC-DC converter, three phase inverter, AC lter, transformer and utility grid equivalent model, as shown in Fig. 1. In this section, the tasks and some of the important parameters which dene each components model will be discussed. The tasks of the different additional control circuits which are used to control some components in the main model will be also discussed. The effects of variable atmospheric operation conditions (irradiation and temperature) will be discussed for some components. A. PV Array Photovoltaic cell is the basic semiconductor device that generates electricity by the photovoltaic effect when exposed to radiant energy such as sunlight [3]. The default parameters values were used to dene the PV module in PSCAD. The
Ipv Vpv
Icon 0.01 [H] T1 dcVltg
Idc Timed Breaker Logic Closed@t0 gt1
5 gt5
3 gt3
BRK_cap
R=0
V+ G
Ia_motor Eab
Ebc 2e-4 #1 #2
SolRad
Tcell CellTemp V-
P = 0.1467 Q = -0.0006585 V = 11.06 A V
Icharge
model enables the user to specify the number of series and parallel cells per module and the number of modules connected in series and in parallel which helps in building PV systems with high power rating. By using the default values, the nal output of the single module is 650 watt and 260 kilo-watt for the total 400 modules. Increasing the input irradiation value would increase the short circuit current of the PV while increasing the input temperature value would decrease the open circuit voltage of the PV. B. DC Link Capacitor The DC link minimizes the ripple of the PV source current by using a large capacitor. It is assumed in determining the size of DC link capacitor that the output current is ripple free. The voltage across the Dc link capacitor is controlled by the DC-DC converter as discussed next. C. DC-DC Converter for MPP Tracking DC-DC converter is used for Maximum Power Point Tracking (MPPT) by controlling the voltage across the DC link capacitor and the PV array. This is achieved by rst creating a reference voltage that is then supplied to a PI controller which creates switching signals that force the voltage across the PV array to follow the reference voltage. These two stages are discussed next. 1) Maximum Power Point Tracking (MPPT): The model used for creating the reference voltage is shown in Fig. 2. First, photovoltaic output current (Ipv) and output voltage (Vpv) are passed through a rst order low pass lter with a magnitude of G = 1 and a time constant of T = 0.01 seconds in order to lter out the high frequency components or harmonics from these signals. The ltered current and voltage signals (IpvF and VpvF ) are then fed into the MPPT control block that uses the Incremental Conductance Tracking Algorithm. An algorithm that is based on the fact the slope of the PV array power curve shown in Fig. 3 is zero at the Maximum Power Point (MPP), positive on the left of the MPP, and negative on the right. The MPP can thus be tracked by comparing the instantaneous conductance (I/V) to the incremental conductance (I/V) [4] as in (1): I/V = I/V, at MPP I/V > I/V, left of MPP (1) I/V < I/V, right of MPP
10000
80000
Fig. 1. Grid-Tied PV model in PSCAD.
BRK_cap
2 2 gt2 6 6 gt6 4 4 gt4 Ec Eb Ea
Vpv Ipv
G 1 + sT G 1 + sT
Vpv Ipv
MPPT control
Vmpp Vmppt
Ipv_F
Vpv_F
Fig. 2. Maximum power point tracking model in PSCAD.
!
Fig. 3. Characteristic PV array power curve at STC. PV array power output in kWatt (y-axis) and PV array voltage in kV (x-axis).
Based on the previous three cases, the MPPT generates a reference voltage (Vmppt) at which the PV array is forced to operate. The algorithm decrements or increments Vmppt to track the maximum power point when operating under varying atmospheric conditions. This reference voltage Vmmpt is used as an input to the DC-DC Converter Control model discussed next. 2) DC-DC Converter Control: DC-DC converter is an electronic circuit that is used either to step down the input voltage (buck converter) or to step up the input voltage (boost converter). In this PSCAD model, buck converter was used that consists of a Pulse Width Modulation circuit (shown in Fig. 4), Insulated Gate Bipolar Transistor (IGBT) switch, inductor, capacitor and free-wheel diode [5], as shown in Fig. 1. The difference between the solar panel output voltage (Vpv) and the reference voltage (Vmppt) is used as an input to the Proportional-Integral (PI) controller, shown in Fig. 4, which
P
P Vpv
Vmppt
D + F
A B
dcVltg 0.5
Ctrl = 1
Ang
Ctrl
Vpv_ref
A B Comparator
T1
Mppt_ON/OFF
P 0.0 Q D + F I Mag
Fig. 4. DC-DC Converter Control model in PSCAD.
Fig. 6. Simple P and Q regulation model in PSCAD
D. Three Phase Inverter In order to be able to tie a PV system with the utility grid, the DC output power of the DC-DC converter should be converted into a three phase AC power using a three phase inverter. IT is part of inverters task to keep the DC voltage across its input (DC-DC converters output) at a constant value. In this PSCAD model, the three phase inverter consists of a simple P and Q regulation circuit, a ring pulse generator and a three phase inverter bridge. 1) Simple P and Q Regulation: In order to establish a constant DC bus voltage (dcVltg) between the DC-DC converter and the inverter, a PI controller, shown in Fig. 6, is used to set this voltage at 0.5 kV. The output of the controller (Ang) will be used as an input to the ring pulse generator which will be discussed next. The second PI controller sets the reactive power (Q) of the grid to zero which forces the inverter to operate at unity power factor so that it produces sinusoidal voltage and current which are in phase. The output of this controller (Mag) will be also used as an input to the ring pulse generator. 2) Firing Pulse Generation: The switching signals of the 6 IGBT switches of the 3-legged inverter bridge shown in Fig. 1 were generated using a Sinusoidal Pulse Width Modulation (SPWM) technique shown in Fig. 7. It starts with creating three sinusoidal modulating waves with a frequency of 60 Hz and a phase shift equal to the output of the previous PI controller (Ang) with additional shifting of -120 and 120 degrees. The magnitude of the modulating waves is equal to (Mag) from the previous PI controller. Then, the three sinusoidal modulating waves were compared with a triangular carrier wave with magnitude ranging between -1 and 1. Switching signals gt1, gt3 and gt5 were generated by setting the output of the comparator to 1 whenever the modulating wave is greater than the carrier wave and 0 otherwise. Since the operation of the two switches in each of the three legs of the inverter bridge should be complementary to produce the nal sinusoidal wave, the switching signals gt4, gt6 and gt2 were generated by inverting the switching signals gt1, gt3 and gt5, respectively [7]. 3) Three Phase Inverter Bridge: By applying the previously generated switching signals (gt1 to gt6) to the 6 IGBT switches shown in Fig. 1, the inverter
!
Fig. 5. MPPT reference voltage (Vmppt) and PV array output voltage (Vpv) at 1000 W/m2 and 500 W/m2 .
then, based on this difference, controls the duty cycle of the PWM pulse. The duty cycle, dened as the fraction of the period during which the switch is on, ranges between 0 and 1. A duty cycle value of 0.5 means on and off time are equal, a value greater than 0.5 means on time is greater and a value less than 0.5 means off time is greater [6]. The PWM signal (T1) was generated by using a comparator which has the duty cycle signal at port A and a saw-tooth wave at port B which ranges from 0 to 1. The comparator sets its output to 1 whenever A is greater than B and 0 otherwise creating pulses with a magnitude of 1 and with pulse widths which depend on the duty cycle. By supplying the gate terminal of the IGBT switch with the PWM signal (T1), the converter could be switched on (when T1 = 1) and off (when T1 = 0) and for the time durations which are determined by the widths of the pulses. When the IGBT switch is on, the free-wheel diode is reverse biased (open circuit) and current ows through the inductor causing it to be charged with energy which helps in limiting the slew rate, maximum rate of change of the output voltage, of the switch. The capacitor is also charged and provides a ltering action by minimizing the voltage ripple produced at the output of the buck converter. When the IGBT switch is off, the free-wheel diode is forward biased (short circuit for ideal diode) providing a path for the discharge current from the inductor. The capacitor is also discharged. This continuous charging and discharging process of the inductor and the capacitor forces PV output voltage (Vpv) to track and follow the reference voltage (Vmppt) to operate at the MPP, as shown in Fig. 5, even when the irradiation decreased from 1000 W/m2 to 500 W/m2 .
Mag
Phase Ang Sin Mag Freq Freq D + 2.094395102 F (2*pi)/3 Phase Sin Mag Freq Freq Phase Sin Mag Freq Freq gt1
gt3
A B Comparator
gt1
A B Comparator
gt3
+ F
A B Comparator gt4
gt6
2.094395102 (2*pi)/3
gt5
!
Fig. 10. Close up view of the AC lter output Line-Neutral voltage at 1000 W/m2 . TABLE I VOLTAGE R ATING FOR E LECTRIC P OWER S YSTEM AND E QUIPMENT Voltage Class Low Voltage Medium Voltage High Voltage Extra High Voltage Ultra High Voltage Nominal Line-Line RMS Voltage < 600 V 600 V 69 kV 69 kV 230 kV 230 kV 1100 kV > 1100 kV
gt5
gt2
Fig. 7. Firing pulse generation model in PSCAD
!
Fig. 8. PV array output voltage (Vpv) and DC-DC converter output voltage (dcVltg) at 200 W/m2 and 1000 W/m2 .
!
Fig. 9. Close up view of the inverter output Line-Neutral voltage at 1000 W/m2 .
kept its input DC voltage (dcVltg) at a constant value of 0.5 kV, as shown in Fig. 8, even when the irradiation increased from 200 W/m2 to 1000 W/m2 ). It also converted the constant DC voltage at its input (dcVltg) to an AC voltage which is shown in Fig. 9 which clearly shows the distortion present in inverters output. E. AC Filter Based on the previous observations related to the output voltage of the inverter and the distortion involved, an AC ltering stage is required to further smoothen the output [8]. In this model, the AC lter was implemented using the inductor shown in Fig. 1. The inductor improved the shape of the output voltage of the inverter to an almost sinusoidal wave as shown in Fig. 10. F. Transformer Transformers in grid connected PV systems act as galvanic isolation and can be used for voltage adjustment if required. There are three main methods used by the inverters for
galvanic isolation: low frequency transformer, high frequency transformer and transformer-less inverters. Most commonly used method for galvanic isolation is using the conventional low frequency transformer operating on grid frequency. By controlling AC current, the power that is fed into the grid can be controlled. This is a tried and tested method and is being used right from the start of the PV technology. But this has some disadvantages like heavy weight, high cost, additional losses and non-unity power factor, especially at low load conditions. One way to omit the bulky transformer is to use high frequency transformers. Another emerging topology is the transformer-less inverter which has less overall losses, lighter in weight and it is cheaper than conventional grid frequency transformer topology. In addition, topology without transformer increases the control over the system voltage and power since transformer limits the control of the grid current [9] [10]. A conventional, operating on grid frequency (60 Hz), step up, wye-wye, three phase transformer (230 V / 11 kV) was used in the PSCAD model as shown in Fig. 1. The low voltage side of the transformer (230 V) was connected to the inverter while the high voltage side (11 kV) was connected to the grid. G. Utility Grid The utility grid system is represented only as an equivalent 11 kV and 60 Hz source behind the system inductive impedance as shown in Fig. 1. Based on [11], which provides ranges of voltage ratings for electric power systems in the US as shown in Table I, the utility grid (11 kV) in this PSCAD model falls within the medium voltage (MV) power systems range. III. T OTAL H ARMONIC D ISTORTION (THD) A NALYSIS Harmonics are sinusoidal components of a periodic wave having a frequency that is at multiples of the fundamental
TABLE II C URRENT D ISTORTION L IMITS AT PCC FOR SIX - PULSE CONVERTERS AS RECOMMENDED IN IEEE S TD 519-1992 Odd Harmonic 3rd 9th 11th 15th 17th 21st 23rd 33rd Above the 33rd Distortion Limit < 4.0 % < 2.0 % < 1.5 % < 0.6 % < 0.3 %
I_pcc
15
Total Harmonic Distortion 15 Individual
15

15
X1
Mag1 Mag2 Mag3 (15) (15) (15) Ph1 (15) FFT Ph2 (15) F = 60.0 [Hz] dc1 dc2 Ph3 (15) dc3
X2 X3
Fig. 11. PSCAD circuit to calculate total current harmonic distortion at PCC.
frequency. Harmonics in PV systems are generated by the converters which use switching techniques that generate signals that are not perfect sinusoidals. Connecting PV systems to the utility grid, which is already being injected with harmonics by the non-linear loads connected to the power network, will add a stress on the power quality of the grid. In order to maintain acceptable levels of grid power quality, standards that regulate the effects of PV systems on the utility grid should be developed. One such standard is IEEE Std 9292000 IEEE Recommended Practice for Utility Interface of Photovoltaic (PV) Systems [12] which ensures compatible operation of photovoltaic (PV) systems that are connected in parallel with medium and high voltage level electric utility. It is recommended by the previous standard that the harmonic distortion at the Point of Common Coupling (PCC), which is the point at which the PV system is tied with the grid, should comply with IEEE Std 519-1992 [13]. In this PSCAD model, PCC lies between the transformer and the grid. The PV system output should have low current distortion levels to ensure that no adverse effects are caused to other equipment connected to the utility system. The key requirements of clause 10 of IEEE Std 519-1992 are summarized as the following:
!
Fig. 12. THD (%) of the current in phase a at PCC and measured at STC.
!
Fig. 13. Individual harmonic distortion (%) of the current in phase a at PCC and measured at STC. In the x-axis is shown the index of the harmonic with the fundamental frequency having harmonic index that is equal to 1.
Total harmonic current distortion shall be less than 5% of the fundamental frequency current at rated inverter output. Each individual harmonic shall be limited to the percentages listed in Table II. The limits in Table II are a percentage of the fundamental frequency current at full system output. All even harmonics in these ranges shall be <25% of the odd harmonic limits listed.
In order to calculate the Total Harmonic Distortion (THD) of the current at the PCC, the PSCAD standard blocks shown in Fig. 11 were used. The Fast Fourier Transform (FFT) block was used to determine the harmonic magnitude of the line current at PCC in each phase (a, b and c) as a function of time. The resulting harmonic magnitude was then used as an input to the Harmonic Distortion block which measures both the total and individual harmonic distortion in percentage (%). The FFT block was congured to output the magnitude of 15 harmonics with the fundamental frequency at 60 Hz. Inputs to this block were the three RMS values of the three line currents at PCC. Harmonic Distortion block was congured to output THD and individual harmonic distortions in percent.
In Fig. 12 is shown the THD of the current in phase a at PCC and measured at STC which was plotted as a percentage that varies with time. It is noticed from this gure that the THD is oscillating around 4% which is less than the 5% limit established by the previously discussed standard. In Fig. 13 is shown the harmonic distortion of the current in phase a at PCC and measured at STC which is visualized using a polymeter in PSCAD that associates the harmonic distortion (%) with its harmonic index (1 to 15). By comparing the value of each harmonic in Fig. 13 with the range given in Table II, it was found that none of the harmonics violates the specied limits. By expanding the harmonic index to 63 as shown in Fig. 14, it was found that harmonics with indices from 38 to 46 were violating the distortion limits, which is 0.3%. Based on this, one can conclude that these distortions were introduced to the PCC because of the insufcient ltering stages implemented in the PSCAD model. Another reason is the simplicity of the controls used in generating the switching signals of the inverter.
!
Fig. 14. Individual harmonic distortion (%) of the current in phase a at PCC and measured at STC. The number of the monitored harmonics was expanded to 63.
Fig. 16. Current THD (%) in phase a at PCC (y-axis) measured at solar irradiation varying between 0 to 1200 W/m2 (x-axis).
using PWM and SPWM for DC-DC converter and inverter, respectively, were tracked using the PSCAD model. Some of the standards which regulate the interfacing of PV systems with the utility grid were discussed. Finally, total harmonic distortion analysis on the output current of the inverter at PCC and at STC was conducted using standard PSCAD blocks (FFT and Harmonic Distortion blocks) and then compared with the limits specied by IEEE Std 519-1992 standard.
Fig. 15. Voltage THD (%) in phase a at PCC (y-axis) measured at solar irradiation varying between 0 to 1200 W/m2 (x-axis).
R EFERENCES
[1] G. K. Venayagamoorthy, Comparison of power system simulation studies on different platforms rscad, pscad/emtdc, and simulink simpowersystems, in Int. Conf. on Power Systems, Operation and Planning, 2005, pp. 3841. [2] A. Rajapakse and D. Muthumuni, Simulation tools for photovoltaic system grid integration studies, in Electrical Power Energy Conference (EPEC), 2009 IEEE, Oct., pp. 15. [3] Standard Terminology of Solar Energy Conversion, ASTM Std. E772, 2011. [4] T. Esram and P. Chapman, Comparison of photovoltaic array maximum power point tracking techniques, Energy Conversion, IEEE Transactions on, vol. 22, no. 2, pp. 439449, Jun. 2007. [5] M. M. Elamari, Optimisation of photovoltaic- powered electrolysis for hydrogen production for a remote area in libya, Ph.D. dissertation, School of Mech., Aero. and Civil Eng., Univ. of Manchester, Manchester, UK, 2011. [6] M. H. Rashid, Power Supplies in Power Electronics Handbook: Devices, Circuits, and Applications Handbook, 3rd ed. Massachusetts: Elsevier, 2011, pp. 623624. [7] B. Wu, High-Power Converters and AC Drives, 1st ed. New Jersey: Wiley, 2006, ch. 6, pp. 9597. [8] S. Rahman and R. Varma, Pscad/emtdc model of a 3-phase grid connected photovoltaic solar system, in North American Power Symposium (NAPS), 2011, Aug., pp. 17. [9] D. Jose, Comparison of a three phase single stage pv system in pscad and powerfactory, Degree project, School of Elect. Eng., KTH Royal Institute of Technology, Stockholm, Sweden, 2012. [10] F. Schimpf and L. E. Norum, Grid connected converters for photovoltaic, state of the art, ideas for improvement of transformer-less inverters, in Nordic Workshop on Power and Industrial Electronics, 2008, pp. 16. [11] Voltage Rating for Electric Power Systems and Equipment, ANSI Std. C84.1, 1989. [12] IEEE Recommended Practice for Utility Interface of Photovoltaic (PV) Systems, IEEE Std. 929, 2000. [13] IEEE Recommended Practices and Requirements for Harmonic Control in Electrical Power Systems, IEEE Std. 519, 1992.
Standards which impose limits on the voltage THD levels in power networks were established, but it is a common practice, especially in the case of grid-tied PVs, to pay more attention to current THD analysis. In order to justify this practice, the voltage and current THD values of the PSCAD model at PCC were plotted at variable solar irradiation values as shown in Fig. 15 and Fig. 16, respectively. As shown in Fig. 15, there is no noticeable effect of variable solar irradiation values on the voltage THD values which remains almost constant. On the other hand, it is shown in Fig. 16 that by increasing the solar irradiation values, the current THD values were decreasing. One could conclude that current THD analysis are more important than voltage THD analysis in the case of grid-tied PV systems that are operating under varying solar irradiation. Also, PV systems operating under low solar irradiation values inject more current harmonics into the utility grid than at high irradiation values. This might force PV system operators to either use better ltering techniques or even disconnecting the PV system from the grid to avoid paying the high THD levels penalty specied by the utility operator. IV. C ONCLUSION In this paper, the tasks of the different components of gridtied PV system PSCADs model were discussed. Also, the effect of varying the atmospheric conditions (irradiation and temperature) on the PV system was justied with graphs. The control stages involved in generating the switching signals
Smart Charging of Electric Vehicles using Adaptive Fuzzy Logic

Muhammad Ansari, Ali T. Al-Awami, M. A. Abido
Department of Electrical Engineering King Fahd University of Petroleum & Minerals Dhahran, Saudi Arabia Email: g201103030, aliawami, mabido@kfupm.edu.sa
Eric Sortomme
Alstom Grid Redmound,WA 98052 USA Email: eric.sortomme@alstom.com
AbstractVehicle-to-grid (V2G) is a way to effectively integrate the electric vehicles with the power system network. Unidirectional vehicle to grid charging is especially attractive because it is easy to implement and requires little additional hardware. Electric vehicle can be effectively charged during its parking time, and can potentially generate revenues for the electric vehicle owner. The aggregator combines the capacity of many electric vehicles and bid into the electricity market. Several smart charging algorithms have been proposed previously. In this work, a novel smart charging algorithm based on the adaptive fuzzy logic is proposed. The proposed algorithm considers several uncertain parameters in the charging schedule. Simulations are performed on a hypothetical group of 10,000 EVs operating in the real electricity market, ERCOT. The proposed fuzzy logic based algorithm shows an improvement over the previously proposed algorithms in terms of system, customer and aggregator benets. Index TermsElectric vehicles (EV), Electricity market, ERCOT, Fuzzy logic, Regulation service, Smart grid, Vehicle-to-Grid (V2G).
P OPi SOCi V V 2G
Preferred operating point State of charge of the battery Regulation signal from the ISO Vehicle to grid I. I NTRODUCTION
N OMENCLATURE CRi EFi ERCOT EV H ISO L1 L4 LDA Mn Mx M ci M nAPi M Pi M P OP MV M xAPi P1 P4 PDA P Di Charge remaining in the EV Efciency of the battery charger Electric Reliability Council of Texas Electric vehicle Number of charging hours Independent system operator Intermediate system load values for membership classication Day-ahead system load Minimum day-ahead energy price/system load Maximum day-ahead energy price/system load Maximum charger of the battery Minimum additional power draw of an EV Power rating of the battery charger Maximum preferred operating point Maximum regulation signal from the ISO Maximum additional power draw of an EV Intermediate energy price values for membership classication Day-ahead energy price Power draw of an EV
Electric vehicles (EVs) have the potential to provide energy and ancillary services to the electrical power grid using the vehicle to grid (V2G) concept [1][3]. Utilities and researchers have been developing the needed technologies for large scale integration of EVs. Despite their several advantages, EVs face many challenges, such as high initial cost and limited energy capacity. In addition, if the EVs are not properly controlled, they can cause power shortage during heavy charging periods [4]. To integrate the EVs with the electrical grid efciently, V2G has been proposed [2]. Earlier work on the integration of EVs with the grid largely focused on the basic concepts and simulation to show the potential benets of V2G for different electricity markets [1], [2], [5]. The V2G power ow can be either unidirectional or bidirectional. The unidirectional V2G can only provide regulation service and can enhance spinning reserves capacity [3], [5][7] while the bidirectional V2G can also provide energy [8]. Several studies and demonstrations have conrmed that EVs are well suited for regulation services [9][11]. Through V2G, an EV owner can generate revenues from their EV while charging. Bidirectional V2G has many advantages as compared to the unidirectional V2G, but there are many challenges for its implementation. In order to supply energy from the EV, additional hardware is required. Many customers will not be willing to allow the utility to drain their battery for any unexpected departure [12]. Moreover battery degradation issues due to repeated cycling on the battery [6], [13], [14]. From the electrical grid side, protection and control schemes has to be modied to handle the bidirectional power ow [15]. The likely step to avoid the challenges of bidirectional V2G is to start with the unidirectional V2G. Electric vehicles will be able to participate in most markets without the need of additional hardware and infrastructure requirements and the utilities will not be able to withdraw power from the customer batteries.
A considerable amount of work in unidirectional V2G has been done previously. In [3], a unidirectional regulation algorithm to be followed by an EV is proposed. Several smart charging algorithms were proposed in [3], [16]. These algorithms are load only regulation that controls the charging level of an electric vehicle. The charging schemes are based on the real time communication of system load, energy price, regulation signal deployments and the charging hours of the electric vehicles. In this work, a novel fuzzy logic based algorithm is proposed that combines the individual smart charging algorithms in a fuzzy logic framework. An aggregator prot scheme is developed that receives the revenues from participating in the ancillary service market and from charging the EVs. These algorithms are simulated over a hypothetical group of 10,000 EVs in the electric reliability council of Texas (ERCOT) area. Commuter cars are used in the simulations and it is assumed that all the EVs are available during the charging hours from 8 A.M. to 5 P.M. Results shows the benet of the proposed fuzzy logic based charging from the aggregator point of view, while each algorithm has its own specic advantages that are explained later in this paper. II. R EGULATION A LGORITHM An Electric vehicle (EV) can perform the regulation service by varying its charging rate above or below its scheduled charging rate, which is called the preferred operating point (POP). The value of the POP in the system is scheduled by the aggregator. By varying the charging rate of an EV below/above its POP, regulation up/down capacity is provided. As a single EV has insufcient capacity for providing this regulation service, the aggregator aggregates many EVs and bid their combined capacity in the electricity market. The aggregator controls the charging of EVs according to the regulation signal provided by the system operator. More details on the regulation service algorithm can be found in [3]. Note that in order to perform the regulation service by the aggregator, communication between the aggregator and the system operator is required. III. S MART C HARGING A LGORITHMS The regulation capacity is based on the charging rate of the EV. The charging rate is selected by varying the POP; therefore the aggregator must schedule the value of POP smartly to maximize the prots. Previously, price based and load based smart charging algorithms were proposed for the EV charging [16]. However, these were not considered for regulation capacity bidding. Later, a modied algorithms maximum-regulationbased charging was proposed [3]. In this paper, a new fuzzy based algorithm is proposed in a fuzzy logic framework to overcome the shortcomming of the previous algorithms. Fig 1 shows the three smart charging algorithms suggested in [16] and [3] and the one to be proposed in this paper. A. Price Based Charging The price based algorithm sets the value of the POP based on the energy price in the system. The EVs charge more and
Smart Chargers
Price Based
Load Based
MaxReg Based
Fuzzy Logic Based
Fig. 1.
Smart Charging Algorithms
provide more regulation capacity when the energy price is low. The POP is selected based on: Mx P (t) P OPi (t) = M Pi (1) Mx Mn where, Mx = max(PDA (t0 : t0 + 24 60/n)) Mn = min(PDA (t0 : t0 + 24 60/n)) B. Load Based Charging This algorithm sets the value of the POP based on the load in the system. The EVs charge more and provide more regulation down capacity when the load on the system is low. The POP is selected based on: Mx L(t) P OPi (t) = M Pi (4) Mx Mn where, Mx = max(LDA (t0 : t0 + 24 60/n)) Mn = min(LDA (t0 : t0 + 24 60/n)) C. Maximum Regulation (MaxReg) Based Charging The MaxReg based algorithm sets the value of the POP based on the battery level and the number of hours available for battery charging. The MaxReg based bids the regulation up/down capacity during the whole charging period. The POP is selected based on: M ci SOCi P OPi (t) = (7) H D. Proposed Fuzzy Logic Based Charging A novel adaptive fuzzy logic based charging is proposed that combines the features of the previous published chargers in a fuzzy logic framework. Fig. 2 shows the block diagram of a typical fuzzy logic controller (FLC) as described in [17]. A FLC reects the mechanism implemented by the humans without any complete knowledge of the control object in a mathematical form. Here the control system works on the set of rules dened by humans past experience about the system. There are ve main components of a fuzzy logic controller [17]: Fuzzication module (fuzzier) Knowledge base Rule base (5) (6) (2) (3)
Knowledge Base
Rule Base
EP
Defuzzification Denormalization
Output System
Input
Scaling factor Normalization
Fuzzification
Interface
Fig. 2.
Typical Fuzzy Logic Controller
Mpn
P1
P2
P3
P4
Mpx
Interface engine Defuzzication module (defuzzier) In the fuzzier, the values of input variables are measured and it converts the crisp data into fuzzy linguistic values. The knowledge base consists of a database and linguistic control rule base. It provides necessary denitions for the fuzzication process such as membership functions, fuzzy set representation, etc. The rule base is the control strategy of the fuzzy logic system. It is usually obtained from the expert human knowledge and expressed as a set of IF-THEN rules. The defuzzication process converts the output variable into corresponding universe of discourse. Various techniques are used for the defuzzication such as maximum method, height method and the centroid method [17]. Automatic changes in the design parameters of any of the above elements create an adaptive fuzzy system. In this study, the membership function of the input and the output variables are made adaptive in nature i.e. the universe of discourse and the membership functions varies based on the energy price and system load inputs. 1) Membership Function: The input of the fuzzy logic controllers are the energy price, system load and the charging hour. The output of the fuzzy logic controller is the POP for the EVs. The triangular membership functions are assumed for every input and output variable. The membership functions for energy price, system load and the POP will be adaptive as the universe of discourse will change for these functions each day. Three membership functions low (L), medium (M) and high (H) are dened for each input/output variable. a) Energy Price: Fig. 3 shows the membership function of the energy price. The membership function is made adaptive by varying the universe of discourse of energy price based on a particular day. b) System Load: Fig. 4 shows the membership function for the system load. The membership function is made adaptive by varying the universe of discourse of system load based on the system load for a particular day. c) Charging Hours: Fig. 5 shows the membership function for the charging hours. d) Preferred Operating Point (POP): Fig. 6 shows the membership function for the output i.e. POP. The membership function of POP will vary based on the maximum POP in the batteries remaining. 2) Rule Base (Decision Matrix): A set of rules which relates the relationship between the inputs and output variables

Fig. 3.
Membership function of energy price
SL L M H
Mln
L1
L2
L3
L4
Mlx
Fig. 4.
Membership function of system load
Hours H
8:00 AM 10:00 AM
11:30 AM 1:30 PM 3:15 PM
5:00 PM
Fig. 5.
Membership function of charging hours
POP L M H
0.2
0.4
0.6
0.8
1 x MPOP
Fig. 6.
Membership function of preferred operating point
TABLE I POP FUZZY VALUE WHEN THE CHARGING HOUR IS LOW Price\Load L M H L H H M M H M M H M M L
TABLE II POP FUZZY VALUE WHEN THE CHARGING HOUR IS MEDIUM Price\Load L M H L M M L M M M L H L L L
Tesla Model-S. It is also assumed that each EV has a charging efciency of 90%. Each EV arriving at the workplace with an SOC of greater than 95% will not participate in the regulation service. The aggregator prot comes from two different sources, regulation revenues and the markup on the energy sale supplied to the customer [3]. The aggregator gets 20% of the regulation up and down revenues and $ 0.05/kWh over the energy purchased from the market for the EV battery charging. In this way, the aggregator is not exposed to the variations in market energy prices and passes the energy cost to the EV owners. V. R ESULTS A ND D ISCUSSIONS The simulations are performed for each hour from 8 A.M. to 5 P.M. for the period of three months. Comparison of charging prole for each smart charging algorithm is shown in Fig. 710 for 2nd Aug, 2010. The price and the load algorithm follow almost the same pattern and schedules the POP to be highest at the start of the charging period as the system load and energy prices are low in the morning and in the afternoon, the POP decreases to a lower value due to the low energy price and higher load and also a large number EVs are charged during the early hours. The MaxReg algorithm keeps the POP to almost constant value during the whole charging period and sells both regulation up and down capacity during each hour. However, the proposed fuzzy algorithm efciently varies the value of the POP based on the energy price, system load and the charging hours. During initial hours it keeps the POP to nearly constant value similar to MaxReg while during the last hours it follows the pattern of price and load algorithms. It combines the advantages of other methods and overcome their shortcomings.
TABLE III POP FUZZY VALUE WHEN THE CHARGING HOUR IS HIGH Price\Load L M H L M L L M L L L H L L L
are dened in the fuzzy rule database. These rules are dened based on the human intelligence. The three inputs, energy price, system load and the charging hours, result in a total of 27 rules. These rules are shown in Table. I-III. 3) Defuzzication: The defuzzication is the process of producing quantitative results in fuzzy systems, given the fuzzy set and corresponding membership function. The most popular defuzzication method is the centroid calculation, which returns the center of area under the curve and, therefore, is considered here for defuzzication. The general formula for the defuzzication is taken from [17]: f (y ) = (y ) y (y ) (8)
IV. S IMULATIONS The four different algorithms price, load, MaxReg and the proposed fuzzy logic based, are simulated over a hypothetical group of 10,000 EVs in the ERCOT area. The simulations are performed for the commuter cars that are available during the day time from 8 A.M. to 5 P.M. at the workplace. It is assumed that all the EVs are available during this nine hours period and the aggregator can potentially sell regulation services during this period. For this study, different system data such as system load, energy price, regulation up/down prices and regulation deployments are taken from ERCOT archives for a period of three months from 21st July, 2010 to 20th Oct, 2010 [18]. In this study, ve minutes resolution is considered. The day ahead load and price forecasts are generated to match the load forecast errors in [19] and to match the error distribution found in [20] for the energy price. The EVs are a hypothetical group of three different types of EVs that are available in the market; Nissan Leaf, Mitsubishi i-MiEV and the Tesla Model-S. Battery specication, EV performance and other specications are given in [21][24]. Among this hypothetical group, it is assumed that 50% of EVs are Nissan Leaf, 20% are Mitsubishi i-MiEV and 30% are
90 80 70 60 50 40 30 20 10 0 8 9 10 11
PriceAlgorithm
MW
POP Rdown POP Rup POP
12 Time(hours)
13
14
15
16
Fig. 7.
Price based POP selection algorithm
By examining the different algorithms from the aggregator point of view, the proposed fuzzy algorithm results in the highest prots. The prot is 0.21% higher as compared to MaxReg which increase the prot by $ 3300. The price and load algorithms performance is poor in terms of aggregator prot. Table IV shows the prots of aggregator for three months duration, by each algorithm. From the power system point of view it is desirable that the EVs should not burden the power system network. It is evident from Fig. 11 that the load based algorithm results in the lowest
LoadAlgorithm
90 80 70 60 50 40 30 20 10 0 8 9 10 11 12 Time(hours) 13 14 15 16
45 40 35 Load(MW) 30 25 20 15 10 5 0 Price
PeakLoadIncrease
MW
Load
MaxReg ChargingAlgorithms
Fuzzy
Fig. 8.
Load based POP selection algorithm
Fig. 11.
Peak Load Increase by each algorithm
MaxRegAlgorithm
90 80 70 60 50 40 30 20 10 0 8 9 10 11 12 Time(hours) 13 14 15 16
up capacity. The MaxReg algorithm bids almost constant regulation up capacity during the whole charging period. The Fuzzy algorithm bids the high regulation down capacity during the initial and middle of the charging period while the MaxReg algorithm bids almost in every hour except for the last hour, as shown in Fig. 13. The price and the load algorithms follow the MaxReg algorithm pattern but bids considerable less amount. This is evident from Fig. 13.
AverageRegulationUp
80 70 Reg gulationUp(MW) 60 50 40 30 20 10 0 8 9 10 11 12 Time(hours) 13 14 15 16 Price Load MaxReg g Fuzzy
MW
Fig. 9.
MaxReg based POP selection algorithm
TABLE IV AGGREGATOR P ROFITS F OR T HREE M ONTHS P ERIOD Algorithms Prots ($1000) Price 1438.1 Load 1345.72 MaxReg 1519.2 Proposed Fuzzy 1522.5
Regu ulationDown(MW)
peak and the proposed fuzzy based algorithm results in almost similar load as that of load based algorithm. It can also be seen that the price based algorithm results in the highest value of the peak load. The amount of average regulation up and down capacities during each hour is shown in Fig. 12 and Fig. 13, respectively. It is evident from Fig. 12 that the price, load and fuzzy algorithms have high regulation up during the rst ve hours and after hour 12, most of the electric vehicles are charged so the price and load algorithms bid very low after hour 12, while the fuzzy algorithm stills bid some amount of regulation
Fig. 12.
Average regulation up capacity by each algorithm
AverageRegulationDown
80 70 60 50 40 30 20 10 0 Price Load MaxReg g Fuzzy
ProposedFuzzyAlgorithm
90 80 70 60 50 40 30 20 10 0 8 9 10 11 12 Time(hours) 13 14 15 16
10
11
12 Time(hours)
13
14
15
16
MW
Fig. 13.
Average regulation down capacity by each algorithm
Fig. 10.
Proposed Fuzzy Logic based POP selection algorithm
Although the main focus of this study is to generate maximum prots for the aggregator, but it is also desirable to verify that the charging of the EVs are not done at very high energy prices. From the EV owner perspective, the vehicle should be charged at the lowest possible cost. Fig. 14 shows the average price of energy per kWh for different algorithms.
The price based algorithm results in the lowest energy cost and the fuzzy based algorithm also results in a very close cost, while the MaxReg algorithm results in the highest energy price.
$0.0860 $0.0840 $0.0820 EnergyPrice($/kWh) $0.0800 $0.0780 $0.0760 $0.0740 $ $0.0720 $0.0700 $0.0680 Price Load MaxReg ChargingAlgorithms Fuzzy
Fig. 14.
Average energy price per kWh charged to EV owner
VI. C ONCLUSION This work presents a novel adaptive fuzzy smart charging algorithm for the unidirectional electric vehicles. Simulations are performed on a hypothetical group of commuter EVs. Previously different algorithms were proposed for the smart charging and each has its own specic benets while the proposed fuzzy based algorithm combines their benets and generates higher revenues for the aggregator. It also results in a lower energy price for charging the electric vehicles and the impact on the power system is also reduced as compared with the previous algorithms. One of the main advantages of the smart chargers is that they can be easily implemented for real time systems and the proposed fuzzy algorithm can be easily incorporate in any previous real time system as the fuzzy logic controllers are easy to implement and requires little or no additional hardware. ACKNOWLEDGMENT The authors would like to acknowledge the support of King Fahd University of Petroleum and Minerals. R EFERENCES
[1] W. Kempton and J. Tomi c, Vehicle-to-grid power implementation: From stabilizing the grid to supporting large-scale renewable energy, Journal of Power Sources, vol. 144, pp. 280294, June 2005. [2] W. Kempton and J. Tomi c, Vehicle-to-grid power fundamentals: Calculating capacity and net revenue, Journal of Power Sources, vol. 144, pp. 268279, June 2005. [3] E. Sortomme and M. A. El-Sharkawi, Optimal Charging Strategies for Unidirectional Vehicle-to-Grid, IEEE Transactions on Smart Grid, vol. 2, pp. 131138, Mar. 2011. [4] S. W. Hadley and A. a. Tsvetkova, Potential Impacts of Plug-in Hybrid Electric Vehicles on Regional Power Generation, The Electricity Journal, vol. 22, pp. 5668, Dec. 2009. [5] J. Tomi c and W. Kempton, Using eets of electric-drive vehicles for grid support, Journal of Power Sources, vol. 168, pp. 459468, June 2007. [6] S. B. Peterson, J. Apt, and J. Whitacre, Lithium-ion battery cell degradation resulting from realistic vehicle and vehicle-to-grid utilization, Journal of Power Sources, vol. 195, pp. 23852392, Apr. 2010. [7] E. Sortomme and M. El-Sharkawi, Optimal Combined Bidding of Vehicle-to-Grid Ancillary Services, IEEE Transactions on Smart Grid, vol. 3, pp. 7079, Mar. 2012.
[8] E. Sortomme and M. a. El-Sharkawi, Optimal Scheduling of Vehicleto-Grid Energy and Ancillary Services, IEEE Transactions on Smart Grid, vol. 3, pp. 351359, Mar. 2012. [9] W. Kempton, V. Udo, K. Huber, K. Komara, S. Letendre, S. Baker, D. Brunner, and N. Pearre, A test of vehicle-to-grid (V2G) for energy storage and frequency regulation in the PJM system, Results from an Industry-University Research Partnership, no. November, 2008. [10] E. Larsen, D. K. Chandrashekhara, and J. sterg ard, Electric vehicles for improved operation of power systems with high wind power penetration, in Proc. Energy 2030 Conference, pp. 16, 2008. [11] P. Kadurek, C. Ioakimidis, and P. Ferr ao, Electric vehicles and their impact to the electric grid in isolated systems, in Proceedings of the International Conference on Power Engineering, Energy and Electrical Drives, Lisbon, Portugal, 18-20 March, pp. 4954, 2009. [12] B. K. Sovacool and R. F. Hirsh, Beyond batteries: An examination of the benets and barriers to plug-in hybrid electric vehicles (PHEVs) and a vehicle-to-grid (V2G) transition, Energy Policy, vol. 37, pp. 1095 1103, Mar. 2009. [13] A. Brooks, Vehicle-to-grid demonstration project: Grid regulation ancillary service with a battery electric vehicle, Tech. Rep. 01, 2002. [14] S. B. Peterson, J. Whitacre, and J. Apt, The economics of using plug-in hybrid electric vehicle battery packs for grid storage, Journal of Power Sources, vol. 195, pp. 23772384, Apr. 2010. [15] E. Sortomme, S. Venkata, and J. Mitra, Microgrid protection using communication-assisted digital relays, IEEE Transactions on Power Delivery, vol. 25, pp. 27892796, Oct. 2010. [16] T. Markel, M. Kuss, and P. Denholm, Communication and control of electric drive vehicles supporting renewables, in Vehicle Power and Propulsion Conference, no. August, 2009. [17] R. Ross, Fuzzy Logic with Engineering Applications. McGraw-Hill, 1995. [18] ERCOT, Electric Reliability Council of Texas, Market information, http://www.ercot.com/mktinfo, Accessed:15-Feb-2013. [19] S. Fan, L. Chen, and S. Member, Short-Term Load Forecasting Based on an Adaptive Hybrid Method, vol. 21, no. 1, pp. 392401, 2006. [20] N. Amjady, Day-Ahead Price Forecasting of Electricity Markets by a New Fuzzy Neural Network, IEEE Transactions on Power Systems, vol. 21, pp. 887896, May 2006. [21] Mitsubishi, Mitsubishi Motors, Mitsubishi Motors to bring new-generation EV i-MiEV to market, http://media.mitsubishimotors.com/pressrelease/e/products/detail1940.html, Accessed:15-Feb-2013. [22] TESLA, Tesla Motors, Tesla Roadster spec sheet, Accessed:15-Feb2013. [23] TESLA, Tesla Motors Model S, http://www.teslamotors.com/models/features#/performance, Accessed:15-Feb-2013. [24] NISSAN, Nissan Zero Emission Website, Leaf specs, http://www.nissan-zeroemission.com/EN/LEAF/specs.html, Accessed:15Feb-2013.
A Novel Single-Phase Soft Switching Microinverter for Photovoltaic Applications

Saad Pervaiz, Muneeb Ur Rehman, Ahmed Bilal Asghar, Nauman Zaffar
SBA School of Science & Engineering, LUMS Email: (saad.pervaiz, muneeb.rehman, bilal.asghar, nauman.zaffar)@lums.edu.pk
AbstractMicroinverters provide a direct way to integrate an individual PV panel to the AC grid. We introduce a novel microinverter design comprising a high frequency full-bridge converter, a pseudo dc link lter, and a line commutated inverter. The approach is well matched to the requirements for high efciency DC to AC grid-tied and stand-alone converters. We present a detailed working analysis of the proposed design in continuous, boundary, and discontinuous conduction mode. We also show that our design enables reduced device losses, exible control, and low current harmonic distortion. We present a soft switching scheme for the high frequency full bridge converter stage. Results for design simulation in PSIMT M and implementation in hardware are presented. Index TermsModule integrated converter, photovoltaic, fullbridge, soft-switching.
Fig. 1.
Proposed DC-AC Inverter Design
I. I NTRODUCTION Photovoltaic (PV) provide a clean source of energy and are capable of satisfying our growing electricity demand. They are expected to become major contributor to electricity generation among all renewable energy candidates by 2040 [1]. Grid parallel inverters are a key component required to integrate PV systems in electric grid. Various inverter system architectures have been studied for PV applications over the course of past few decades [2], [3], [4]. One approach to PV integration is the use of module integrated converters or microinverters. These converters provide a direct way to integrate an individual PV panel to the AC grid. Such integration allows MPPT of individual solar panels which yields greater energy extraction than centralized MPPT of series-connected string of PV modules [5], [6]. Microinverters also provide system scalability and better performance in partially shaded regions [7]. Multiple microinverter topologies are possible depending upon the dc link conguration. These topologies are classied in three different arrangements [6]: Microinverter with a DC link Microinverter with a pseudo DC link Microinverter without a DC link A microinverter interfaced with a single PV panel needs to provide high voltage transformation under varying input voltage conditions. A single PV panel provides a low voltage (20-40 V) at input which needs to be transformed to AC grid voltage (220 V, rms). In such applications, efciency, size and cost are the driving design considerations for microinverters. High efciency can be achieved through soft switching and reduced reverse recovery losses in rectier diodes. Cost of inverter is proportional to the size of energy storage components
and transformer. High switching frequency design is one way to bring down the size and cost of inverter. In this paper, we investigate a microinverter based on the architecture of Figure 1, comprising a high frequency fullbridge converter, a pseudo dc link lter, and a line commutated inverter. Similar general architectures ([8], [9]) have long been known but this specic design has not been explored in previous work (e.g. see topology review in [3], [4]). We propose a novel design that reduces device losses, enables exible control, and low current harmonic distortion. All switching devices operate under zero voltage/zero current switching enabling high efciencies to be achieved. A trailing edge pulse width modulation scheme is employed for the high frequency full bridge converter stage. Rest of paper is organized as follows: Section II provides converter design and working. Section III and IV provide simulation and experimental results to validate the working of proposed converter. We conclude in Section V by discussing the benets of our inverter design. II. P ROPOSED DC-AC I NVERTER A. Design The proposed inverter topology consists of a full bridge converter cascaded with a line commutated inverter, as illustrated in Figure 1. The design consists of an active H-bridge, S1S4, on the low voltage side of a transformer T. A full-wave rectier, D5-D8, on the high voltage side of the transformer is cascaded with another active H-bridge, S9-S12, which is connected to the grid supply. The body diodes, D1-D4, of MOSFETs, S1-S4, are explicitly shown. Cin and C1 are lter capacitors at the input and pseudo DC link respectively. Vin
and Vg mark the input PV voltage and grid supply at output respectively. B. Working & Analysis The proposed converter, shown in Figure 1, consists of two stages: boost full-bridge stage and line commutated inverter stage. The full-bridge stage operates at high frequency and produces a rectied sinusoidal PWM waveform which is ltered by the pseudo-dc link capacitor. This stage is responsible for providing the required boost in dc voltage for grid integration. The second stage operates at line frequency (50/60 Hz) and opens the rectied output waveform of previous stage. Since this stage operates at line frequency which is much smaller than operating frequency of rst stage ( 40 kHz), we can divide the inverter operation into two steps; positive half cycle of grid voltage during which S5 and S8 are closed and negative half cycle during which S6 and S7 are closed. Without any loss of generality, complete inverter operation can be explained by just considering the positive half cycle of grid voltage and the working principle of full-bridge converter during this cycle. PWM waveforms for the operation of full-bridge converter are given in gure 12. In this scheme, the lower switches (S3, S4) are operated at xed (50%) duty cycle while the top switches (S1, S2) are trailing edge pulse width modulated. The converter produces a sinusoidal PWM waveform which is rectied and ltered at the pseudo-dc link stage. At this link, a rectied line frequency AC waveform is produced. We can assume the converter to be operating as a dc-dc converter for small signal analysis since the operating frequency of fullbridge converter is much higher than line frequency (in other words change in line frequency waveform is approximately very small when viewed at 40 kHz operating frequency of full-bridge converter). The full-bridge converter can be operating in Continuous Conduction Mode (CCM), Boundary Conduction Mode (BCM) or Discontinuous Conduction Mode (DCM). The circuit operation in CCM, BCM, and DCM depends on the value of duty cycle. Duty cycle is a function of output load, however in our proposed technique, the duty cycle also depends on grid voltage at output and therefore varies continuously as it is being modulated by the grid voltage. Using steady state analysis, it can be shown that converter might be working in CCM, BCM and DCM during one complete cycle of grid voltage, even for a xed amount of output power delivered. The converter operation can be divided in three operating intervals for positive half cycle of grid voltage. These intervals are determined by the on/off states of the four primary switches (S1-S4). During the negative half cycle of grid voltage, the operation of full-bridge stage remains unchanged while the line commutated inverter ips the states of diagonal bridge switches. In the analysis that follows, MOSFET switches have been modeled with parallel diodes and parasitic capacitances. The transformer leakage inductance is shown explicitly,as it is fundamental to circuit operation.
(a) Continuous Conduction Mode
(b) Discontinuous Conduction Mode Fig. 2. PWM Gate Signals (G1-G4) of Switches (S1-S4), Voltage waveform across transformer primary winding, and current through leakage inductance of transformer in CCM & DCM mode
The detailed circuit operation in all three modes is as follows: [t0-t1] - Energizing Leakage Inductance t0-t1 marks the period when S1 and S4 are closed till the opening of S1. In this time period voltage, Vin , appears across the primary side of the transformer and a current starts building in the leakage inductance. The corresponding secondary current completes its path through the grid/load and thus transfers power to the output. The path of current for this time period is shown in Figure 3. During this interval, current builds in the leakage inductance and power is transferred to output. The circuit operation is similar in CCM, BCM, and DCM for this interval. [t1-t2] - De-energizing leakage inductance During t1-t2, switch S1 is opened while S4 remains closed. Since the current at the primary cannot stop instantly due to transformer leakage inductance, the inductor current discharges switch S3 capacitor until diode D3 is turned on and makes an alternate freewheeling path. During this time interval
Fig. 3.
[t0-t1] - Energizing Leakage Inductance
Fig. 5.
[t2-t3] Dead band mode
transformer, R is the load resistance corresponding to current delivered to the grid, and Ts is switching period. MDCM = Vo = Vin 2n 1+ 4k 1+ 2 D (1)
Fig. 4.
[t1-t2] - De-energizing leakage inductance
MBCM =
Vo = Dn Vin
(2)
the input is disconnected, therefore current dies out as energy stored in leakage inductance is transferred to the output. The circuit operation differs depending on the conduction mode (CCM, BCM, DCM). In CCM and BCM, the current is still non-zero at the end of this interval; therefore a corresponding current is owing through the secondary side as well. In DCM, the inductor is completely de-energized and the current becomes zero before the end of this interval. This interval ends when S4 is opened. [t2-t3] Dead band mode This mode begins when S4 is opened at time t2. In DCM, the current is already zero so this interval is not eventful for DCM. However, in CCM the current through the leakage inductance is not zero. So the inductor current discharges capacitor C2 and turns diode D2 on. The current ows through D3 and D2 during this interval. The rectier current on secondary side is zero as there is no current through the primary winding of transformer. Therefore no power is transferred to the output during this mode. The circuit operation is similar in BCM as compared to CCM, however the current becomes zero at the end of this interval in BCM. [t3-t6] Complementary Intervals These intervals are similar in operation to those of [t0-t3]. The transformer leakage inductance is energized in opposite direction as compared to the rst interval [t0-t1]. Complimentary diagonal rectier diodes are turned on during these three intervals. The full-bridge converter DC gain in DCM and BCM is given by equation 1 and 2, respectively. These equations have been derived in [10]. Here n is the transformer turn ratio, D is the duty cycle of top switches (S1, S2), k is the normalized time constant of the converter, L is the leakage inductance of
The normalized time constant of converter is given by: k= 4n2 L RTs (3)
C. Device stresses and Switching Losses All full-bridge converter switches are zero voltage turned off due to the drain-to-source capacitance of the MOSFET switches. So turn off losses for all the switches are minimal. Furthermore, in discontinuous conduction mode, during the interval t1-t2, the inductor current falls to zero so all the switches turn on at zero current. The switches S3 and S4 turn off at zero current in DCM. In CCM, all the switches achieve zero voltage switching. The line commutated inverter stage has minimal losses due to following reasons:

It is being switched at line frequency. So the switching losses are negligible. The switching takes place at zero crossings of the grid voltage/current. The voltage at dc-link is very small (close to 0V). Therefore the switching takes place at very low voltage.
At turn on, the only losses are the conduction losses that can be reduced by using devices with low on resistance or by paralleling switching devices. III. R ESULTS The proposed design was validated in simulation experiments and hardware prototype testing. The converter was designed to meet the specications given in Table I. The components used in making hardware prototype are listed in Table II.
Fig. 6.
Block Diagram of Feedback Control Blocks for Proposed Converter
TABLE I S IMULATION AND H ARDWARE C ONVERTER D ESIGN S PECIFICATIONS No. 1 2 3 4 5 Parameter Input Voltage (DC) Output Voltage (AC) Rated Output Current Rated Output Power Switching Frequency (Full-bridge) Value 20-40 (V) 220 (Vrms) 2.8 A 600 W 48 kHz Fig. 7. 220 V Test Case 1: Input Voltage 26V; Output Current 2 A; Grid Voltage
TABLE II C OMPONENTS U SED IN H ARDWARE P ROTOTYPE Component MOSFET (Low Voltage side) MOSFET (High Voltage side) Power Rectier Diodes Input Filter Capacitor Pseudo-dc Link Capacitor Transformer Part # IRF3710 IRF740 RHR15120 10,000 uF/100 V 7.5 F Ferrite Core (L=2.5 H)
A. Simulation Experiments PSIMT M was used for simulating the proposed converter design. The model parameters used in simulation were actual component parameter that are given in table II. The simulation was done both for stand-alone and grid tied system. Here, we discuss the results for grid connected circuit simulation. Feedback control for grid connected operation is shown in gure 6. A number of test case experiments were done in simulation experiments. We varied the input voltage to see the behavior of inverter. Figure 7 shows the output current waveform of the inverter at an input voltage of 26 V. The converter was operating in grid-parallel mode in this simulation experiment. Figure 8 and 9 show similar waveforms for input voltage of 30 V and 36 V respectively. The output current was xed at 2 A in all three cases. % THD of the current fed to the grid was less than 4% even in all experiments. We also changed the output current fed to the grid during
Fig. 8. 220 V
Test Case 2: Input Voltage 30V; Output Current 2 A; Grid Voltage
online operation. Output current could be changed at zero crossings to minimize current THD. Figure 10 shows the results for output current change. The smooth transition in output current shows the agility of feedback loop. B. Hardware Prototype A 600 Watt prototype, of the proposed converter, was made to validate the working of proposed converter topology and control scheme. The components values are listed in Table II. MOSFETs are used in the line-commutated stage because of the low power levels, ease of control and low conduction losses as compared to thyristors. The hardware prototype picture is shown in gure 11. The prototype inverter was built and tested for standalone systems. The predicted and simulated waveforms were conrmed. Some of the operating waveforms are given in
Fig. 9. 220 V
Test Case 3: Input Voltage 36V; Output Current 2 A; Grid Voltage
Fig. 11.
Hardware Prototype
Fig. 10. Test Case 4: Output Current changed from 2.8 A to 2 A; Input Voltage 30V; Grid Voltage 220 V
(a) Voltage waveform at pseudo dc-link (100 V/ division)
Figure 12(b) and 12(a). The voltage waveform at pseudo dc link is shown in gure 12(a). The experimental results of converter are given in stand-alone and open loop conguration with a rst order lter. Because of the absence of the control loop, the pseudo-dc link voltage is not sinusoidal. The output current and voltage waveforms are shown in gure 12(b). IV. C ONCLUSION A new design topology and control scheme is proposed for a grid-tied photovoltaic micro inverter. The proposed inverter can be utilized in grid-tied and stand-alone congurations for PV systems.The topology is based on a high frequency fullbridge converter, a pseudo dc link lter, and a line commutated inverter. The operational characteristics of the converter are analyzed in detail for continuous, boundary, and discontinuous conduction mode. A soft switching technique is presented for the proposed converter. Simulation and experimental results are demonstrated for proof-of-concept design specications. R EFERENCES
[1] E. R. E. Council, Renewable Energy Scenario to 2040, http://www.erec.org/leadmin/erec docs/Documents/Publications/ 2040Exec Sum.pdf, 2004, [Online; accessed 07-Mar-2013]. [2] S. Kjaer, J. Pedersen, and F. Blaabjerg, Power inverter topologies for photovoltaic modules-a review, in Industry Applications Conference, 2002. 37th IAS Annual Meeting. Conference Record of the, vol. 2, 2002, pp. 782788 vol.2. [3] Y. Xue, L. Chang, S. B. Kjaer, J. Bordonau, and T. Shimizu, Topologies of single-phase inverters for small distributed power generators: an overview, Power Electronics, IEEE Transactions on, vol. 19, no. 5, pp. 13051314, 2004. [4] S. Kjaer, J. Pedersen, and F. Blaabjerg, A review of single-phase gridconnected inverters for photovoltaic modules, Industry Applications, IEEE Transactions on, vol. 41, no. 5, pp. 12921306, 2005.
(b) Yellow: Output Voltage (scale 100 V / division); Blue: Output Current (Sensed via 0.1 Resistor) (scale 2 A / division) Fig. 12. Hardware Prototype Results
[5] V. Quaschning and R. Hanitsch, Inuence of shading on electrical parameters of solar cells, in Photovoltaic Specialists Conference, 1996., Conference Record of the Twenty Fifth IEEE, 1996, pp. 12871290. [6] A. Trubitsyn, B. Pierquet, A. Hayman, G. Gamache, C. Sullivan, and D. Perreault, High-efciency inverter for photovoltaic applications, in Energy Conversion Congress and Exposition (ECCE), 2010 IEEE, 2010, pp. 28032810. [7] A. Lohner, T. Meyer, and A. Nagel, A new panel-integratable inverter concept for grid-connected photovoltaic systems, in Industrial Electronics, 1996. ISIE 96., Proceedings of the IEEE International Symposium on, vol. 2, 1996, pp. 827831 vol.2. [8] C. Prapanavarat, M. Barnes, and N. Jenkins, Investigation of the performance of a photovoltaic ac module, Generation, Transmission and Distribution, IEE Proceedings-, vol. 149, no. 4, pp. 472478, 2002. [9] A. K. S. Bhat and S. Dewan, Resonant inverters for photovoltaic array to utility interface, Aerospace and Electronic Systems, IEEE Transactions on, vol. 24, no. 4, pp. 377386, 1988. [10] D. Gautam, F. Musavi, M. Edington, W. Eberle, and W. Dunford, A zero voltage switching full-bridge dc-dc converter with capacitive output lter for a plug-in-hybrid electric vehicle battery charger, in Applied Power Electronics Conference and Exposition (APEC), 2012 Twenty-Seventh Annual IEEE, 2012, pp. 13811386.
Design of an IIR Digital Filter for Phasor Measurement Improvement Based on a Hybrid Optimization Algorithm
Abdelmadjid Recioui, Abderrahmane Ouadi and Hamid Bentarzi
Laboratory Signals & Systems, Institute of Electrical Engineering and Electronics University of Boumerdes, Independence Avenue, 35000, ALGERIA sisylab@yahoo.com
Abstract A complete description of the phasor estimation process leads to the ideal multiobjective function that a digital recursive filter must fulfill in order to provide more accurate phasor estimates under power system disturbances. The raised cosine filter, widely used in digital transmission, is described and proposed to approach those criteria. In this work, a hybrid Taguchi Invasive Weed algorithm (HTIW) is employed to solve the problem of designing optimal digital infinite-impulse response (IIR) filters. The HTIW approach is a method of combining the conventional IWO, which has a powerful global exploration capability, with the Taguchi method that can exploit the optimum offspring. Simulation results corroborate an improvement in the performance (accuracy and speed) of these phasor estimates as compared with those obtained with the cosine filter.)
Keywords- Invasive weed optimization, Taguchi method, Digital IIR filters; Multiobjective optimization, phasor estimation I. INTRODUCTION
During the last three decades, digital relays have been widely used in power grid due to their flexibility and enhanced security and dependability properties. Their functions consist mainly on isolating the faulted network if an abnormal condition occurred in the electric power system. The decision of the digital protective device highly depends on the digital signal filtering computing the fundamental phasors of line voltages and currents. Several digital algorithm have been proposed in the literature for estimating the fundamental phasors, such as the discrete Fourier transform [1,2], cosine algorithm [3], least square algorithm [4],Kalman filter [5] and wavelet transform [6]. The discrete Fourier transform (DFT) filter which has proven reliability and robustness in relaying application was considered as the most widely used and most popular algorithm for estimating phasors [5,6]. The full cycle DFT (FCDFT) algorithm is applied for samples obtained during one cycle rectangular window. The half cycle DFT (HCDFT) filter is introduced by Phadke et al. [7] is applied to data samples obtained during half cycle window. Even though that it offers the advantage that it is faster than the FCDFT by a factor of two, it lacks the immunity to dc offset and does not filter out even harmonics as FCDFT filter.
However, during a transient time of power grid condition, line fault current may include unwanted components that are associated to the fault impedance level as well as to the power transmission line type. The first type of disturbance mainly appears in a form of dc exponentially decaying component associated with the line fault current signal waveforms. The initial amplitude and the time constant of this component is closely related to the fault position (transmission line circuit parameter X/R) and to the fault starting time (fault inception angle). The duration period of this dc decaying component may take 0.5 to 5 cycles [8] to die from the faulted line current. In this case of signal waveforms including dc components, phasors estimated through application of the DFT filter may have large errors during the first cycles. This problem stems from the fact that the signal containing dc exponential decaying components has non integer harmonics (using the Fourier transform of this dc component) while the DFT filter does not provide the ability to reject non integer type of harmonics. The second kind of disturbance that may occur in power transmission line including series compensation devices in presence of relatively high fault impedance, where a significant distortion is introduced in the fault line current waveforms. This is due to the behavior of non linear components integrated in series compensated devices. A considerable exponentially decaying sub-harmonic or subsynchronous is associated to fundamental and integer harmonic of the line current waveform. The sub-harmonic component is associated with line parameter (series capacitance and line inductance) while its amplitude is closely related to either the fault position and to the fault impedance level. The time constant is about twice as compared to that of short fault impedance (dc decaying component). This component may last 10 cycles to disappear. This type is more severe since it reduces the speed of convergence of the DFT algorithm to correctly estimate the fundamental phasors. The digital filter has been specified to fulfill multi-objective filter requirements as the high speed output time response involved in this real-time application, the frequency domain response requested to reject unwanted components. Besides the filter dynamic response objectives function, the magnitude response objective function is desired to pass only and ideal narrow band of frequency signals centered in the vicinity of
the fundamental frequency of the power system. In addition to the allowed narrow band, this digital band-pass filter must reject signal components located in the lower and higher frequency bands. The lower stop band includes unwanted dc decaying and sub-harmonic frequency components. The higher stop band concerns all harmonics of the power system fundamental frequency that are considered undesired components. This sustains the use of HCDFT filter phasor algorithm, which is inherently unable to remove even harmonic components. While this latter has the advantage of enhancing computing time and convergence speed of phasor calculations. The Invasive weed optimizer (IWO), inspired from the phenomenon of colonization of invasive weeds in nature, has been introduced by Mehrabian and Lucas in 2006[11] and is based on weed biology and ecology. One important property of the IWO is that it allows all of the agents or plants to participate in the reproduction process. The IWO, gives a chance to the less fit plants to reproduce and if the seeds produced by them have good finesses in the colony, they can survive. Another important feature of IWO is that weeds reproduce without mating. This property adds a new attribute to the algorithm that each agent may have different number of variables during the optimization process. Comparing some aspects of the IWO with some common and standard optimizers, such as Genetic Algorithms and Particle Swarm Optimizer, the parameters of the IWO need to be tuned and this is considered as a key in the convergence and accuracy of the algorithm. The Taguchi method involves reducing the variation in a process through robust design of experiments. The overall objective of the method is to produce high quality product at low cost to the manufacturer. The Taguchi method was developed by Dr. Genichi Taguchi of Japan who maintained that variation. Therefore, poor quality in a process affects not only the manufacturer but also society. The Taguchi optimization method is developed based on the orthogonal array (OA) concept to offer a systematic and efficient way to select design parameters. In addition, it reduces the number of tests required in the optimization process compared to Genetic Algorithms or Particle Swarm Optimizers [12]. Compared with the other optimization techniques, Taguchis optimization method is easy to implement and very efficient in reaching optimum solutions II.
PROBLEM FORMULATION
frequency is N times the fundamental frequency
s=N.
0,
and .
T obviously the sampling period Ts = 0 N
where T0 =
1 20
Assuming that the signal x(t) is corrupted with DC offset decaying having sub-harmonic component and M harmonic components, which can be expressed as:
t M +1
x(t ) = A0 e cos( wh t + sh ) +
n =1
An cos(n 0 t + n )
(1)
Here A0 and
are the amplitude and time constant of dc
decaying component,
sh
and
sh
are the frequency and
arbitrary phase of the sub-harmonic component, A1 and 1 are amplitude and arbitrary phase angle of the fundamental component and An and n are amplitudes and arbitrary phase angles of the harmonic components. The signal can be represented in discrete form by: M +1 2n x(k ) = A0e k .Ts cos(sh .k.Ts + sh ) + An cos( k + n ) (2) N n=2 The full cycle DFT filters, which are among the most popular in relaying, are used to compute fundamental phasor of this signal x(k). The fundamental DFT components of this discretized signal x(k) is provided by [21, 22] :
X (k ) = 2 L
L
x( k ).e
k =1
2 k N
= X r + jX i
(3)
Where X r and X i are real and imaginary part of
X (k ) and L
is the window length of the DFT, for L=N case the full cycle DFT computation is considered. The ideal fundamental DFT of x(t) of Eq.(3) may be calculated for a full cycle as : 2 N 2 2 (4) Xr = A1 cos( k + 1 ). cos( k ) L k =1 N N 2 N 2 2 (5) Xi = A1 cos( k + 1 ). sin( k) L k =1 N N The fundamental phasor can be represented in polar form, magnitude and angle, as:
A=
X r2 + X i2
(6.a)
Xi (6.b) ) Xr The fundamental frequency complex phasor can be obtained for a half-cycle data window length L=N/2 using Equations (4) and (5).
= tan 1 (
A. Full and half cycle DFT filter Analysis that is based on the experimental and simulation results obtained from the power system, shows that both voltage and current signal components are corrupted or contaminated by undesired signals such as the DC decaying and sub-harmonic components. Whereas the harmonic components may not be ignored and should be considered if a better measurement precision is needed. Consider a waveform signal x(t) having a fundamental frequency 0 and sampled at uniform and distinct time intervals x(k) for discrete computation. The sampling
B. Digital filter design The digital filter solution is proposed to remove the unwanted disturbances using digital filter design techniques. The filter design problem requires different kind of conflicting objective functions or specifications. The first objective function is the magnitude response which is desired to have an ideal passband having a narrow bandwidth with sharp edges centered at the fundamental power system frequency. The second objective is tied to time domain response, where a high speed filter output response (fast settling time) inquired in the present fast real-time phasor measurement applied in the
digital protective device. This required feature can be accomplished by requesting a filter having minimum phase or minimum group delay objective function. A quasi-linear phase filter or approximately constant group delay is a third desired objective function involved to avoid filter output waveform distortion. From the frequency response point of view, the filter can be designed using a digital non recursive (FIR) filters. However, a relatively high filter order is required for this class of filter (the high selectivity filtering application), and hence involve increased computational effort with lower speed and a large group delay. To mitigate these problems, researchers have devoted considerable effort in recent years, quite a variety of method for designing recursive digital filters (IIR) to obtain some desirable filter properties as nearly linear phase filter response with respect to the pass-bands since the filter phase can be discarded, or with no consequence in the stop-bands. III.
THE HYBRID TAGUCHI-INVASIVE WEED OPTIMIZER
Fig. 1 Seed production procedure in a colony of weeds.
A. Invasive weed optimization The Invasive Weed Optimization (IWO), inspired from the phenomenon of colonization of invasive weeds in nature, is based on weed biology and ecology. It has been shown that capturing the properties of the invasive weeds, leads to a powerful optimization algorithm. The behavior of weed colonization in a cropping field can be explained as follows: Weeds invade a cropping system (field) by means of dispersal and occupy opportunity spaces between the crops. Each invading weed takes the unused resources in the field and grows to a flowering weed and produces new weeds, independently. The number of new weeds produced by each flowering weed depends on the fitness of that flowering weed in the colony. Those weeds that have better adoption to the environment and take more unused resources grow faster and produce more seeds. The new produced weeds are randomly spread over the field and grow to flowering weeds. This process continues till the maximum number of weeds is reached on the field due to the limited resources. Now, only those weeds with better fitness can survive and produce new weeds. This competitive contest between the weeds causes them to become well adapted and improved over time.
A-1 Invasive weed optimization procedure
The process continue until maximum iterations is reached and hopefully the plant with best fitness it the closest to the optimal solution. Fig. 1 shows a flowchart of the algorithm. The process is addressed in details as follows: - Spatial dispersal Randomness and adaptation in the algorithm is provided in this part. The generated seeds are being randomly distributed over the d dimensional search space by normally distributed random numbers with mean equal to zero; but varying variance. This means that seeds will be randomly distributed such that they abode near to the parent plant. However, standard deviation (SD),of the random function will be reduced from a previously defined initial value, initial, to a final value, final, in every step (generation). In simulations, a nonlinear alteration has shown satisfactory performance, which is given in Eq. (7) (itermax iter ) n (7) iter = ( initial final ) + final (itermax ) n Where itermax is the maximum number of iterations,
iter is
To simulate colonizing behavior of weeds some basic properties of the process is considered 1) A finite number of seeds are being dispread over the search area (initializing a population) 2) Every seed grows to a flowering plant and produces seeds depending on its fitness (reproduction) 3) The produced seeds are being randomly dispread over the search area and grow to new plants (spatial dispersal), 4) This process continues until maximum number of plants is reached; now only the plants with lower fitness can survive and produce seeds, others are being eliminated (competitive exclusion).
the SD at the present time step and n is the nonlinear modulation index. This alteration ensures that the probability of dropping a seed in a distant area decreases nonlinearly at each time step which results in grouping fitter plants and elimination of inappropriate plants. - Competitive exclusion If a plant leaves no offspring then it would go extinct, otherwise they would take over the world. Thus, there is a need of some kind of competition between plants for limiting maximum number of plants in a colony. After passing some iterations, the number of plants in a colony will reach its maximum by fast reproduction, however, it is expected that the fitter plants have been reproduced more than undesirable plants. By reaching the maximum number of plants in the colony, pmax a mechanism for eliminating the plants with poor fitness in the generation activates. The elimination mechanism works as follows: when the maximum number of weeds in a colony is reached, each weed is allowed to produce seeds according to the mechanism mentioned earlier. The produced seeds are then allowed to spread over the search area. When all seeds have found their position in the search area, they are ranked together with their parents' (as a colony of weeds). Next, weeds with lower fitness are
eliminated to reach the maximum allowable population in a colony. In this way, plants and offspring are ranked together and the ones with better fitness survive and are allowed to replicate. This mechanism gives a chance to plants with lower fitness to reproduce, and if their offspring has a good fitness in the colony then they can survive. The population control mechanism also is applied to their offspring to the end of a given run, realizing competitive exclusion.
A-2 Parameter tuning
Among the parameters affect the convergence of the algorithm three parameters, the initial SD, , the final SD, , and the nonlinear modulation index, , should be tuned carefully in order to achieve the proper value of the SD in each iteration. The IWO works well if the initial SD is set around a few percent (1 to 5 percent) of the dynamic range of each variable. The final SD should be selected carefully to allow the optimizer to find the optimal solution as accurate as possible. The final SD in each dimension should be selected based on the precision effect of that variable on the objective function. The value of nonlinear modulation index has a considerable effect on the performance of IWO. It was suggested that the best choice for is 3 [12]. Maximum and minimum numbers of seeds are the two other important parameters needed to be selected. The maximum number of seed between 3 and 5 leads to a good performance of the optimizer. The maximum number of plants is another parameter that should be chosen in the IWO. It was shown that the best performance can be achieved for many problems when the maximum number of plants is set between 10 and 20. B. The Taguchi method Taguchis method was developed based on the concept of the orthogonal array (OA), which can effectively reduce the number of tests required in a design process [9]. It provides an efficient way to choose the design parameters in an optimization procedure. Before presenting the Taguchi procedure, it is worth understanding what OAs are and how are they generated [1215]. Let S be a set of s symbols or levels (the simplest symbols are integers 1, 2, 3). A matrix A of N rows and k columns with entries from S is said to be an OA with s levels and strength t (0< t< k ) if in every Nt subarray of A, each t-tuple based on S appears exactly the same times as a row. The notation OA(N, k, s, t) is used to represent an OA.
- Initialization procedure
The input parameters need to be selected to conduct the experiments. When the OA is used, the corresponding numerical values for the three levels of each input parameter should be determined. In the first iteration, the value for level 2 is selected at the center of the optimization range. Values of levels 1 and 3 are calculated by subtracting/adding the value of level 2 with a variable called level difference (LD). The level difference in the first iteration (LD1) is determined by the following equation: Max Min (8) LD 1 = Number of levels + 1 Where Max and Min represent the upper and lower bounds of the optimization range, respectively.
- Conduct Experiments and Build a Response Table
After determining the input parameters, the fitness function for each experiment can be calculated. These results are then used to build a response table for the first iteration by averaging the fitness values for each parameter n and each level m using the following equation:
Fav =
s N
i i ,OA ( i , n ) = m
(9)
- Identify Optimal Level Values and Conduct Confirmation Experiment
Finding the largest fitness value ratio in each column can identify the optimal level for that parameter. When the optimal levels are identified, a confirmation experiment is performed using the combination of the optimal levels identified in the response table. This confirmation test is not repetitious because the OA-based experiment is a fractional factorial experiment, and the optimal combination may not be included in the experiment table. The fitness value obtained from the optimal combination is regarded as the fitness value of the current iteration.
- Reduce the Optimization Range
If the results of the current iteration do not meet the termination criteria, the process is repeated in the next iteration. The optimal level values of the current iteration are used as central values (values of level 2) for the next iteration. To reduce the optimization range for a converged result, the LDi is multiplied with a reduced rate (rr) to obtain LDi+1 for the (i + 1)th iteration: (10) LDi +1 = rr LDi = RR(i ) LDi Where RR(i ) is called reduced function. When a constant rr is used, RR(i) =rri . The value of rr can be set between 0.5 and 1 depending on the problem. The larger rr is, the slower the convergence rate. If LDi is a large value, and the central level value is located near the upper bound or lower bound of the optimization range, the corresponding value of level 1 or 3 may reside outside the optimization range. Therefore, a process of checking the level values is necessary to guarantee that all level values are located within the optimization range. If an excessive situation happens, reassigning the level value for the
The optimization procedure starts with the problem initialization, which includes the selection of a proper OA and the design of a suitable fitness function. The selection of an OA(N, k, s, t) mainly depends on the number of optimization parameters. In general, to characterize the nonlinear effect, three levels (s = 3) are found sufficient for each input parameter. Usually, an OA with a strength of 2 (t = 2) is efficient for most problems because it results in a small number of rows in the array [12].
- Design of input parameters
parameter will be performed. A simple way is to use the boundary values directly.
- Check the Termination Criteria
When the number of iterations is large, the level difference of each element becomes small from equation (10). Hence, the level values are close to each other and the fitness value of the next iteration is close to the fitness value of the current iteration. The following equation may be used as a termination criterion for the optimization procedure: LDi (11) < converged value
LD1
design variable which is obtained from the optimization engine. After working out with the software developed for optimization filter design, and in order to achieve filter rejection requirements better than the cosine filter, the filter cut-off frequencies have been modified to 55 Hz and 64 Hz. To optimize the computational power needed in the filter implementation, only 3 second order sections has been considered in the optimization process.
Usually, the converged value can be set between 0.001 and 0.01 depending on the problem. The iterative optimization process will be terminated if the design goal is achieved or if equation (11) is satisfied. C. The Hybrid Taguchi Invasive weed optimizer The Algorithm is summarized as follows: Begin Generation 0 1. Start with a random population of size N (preferably even). 2. Evaluate the fitness of each individual. 3. Rank the individuals based on their fitness (from best to lowest) 4. Keep record of the best individual (elitism). 5. Split the population into two equal subgroups (the most fit in one group and the least fit in another one). Generation 1 and on 6. Apply Invasive weed operators on the least fit individuals. 7. Apply the Taguchi operators on the fittest individuals. 8. Group back the newly generated individuals from both invasive weed and Taguchi methods 9. Repeat steps 2 through 4. Update fittest individual if a better one is found. 10. Repeat steps 6 through 8. 11. Check stopping criteria End IV.
RESULTS AND DISCUSSIONS (c)
(a)
(b)
The optimization algorithm was implemented using MATLAB code in which the p-norm parameter p variation has been bounded between 2 and 128. A least square optimization design is considered by setting p=2. A mini-max or optimization design is assumed for the case of p=128. The desired digital filter in the present application is band-pass filter type having an ideal magnitude response of the form: 1 fL < f < fH (12) H ( ) =
elsewhere
(d) Fig.2 frequency response of the digital filter: (a) Magnitude response, (b) phase response, (c) Group delay in the pass-band, (d) pole-zero plot
Where f L and f H are the desired bandwidth limit or cut-off frequencies fixed respectively to 45Hz and 55 Hz. The desired minimum group-delay has been included in the optimization
The optimization was carried out for the required design specifications and the obtained filter coefficient results are summarized in table1. The designed amplitude frequency response is plotted in Fig.2.a. The phase response and group delay response are shown in Fig. 2.b and Fig.2.c in the pass band region. The step response and the step cosine responses are plotted in Fig.3.a and Fig. 3.b. The different designed filter characteristics show a compromise or a trade-off between the filter multi-objective functions, which nearly approximate the desired filter requirements.
V.
CONCLUSIONS
In this paper, a hybrid optimization technique between invasive weed optimization and Taguchi method has been used to design a digital filter to enhance the phasor measurements in a power system protection system employing phasor measurement units. The designed filter has proved to be better than full cycle DFT filter. The hybrid invasive weedTaguchi optimization technique hence has reached the best possible design and proved to be useful in engineering applications. REFERENCES
[1] B A. G. Phadke, Synchronized phasor measurements in power systems, IEEEComput. Appl. Power, vol. 6, no. 2, Apr. 1993, pp. 1015. [2] A.G. Phadke and J.S. Thorp, Synchronized phasor measurement and their applications, Springer press, 1990. Science+Business Media, LLC, 2008. [3] G. Phadke, J. S. Thorp, and M. G. Adamiak, A new measurement technique for tracking voltage phasors, local system frequency, and rate of change of frequency, IEEE Tran. on Power Systems and App., Vol. PAS-102, 1983, pp. 1025- 1038. [4] A. A. Girgis, A. A. Sallam and A. Karim El-Din, An adaptive protection scheme for advanced series compensated (ASC) transmission lines, IEEE Trans. on Power Del., Vol. 13, , 1998, pp. 414-420. [5] T. Lin and A. Domijan, Recursive algorithm for real-time measurement of electrical variables in power systems, IEEE Tran.Power Del., vol. 21, 2006, pp.15-22. [6] Y. Guo, M. Kezunovic and D. Chen, Simplified Algorithm for Removal of the Effect of Exponentially Decaying DC-offset on the Fourier Algorithm,IEEE Trans. Power Deliv. Vol.18, no.3, pp. 711, July 2003 [7] J. Li, X. Xie, J. Xiao and J. Wu, The framework and algorithm of a new phasor measurement unit, IEEE Int. Conf. on Elec. Util. Dereg., Restruc. And Power Tec. (DRPT2004), 2004, pp. 826-831. [8] Jyh-CherngGu, Sun-Li Yu, Removal of DC offset in current and voltage signals a novel Fourier Algorithm, IEEE trans. Power Deliv. Vol.15, no.1, pp. 73-79 , Jan. 2003. [9] G. Benmouyal, Removal of DC-offset in Current Waveforms Using Digital Mimic Filtering, , IEEE trans. Power Deliv. Vol.10, no.2, pp. 621630 , April. 1995. [10] M.S.Sachdev and M. A. Baribeau,A New Algorithm for Digital Impedance Relay, IEEE trans. Power Apparatus and systems.Vol.98, pp.2232-2240, November/December 1979. [11] Ali Reza Mehrabian, and Caro Lucas, "A novel numerical optimization algorithm inspired from weed colonization", Ecological Informatics 1(4):355366 (2006) [12] Weng, W., F. Yang, and A. Elsherbeni, 2007, Linear antenna array synthesis using Taguchi's method: A novel optimization technique in electromagnetic", IEEE Trans on Antennas and Propagation, Vol. 55, 723730. [13] Wei-Chung Weng and Charles T. M. Choi, 2009; Optimal Design of CPW Slot Antennas Using Taguchis Method, IEEE Transactions on magnetic, Vol. 45, NO. 3. [14] A. S. Hedayat, N. J. A. Sloane, and J. Stufken, 1999, Orthogonal Arrays: Theory and Applications, Springer-Verlag: New York. [15] N. Dib, S. K. Goudo, H. Muhsen, 2010, Application of Taguchi's optimization method and self adaptive diffrrential evolution to the synthesis of linear antenna arrays, Progress In Electromagnetics Research, PIER 102, 159-180.
(a)
(b) Fig.3 Filter output response: (a) for step cosine input signal and (b) for step input signal
Table 1 Direct-form 2 filter second order coefficients NUM Coefficients DEN Coefficients Section a01 = 1 b01 =1 1 a 11 = -1.9993984 b 11 =-1.45738001 a 21 = 0.9999998 b 21 =0.5625 Section 2 Section 3 Section 4 Scale Value: a 02 = 1 a 12 = -1.8817117 a22 = 0.9999998 a 03 = 1 a13 = -1.6978919 a 23 =0.9999998 a 04 =1 a 14 =-.38841470 a 24 =0.9999998 H0 =0.0135986352 b 02 =1 b12 =-1.45738001 b 22 =0.5625 b03 =1 b 13 =-1.45738001 b 23 =0.5625 b 04 =1 b 14 =-1.45738001 b 24 =0.5625
New Algorithm to Detect Balanced Three Phase Fault during Power Swing on Transmission Lines with Communication Links
Mohammed H. Qais
Electrical Engineering Department King Saud University Riyadh, Saudi Arabia Qais65_2002@yahoo.com
Abstract distinguishing a balanced three phase fault from a power swing is difficult since both have similar characteristics. In this paper a new algorithm to detect three phase faults during power swings is proposed. The algorithm is based on separating the power swing blocking (PSB) schemes into two distinct schemes, namely PSB1 and PSB2. The first scheme (PSB1) is dedicated to blocking of zones 2 and 3, whereas the second scheme (PSB2) is dedicated to blocking of zone 1. Every relay send signal to the other relay, then coordination is applied between relays at both ends of protected transmission line using communication link. The proposed algorithm increases the reliability and speed of distance protection to detect three phase faults during power swings. Keywords-component; Power swings, transmission protection, distance relays, relay blocking schemes. line
Saad M. Alghuwainem
Electrical Engineering Department King Saud University Riyadh, Saudi Arabia saadalgh@ksu.edu.sa protection zones of the relay. This is known as decrease impedance method [2]. II. IMPEDANCE CALCULATION
If the power swing trajectory enters the operating characteristics of a distance relay, the distance relay will detect the power swing as a three phase fault. The impedance seen by a distance relay during a power swing condition for the simple two-source system shown in Fg.1, will be calculated. The receiving-end source R will be used as a reference ER 0. We can express the sending-end source as Es . The current flowing on the transmission line is, (1) I = (ES ER ) / X Where X is the total inductive reactance of circuit. The voltage measured at the sending-end bus can be found as following: V = ES XS . I (2)
I.
INTRODUCTION
Distance relays see impedances during power swings which are similar to the impedance seen during a balanced threephase fault with the exception that due to power swing oscillations the impedances oscillate back and forth, entering and leaving the relay protection zone. However if the relay time setting is fast, and the power swing oscillation is slow, it might operate unnecessarily due to the power swing. The power swing oscillation frequency is determined by the inertia of the systems and the line impedances (lengths) between different generators [1]. Power swings can cause undesirable operation of distance relays which might lead to a complete blackout especially for stressed power systems. By analyzing the data obtained from the national control center of Yemeni Power System. The system blackout is occurring in Yemeni Grid repeats about three times per year. The outage of some generating units in overloaded system result to power swing in the system. Power Swing is seen by some distance relays as a three phase fault which trip the protected transmission lines (TLs). Tripping TLs leads to system blackout. The most popular method to detect a power swing is to measure the rate of change of impedance as it travels into the
Then the impedance seen by a relay located at the sending bus is given by: (3) Z = V / I = - XS+X (ES / (ES ER)) Suppose that |Es|=|ER| then the impedance is calculated as the following Z= X/2 XS j (X/2)cot ( / 2) (4)
Figure 1. Figure 1. Two source system
III.
POWER SWING TRAJECTORY DETECTION
A power swing can be best detected by analyzing its impedance trajectory behavior in a certain time window. The impedance is calculated Every 5 milliseconds to check its continuity. A power swing suspicion is established, if the impedance changes at least 50 milliohms during each calculation for six consecutive calculations. The continuity test checks that the rate of change of the impedance in R and in X is not beyond a limit. Thus it guarantees that the impedance trajectory has a uniformly smooth movement without abrupt changes. There is alternative limit used instead of fixed limit. The limit is calculated based on previously calculated values, this leads to a dynamic calculation of the limits [3]. Fig. 2 shows the moving of the impedance trajectory from the load area towards the characteristic zones of relay. During the fault the impedance trajectories jump quickly into protected zones area. During power swing the impedance trajectory takes longer time to enter protected zones area. According to the time of this movement the relay algorithms can decide whether it is a fault or a power swing.
see power swing impedance locus inside zone 3 only or outside all protected zones. As long as distance relays R1 and R2 see the power swing impedance in different locations, they can operate together with the presence of a communication link. Every Relay sends and receives signal transmitted by communication link to enhance the reliability and speed of distance relays. Power swing affects distance relays according to the location of the disturbance. The disturbances that can lead to power swing are: generation units outage and tripping faulted transmission lines. The main and common protected zone for both relays R1 and R2 at both sides of TL is zone 1. The power swing impedance commonly enters zone 3 of both relays. Sometimes power swing impedance trajectory enters zones 2 & 3 at one side and enters all zones at other side. The power swing blocking (PSB) will block all protected zones of distance relay during power swing. If three phase fault incepted in the presence of PSB the relay will not operate because it is blocked. In this paper, New Algorithm is presented to detect faults incepted in the presence of power swing blocking as shown in the flowchart in Fig. 5. Two proposals are presented in the proposed algorithm: The first proposal is to separate the power swing blocking PSB area into two areas PSB1 and PSB2 as shown in Fig. 4. PSB1 is to block zones 2 and 3 only, whereas PSB2 is to block zone 1. The second proposal is to send trip signals (SR1 and SR2) between R1 and R2 through communication link. The sent trip signal from any side will unblock the blocked relay due to power swings. If relay R1 is totally blocked by PSB1 and PSB2, then in the other side only zones 2 and 3 of relay R2 are blocked. Therefore, if any fault is incepted to the protected line (zone 1) the relay R2 will operate.
Figure 2. Figure 2. Continuous monitoring of the impedance trajectory [3]
IV.
LITERATURE REVIEW OF PUBLISHED ALGORITHMS TO DETECT POWER SWING
Many algorithms have been proposed to detect symmetrical faults during power swings. In [4-7] the authors presented new methods to detect symmetrical and asymmetrical faults during power swings. The presented methods are based on the current, voltage, three-phase active and reactive power signals, and based on the damping frequency component of 50 (or 60) Hz. In [8] a new fast detector of symmetrical fault during power swing is presented, based on the sudden reduction of absolute value of the rate of change of power swing center voltage. V. NEW PROPOSED ALGORITHM TO DETECT FAULTS
DURING POWER SWING
Figure 3. Figure 3. Single line diagram of modeled power system
Transmission lines (TLs) are protected by directional distance relays which are located at both ends of line. The relays at both ends look in one direction, so both relays see the protected TL in opposite direction. In Fig. 3 relay R1 at the left side of TL can see the power swing impedance locus inside the protection zones (1, 2 and 3). Relay R2 at the right side of TL
Figure 4. Figure 4. PSB1 and PSB2 areas
VI.
SIMULATION AND RESULTS
The power system model is simulated using PSCAD as shown in Fig. 6. The simulated system consists of one equivalent large 11-kV, 206-MVA generator, 230-kV infinite bus-bar, two parallel transmission lines, and 11-kV/230 kV transformer. The length of transmission line TL1 is 300 km. The positive sequence impedance of TL1 is Z1=10.4+j113 ohm, and the zero sequence impedance Z0=74.1+j387 ohm. The TL1 is protected by two distance relays R1 and R2 located at both sides of TL1. The zone 1 impedance setting (ZR1) is 80% of total impedance Z1 of TL1 (ZR1 < 91 ohm). Zone 2 impedance setting (ZR2) is 120 % of Z1 of TL1 (ZR2 < 136.5 ohm). Zone 3 impedance setting (ZR3) is 100% of Z of TL1 plus 120% of adjacent transmission line (ZR3 < 250.25 ohm). The measured voltage and current signals are filtered by Digital Fast Fourier Transform FFT block in PSCAD. The fundamental signals of V and I are used to calculate the impedance seen by relays R1 and R2. Power swing occurs after tripping the faulted TL2 (400 km) by CBs located at both ends of TL2. The impedance trajectory locus seen by relays R1 and R2 located at the ends of TL1 will enter and leave the protected zones as shown in Fig. 8, Fig. 9. A mho characteristic of zone 1 is the reference for all figures from Fig. 7 to Fig. 12. It is clear in Fig. 9 at left side that the impedance trajectory is entering all zones 1, 2 and 3 of relay R1. The relay R1 will be blocked by the power swing blocking PSB1 and PSB2. The impedance trajectories seen by relay R2 as shown in Fig. 9. Only zones 2 and 3 will be blocked by PSB1 whereas zone 1 will not be blocked. Fig. 10 to Fig. 12 shows the occurrence of a fault during power swing. The impedance seen during fault is inside zone 1 of relays R1 and R2. It is clear from Fig. 11 that zone 1 of R1 is blocked by PSB2 but zone1 of R2 is not blocked because the power swing is outside zone 1. The unblocked zone 1 of relay R2 will decide to trip the fault instantly and send signal SR2 through communication link to unblock the relay R1 at left side. The sent signal will unblock zone 1 only of relay R1 so the security of the relays will not be affected and the reliability of relays increased. The proposed algorithm in this paper should work in parallel with other published algorithms, then by changing the location of source of power swing and the location of faults will not affect the security of relays. The advantage of the proposed algorithm is its fast detection of faults during power swing keeping the reliability of relays. The simulation is repeated for different locations of source of power swing. It is rare that all zones of both relays see the power swing trajectory.
Figure 5. Flow Chart of new algorithm for detecting fault during power swing
Figure 6.
Simulated power system model using PSCAD Figure 9. Measured impedance at R2 of protected TL1
Figure 7.
measured voltage and currents Figure 10. measured voltage and currents
Figure 8. Measured impedance at R1 of protected TL1.
Figure 11. Measured impedance at R1 of protected TL1.
Characteristics," International Conference on Advancements in Electronics and Power Engineering, Bangkok, 2011. [7] J. Shaohua, B. Zhiqian, L. Wanshu, and Y. Qixun, "New Principles to Detect Faults During Power Swing," IEEE Seventh International Conference on Developments in Power System Protection, 2001, pp. 515518. [8] B. Su, X. Z. Dong, Z. Q. Bo, Y. Z. Sun, B. R. J. Caunce, D. Tholomier, and A. Apostolov, "Fast Detector of Symmetrical Fault During Power Swing for Distance Relay," IEEE Power Engineering Society General Meeting, 2005, pp. 1836-1841 Vol. 2.
Figure 12. Measured impedance at R2 of protected TL1 with fault
VII. CONCLUSION The simulation shows that Relays at both sides of protected transmission line can see the measured impedance during power swing in different locations. The impedance trajectory of power swing can enter all zones of one relay at one side but in the other side will enter third zone or second zone. If there is a fault at protected transmission line during power swing it will be exactly inside the first zone, which is not blocked by power swing blocking. The unblocked zone 1 of relay will send trip signal through communication link to the relay at the opposite side of transmission line. The sent signal will unblock the first zone of the other relay which directly detect the fault. According to the results and proposed algorithm, both relays will trip the faulted transmission line during power swing. REFERENCES
[1] H. Khoradshadi-Zadeh, "Evaluation and performance comparison of power swing detection algorithms," in Power Engineering Society General Meeting, 2005. IEEE, 2005, pp. 1842-1848 Vol. 2. [2] H. K. Karegar and B. Mohamedi, "A new method for fault detection during power swing in distance protection," in Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology, 2009. ECTI-CON 2009. 6th International Conference on, 2009, pp. 230-233. [3] L. Martuscello, E. Krizauskas, J. Holbach, and L. Yuchen, "Tests of distance relay performance on stable and unstable power swings reported using simulated data of the August 14<sup>th</sup> 2003 system disturbance," in Protective Relay Engineers, 2009 62nd Annual Conference for, 2009, pp. 236-255. [4] B. Mahamedi and Z. Jian Guo, "A Novel Approach to Detect Symmetrical Faults Occurring During Power Swings by Using Frequency Components of Instantaneous Three-Phase Active Power," Power Delivery, IEEE Transactions on, vol. 27, pp. 1368-1376, 2012. [5] N. Z. Mohamad, A. F. Abidin, and W. N. W. A. Munim, "A new tracking method of symmetrical fault during Power Swing Based on S-Transform," in Power Engineering and Optimization Conference (PEDCO) Melaka, Malaysia, 2012 Ieee International, 2012, pp. 141-146. [6] S. Saeid Taheri and HoseinAskarian, "Detecting of a Power Swing Phenomenon duringTwo Worst Simultaneous Faults Using System
Stable Operation of Grid Connected Cascaded H-Bridge Inverter under Unbalanced Insolation Conditions
Abbas Eskandari
School of Electrical and Computer Engineering University of Tehran Tehran, Iran Abbasskandari@ut.ac.ir
Vahid Javadian
Department of Electerical Engineering Sharif University of Technology Tehran, Iran Javadian@alum.sharif.edu
Hossein Iman-Eini
School of Electrical and Computer Engineering University of Tehran Tehran, Iran Imaneini@ut.ac.ir
Milad Yadollahi
School of Electrical and Computer Engineering University of Tehran Tehran, Iran M.yadollahi65@gmail.com
Abstractthis paper presents a single-phase Cascaded H-Bridge (CHB) inverter for photovoltaic applications. The proposed control strategy permits the optimum control of each string of PV panels in different insolation conditions. Based on the presented mathematical analysis, an analytic constraint is derived to check the stable operation of the inverter. A modified MPPT control strategy is also proposed based on this new condition. This strategy enhances the stability margins of the system by changing the operating point of some PV arrays. In this approach, the extracted power from high power cell is reduced to prevent instability issue and to avoid injection of highly distorted current into the grid, keeping the total harmonic distortion (THD) below 5%. The performance and effectiveness of the proposed method is verified by simulations in PSCAD/EMTDC environment, under severe imbalance insolation conditions. Keywords-cascaded H-bridge; MPPT; multilevel inverter; grid connected inverter
needs isolated dc-links which makes it appropriate for photovoltaic applications. Traditionally CHB inverters need three blocks of controllers in PV applications. In the first block, a PI controller is used to regulate the sum of the dc-link voltages to the appropriate reference voltage, so that the maximum power is derived from each PV panel [7, 8]. Second block is used for current control and is necessary to achieve a low distortion current in the grid side [9, 10]. The third block determines the modulating strategy in the line side inverter, regarding the employed control strategy [11]. Based on these requirements, a gird connected CHB inverter should control N+1 variables (N dc link voltages and one ac current) while there is only N switching functions, which leads to inherent instability problem especially under unbalanced insolation conditions. Nevertheless, several control methods have been proposed for CHB inverters. In [12, 13], the reference signal for ac current is adjusted using a factor proportional to dc link voltage of each PV panel. In some control methods, the performance of the system is only evaluated under balanced insolation conditions [14]. The proposed structure in [15], uses a simple way to control the dc link voltages and to detect the maximum power. All previous strategies do not consider stability issues or do not study the system behavior in non-ideal conditions. Instability issues in a symmetric CHB inverter have been studied in [16] and a solution has been offered. This method uses the power factor variable to stabilize the operation of CHB inverter in different insolation conditions. In this approach, when the CHB inverter becomes unstable, the amplitude of ac current increases and the reactive power is injected to the utility grid. Using this strategy, the system is kept stable and brought back to linear mode. In this paper, a novel control method is presented to stabilize the grid connected CHB inverters under unbalanced insolation conditions. Based on the mathematical analysis, stability margin is extracted for the CHB inverter and a novel technique is introduced to keep the converter operation in linear mode. This control method extends the stable operating range of CHB inverters and keeps THD of the output current in acceptable range.
I.
INTRODUCTION
Power electronic inverters are used in photovoltaic applications for extracting the maximum power from the photovoltaic panels and injecting the power into load or grid. The inverter is also used to convert the output dc voltage of the PV arrays into desired ac voltage, suitable for ac costumers. Among the proposed power electronic converters in singlephase grid-connected PV systems, two stage power processing structures are commonly used. This topology employs a dc-dc converter to track the maximum power point (MPP) and a dcac inverter to synthesize the output ac voltage. This method is simple, modular, and facilitates the control of two successive converters [1,2]. Nonetheless, in recent years, attention to single stage power processing methods has increased. With a single stage converter, the total efficiency increases and the system cost reduces. One of these approaches is to use a Cascaded H-bridge (CHB) multilevel inverter to inject a high quality ac power into grid [3-6]. Using a CHB inverter, it is also possible to generate a low distortion stepwise waveform with a low switching frequency. This feature enhances the output power quality of the system, reduces the size of output filters and decreases EMI noise. Moreover, a CHB inverter
II.
TOPOLOGY DESCRIPTION
A. Operating principle of grid connected CHB inverter The structure of a seven-level cascaded H-bridge inverter has been shown in Fig. 1. In this structure, an inductive filter is necessary between the inverter ac terminals and the utility gird. This topology is made of a series connection of H-bridge cells with isolated dc links. In the CHB inverter, each H-bridge cell can generate three voltage levels, i.e., Vdc, 0 and +Vdc. Therefore, using N (here N=3) cells, 2N+1 voltage levels can be synthesized at the ac terminals of the inverter. Increasing the number of voltage levels will lead to a better voltage waveform and lower THD at the injected grid current. But, increasing the number of voltage levels will increase the control system complexity and therefore N=3 is chosen for the study in this paper. In order to inject active power from PV modules into the grid, it is necessary to have a dc link voltage higher than grid peak voltage. Therefore, the number of series connected PV modules in each H-bridge cell is selected so that the total dc voltage to be 343V which is almost 10 percent higher than grid peak voltage, i.e., 310 V. The specifications of the PV modules which are employed in the study are given in Table I. It is seen that the corresponding voltage at MPP is approximately 16.3 V, so seven modules should be connected in series, in each PV string. The base of control system which is employed in the CHB inverter is demonstrated in Fig. 2 [17]. In this paper, the MPPT algorithm will be modified to improve the system performance. The control system includes three voltage control loops, one internal current controller and a multilevel PWM modulator. It is seen that the current and voltage of each PV string is sensed and the corresponding voltage of MPP, i.e. Vpv* is determined according to P&O algorithm. Then, the reference value is compared with the string voltage Vpv and the error values of all H-bridge cells are summed together. In the next stage, the total error is entered to a PI compensator and it determines the amplitude of grid current (or the amount of active power) which should be injected into the grid. Since the injected current must be in-phase with the grid voltage, its amplitude is multiplied by a sinusoidal waveform derived by a PLL from grid voltage Vg. Finally, the reference current ig* is compared with the grid current ig and a Proportional Resonant (PR) controller is employed to zero the current error. The output of current controller is summed with the grid voltage and generates the modulator command, i.e. Vg*. This voltage reference is synthesized by the 7-level CHB inverter at the ac terminals, using the multicarrier phase shift modulation technique (or PSPWM) [11]. In this modulation, the phase difference among the carriers is set to 60 degrees in order to achieve the minimum THD at the injected current. B. CHB inverter equations Noting to the CHB structure in Fig.1 and neglecting the series resistance of the line side filter, one can derive the following equations for the grid connected CHB inverter.
Figure 1. Structure of 7-level cascaded H-bridge inverter in grid connected PV applications
i pvi -C
dv pvi dt
=si ig .
(1)
si =
v hi . Vpvi
(2)
s V
i i
pvi =vg +L
di g dt
where i = 1, 2, 3.
(3)
Where, vhi represents the fundamental harmonic waveform of the ac terminal at the ith H-bridge cell and si is the modulating waveform of the ith H-bridge cell. ipvi and vpvi are current and voltage waveforms of the ith PV string (connected to the ith H-bridge cell), Ig and Vg indicate respectively the current and voltage of the utility grid. Moreover, L and C represent the amount of output inductance and the dc-link capacitor of each H-bridge cell. It is worth noting that, small letters are used to demonstarte ac vaiables (or waveforms) and capital letters are used to demonstrate average (or rms) value of corresponding variables. For correct operation of the inverter in linear zone and keeping stability of the system, the amplitude of modulating waveform, i.e. |si| of each cell must be bounded inside [-1,+1] span. Equation (4) determines the average value of PV current at each H-bridge cell derived from (1), in steady state condition.
Figure 2. General structure of control system in 7-level CHB inverter in PV applications
I pvi =
1 si ig cos. 2
(4)
H-bridge cells go into the over-modulation region. In this case, the converter must be isolated from the gird. In steady state condition, the injected current into the grid has the unity power factor and assuming that the voltage drop across the output filter is negligible, cos can be approximated by unity (or cos 1). Then, (4) can be modified as follows to demonstrate the stability constraint for each PV array:
Where is the phase difference between the grid current and the fundamental harmonic of the ac-side voltage, i.e. vhi. By neglecting the series resistance and the converter loss, total generated power by PV panels will be equal to the injected power into the grid. This is shown in (5).
2I pvi ig
i = 1, 2,3.
(6)
P=
I
i
pviV pvi
1 v g ig . 2
(5)
III.
DERIVATION OF STABILITY MARGIN
Because, in stable operating mode, the amplitude of modulating waveform (|Si|) of each cell must be kept lower than unity. Noting to (5) and (6), the following condition is derived for the stability boundary of the system:
I pviV g I pv 1 V pv 1 + I pv 2V pv 2 + I pv 1 V pv 3
Adding a current or voltage loop to power electronic system controller can result in instability [18]. This issue becomes more critical when the grid connected CHB inverter operates in unbalanced conditions such as non-uniform irradiation condition of PV strings. The control system demonstrated in Fig. 2 has a stability margin whose range depends on dc link voltage and PV array characteristics. If the insolation condition of PV arrays changes so that the operating point of some PV arrays move outside the stability margin (or the modulation index of some H-bridge cells increases higher than unity), the injected current will distort considerably. On the other hand, the instability begins when the operating point of some H-bridge cells enters to the non-linear zone. This problem occurs when one or two of the
i = 1, 2,3.
(7)
When the above condition is satisfied for each H-bridge cell, the CHB inverter will remain in stable region and the modulation index will be lower than unity. According to (7), if the input power into the some cells decreases and consequently the modulation index of high power cell increases and goes beyond unity, the inverter will become unstable and will inject a high distorted current into the grid. To solve this problem in imbalance condition, the operating point of high power cell must be shifted from MPP to the voltage source region of I-V
TABLE I. THE INITIAL CONDITION AND DATA OF SIMULATION FOR PV
ARRAYS
Initial insolation level Initial temperature Output filter capacitance Output filter inductance Grid peak voltage Maximum peak power of PV module Corresponding voltage of MPP in the module Figure 3. The modified MPPT block for improving the stability of system Number of series modules in a PV array Number of parallel modules in a PV array
Einitial Tinitial C L Vg PMPP VMPP NSeries Nparallel
1000 [w/m2] 298 [k] 3 [mF] 4 [mH] 311 [v] 45 [w] 16.3 [v] 7 5
curve. Using this approach, the current of high power cell will decrease and leads to reduction of modulation index. According to (7), reduction rate of numerator term is higher than denominator when the current of high power cell decreases. Hence, by decreasing the input energy from the high power cell, the modulation index would remain inside the specified [-1,+1] span. This method can be generalized for any number of H-bridges in a multilevel CHB converter and it can be used as a stability constraint in the MPPT block. The modified MPPT block in this paper use this constraint (7) to extend the operating range of grid connected CHB inverter. According to Fig. 3, while the amplitude of modulation index (in each H-bridge cell) is lower than unity, the maximum power point of each array is tracked by P&O algorithm. But, in the unbalanced insolation conditions which lead to modulation index of some cells become higher than unity, the proposed algorithm shifts the operating point of high power cell to the point in which the constraint of (7) is satisfied. This function avoids any interruption of the CHB inverter and keeps the system stability. IV. SIMULATION RESULTS
(a)
In this part, the performance and effectiveness of the proposed method is verified by several simulations on a 7-level CHB converter in PSCAD/EMTDC environment. The system parameters in the simulations are listed in Table I. Since among the various methods of MPPT, P&O method seems to have good compatibility with the proposed idea of this paper, Perturb and Observe (P&O) algorithm is used for maximum power point tracking of each PV arrays. A Proportional Resonant (PR) current controller is also employed to compensate the grid current because of its good performance and simplicity. The first simulation investigates the performance of the proposed control scheme in imbalance irradiation condition. In this study, the irradiation level of second array, i.e., E2 changes from initial value 1000 W/m2 to 500 W/m2 at t=1 s and the irradiation level of third cell, i.e., E3 reduces from 1000 W/m2 to 700 w/m2 at t=2 s. The proposed algorithm then is applied at t=3 s to investigate its behavior before and after applying the algorithm. Corresponding simulation results for this study are demonstrated in Fig. 4 and Fig. 5.
(b) Figure 4. Investigation of grid current under unbalanced insolation condition, a) before applying the proposed algorithm, b) after applying the proposed algorithm (with inhancing the operation of system)
Fig. 4 illustrates the waveform of the grid current before and after applying the proposed control scheme, in steady state. From Fig. 4(a) and 4(b), it is concluded that the total harmonic distortion (THD) of the grid current increases from 1.7% to 12% after first change in the irradiation level. This distortion even becomes higher (20%) after second change in the irradiation level. But, when the proposed algorithm is applied at t=3 s, the controller increases the voltage of the first PV array to reach a modulation index lower than unity. As a result, THD of the injected current reduces to 4.5% which satisfies the standard. Fig. 5 also demonstrates dynamic behavior of the dclink voltages, first PV array power into the first H-bridge cell, and the modulation index of the first H-bridge cell.
(a)
(a)
(b)
(b)
(c)
(c) Figure 6. Investigation of inverter performnace in imbalance temperature conditions a) waveform of grid current before applying the proposed algorithm(unstable), b) after applying the proposed algorithm(stable), c) corresponding dc-link voltages of PV arrays.
control method is applied at t=2 s, THD decreases form 25% to 3%. Fig. 6 shows the corresponding simulation results and the behavior of dc link voltages, before and after temperature change in the first array.
(d) Figure 5. Dynamic behavior of CHB inverter under unbalaned insolation condition: a) dc-link voltages, b) Corresponding power of PV arrays, c) modulation waveform of first cell, d) modulation waveform of cells after applying the new algorithm (especially for the first cell)
The simulation results in this part confirm the correct behavior of new algorithm under unbalanced insolation conditions and when the temperatures of PV arrays are different. V. CONCLUSIONS
Second simulation investigates the performance of the proposed algorithm when the module temperature changes. In this simulation, temperature of the first panel increases from T=25 to T=77 at t=1 s. With this change THD of ac current increases from 1.7% to 25%. But, when the proposed
In this paper, a novel control strategy has been introduced for grid connected CHB inverters. Using this control scheme, the converter can continue its operation under heavy unbalanced conditions. An analytic constraint has been derived to check the stable operation of the CHB inverter. In fact, in the unbalanced insolation conditions which lead to modulation
index of some cells become higher than unity, the proposed algorithm shifts the operating point of high power cell to bring all H-bridge cells in the linear (or stable) region. This algorithm prevents any interruption of the CHB inverter and keeps the THD of injected current below 5%. The validity and effectiveness of the proposed control method has also been confirmed by several simulations.
REFERENCES
[1] M. Meinhardt and G. Cramer, Multi-string converter: The next step in evolution of string converter technology, in Proc. 9th Eur. Conf. Power Electron. Appl 2001. S. Khajehoddin, A. Bakhshai, and P. Jain, A novel topology and control strategy for maximum power point trackers and multi string grid connected PV inverters, Applied Power Electronics Conference and Exposition, 2008. APEC 2008. Twenty-Third Annual IEEE. S. Daher, J. Schmid, and F. Antunes, Multilevel inverter topologies for stand-alone PV systems, IEEE Trans. Ind. Electron., vol. 55, no. 7, pp. 27032712, Jul. 2008. S. Busquets-Monge, J.Rocabert, P. Rodriguez, S. Alepuz, and J. Bordonau, Multilevel diode-clamped converter for photovoltaic generators with independent voltage control of each solar array, IEEE Trans. Ind. Electron. vol. 55, no. 7, pp. 27132723, Jul. 2008. Z. Wang, S. Fan,Y. Zheng and M. Cheng, Design and Analysis of a CHB Converter Based PV Battery Hybrid System for Better Electeromagnetic Compability, IEEE Transaction on Magnetics, vol. 48, no. 11, pp.4530- 4533, November 2012. S. Kouro, C. Fuentes, M. Perez and J. Rodriguez, Single DC-link Cascaded H-bridge Multilevel Multistring Photovoltaic Energy Conversion System with Inherent Balanced Operation, IECON 201238th Annual Conference on IEEE Industrial Electronics Society, pp. 4998- 5005, October 2012. S. Jain, V. Agarwal, Comparison of the performance of maximum power point tracking schemes applied to single-stage grid-connected photovoltaic systems, Electric Power Applications, IET, Page (s): 753 762, 2007. T. Esram, PL. Chapman, Comparison of Photovoltaic Array Maximum Power Point Tracking Techniques, IEEE Transactions on Energy Conversion, vol. 22, no. 2, June 2007. M. Mohapatra, BC. Babu. Fixed and Sinusoidal-Band Hysteresis Current Controller for PWM Voltage Source Inverter with LC Filter, Students' Technology Symposium (Tech Sym), IEEE 2010. G. Xiaoqiang, Z. Qingli, A New Current Feedback PR Control Strategy for Grid-Connected VSI with an LCL Filter, Power Electronics and Motion Control Conference 2006. IPEMC 2006. CES/IEEE 5th International. S. Khajehoddin, A. Bakhshai, and P. Jain, The application of the cascaded multilevel converters in grid connected photovoltaic systems, in Proc. IEEE EPC, Montreal, QC, Canada, Oct. 2007, pp. 296301. O. Alonso, P. Sanchis, E. Gubia, and L. Marroyo, Cascaded H-bridge multilevel converter for grid connected photovoltaic generators with independent maximum power point tracking of each solar array, in Proc. 34th Annu. IEEE PESC, Jun. 2003, vol. 2, pp. 731735. J. Rodriguez, J. S. Lai, F. Z. Peng, Multilevel Inverters: A Survey of Topologies, Controls, and Applications, IEEE Trans. On Industrial Electronics, Vol. 49, n 4, pp. 724-738, Aug. 2002. B. R. Lin and H. H. Lu, New multilevel rectier based on series connection of H-bridge cell, Proc. Inst. Elect. Eng. Electr. Power Appl, vol. 147, no. 4, pp. 304312, Jul. 2000. E. Villanueva, P. Correa, J. Rodriguez, and M. Pacas, Control of a single-phase cascaded h-bridge multilevel inverter for grid-connected photovoltaic systems, IEEE Transactions on Industrial Electronics, vol. 56, no. 11, pp. 43994406, Nov. 2009.
[16] MA. Rezaei, S. Farhangi, H. Iman-Eini, Enhancing the reliability of single-phase CHB-based grid-connected photovoltaic energy systems, power electronics, drive systems and technologies conference (PEDSTC), 2011 2nd. [17] S. Kouro, A. Moya, E. Villanueva, P. Correa, J. Rodriguez, Control of a cascaded H-bridge multilevel converter for grid connection of photovoltaic systems , Industrial Electronics, 2009. IECON '09. 35th Annual Conference of IEEE. [18] M. A. Rezaei, S. Farhangi, and H. Iman-Eini, Extending the operating range of cascaded h-bridge based multilevel rectier under unbalanced load conditions, in Proc. of PECON 2010, Kuala Lumpur, Malaysia, Nov. 2010, pp. 780785.
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
Harmonic Filtration by Current Injection and Shunt Capacitors Technique

Mohamed Y. Tarnini Department of Electrical Engineering Beirut Arab University Al-Debieh Lebanon m.tarnini@bau.edu.lb harmonic current and therefore to reduce the total harmonic distortion THD of the line current to meet IEEE standards and all international standards. Knowing that the overall harmonic filtration system is easy to implement and to functioning normally. The Zig-zag transformer and the boost are work together to shape the input current, near pure sinwave current, by injecting a third harmonic current at the ac side of the diode rectifier, knowing that the existence of three capacitors in the ac side acts as effective passive filter which reduces the THD more and more. In the paper a new three-phase rectification scheme is proposed to draw pure sinusoidal input currents. The proposed approach, based on the triplet harmonic current injection technique, incorporates one bridge inverters and one singlephase transformer. The actively shaped zero-sequence harmonic currents are then circulated through the ac side of the rectifier. This results in pure sinusoidal input current in the three-phase diode rectifier. The scheme has the following characteristics Pure sinusoidal input currents due to the proposed zero sequence or triplet harmonic current injection. No switching devices in series with the main power flow path. No increased dc link voltage small component rating of the additional control circuits
Boost circuit S
Abstract---- The controlled and uncontrolled converters or inverters loads are widely used in industrial applications and especially in producing electric energy from renewable sources. These types of nonlinear loads generate harmonics, with damaged effects, lead to voltage and current distortion at utility and load sides, in addition to high amount of losses in distribution system. The source currents of single phase and three-phase controlled or uncontrolled bridges contains high level amount of harmonics and consequently the total harmonic distortion is greater than the allowed values according to all international standards. This paper is proposed a new scheme based on triplet injection technique in addition to a passive filter connected to the ac utility side. This technique is based on Injecting the harmonic components from the DC side back to the line through the boost circuit and zigzag transformer in presence of line capacitors, as shown in Fig. 1 to control the injection of the harmonics, the final scheme will reduce the total harmonic distortion of the line current to an acceptable value that meet the requirements of international standards and especially IEEE standard.
Keywords----harmonic reduction, distortion (THD, current injection

I. INTRODUCTION
total
harmonic
The nonlinear loads insert large contents of harmonics into the power system distribution which cause heat losses, defect in waves nature and change of real power factor of the circuit at utility system and costumers side. There are many techniques to mitigate the harmonics in the power system harmonic correction methods. They may include active power filters, passive filter; boost converter filter, multi-pulse rectifier technique and harmonic current injection technique. We can select one or more technique according to cost, efficiency, performance, reliability, application, size and many other factors. The proposed injection method of Ametani [1] has low cost and high efficiency it is applicable to high power rectification. Our technique is based on a boost converter and a zigzag transformer as a current injection device as shown in fig.1 in addition to a small line capacitors act as passive filter at ac side of utility, all are used to circulate the injected 1
Non-linear load
O X
V0 Zig-zag Transformer Firing circuit N
Figure \ 1. Proposed circuit with complete filter
II.
HARMONIC CURRENT
A. Effect of harmonics currents on some electric equipments We can observe some important damaged effects on some specific important devices or equipment like as:
Computers: sensitive to threshold vo oltages of digital

circuit. Manufactures impose limits o on supply-voltage harmonics distortion. Television: distorted waveforms cause e fluctuations in TV picture size and brightness which leads to damage before the end of expected life time Converters: are sensitive to voltage so, misfiring angles for these converters. B. Simulation of proposed circuit without con ntroller To show the effect of the proposed technique we consider first the nonlinear load without any controller as sh hown in fig. 2
cos
ontains harmonic affects the It is clear that this output voltage co line current IA with respect to the load current IDC as follows:
IA =
1 1 2 3 I DC (cost + cos5 t + cos7t...) 5 7
(2)
Where 32
62 1
6,12,18
Figure 2. Nonlinear load across three phase ac c circuit
It is formed of 3 phase diode bridge and d pure resistance R=50. The simulation of the line current c can be shown in fig. 3. The THD of the line current IA is 27% which is very high and the main target in this paper to decrease this value below 5% to meet IEEE 519-1992 standard. And the Fourier analysis of the line current IA is shown in fig. 4. It is formed of the fundamental current beside the 5th and 7th.
Since the output voltage is periodic c with period one-six of the ac supply voltage, the harmonic in the output are of the order 6 1,2 2,3 . The harmonic currents may present problems in the ac system, filters are frequently necessary to o prevent these harmonics from entering ac system [6]. By cha anging the load resistor we can observe that this simulation n technique is still valid effectively up to up to R=300 or o the load current ranges from 2A up to around 70A. III. T INJECTION TECHNIQUE
IA (A) 10 5 0 -5 -10 0.1 0.12 0.14 0.16 Time (s) 0.18 0.2
A. Injection principle: The transformation of ac into dc cu urrent introduces harmonics due to use of power electronic devices in line current. Therefore, a harmonic reduction is needed to prevent the whole electric system losses and pr roblems or faults. Hence an improved current injection techniqu ue have been used depends on the 3rd harmonic current only since s the effect of the rest triplet components is very small compared with the 3rd harmonic component in the propo osed load. Third harmonic current injection can be achieved by using LC circuit tuned around 3rd harmonic frequency an nd zigzag transformer. The duty ratio, D of boost converter is used to control the amplitude of 3rd harmonic filter cu urrent, If [2]. A generalized analysis for the three-phase conv verter with the following injection technique is proposed to be b used in case of rectifier or inverter. B. Proposed filter subcircuits: The circuit of Fig.1 is formed of ac power supply connected to nonlinear load and two sub-circuits s one of them represent the connection of Zig-zag transformer r as shown in Fig. 5 and control-firing circuit of Fig. 6
Figure 3. Simulation of line current IA

IA (A) 12 8 4 0 100 200 300 Frequency (Hz) 400
Figure 4. Fourier analysis of IA
C. Mathematical equations and harmonics co omponents: The uncontrolled load voltage output Vu(t) c can be expressed as: 2
C. Mathematical eqution of filter cu urrent IF3: The voltages V0 and VSX are the vo oltages that have been used to circulate the third harmonic cur rrent through the injection path where V0 represents the rms value of (3k) harmonic of voltage between points O and N where w k=1,2,3.. By using
Fourier analysis we can write V0 for odd valu ues of k from the following equation: 3 5 9 1
exactly 150Hz and the maximum value is proportion to the DC link current I0. IV. SIMULATION WORK FOR PROPOSED TECHNIQUE The simulation analysis by PSIM M program has been carried out to reduce total harmonic dis stortion of the utility line current of three-phase Diode Bridge e with boost rectifier, three line to line capacitors and one to on ne Zig-zag transformer. The harmonic components of the cu urrent are then circulated through the ac side of the bridge via a zigzag transformer, rrents as shown in fig. 7. A resulting in pure sinusoidal line cur design example with variable load power, up to 5KVA in addition to three line capacitors of 100F connected across ac power supply of 220v line-voltag ge and 50Hz frequency as shown in Fiq.1 The simulation resu ults of the proposed system in Fig.7 and Fig.8 shows the valid dity of the injection current technique at a specific load resistance R=50 and two DC capacitors of 1300F of each. The Fourier analysis of IA in n Fig. 8 shows that the harmonics are diminished to a lo ow values and gives THD around 3.2% which meet IEEE requirement. r Note that the average voltage across DC link capacitors is 560v as shown in Fig. 10 which means that the poten ntial difference across each capacitor is around 280V. These e capacitors are relatively small, then they have low price in addition to a low price of one to one Zig-zag transformer. Th his means that the proposed technique is very simple and has a low price compared by other techniques. In addition the rating of lin ne capacitors, of 100F capacitance, across the ac lines is al lso small and the rating line voltage is around 220V. Fig. 9 show ws the variation of the total filter current IF, with frequency f= =150Hz, represents the 3rd harmonic that must injected int to the lines to perform sinusoidal lien currents, knowing th hat injected amount on each line is IF divided by three. Fig. 11 shows the variation of the transformer voltage V0 versus time, , it is clear that this voltage is a pulsating voltage to control th he injection current IF. The simulation result verified the propos sed (6) of V0. I(A) 40
20 0
Figure 5. Zig-zag transformer connection
Figure 6. Firing circuit sub-circuit
Tuning the harmonic injection path around triplet frequency will circulate the third harmonic current freely. So, the following analysis is focused on the third harmonic component at k=1 then (1) tends to: 3 8 6
The angle between the third harmonic compo onent of injection current, IF3 and supply current of phase 'A'' with respect to frequency f3=150 Hz, IA3 is 180 . [3][4]. T The amplitude of injection current for minimum THD is a fun nction of dc-link current, I0 and it is assumed to be MI0, where M is a proportional constant. Then the equation of injected current, IF3, where IF3 represents the IF through the tran nsformer divided by 3 multiplied by negative sign, can be writte en as: 3 sin 3 7
n, the current of We consider VA is a reference voltage, then phase A can be determined as: 8 through the line The current IL represents the line current t reactor. Fig. 9 shows that the frequency of th he filter current is
0.2
0.22
0.24
0.2 26 Time(s)
0.28
0.3
Figure 7. Simulation of line current IA
40
I(A)
V.
EXPERIMENTAL WORK
The experimental work on a 1kVA laboratory prototype has been constructed in the laboratory and the results are discussed in this section. Fig.12 shows the laboratory connection of the overall circuit.
20
100
200
300
400
Frequency (Hz)
Figure 8. Fourier analysis of IA

IA IF (A) 40 20
Filter current IF
Load current IA
Figure 12. 1 KVA prototype circuit

-20 -40 0.2 0.22 0.24 0.26 Time (s) 0.28
Figure 9. Filter current IF3 injected on line A

600 V(DC) V
Fig. 13 shows the variation of line current IA without fitter which is similar to that of Fig. 3 made by PSIM simulation program, with the same THD of 27%. Fig. 14 shows the experimental current IA of the line a, with THD around 3.7% which is more than that of the simulation result of Fig. 7, 8 due to many experimental factors and losses. Fig.15 shows the Fourier analysis of residual harmonics on line a, it is clear that all harmonics diminished and the line current behaves as pure sinwave. But in general it is very small compared with that of the uncontrolled one.
IA (A)
400
200 0.2
0.22
0.24
0.26 Time (s)
0.28
0.1 0.12
t(s)
Figure 10. DC link voltage

V0 (V) 1k 0.5k 0 -0.5k 1k 0.28 0.29 Time (s) 0.3
t(s)
Figure 13. Experimental IA without filtration

IA (A)
Figure 11. Transformer voltage V0 Figure 14. Experimental IA with filtration
Figure 15. Fourier analysis of experimental IA
This technique is still valid if we change the load or if we decrease the load. Fig. 16 shows the variation of the current IA at R=100. It shows that the line current is approximately pure sin wave and the total harmonic distortion is 3.7%. We repeat the above experiment at low voltages for protection purposes, we found that this technique is still valid and correct the line current effectively as shown in fig. 17 before any correction and fig. 18 after correction by Zig-zag transformer beside filtered capacitors and boost circuit when the input voltage is 25v.
IA (A)
Fig. 18 Filtered line currents at 25v input voltage measured across 1 ohm power resistor
VI.
CONCLUSION
t(s)
Figure 16. Line current IA at R=100 ohms.
Harmonic injection technique plays the main rule in reducing the THD of the line current when we use converters or any electronic device. This paper presented an injection current scheme by using a special connection of transformer (Zig-zag) in addition to a three line capacitors with small rating, for reducing THD in the line currents and hence to reduce the neutral current in distribution system. The scheme is very effective reducing the THD from 27% to 3.5%. The THD of the line current from simulation results and experimental results proves the mathematical equations and results of this technique. The experimental results shows the validity of this technique for variable load current As shown in fig. 16 at new resistance of 100 ohms, and at different input voltages as shown in fig. 17,18. Finally we conclude that this technique is valid under different load conditions and different input voltages which reduce harmonics to values meet with IEEE 519-1992 standard. VII. REFERENCES
[1] A. Ametani, Harmonic reduction in thyristor converters by harmonic Current injection, IEEE Trans. Power App. Syst., vol. PAS-95, no. 2, pp. 441450, Mar./Apr. 1976. [2] K. S. Chandragupta Mauryan, K. Govindaraj, C. Kayalvizhi, B. Deepa, V. S. Divyaa, IOSR Journal of Electrical and Electronics Engineering (IOSR-JEEE) e-ISSN: 2278-1676 Volume 5, Issue 2 (Mar. - Apr. 2013), PP 01-06 [3] A.M. El-Tamaly, P. N. Enjecti, H.H. El-Tamaly, An improved approach to reduce harmonics in the utility interface of wind, photovoltaic and fuel cell power systems, Proceedings of IEEE Applied Power Electronics Conference and Exposition, APEC 2000, vol.2, 2000, pp. 1059-1065. [4] Ali M. Eltamaly, Harmonics reduction techniques in renewable energy interfacing converters, Renewable Energy, Intechweb, December 2009, ISBN 978-953-7619-52-7 (Renewable Energy). [5] B. Singh, B. N. Singh, A. Chandra, K. Al-Haddad, A. Pandey, and D. P.Kothari, A review of three-phase improved power quality ac-dc converters,IEEE Trans. Ind. Electron., vol. 51, no. 3, pp. 641 660, Jun. 2004 [6] Daniel Hart power electronics international edition 2011
Fig. 17 Unfiltered line current at 25v input voltage
Optimal PID Tuning for Load Frequency Control Using Lvy-Flight Firefly Algorithm
Mahmoud Reza Shakarami Electrical Engineering Dept, Faculty of Engineering Lorestan University Lorestan, Iran m.r.shakarami@gmail.com Iraj Faraji Electrical Engineering Dept, Faculty of Engineering Lorestan University Lorestan, Iran Faraji.iraj@ieee.org Iraj Asghari Clean & Renewable Energies Dept, Institute of Natural and Applied Science Hacettepe University, Ankara, Turkey Asghari.iraj@gmail.com
Mohsen Akbari
Electrical Engineering Dept, Faculty of Engineering K.N. Toosi University of Technology, Tehran, Iran
mohsenakbari@ee.kntu.ac.ir
AbstractNature-inspired algorithms are among the most powerful algorithms for optimization. In this paper, a load frequency control (LFC) in two-area power systems using the lvy-flight firefly optimization algorithm (LFOA) is presented. The system simulation is realized by using MatLab/Simulink. The proposed LFOA-based PID controller has been compared with the firefly optimization algorithm (FOA)-based and particle swarm optimization (PSO) algorithm-based PID controllers. Simulations and results indicate that the proposed LFOA is superior to other ones for the LFC in two-area power systems. Keywords- Firefly optimization algorithm; lvy-flight firefly optimization algorithm; particle swarm optimization; load frequency control; two-area power system
I.
INTRODUCTION
This Large scale power systems are normally composed of control areas or regions representing coherent groups of generators. Various areas are interconnected through tie lines. The tie lines are utilized for contractual energy exchange between areas and provide inter-area support in case of abnormal conditions. Area load changes and abnormal conditions lead to mismatches in frequency and scheduled power interchanges between areas. These mismatches have to be corrected by load frequency control (LFC), which is defined as the regulation of the power output of generators within a prescribed area [1]. The regulation is performed so as to maintain the scheduled system frequency and/or established interchange with other areas within predetermined limits in response to changes in system frequency and tie line loading [2]. Several strategies for the LFC of power systems have been proposed by researchers over the past decades. A robust decentralized power system LFC controller design approach using structure singular value has been designed in [3]. Two
robust decentralized LFC controllers are introduced in [4]. The first one is based on H theory, and results in a high order controller. The second controller is a PI controller tuned by genetic algorithm (GA) to achieve the same robust performance as the first one. A decentralized adaptive control scheme for LFC of multi-area power systems to deal with variations of system parameters is introduced in [5]. An approach based on the tabu search (TS) algorithm for optimal design of a fuzzy logic based Proportional Integral (FLPI) LFC in a two-area interconnected power system is presented in [6]. The PI and I control parameters are tuned based on hybrid particle swarm optimization (HPSO) algorithm method for LFC control in a two-area power system in [7]. PSO based multi-stage fuzzy controller is proposed for solution of LFC problem in power system in [8]. Designing of PID controller for LFC in interconnected power system using PSO has been discussed in [9]. Hybrid neuro-fuzzy (HNF) approach is employed in [10] for an automatic generation control (AGC) of interconnected power system with and without generation rate constraint (GRC). Application of real coded GA for optimizing the gains of a PI controller for two-area thermal reheat power system has been discussed in [11]. Fuzzy logic controller is designed for automatic LFC of two-area interconnected power system in [12]. Nature-inspired metaheuristic algorithms are becoming powerful in solving modern global optimization problems [1314]. For example, PSO was developed by Kennedy and Eberhart in 1995 [13], based on the swarm behavior such as fish and bird schooling in nature. It has now been applied to find solutions for many optimization applications. Another example is the firefly optimization algorithm (FOA) developed by Yang [14] which has demonstrated promising superiority over many other algorithms.
The search strategies in these multi-agent algorithms are controlled randomization, efficient local search and selection of the best solutions. However, the randomization typically uses uniform distribution or Gaussian distribution. On the other hand, various studies have shown that flight behavior of many animals and insects has demonstrated the typical characteristics of lvy flights [15]. A recent study by Reynolds and Frye shows that fruit flies or Drosophila melanogaster, explore their landscape using a series of straight flight paths punctuated by a sudden 90 turn, leading to a lvyflight-style intermittent scale free search pattern. Even light can be related to lvy flights [16]. Subsequently, such behavior has been applied to optimization and optimal search, and preliminary results show its promising capability [15]. This paper aims to implement a lvy-flight firefly optimization algorithm (LFOA) and to provide the comparison study of the LFOA with FOA and PSO algorithms in LFC of two-area power systems. We will first outline the firefly algorithms, then formulate the LFOA and finally give the comparison about the performance of these algorithms. The LFOA optimization seems more promising than PSO and FOA in the sense that LFOA has a less fitness function value and converges quickly and deals with global optimization more naturally. In addition, PSO is just a special class of the LFOA as it is demonstrated in [17]. II. FIREFLY ALGORITHM
A. Behavior of Fireflies The flashing light of fireflies is an amazing sight in the summer sky in the tropical and temperate regions. There are about two thousand firefly species, and most fireflies produce short and rhythmic flashes. The pattern of flashes is often unique for a particular species. The flashing light is produced by a process of bioluminescence, and the true functions of such signaling systems are still debating. However, two fundamental functions of such flashes are to attract mating partners (communication), and to attract potential prey. In addition, flashing may also serve as a protective warning mechanism. The rhythmic flash, the rate of flashing and the amount of time form part of the signal system that brings both sexes together. Females respond to a males unique pattern of flashing in the same species, while in some species such as photuris, female fireflies can mimic the mating flashing pattern of other species so as to lure and eat the male fireflies who may mistake the flashes as a potential suitable mate. The flashing light can be formulated in such a way that it is associated with the objective function to be optimized, which makes it possible to formulate new optimization algorithms. B. Firefly Algorithm Now we can idealize some of the flashing characteristics of fireflies so as to develop firefly-inspired algorithms. For simplicity in describing our FOA, we now use the following three idealized rules: 1) all fireflies are unisex so that one firefly will be attracted to other fireflies regardless of their sex; 2) Attractiveness is proportional to their brightness, thus for any two flashing fireflies, the less brighter one will move
towards the brighter one. The attractiveness is proportional to the brightness and they both decrease as their distance increases. If there is no brighter one than a particular firefly, it will move randomly; 3) The brightness of a firefly is affected or determined by the landscape of the objective function. For a maximization problem, the brightness can simply be proportional to the value of the objective function. Other forms of brightness can be defined in a similar way to the fitness function in genetic algorithms or the bacterial foraging optimization algorithm (BFOA) [18]. In the FOA, there are two important issues: the variation of light intensity and formulation of the attractiveness. For simplicity, we can always assume that the attractiveness of a firefly is determined by its brightness which in turn is associated with the encoded objective function. In the simplest case for maximum optimization problems, the brightness I of a firefly at a particular location x can be chosen as I(x) f(x). However, the attractiveness is relative, it should be seen in the eyes of the beholder or judged by the other fireflies. Thus, it will vary with the distance rij between firefly i and firefly j. In addition, light intensity decreases with the distance from its source, and light is also absorbed in the media, so we should allow the attractiveness to vary with the degree of absorption. In the simplest form, the light intensity I(r) varies according to the inverse square law I(r) = Is/r2. where Is is the intensity at the source. For a given medium with a fixed light absorption coefficient , the light intensity I varies with the distance r. That is I = I e r (1)
0
where I0 is the original light intensity. As a fireflys attractiveness is proportional to the light intensity seen by adjacent fireflies, we can now define the attractiveness of a firefly by
= 0 e r
(2)
C. Lvy-Flight Firefly Algorithm If we combine the three idealized rules with the characteristics of lvy flights, we can formulate a new LFOA which can be summarized as the pseudo code shown in [17]. In the implementation, the actual form of attractiveness function (r) can be any monotonically decreasing functions such as the following generalized form
(r ) = 0 e r
m ( m 1)
(3)
For a fixed , the characteristic length becomes = -1/m 1 as m. Conversely, for a given length scale in an optimization problem, the parameter can be used as a typical initial value. That is =1/m. The distance between any two fireflies i and j at xi and xj, respectively, is the Cartesian distance
rij = x i x j =
(xi,k x j,k )2 k =1
d
(4)
where xi,k is the kth component of the spatial coordinate xi of ith firefly. The movement of a firefly i is attracted to another more attractive (brighter) firefly j is determined by
x i = xi + 0 e
r 2
ij
1 x j x i + sign rand Lvy 2
IV. (5)
SIMULATIONS AND RESULTS
where the second term is due to the attraction while the third term is randomization via lvy flights with being the randomization parameter. The product means entry wise multiplications. The sign[rand1 ] where rand [0,1] 2
essentially provides a random sign or direction while the random step length is drawn from a lvy distribution
Lvy u = t - (1 < 3)
(6)
The study system is simulated in MatLab/Simulink. Each of three algorithms (PSO, FOA and LFOA) is implemented to optimize the PID controller parameters of two areas. PID controllers are similar and simultaneously having the same parameters along with optimization process. The fitness function for the search algorithms is chosen as the integral of the time multiplied absolute value of the error (ITAE) index. ITAE penalizes long duration transients, and it is much more selective than other indices. A system designed using this criterion exhibits small overshoots and well damped oscillations. ITAE is defined
which has an infinite variance with an infinite mean. Here the steps of firefly motion is essentially a random walk process with a power-law step-length distribution with a heavy tail. In this study, we took 0 = 1, = 0.7, = 1, and =1.5. The parameter now characterizes the variation of the attractiveness, and its value is crucially important in determining the speed of the convergence and how the FOA algorithm behaves. III. TWO-AREA POWER SYSTEM
f ITAE =
t e (t ) d t
0
(7)
A two-area interconnected power system consists of two single areas connected through a power line called the tie line. Each area feeds its user pool, and the tie line allows electric power to flow between areas. Since two areas are tied together, a load perturbation in one area affects the output frequencies of all two areas as well as the power flow on the tie line. The control system of each area needs information about the transient situation of two areas in order to bring the local frequency back to its steady state value. Information about the other areas is found in the output frequency fluctuation and in the tie line power fluctuation of that area. Therefore, the tie line power is sensed, and the resulting signal is fed back into related areas. A block diagram related to the two-area interconnected power system is given in Fig. 1. The two-area power system parameters are available in appendix.
where, e(t) is defined as the sum of ACE1 and ACE2 shown in Fig. 1, and t is also the simulation time. The system is stable, and the control task is to minimize the system frequency deviation F1 in area1, F2 in area2 and the deviation in the tie line power flow Ptie between the two areas under the load disturbances PD1 and PD2 in the two areas. Since the system parameters for the two areas are identical and Ptie is caused by (F1 - F2) the system performance can be mainly tested by applying a disturbance PD1 to the system and observing the time response of F1 [19]. System frequency deviation F1 in area1 under 0%, 15% and 30% disturbance PD1 is demonstrated in Figs. 2, 3 and 4, respectively, using the optimization algorithms. Since the power flow oscillation of the tie line is exactly zero under PID controllers optimized by three aforementioned algorithms, so it is not demonstrated in this paper. Furthermore, ITAE value is calculated to compare the performance of these three optimization algorithms tabulated in Table 1. The obtained results show better performance of LFOA than the other ones.
Figure 2. The frequency variation of two areas when PID controllers are optimized by PSO, FOA and LFOA under 0% load disturbance PD1 Figure 1. Block diagram of two-area power system
APPENDIX Two-area power system parameters: TT1=TT2=0.3s, TG1=TG2=0.08s, TP1=TP2=20s, T12=0.0866 p.u., B1=B2=0.425 p.u. MW/Hz R1=R2=2.4 Hz/p.u. MW, KP1=KP2=120 Hz/p.u. MW. REFERENCES
[1] [2] N. Jaleeli, et al., Understanding automatic generation control, IEEE Trans. Power Syst., vol. 7, no. 3, pp. 11061122, August 1992. E. Yesil, M. Guuzelkaya, and I. Eksin, Self tuning fuzzy PID type load and frequency controller, Int. J. Energy Conversion and Manag., vol. 45, no. 3, pp. 377390, 2004. T. C. Yang, H. Cimen, and Q. M. Zhu, Decentralized load frequency controller design based on structured singular values, IEE Proc. Gener., Transm., Distrib., vol. 145, no.1, pp. 7-14, January 1998. D. Rerkpreedapong, A. Hasanovic, and A. Feliachi, Robust load frequency control using genetic algorithms and linear matrix inequalities, IEEE Trans. Power Syst., vol. 18, no. 2, pp. 855-861, May 2003. M. Zribi, M. Al-Rashed, and M. Alrifai, Adaptive decentralized load frequency control of multi area power systems, Int. J. Elect. Power and Energy Syst. Eng., vol. 27, pp. 575- 583, 2005. S. Pothiya, et al., Design of optimal fuzzy logic based PI controller using multiple tabu search algorithm for load frequency control, Int. J. Control, Automation, and Syst., vol. 4, no.2, pp. 155-164, April 2006. S. Taher, R. Hematti, A. Abdolalipour, and S. H. Tabei, Optimal decentralized load frequency control using HPSO algorithms in deregulated power systems, Amer. J. Appl. Sci., vol. 5, no. 9, pp 11671174, 2008. H. Shayeghi, A. Jalili, and H. A. Shayanfar, Multi stage fuzzy load frequency control using PSO, Int. J. Energy Conversion and Manage., vol. 49, 2008, pp. 2570-2580. K. Sabahi, A. Sharifi, M. Aliyari, M. Teshnehlab, and M. Aliasghary, Load frequency controller in interconnected power system using multiobjective PID controller, J. Appl. Sci., vol. 8, no. 20, pp. 3676-3682, 2008. G. Panda, S. Panda, and C. Ardil, Automatic generation control of interconnected power system with generation rate constraints by hybrid neuro fuzzy approach, Int. J. Elect. Power and Energy Syst. Eng., vol. 2, no. 1, pp. 13-18, 2009. S. Ramesh, and A. Krishnan, Modified genetic algorithm based load frequency controller for interconnected power system, Int. J. Elect. and Power Eng., vol. 3, no. 1, pp. 26-30, 2009. P. Aravindan, and M. Y. Sanavullah, Fuzzy logic based automatic frequency control of two area power system with GRC, Int. J. Computational Intell. Research, vol. 5, no. 1, pp. 37-44, 2009. J. Kennedy, R. Eberhart, and Y. Shi, Swarm Intelligence, Academic Press, 2001. X. S. Yang, Nature-Inspired Metaheuristic Algorithms, Luniver Press, 2008. I. Pavlyukevich, Lvy flights, non-local search and simulated annealing, J. Computational Physics, vol. 226, no. 2, pp. 1830-1844, October 2007. P. Barthelemy, J. Bertolotti, and D. S. Wiersma, A Lvy flight for light, Nature 453, pp. 495-498, May 2008. X.-S. Yang, Firefly algorithm, lvy flights and global optimization, in: Research and Development in Intelligent Systems XXVI (Eds M. Bramer, R. Ellis, M. Petridis), Springer London, pp. 209-218 (2010). K. Gazi, and K. M. Passino, Stability analysis of social foraging swarms, IEEE Trans. Syst., Man., Cybern. B, Cybern., vol. 34, no. 1, pp. 539-557, Feb. 2004. T. C. Yang, H. Cimen, and Q. M. Zhu, Decentralised load-frequency controller design based on structured singular values, IEE Proc Gener., Transm., Distrib., vol. 145, no. 1, pp. 714, Jan. 1998.
[3] Figure 3. The frequency variation of two areas when PID controllers are optimized by PSO, FOA and LFOA under 15% load disturbance PD1 [4]
[5]
[6]
[7]
[8]
[9] Figure 4. The frequency variation of two areas when PID controllers are optimized by PSO, FOA and LFOA under 30% load disturbance PD1 [10] TABLE I. ITAE VALUE FOR OPTIMIZATION ALGORITHMS UNDER DIFFERENT LOAD DISTURBANCES PD1 0% load disturbance 0.8693 0.0620 0.0349 15% load disturbance 0.0499 0.0438 0.0404 30% load disturbance 0.0560 0.0544 0.0530 [11]
Optimization Method PSO FOA LFOA
[12]
[13]
V.
CONCLUSION
[14] [15]
Nature-inspired metaheuristic algorithms are becoming powerful in solving modern global optimization problems. Therefore, in this paper, PID controllers are optimized by PSO, FOA and LFOA for the LFC of two-area interconnected power systems. It is seen LFOA-optimized controllers have the better performances than two other algorithms-optimized controllers for the LFC in two-area power systems.
[16] [17]
[18]
[19]
Command Filtered Adaptive Backstepping Control of Linear Servo

Jacek Kabziski
Institute of Automatic Control Lodz University of Technology Lodz, Poland jacek.kabzinski@p.lodz.pl
AbstractThe paper develops new control methods for linear servo with significant friction with Stribeck-type nonlinear characteristics. Different adaptive backstepping approaches to control tubular permanent magnet motor working with an inverter controlling the motor current are compared. Keywords- adaptive control, linear drives, friction compensation
be approximated by linear combination of several previous command values. Using this approach we have to assume that the difference between the exact derivative and its approximation is bounded. The other technique to avoid commend derivative calculation was introduced in [6,7] and is called command filtered adaptive backstepping (CFAB). This approach uses filtering methods to produce certain command signals and their derivatives which eliminates the requirement of analytic differentiation. The method is conceptually attractive and allows rigorous derivation of Lyapunov stability conditions. Several implementations of CFAB were reported recently [8,9]. The purpose of this contribution is to compare performance of different approaches to command derivative calculation in adaptive backstepping applied for control of linear servo with permanent magnet tubular motor. We assume that all parameters are unknown and that the main load of the motor is friction with Stribeck-type nonlinear characteristics. In section II we formulate the control problem and drive adaptive backstepping control with linearly parameterized models of plant nonlinearities. Next we derive and discuss two methods to avoid analytical differentiation of a current command. The first approach is based on modelling the derivative, the second on CFAB approach with first order filter. In section IV we present an example of implementation and compare performance of both approaches. Finally we address some practical problems like motor current saturation, and we come to conclusions in the last section. II. ADAPTIVE BACKSTEPPING CONTROL OF LINEAR SERVO We consider a tubular linear motor with permanent magnets build in the inner part, longer then the outer part containing the armature (so called construction type). The motor thrust force is usually modelled as proportional to the armature current , but in several constructions the force coefficient is position dependent:
I.
INTRODUCTION
Permanent magnet linear motors are becoming more and more popular in many industrial applications, especially those requiring high precision position tracking of linear motion. The main difficulties with such control systems are caused by several nonlinear and time varying disturbances, like friction components, force ripples etc. Also almost all linear motor model parameters are subject to serious changes during system operation. Although adaptive backstepping control [1] has proven to be useful and effective tool to obtain fine control performance in many mechatronic systems including electric drives [2], it is not free from several difficulties. The first weakness of adaptive backstepping approach is the necessity of liner parameterization by unknown parameters. This assumption is obviously difficult to fulfill, for example if we consider nonlinear phenomena as friction what is very important in linear servo modeling. The second limitation is the fact that implementation of adaptive backstepping controllers requires analytic calculation of the partial derivatives of certain stabilizing functions called virtual control commands. If the system order increases analytical derivation of these derivatives becomes more and more tedious, obtained formulas are more and more complicated making the whole system inapplicable in practice. Even for moderate order of nonlinear drive model this problem is serious [2]. The stability and performance of adaptive backstepping controlled nonlinear system is guaranteed by rigorous Lyapunov analysis, that requires both assumptions mentioned above. To cope with the problem of linear parameterization we have proposed [3,4] the idea of approximation of plant nonlinearities by linearly parameterized model, with artificial parameters. These models may be obtained by fuzzy [5] or neural approach, or by any other form of approximation. To a certain extend this modelling approach may also cover the problem of virtual command derivative calculation, as it may
The most important part of a linear servo load is friction. Friction forces are relatively bigger then in rotational motors. Eccentricity increases the friction in linear bearings, causing
losses and reducing the effective force. The presence of friction may cause severe deterioration of tracking performance by being the reason for oscillations and steady state errors. The nature of friction is complicated and difficult in identification and modeling. The data obtained for friction modeling are usually corrupted by noise and outliers, and even if we decide to use sophisticated model the model accuracy is limited. That is why we consider a very simple servo model
Current i will be the virtual control for the speed tracking error, and as it is obvious from (5) that to stabilize the desired current trajectory will be:
where v is the forcer velocity, m the forcer mass, represents the coefficient converting the motor current i into the thrust force, and Ff is an external load mainly a friction force, bat also external load and ripple force. The motor current i is supplied by a PWM inverter working in a current control mode and we assume the this loop dynamics may be described by a first order system with time constant T, gain G and input :
with positive design parameter Kv. As the current is not an available input we have to consider current tracking error ei ir i . Making use of (6-10) we may rewrite (5) as with bounded
Using (3) we are able to describe current error dynamics by Assuming that we know for a function with 2 parameters and good model approximating and respectively we may construct a model
We assume that the parameters m, , T, G are unknown, but constant or slowly varying. The control objective is that the motor speed has to follow a smooth reference vref. We denote the tracking error by
We apply adaptive backstepping scheme [1], so first we consider the tracking error dynamics:
We assume that we are able to construct linear in parameters models for functions
. As previously we assume existence of a
with a bounded modelling error and we denote the error between good and actual adaptive parameters by: Selecting the control input as
respectively. and are adaptive parameters and , are known functions of measurable variables. The idea of the first construction may be based on fuzzy friction approximation [5], polynomial friction curve approximation, or Maxwell slip model [10]. The second model is based on assumed approximation of force coefficient ki as a function of position or temperature, or simply if we have no additional knowledge about ki. We assume that there exists good choice of parameters , generating models
where Ki is a positive design parameter, we get
We verify the system stability by Lyapunov function
with a bounded modelling error. We denote the error between good and actual adaptive parameters by:
with positive definite matrices * , * fv, fi , gv . Calculation of Lyapunov function system derivative gives
compensate changes of T and G, but also differences between accurate and numerical derivative. Of course many more sophisticated approximations of the derivative may be used. B. Virtual Control Command Filtering The assumption that function (15) may be represented by a linearly parameterized model with measurable input is more heuristic than rigorous. Below we present another approach based on command filtering adaptive backstepping concept [6,7], that allows rigorous stability analysis. There are several possibilities to conduct the derivation of stability conditions using command filtering approach. First [11] or second order filter [6] may be used, an extra state variable may be introduced to represent the error due to imperfect command differentiation and different type of adaptive laws may be applied. Below we present one of many possible derivations. Let us introduce a linear filter with state variable positive parameter and
As good parameters are constant we have
and therefore under adaptive laws: we get
When a transient response of the filter disappears we get Let us denote It follows from (31) that so from (3) and (29)
so the system (5), (13) is uniformly ultimately bounded (UUB) and we may tighten the bound for ev ,ei by increasing Kv and Ki. Similar result may be obtained by using a little more complicated, but safer adaptive laws:
III.
VIRTUAL CONTROL COMMAND DERIVATIVE , i.e.
Constant, but unknown parameters in (33) are
A. Virtual Control Command Derivative Modeling The derivation presented above requires calculation of we have to obtain
We introduce adaptive estimators of these parameters and denote adaptation errors by Let us consider Lyapunov function
Although all signals under the derivative operator are taken from the control algorithm, analytical differentiation is complicated. Instead we may assume that we are able to build linearly parameterized model (14) for function (15), but with different inputs than proposed before. The simplest approach is to construct the regressor function as where interval
states for change of command signal in time . In this case adaptive parameters will not only
with design parameters: positive definite matrices
and
leading to
positive scalars , . The unknown, constant and are always positive. Using (11) and (33) we are able to calculate derivative of V along trajectories of and :
and also giving uniformly ultimately bounded errors ev and eif. Summing up: the controller is described by:
and to transform it into equivalent form
and adaptation laws are:
Now, adaptive laws may be used to cancel expressions in square brackets:
Adaptation laws (48) describe a response of stable linear system to a bounded input, so adaptive parameters are bounded. Modification of adaptation laws to
To cope with the last component of (38) we will design such control law , that will give similar results. We are able to modify the tracking errors by proper choice of design parameters. Error bound may be decreased by increasing Kv and Ki . Adaptation speed is influenced mostly by , . We have to stress that the number of adaptive parameters in models (6) is not settled yet. If models are more accurate the tracking error bound will be smaller and the control effort will be lower. Increasing the number of adaptive parameters results usually in improvement of model accuracy, but each parameter means one more differential adaptive law to be solved on-line, so we should compromise between these criteria. Sometimes, if fast adaptation is possible just one adaptive parameter in a model is enough [4].
where will be defined soon. Under control (40) the system derivative of Lyapunov function is Assuming that choose and in this case so is negative outside the set defined as is sufficiently big, say
we may
IV.
COMPARISON OF CONTROL STRATEGIES
According to the well known generalization of Lyapunov theorem due to LaSalle and Lefschetz [1] the tracking errors ev and eif are uniformly ultimately bounded. For small we may apply
Command derivative modeling and command filtering approaches were compared by numerous experiments with a linear permanent magnet motor with the motor model parameters m=7.04 kg, ki=39 N/A, T=0.445 ms, G=2. friction was modeled by Stribeck curve
( ), Layapunov function parameters ( ) and filter time constants ( ms) were similar in both cases.
with parameters fc = 5 N, fs - fc =1 N, = 0.15 m/s, B = 3 Ns/m, = 2. The curve was identified from the measurement data and modeled by a 5-th order polynomial. So model (6) is constructed with 7 parameters : the first parameter corresponds to the motor mass m and consecutive 6 are coefficients of the approximating polynomial. Vector consists of speed reference derivative and 6 powers of speed. Model is just a single parameter. The initial values of parameters were disturbed by 20% error. We have tested control algorithms with sinusoidal speed reference with multiple speed reversal making the servo to work against the jump in the friction curve at zero velocity. The identified friction force and approximating polynomial are plotted in fig. 1.
Figure 2. Speed tracking errror for modelling (solid) and CFAB (dotted) approach
Figure 1. Friction curve (solid) and approximating polynomial (dotted)
In the derivative modeling approach was approximated by the output of discrete filter with transfer function
Figure 3. Current tracking errror ei=ir-i for modelling (solid) and CFAB (dotted) approach
where is filer time constant and is sampling time. The filter output was used to construct model (14). The control law (18) together with adaptation laws (26) complete the control algorithm in this case. Using CFAB approach we apply equations (47) (48). Several practical aspects of adaptive control have to be considered. We imposed a hard bound on the motor current. Adaptive parameters may be bounded by introducing projection operator into adaptation laws. We decided to block adaptation when the current reference saturation is active. In fig. 2-5 we compare tracking performance of two presented approaches. Control loop parameters, i.e. control gains Kv, Ki,
Both approaches generate satisfactory results. The speed tracking error converges to reasonable limits. The motor current is limited. Rather surprisingly CFAB approach tends to produce higher jumps of current and the bigger speed tracking error in this case results from active current saturation. Adaptive parameters during all experiments remained bounded as it is visible in fig. 6.
the errors are uniformly ultimately bounded. We have compared both approaches in a series of experiments, and have demonstrated that the system behavior was satisfactory in both cases, although for the tested motor modeling approach has resulted in smaller tracking errors and smoother current trajectory.
Figure 4. Integral of the square value of speed tracking errror for modelling (solid) and CFAB (dotted) approach, reseted each period of the speed reference.
Figure 6. Adaptive parameters
history, CFAB approach
REFERENCES
[1] M. Krstic, Kanellakopoulos I., Kokotovic P. V., Nonlinear and Adaptive Control Design, New York: Wiley, 1995. [2] D. M. Dawson, Hu, J., Burg, T. C. (1998). Nonlinear control of electric machinery , CRC Press, 1998. [3] J. Kabziski, "Adaptive backstepping control of a completely unknown permanent magnet motor," Power Electronics and Applications, 2007 European Conference on , pp.1-10, 2007 [4] J. Kabzinski, "Adaptive control of Duffing oscillator with unknown input gain," Methods and Models in Automation and Robotics (MMAR), 2010 15th International Conference on , pp.234-239, 2010 [5] J. Kabzinski, Fuzzy Friction Modeling for Adaptive Control of Mechatronic Systems, Artificial Intelligence Applications and Innovations, IFIP Advances in Information and Communication Technology, , vol. 381, pp.185-195, Springer Berlin Heidelberg 2012 [6] J. A. Farrell, M. Polycarpou, M. Sharma, and W. Dong, Command filtered backstepping, IEEE Trans. Autom. Control, vol. 54, no. 6, pp. 13911395, Jun. 2009. [7] Wenjie Dong; Farrell, J.A.; Polycarpou, M.M.; Djapic, V.; Sharma, M., "Command Filtered Adaptive Backstepping," Control Systems Technology, IEEE Transactions on , vol.20, no.3, pp.566,580, 2012 [8] V. Djapic, Nad D., Command Filtered Backstepping Design in MOOSIvP Helm Framework for Trajectory Tracking of USVs, American Control Conference, pp. 5997-6003, 2010 [9] Xun Gong; Yue Bai; Cheng Peng; Changjun Zhao; Yantao Tian, "Trajectory tracking control of a quad-rotor UAV based on command filtered backstepping," Intelligent Control and Information Processing (ICICIP), 2012 Third International Conference on , pp.179,184, 2012 [10] M. Jastrzbski, Kabziski J., Adaptive Backstepping Control of Linear Motor Based on Maxwell Slip Model of Friction Proceedings of the 19th International Conference on Automation & Computing, Brunel University, Uxbridge, UK, 13-14 September 2013 (submitted) [11] Dong, W.; Farrell, J.A.; Polycarpou, M.M.; Sharma, M., "Command filtered adaptive backstepping," American Control Conference (ACC), 2010 , vol., no., pp.105,110, June 30 2010-July 2 2010.
Figure 5. Integral of the square value of current tracking errror for modelling (solid) and CFAB (dotted) approach, reseted each period of the speed reference.
V.
CONCLUSIONS
We have proposed new control techniques for linear servo with permanent magnet linear motor. Introducing linearly parameterized models of plant nonlinearities we were able to approximate friction and other nonlinearities appearing in a linear motor model and to use the models effectively in adaptive backstepping frame. The same technique may be used to avoid analytical differentiation of the motor current command. Based on the idea of command filtered adaptive backstepping we have derived control algorithm for linear servo using the simplest possible control scheme with the first order linear filter, and have proved the system stability i.e. that
Design and Simulation of a High Power Quality Regenerative PWM Rectifier System for 1 MW Electric Locomotives
Koray Sepetci and Bunyamin Tamyurek
Electrical and Electronics Engineering Department Eskisehir Osmangazi University Eskisehir, Turkey ksepetci@kemsan.com.tr, btamyurek@ogu.edu.tr
AbstractThis paper presents the design and simulation of a high power quality regenerative PWM rectifier system for electric locomotives rated at 1 MW. The demand for efficient use of energy sources and high power quality requirements mandate the use of PWM rectifiers when interfacing the traction converters to the railway power system. The paper first develops the simulation models of the converter power stage and the controller. Then, a design procedure, which determines the critical design parameters, is presented. In addition, a controller that achieves unity power factor and a sinusoidal current that has a total harmonic distortion (THD) less than 2% at the worst case without employing a passive filter is developed. The significant reduction in the switching frequency ripple and the harmonics that enables the filterless design is realized by the interleaved operation of the parallel-connected rectifier units. In conclusion, the concept and the proposed design are verified through simulation studies. Keywords-Electric locomotives; high power quality rectifiers; interleaving; regenerative PWM rectifiers; switchmode rectifiers
I. INTRODUCTION
In modern grid connected power electronic systems, high power quality and regenerative features have become the major and incontrovertible part of specifications in most designs. It is because of the significant benefits of these features to the overall system performance: the efficient use of energy sources, the lower cost, better waveform quality at the point of common coupling, managing and minimization of harmonics; and thus compliance to the governmental regulations [1]. The electric locomotive manufacturers demand these features to be included in their new productions [2]. In addition, the ability to manage the harmonics may allow handling or eliminating the undesired oscillations in the railway power system. The harmonic oscillations are signified especially when the catenary line inductance is varying depending on the distance of the locomotive with respect to the source central [3]. Consequently, the solution is to
use PWM controlled rectifiers since they can realize high power quality and regenerative features effectively [4]. The classical diode rectifiers only allow unidirectional power flow from ac source side to dc load side. Therefore, they are not suitable for the regenerative operation. Moreover, the power quality is very poor. The back-toback connected thyristor controlled rectifier system is one of the solutions at high power regenerative applications. However, it is not practical to operate thyristors at high PWM switching frequency. Similarly, the gate turn-off thyristor (GTO) controlled rectifiers cannot be operated at high PWM switching frequency, as well. Additionally, the gate control of these devices needs complex and costly gate drive circuits [5]. Nevertheless, advances in semiconductor power device technology have produced a hybrid transistor called insulated gate bipolar transistors (IGBT). Today, the IGBTs are available with voltage ratings up to 6.5 kV and the current ratings up to 2.4 kA [6]. They are fully controllable, requiring a very simple gate drive, and achieving switching speeds less than one micro-second. The IGBTs that can be used up to 4 kHz switching frequency efficiently at the Megawatt level are available. With the availability of such devices, realization of the high power quality regenerative PWM rectifiers at 1 MW and above rated electric locomotives has become practical. The ability to switch the IGBTs quickly also allows interleaved operation of several parallel connected rectifiers inside one switching period. Simply, the interleaving means operation of several converter units in parallel, but the switching function generated for each unit is equally phase shifted with respect to each other [7]-[8]. The major advantage of the interleaving technique is that the frequency of the ripple at the voltage or the current as well as the frequency of the harmonics created by switching is increased depending on the number of interleaved units. So, this means that the ripple (harmonics) can be easily filtered and the waveform quality is significantly improved. In addition, the ripple magnitude at the common coupling
point is reduced because of the equal phase shifting, which also contributes to better waveform quality and easy filtering. It will be shown later that this technique will significantly lessen the requirement for passive filtering in order to eliminate the harmonic currents at the source side. The following sections describe the topology of the converter, explain the control system architecture, define the design steps, and finally verify the concept and the proposed design through simulation studies.
II.
SYSTEM DESCRIPTION
This section describes the high power quality regenerative PWM rectifier system in detail. The Simulink block diagram of the system is shown in Fig. 1. The block named Regenerative PWM rectifier incorporates the power circuit of the rectifier units and the load. The content of this block will be described in the next paragraph. The catenary voltage, which is a singlephase ac grid, is represented by an ideal sinusoidal voltage source at 23 kV (nominal voltage), but it is expected to vary between 17 kV and 29 kV as shown in Fig. 1. The given range is likely to be the practical maximum range. The catenary voltage is stepped down to an appropriate level before the rectification.
a single-phase transformer with four secondary windings and used to step down the catenary voltage to a low voltage and provide isolated input for each rectifier unit. The inductors shown in Fig. 2 are used to shape the input currents and can be embedded into the main transformer as part of the leakage inductance. This technique provides easy cooling, smaller size, and lower cost for the whole system. The regenerative PWM rectifiers are based on a fullbridge topology. The topology uses four fully controllable bipolar switches constructed using an IGBT with an antiparallel diode. Since the input to the converter is an effective current source, the output must be a voltage source. To achieve this, a capacitor is placed at the output of each rectifier. This capacitor also provides filtering of the variations at the voltage caused by the switching frequency and the low frequency (100 Hz) components of the rectified current. The sizing of this capacitor is done in the design section. As shown in Fig. 2, the traction inverters are represented by the resistive loads that are sized to the draw the nominal power of 250 kW. The traction motors will absorb this power at the nominal speed of the locomotive. The full-bridge topology shown in Fig. 2 is also known as four-quadrant converter because it can operate both as a rectifier and as an inverter when needed. During the motoring mode, they provide rectification from ac to dc and the power goes from ac side to the dc side. Conversely, during braking or downhill travelling, the motors operate as generators and actually generate electricity. During the generator mode, the converters operate as inverters and convert dc into ac transferring the power from dc side to the ac side (to the catenary). The four-quadrant operation is realized using the fully controllable bipolar switches and the proper control. B. Control System The system in Fig. 1 shows the controllers for each rectifier unit. Each controller takes four inputs: the conditioned catenary voltage, the instantaneous boost inductor current, dc bus voltage, and the triangular carrier waveform. Signal conditioning of the catenary voltage is provided by a single-phase phase lock loop (PLL) as shown in Fig. 1. The output of the PLL is a sinusoidal signal synchronized to the catenary voltage with 1.0 perunit magnitude. This signal is used by the controllers to shape the input current of the rectifiers to achieve high power quality: low THD and unity power factor. The control system for each unit is identical and the inner details are shown in Fig. 3. The control system consists of two control loops compensated by the proportional plus integral (PI) and the proportional plus resonant (PR) regulators. The PR regulator provides a high gain only at the fundamental frequency of the ac grid and therefore very effective in correcting the steady-state error and obtaining a good tracking of the reference signal [9]-[12].
Figure 1. The Simulink model of the high power quality regenerative PWM rectifier system including the controller blocks for each parallelconnected rectifier unit and the THD measurement block.
A. Power Circuit Figure 2 shows the power circuit model of the rectifier system, the load, and the main transformer. The model is constructed using piecewise linear electrical circuit simulation (PLECS) tool, which is a Simulink toolbox for system-level simulations of electrical circuits developed by Plexim. The model includes four regenerative PWM rectifiers each connected to the secondary of the main transformer through an inductor. The main transformer is
Figure 2. The PLECS model of the four high power quality regenerative PWM rectifiers connected to the catenary voltage source through a boost inductor and the main transformer.
Figure 3. The Simulink model of the control system.
The outer loop is the voltage loop where the compensated dc bus voltage error is multiplied by the PLL signal to generate the reference input current. This reference is established by the PI compensator such a way that it regulates a constant output voltage and a sinusoidal input current in phase with the catenary voltage. The inner loop is the current control loop where the instantaneous average of the inductor current is compared to the reference current, and the compensated error is compared to the triangular carrier waveform to produce the
switching function for each IGBT switch. Switching of the IGBTs in the rectifier bridge is performed using the unipolar voltage PWM method.
III. CONVERTER DESIGN

This section presents the design of the power circuit and determines the control system parameters based on the specifications given in Table I. The nominal power of the electric locomotive is 1 MW. However, the total power is distributed to each traction motor independently
through four rectifiers, each rated at 250 kW. The rectifiers are responsible for regulating a constant dc voltage at 1500 V. This voltage is the input to the voltage source inverter (VSI) that drives the traction motors of the locomotive. The main transformer has four isolated secondary windings to supply each rectifier. At the primary of the transformer, the rectifier input currents are summed and drawn from the catenary line.
TABLE I. DESIGN SPECIFICATIONS 17-29 kV Catenary voltage range THD of the catenary < 5% (input) current Power factor of the input > 0.99 Output voltage Nominal output power per rectifier unit Peak-to-peak ripple at the output voltage Switching frequency
_
the duty ratio ( ) is 50% [5]. The following finds the inductance.

Where the is the specified peak-to-peak ripple at the inductor current, and it is found using the formula below: % %

(3)
1500 V 250 kW
4% of nominal = 60 V 2 kHz
Using 250 kW for and 750 V for in (3), is calculated as 666.67 A, and as 133.34 A for 20% ripple. Using these numbers in (2) yields and an inductance of 703 H. This value is rounded to 700 H, and used in the simulation studies. Another important step is to determine the minimum value of the output filter capacitor. The main factor in sizing this capacitor is suppressing the double-ac grid frequency (100-Hz) component of the rectified current. The capacitor current is equal to the rectified input current minus the average output current as given below:
4
A. Design of the Power Circuit The first step is to determine the main transformer turns ratio. In order to achieve high power quality at the input, the PWM rectifier has to boost the secondary voltage to a level greater than its peak, which is similar to the operation of a conventional boost converter. Therefore, the criterion is to select the turns ratio of the transformer such that the peak of the secondary voltage at the worst case must always be less than the desired minimum dc bus voltage. The following finds the minimum value of the turns ratio.

_ _
Where the is the average output current and is the instantaneous rectified input current. If the switching frequency components of the waveforms are ignored, is equal to ( sin 2 ). Therefore, the capacitor current becomes sin 2 A. After knowing the current through the capacitor, the minimum capacitance is found using the following equation, which based on the integral of the current:

(1)
Where and are the main transformer primary and secondary voltages, respectively. If we assume a worstcase minimum of 1400 V dc bus voltage and 29 kV maximum catenary voltage in (1), the turns ratio of the transformer is found as 30. The next step is the sizing of the current shaping (boost) inductor. Generally, there are two criteria in selecting the inductance. One is the desired ripple requirement at the current, and the second is the dynamic response. The low ripple, which needs a larger inductance, is preferred to yield better efficiency and better waveform quality. On the other hand, a low inductance is desired for faster dynamic response. This design selects an inductance that optimizes the both criteria, which corresponds to approximately 20% ripple at the current when the maximum ripple voltage is present across the inductor. The maximum ripple voltage is present across the inductor when the peak of the rectifier input voltage is half of the dc output, which is when
Where is equal to . Using 250 kW for , 1500 V for , 50 Hz for , and 60 V for in (5), the minimum value of output filter capacitor is found as 8842 F. We choose 9000 F as shown in Fig. 2. B. Design of the Control System The Simulink model of the control system is shown in Fig. 3. The control system is a two loop average current mode controller. The controller details were explained in Section 2.2. The scaling of the signals is done assuming an analog controller. The parameters of the PI and the PR compensators are determined via trial error method since the simulation models of the system includes all the major dynamics. The content of each controller for each rectifier is the same except the phasing of the triangular carrier waveforms. As shown in Fig. 4, there are four carrier waveforms with 2 kHz frequency. The optimum phase shift that provides the lowest THD of the input current was determined from the simulations. According to Fig. 4, the phase of the carrier waveforms applied to the
second (green), the third (red), and the fourth (light blue) rectifiers are delayed with respect to the first carrier (dark blue) by an angle of 45, 90, and 135, respectively. Although 2 kHz may seem a low switching frequency, the effective ripple frequency will become four times this at the input current due to phase shifting of carriers. Interleaving allows operation at a low switching frequency with no sacrifice to the waveform quality. The benefit of this is the higher efficiency.
1 0.5 0 -0.5 -1 0 0.2 0.4 Time (s) 0.6 0.8 x 10 1
-3
which is well below the specifications. The results verify that the proposed rectifier design achieves high power quality without needing any filtering effort.
600 465 Current (A)
-465 -600 0.96 80 62 Current (A)
0.965
0.97
0.975
0.98
0.985
0.99
0.995
Figure 4. The waveforms of the triangular carriers.
IV. SIMULATION RESULTS

The simulation studies are performed using Matlab/Simulink and PLECS tools. The results are obtained at rated catenary voltage of 23 kV and at 1 MW total output power. Fig. 5 shows the waveforms of the inductor current for each rectifier (upper trace) and the catenary current (transformer primary winding current) (lower trace) for the case when there is no phase shift among the carrier waveforms. This result is included here for comparison purposes so that the advantage of interleaving method is verified. As shown in the upper trace of Fig. 5, the switching frequency ripple at the inductor currents is 20% of 50 Hz component of the current as designed. The same amount of ripple is also reflected to the catenary current since interleaved operation is not active. This ripple results in 9.934% THD of the current and a slightly low power factor (0.995) at the source. In this case, the only way to reduce THD and improve the power factor at the source side is to use an LC low pass filter with a corner frequency of around 400 Hz. Since placing a filter at the high voltage side may not be practical, a separate filter placed at each rectifier input should be used. However, the interleaving operation will eliminate the need for these filters. The following figures will show the benefit of the interleaving. Figure 6 shows the same waveforms given in Fig. 5, but this time, they are obtained for the case where the carrier waveforms are phase shifted. The inductor currents (upper trace) still have the same amount of ripple as in Fig. 5. The detailed view of these currents is shown in Fig. 7 (upper trace). The catenary current, on the other hand, is free from the ripples as shown in the lower trace of Fig. 6. This improvement is clearly seen in Fig. 7 (lower trace), as well. Comparison of Fig. 6 with Fig. 5 reveals the benefit of interleaved operation. Figure 8 shows the catenary voltage (blue) and the current (green). The waveforms indicate that the catenary current is perfectly sinusoidal and in phase with voltage resulting unity power factor. The THD of the current is 1.257%,
-62 -80 0.96
0.965
0.97
0.975
0.98 0.985 Time (s)
0.99
0.995
Figure 5. The waveforms of the inductor current for each rectifier (upper trace), and the catenary current (lower trace) for the case when there is no phase shift among the carrier waveforms (non-interleaved case).
600 465 Current (A)
-465 -600 0.96 80 62 Current (A)
0.965
0.97
0.975
0.98
0.985
0.99
0.995
-62 -80 0.96
0.965
0.97
0.975
0.98 0.985 Time (s)
0.99
0.995
Figure 6. The waveforms of the inductor current for each rectifier (upper trace), and the catenary current (lower trace) for the case when there is a phase shift among the carrier waveforms (interleaved case).
600 Current (A) 465
0 0.981 80 Current (A) 62
0.9815
0.982
0.9825
0.983
0.9835
0.984
0 0.981
0.9815
0.982
0.9825 Time (s)
0.983
0.9835
0.984
Figure 7. The zoomed in view of the waveforms shown in Fig. 6.
4 3.25 Voltage (V), 200*Current (A)
x 10
excess energy back to the source. The system goes back to normal at 0.69 s.
V. CONCLUSION
1.23 0 -1.23
-3.25 -4 0.96 0.965 0.97 0.975 0.98 0.985 Time (s) 0.99 0.995 1
Figure 8. The waveforms of the catenary voltage (blue) and the current (green). The current waveform is multiplied by 200 for easy viewing.
1600 1530 1500 1470 Voltage (V) 1400
The demand for efficient use of energy sources and the high power quality requirements mandate the use of PWM rectifiers when interfacing the traction converters to the railway power system. In this study, the design and the simulation of high power quality regenerative PWM rectifier system that realizes these objectives for electric locomotives rated at 1 MW have been presented. The paper gives the details of a design that specifies a unity power factor and a current total harmonic distortion less than 5% without using a passive filter. The filterless operation is realized by interleaved operation of the parallel connected rectifier units. In conclusion, the concept and the design are verified through simulation studies. The results show that proposed rectifier system satisfies the high power quality and regenerative operating needs of modern electric locomotives.
REFERENCES
[1] IEEE Recommended Practices and Requirements for Harmonic Control in Electric Power Systems, IEEE Std. 519, 1992. [2] A. Steimel, Electric Traction Motive Power and Energy Supply, Oldenbourg Industrieverlag, Munich, 2008. [3] C. Heising, R. Bartelt, M. Oettmeier, V. Staudt, and A. Steimel, Improvement of Low-Frequency System Stability in 50-Hz Railway-Power Grids by Multivariable Line Converter Control in a Distance-Variation Scenario, Elect. Systems for Aircraft, Railway and Ship Prop., 2010, pp. 1-5. [4] J. R. Rodriguez, J. W. Dixon, J. R. Espinoza, J. Pontt, and P. Lezana, "PWM Regenerative Rectifiers: State of the Art," IEEE Tran. on Power Elect., v.52, no.1, pp.5-22, Feb. 2005. [5] N. Mohan, T. M. Undeland, W. P. Robbins, Power Electronics: Converters, Applications, and Design, John Wiley and Sons, NY, USA, 1995. [6] HVIGBT modules, (2013, June 15). Available: http://www.mitsubishielectric.com/semiconductors/products/power mod/hvigbtmod/. [7] J. Wen, T. Jin and K. Smedley, A new interleaved isolated boost converter for high power applications, IEEE App. Power Elect. Conf. and Exp., vol. 3, 2006, pp. 79-84. [8] B. Tamyurek and D. A. Torrey, A three-phase unity power factor single-stage acdc converter based on an interleaved flyback topology, IEEE Trans. Power Electron., vol. 26, no. 1, pp. 308 318, Jan. 2011. [9] H.-S. Song, R. Keil, P. Mutschler, J. van der Weem, and K. Nam, Advanced control scheme for a single-phase PWM rectifier in traction applications, Industry Applications Conference, vol.3, 2003, pp. 1558 1565. [10] R. Teodorescu, F. Blaabjerg, M. Liserre, and P. C. Loh, Proportional resonant controllers and filters for grid-connected voltage-source converters, in Proc. Inst. Electr. Eng. (IEE), Electric Power Appl., vol. 153, no. 5, pp. 750762, Sep. 2006. [11] G. Gy. Balzs, M. Horvth, I. Schmidt, P. Kiss, Examination of New Current Control Methods for Modern PWM Controlled AC Electric Locomotives, 6th IET Power Electronics, Machines and Drives Conference, Paper c.2.4 (0276), 2012, pp. 1-5. [12] M. X. Xu, D. Xu, P. Lin, M. Chen, J. Ni, and T. Zhang, Understanding repetitive control and resonant control, 3rd IEEE Inter. Sym. on Power Electronics for Distributed Generation Systems (PEDG), pp. 621- 627, 2012.
1300
1200 0.8
0.85
0.9 Time (s)
0.95
Figure 9. The output voltage (dc bus voltage) in steady-state.

x 10 3.25 Voltage (V), 500xCurrent (A)
4
1.05 0 -1.05
-3.25 0.48 0.516 Time (s) 0.69
Figure 10. The waveform of the catenary voltage (blue) and the current (red) during a regenerative operation.
Moreover, Fig. 9 shows the waveform of the output voltage. The result shows that the control achieves an average voltage constant at 1500 V. The ripple is 60 V as designed. The switching frequency ripple is very small compared to 100-Hz ripple. Finally, Fig. 10 shows the waveform of the catenary voltage and the current during a regenerative operation. At time 0.516 s, a charged capacitor is switched in and a large amount of energy is dumped into the dc bus of each rectifier. As shown in Fig. 10, the converter successfully goes into regenerative mode and transfers the
Comparative Design of Direct Drive PM Synchronous Motors in Gearless Elevator Systems

Hicret Yeti, Hande Boztepeli, Yusuf Yaa, Erkan Mee
Yildiz Technical University

Faculty of Electrical and Electronics Engineering Department of Electrical Engineering Istanbul, Turkey hicretyetis@gmail.com, handeboztepeli@gmail.com, yasa@yildiz.edu.tr, emese@yildiz.edu.tr
Abstract In todays world electrical machines become irreplaceable for human beings to continue daily routines. Due to the features such as high efficiency, high starting torque, and silent working, permanent magnet synchronous motors (PMSM) are becoming more common in many applications. One of these applications is gearless elevator system where PMSM seems to be the best candidate so far. In this paper, comparative design of PMSM for gearless elevator has been investigated. Maxwell 2D Finite element analysis software has been used in this study. With comparative design, the influence of pole count on the machine performance is studied. Also, material consumption, weight and costs of two machines have been compared. Keywords; pmsm, machine design, direct drive, gearless elevator system, maxwell
the same weight capacity. The reduction can be as high as 50 %. This paper presents PMSM design used in elevator systems. As a first step, elevator system requirements are determined. Based on these requirements motor torque and speed values are calculated. Then, commonly used stator diameter and axial length numbers are investigated by using commercial product datasheets. These numbers are used for all of alternative designs. In other words, comparative design does not include the comparison of stator diameter and axial length of different designs. In the comparative design technique, pole count, magnet shape, slot geometry can be considered as most influential on the efficiency. II. REQUIREMENTS OF ELEVATOR SYSTEMS
I.
INTRODUCTION
Subject of energy efficiency become more of an issue day by day in many industry and domestic applications. Developing technology in energy efficiency enables new and better solutions instead of classical solutions in elevator systems. Gearless elevator systems driven by permanent magnet synchronous motor have many advantages compared to traditional systems. Due to highly sensitive absolute encoder coupled to motor and closedloop driver system providing sensitive control, PMSM based gearless elevators show high performance at lifting up, stopping and moving. Thanks to torque and speed control which are implemented more efficiently compared to asynchronous motors, programmed speed-time curve under all weight conditions- can be realized precisely. As a result exact and comfortable approach to floor can be achieved. Beside comfort, one of the most important feature of gearless elevator systems driven by PMSM is providing energy conservation compared to systems designed with traditional asynchronous motor and gear box[1]. Gearless elevators can be designed with smaller electric motors for
The major requirements in the motor design of gearless elevator systems are torque and speed. These two parameters can be calculated by operating speed, cabin weight capacity, type of suspension and pulley diameter of designed elevator system. [2] For some given elevator specifications such as 325 kg (for 4 people) weight capacity, 1m/s cabin velocity, 2:1 suspension ratio following motor requirements can be calculated. (2.1) Where, rpulley : The radius of drive pulley (m) (0.12 m), g : The force of gravity (m/ (9.88 m/ ) Mcarry: Maximum carrying capacity (kg) (325 kg) u: The coefficient for suspension type. 1 for direct suspension, 2 for 2:1 suspension. Design was carried out for u=2. : Well and rope system efficiency (80%)
= 238.8 Nm. Motor rated speed; (2.2) v: cabin vertical velocity = 1m/s
(2.3) (rpm) Rated power; (2.4)

Figure 1. Slot geometry of 36-34 PMSM
So following motor parameters are fixed before starting design procedure, Pmotor = 4 kW, T = 238.8 Nm, n = 160 rpm. III. DEFINING CRITICAL DIMENSION OF PMSM
Figure 2. Slot geometry of 36-42 PMSM
To achieve expected performance, PMSMs stator and rotor diameter, length, slot and pole numbers should be properly determined. Comparative design will be performed with two PMSMs. Outer envelopes of these two machines are chosen so that the machines would have similar outer dimensions of other commercial gearless motors used in the elevator industry. The outer envelope dimensions which have been selected from other applications are stator outer diameter to be 290mm, stator inner diameter to be 218mm and axial length to be 178mm. Defining proper slot and pole number combination in machine design process is very important. The aim of this paper is to outline the effect of this combination in gearless elevator motor design. Two combinations are considered for this purpose. First one has 36 slots and 34 poles, second one has 36 slots and 42 poles [3,4]. DEFINING SECONDARY DIMENSIONS OF PMSM After defining critical dimensions in design process, other dimensions for a successful design, are slot dimensions, air gap length, magnet thickness, magnet width, slot opening (Bs0) should be defined. Slot geometry is defined for both machines in Fig 1. and Fig 2. IV.
Values of air gap, magnet thickness, magnet width and slot opening have been optimized by parametric analysis in which Rmxprt was used for this purpose. For both machine, parametric analyses are done to optimize values below: Table I. Parametric analysis variables under consideration Dimensions Limits Increment
g: Air Gap Length Bs0: Slot opening emb: Magnet Embrace Lm: Magnet Thickness 0, 75 mm< g < 1, 25 mm Wire Diameter< bs0 < 3 x wire diameter 0, 6 mm< emb < 0,96mm 3mm < Lm < 6mm g= 0, 05 mm Stator wire diameter emb = 0, 02 mm Lm = 0, 25 mm
In the selection of the best geometric dimension given in Table 1, following performance metrics are considered; efficiency, output torque, torque angle, cogging torque and induced back EMF voltage. From the results of parametric analyses, following observations are done. Condition of decreasing air gap: Cogging torque and torque angle decreased, Efficiency and induced voltage increased.
Condition of increasing magnet width (embrace ratio): Cogging torque, induced voltage and efficiency increased, Torque angle decreased. Condition of increasing magnet thickness; Cogging torque and efficiency increased, Torque angle decreased. Induced voltage increased and also phase shift occurs. Condition of increasing slot opening (bs0); Cogging torque and torque angle decreased, Efficiency decreased until certain slot opening and starts to increased. Induced voltage decreased and also cause phase shift. At the end of parametric analysis process following dimensions are found to be optimum for 36-34 PMSM; Air gap: 0,875mm Magnet width: 0,84 mm, Magnet thickness: 4,50 mm, Slot opening: 2.56 mm. Similarly following dimensions are found to be optimum for 36-42 PMSM; Air gap: 0,75mm Magnet width: 0,88 mm Magnet thickness: 4,75 mm, Slot opening : 3.037 mm. V. COMPARISON OF TWO MACHINES HAVING DIFFERENT NUMBER OF POLES B.
Figure -3. Cogging torque
Induced Voltage
Induced back EMF waveform has effect on the torque ripple performance of the motor. Hence two motors are compared for their back EMF waveforms. Fig.4 shows comparable back EMF waveforms in terms of closeness to sinus waveform.
Figure -4. Induced Voltage
C. Torque Ripple As far as the torque ripple is concerned, 36-42 motor seems performing better as shown in Fig.5.
After designing two motors in RMXPRT, they are analyzed to compare their performance in terms of efficiency, weight and cost. Furthermore, two machines are also analyzed with Maxwell 2D software to estimate their cogging torque, torque ripple, back emf, flux distribution and phase current [5]. A. Cogging Torque
Cogging torque is generated as a result of interaction between slot teeth and magnets. Phase current waveform has no effect on it. For this reason, cogging torque waveforms are generated by Maxwell for zero phase current. Fig. 3 shows that 36-34 motor has less cogging torque value.
Figure 5. Torque and torque ripple
D. Phase Current Phase current comparison for both machine is shown in Fig.6. 36-34 motor is drawing slightly higher current. This will cause some small amount of efficiency difference between two motors.
TABLE II. MATERIAL CONSUMPTION WEIGHT (KG) COMPONENT Armature Copper Weight Permanent Magnet Weight Armature Steel Core Weight Rotor Core Steel Weight Figure 6. Phase A current Total Net Weight Armature Steel Core Consumption 36-34 8.38329 3.28849 23.9805 33.3794 69.0317 65.4231 44.9066 36-42 8.27325 3.67257 24.6551 33.2358 69.8178 65.6093 44.7204
E.
Flux Distribution
Rotor Core Steel Consumption
Flux distribution maps for two machines are shown in Fig.7 and Fig.8. Overall flux distribution pictures indicate that 36-42 motor is slightly less magnetic loading. But, some local spots exist in both machines where heavy saturation occurs.
G.
Material Cost
Material cost comparison has been performed with the available price data of steel, copper and magnets. Table III shows the data. Steel and copper costs are comparable in two machines whereas magnet costs have significant difference. 36-34 machine has less magnet cost due to less pole count.
TABLE II. MATERIAL COST
Price (US $) NAME 36-34 PMSM 8.38329 kg x 11 $/kg =92.21619 $ 36-42 PMSM 8.27325 kg x 11 $/kg =91.00575 $
Armature Copper
Figure 7. Flux Distribution of 36-34 PMSM Permanent Magnet 34 x 10 $/number =340 $ 23.9805 kg x 2 $/kg =47.961 $ 33.3794 kg x 2 $/kg =66.7588 $ 42 x10 $/number =420 $ 24.6551 kg x 2 $/kg =49.3102 $ 33.2358 kg x 2 $/kg =66.4716 $
Armature Core Steel
Rotor Core Steel
Figure 8. Flux Distribution of 36-42 PMSM
F.
Material Consumption
When both motors are compared in terms of their active material weight, it turns out to be steel, copper and magnet consumptions are comparable.
VI. EVALUATING OF THE ANALYSISS RESULTS Overall comparison results are given in Table III. 36-34 motor seems performing better in many performance metrics. Generally, for directly driven PMSM design in elevator systems, aims are low cost, high efficiency, high torque density, as low as possible cogging torque and torque angle. Investigation shows that both machines perform almost at the same efficiency but output torque of 36-34 PMSM is higher. Also 36-34 machine has less cogging torque and smaller torque angle. As far as the costs of two motors are concerned, 36-34 motor is cheaper than 36-42 motor.
TABLE III. COMPARING TWO PMSMs [1] DATA Efficiency Rated Torque Cogging Torque Torque angle Induced Voltage Stator Teeth Flux Density Stator Yoke Flux Density Steel Cost Copper Cost Magnet Cost Total Net Cost 36-34 PMSM % 93.2 310 Nm 1.98 Nm 16.4 deg 364.059 V 1.75 Tesla 0.75 Tesla 114.7198 $ 92.21619 $ 340 $ 546.93599 $ 36-42 PMSM %93.57 290 Nm 8.76 Nm 17.23 deg 394.978 V 1.57 Tesla 0.58 Tesla 115.7818 $ 91.00575 $ 420 $ 626.78755 $ [4] [3] [2]
VII. CONCLUSION In this paper, comparative design of two PMSM motor is presented. Targeted application is gearless elevator system where low torque ripple and cogging torque is desired. In the comparative design, two motors are compared where their outer envelopes are same but they differ in some details such as slot/pole combination, slot shape and magnet shape. 36-34 and 36-42 combinations are selected in two machines. Analysis results show that 36-34 machine is superior to 36-42 machine in several aspects.
REFERENCES
Increasing Energy Efficiency in Elevator Systems by Direct Drive and Permanent Magnet Synchronous Machines (In Turkish), H.T.Duru, R.Demirz, Y.Toktas, 1. EVK Sempozyumu, Kocaeli 2005. Application of Permanent Magnet Excited Synchronous Motor and Direct Drive in Elevator Systems (In Turkish), H.T.Duru, R.Demirz, Y.Toktas; Asansr Sempozyumu, zmir 2006. Design of Brushless Permanent Magnet Machines, J.R. Hendershot, T.J.E. Miller, Motor Design Books LLC; Second Edition edition, 2010. Permanent Magnet Motor Technology, J.F. Gieras,CRC Press, 2002. Maxwell User Manual, ANSYS.
[5]
Analysis of Heteropolar Inductor Machines Based on Space Vectors

Jingzhe Wu
University of Bath Bath, BA2 7AY, UK Email: tw260@bath.ac.uk
AbstractThe concept of heteropolar inductor machines is introduced, including the history, general geometry and electrical properties. Modifications are made to conventional topology for improved effectiveness. The armature space vectors and field dynamic equations are derived. Computer simulations are used to optimize machine design and to estimate important machine parameters. Test machine experiment is performed to verify the theoretical equations under steady state. Keywordsspace vector; heteropolar machine dynamics; steady state; transient inductor machine;
Martin J Balchin
University of Bath Bath, BA2 7AY, UK Email: eesmjb@bath.ac.uk In Fig.1, A B and C represent the three phase coils while F represents field coils. Coil sides with plus and corresponding coil sides with minus signs are carrying current flows in opposite directions. Field coils are connect in series and are supplied from a single DC voltage source and field coils with a same numbering belong to a same winding. Heteropolar inductor machines share some structural similarities with wound field flux switching machines [4-6] and former researchers [1] have suggested that the electrical frequency e of voltage and current in armature windings are determined by both rotor speed S and number of rotor saliencies p . The relationship is e = p S (1) Clearly the 4 rotor saliencies in Fig.1 represents that the electrical frequency is 4 times the rotor angular frequency, similar to a synchronous machine with 8 poles. The electrical angle between two stator slots is 2 / 3 and the armature winding configuration in Fig.1 shows a balanced three phase system. Based on the topology in Fig.1, several modifications are made to improve the effectiveness of the machine. Firstly, instead of the square shaped rotor which generates a square airgap permeance wave, a specially designed rotor shape which gives a sinusoidally varying airgap permeance is used. Secondly, the number of stator slots is tripled to fully utilise the space on stator while the number of rotor saliencies is kept unchanged, as is mentioned by R Nair [7]. The modified machine schematic is shown is Fig.2. The stator in Fig.2 is developed that in Fig.1 by replacing the concentrated windings with distributed windings. The mechanical robustness is one of the most attractive merits of heteropolar inductor machines [8]. As shown in the section view in Figs.1 and 2, all windings of the machines are moved to stator side while the moving parts, i.e. the rotors, are completely passive. Such a topology offers a very high operational reliability since the removal of windings from rotor side reduces greatly the mechanical risks at high rotating speed. Also the absence of permanent magnets (PMs) makes the machine a good candidate for applications in high surrounding temperature and adverse air environment [9], and
I.
INTRODUCTION
The concept of heteropolar inductor machines [1-3] was carried out in 1930s and 1940s and they were once used for generating high frequency alternating currents (AC) before replaced by power electronics. Due to this reason, recent literature related to heteropolar inductor machines is very limited. Generally, a heteropolar inductor machine (Fig. 1) consists of a non-salient stator similar to general induction and synchronous machines and a salient rotor similar to standard switched reluctance machines (SRMs). There are two kinds of windings wound on the stator, one of which is fed by three phase AC power supply, namely armature winding, and the other fed by direct current (DC), namely field winding. Three phase armature windings and field windings are interleaved in connection on stator.
Fig.1. Schematic of a simple heteropolar inductor machine with 4 rotor saliencies and 12 stator slots
makes the manufacturing cost lowered greatly, facilitating mass production.
complex vector represents that the vector is a conjugate vector. is the angular displacement of rotor axis from reference and its initial position is 0 . Rotor axis can coincide with any one of the rotor teeth. Angular relationship is shown in Fig.2. When balanced three phase currents are flowing into armature windings, a current space vector is formed by (5) and a resultant two-pole magneto-motive-force (mmf) is generated and can be expressed in complex form, as in (7). 2 2 j j 3 (5) iS = iA + iB e 3 + iC e 3 = i ph e j = iS e j 2 = 0 + e t (6) kdp1 N S * j ma = iS e j + iS e (7) 4 iA iB and iC are instantaneous phase currents while i ph is
Fig.2. Schematic of improved heteropolar machine with distributed windings and rotor for sinusoidal permeance
With all the advantages prospected, it is important to develop a unified mathematical model for heteropolar inductor machines, which is suitable for both steady state and transient analyses. Also, such a mathematical model can provide a basis for dynamic and accurate control theories for various applications. In this paper, the machine will be analyzed based on space vector theories [10, 11]. Space vectors are often regarded as the method for analyzing dynamics of electrical machines and a solution for field oriented control (FOC). This paper describes the steps to express a heteropolar inductor machine by a set of differential equations for rotating space vectors. The theory will be verified by practical tests. II. DERIVATION OF SPACE VECTORS
denoted as the peak current value in each phase. Angle is the angular displacement of armature mmf waveform from reference and its initial value is 0 . The number of turns per phase winding is N S . The armature winding factor for fundamental harmonic k dp1 is approximately equal to 0.5 since the coil pitch is 6 stator slots, or / 3 . The magnetising flux linkage of phase A can be calculated by integration.
mA = kdp N S ma ( ) rd (8) 0 6 Nf Nf + NS i f r ( ) d + i f r ( ) d 2 2 0 6 i f stands for field current while N f is the number of turns per set of field winding. r is the bore radius of the heteropolar inductor machine. As will be seen, harmonics other than fundamental will be induced by the mmf wave and permeance variation. Therefore the armature windings will function as a filter for different orders of harmonics and hence a general winding factor term kdp is used in the expression. With the following abbreviations for inductances 3 2 2 LmS = kdp (9) 1 N S r 0 2 3 2 LmS 2 = kdp1kdp 5 N S r1 (10) 20 1 2 LmS 0 = kdp1kdp 3 N S r1 (11) 4 9 Lmfa = N S N f r1 (12) 16 The result of magnetising flux linkage calculation is
A. Armature Equations Due to the unsmooth rotor surface, there exists a magnetic permeance variation in the airgap. The variation of permeance is shown in Fig.3 assuming stator bore surface is ideally smooth and the machine is not saturated.
Fig.3. Airgap permeance variation along airgap
The origin of is referenced to the center of phase A, as shown in Fig.2. As the rotor rotates, the permeance variation in Fig.3 can be expressed in complex vector form. 1 * j 4 ( ) = 0 + 1e j 4 + 1 e (2) 2 Where 1 = 1e j 4 (3)
= 0 + S t
(4)
0 and 1 are invariant and fundamental component of permeance variation, respectively. The asterisk on the
mA =
1 1 * * j 4 LmS iS + iS + LmS 2 iS e j 4 + iS e 3 3 2 1 LmS 0 iS + Lmfa i f + i* f 3 3
(13)
Other than magnetising flux linkage, the stator phase windings are also linked by their leakage flux linkage. Therefore the total flux linkage space vector is S = LS iS + LS 2 iS e j 4 + Lmfa i f (19)
Where
j (4 ) 2 if = if e
(14)
Magnetising flux linkages for the other two phases can be calculated in a similar manner. 2 j j 2 1 * mB = LmS iS e 3 + iS e 3 3 2 2 j j 1 * + LmS 2 iS e 3 e j 4 + iS e 3 e j 4 (15) 3 j 2 j 23 2 1 LmS 0 iS + Lmfa i f e 3 + i* f e 3 3 2 2 j j 1 * mC = LmS iS e 3 + iS e 3 3 2 2 j j 1 * + LmS 2 iS e 3 e j 4 + iS e 3 e j 4 (16) 3 j 2 j 23 2 1 LmS 0 iS + Lmfa i f e 3 + i* f e 3 3 The vector terms of mA mB and mC are 2 / 3 apart from each other, proving that three phases are balanced. The magnetising flux linkage space vector is obtained as 2 2 j j mS = mA + mB e 3 + mC e 3 (17) If three phases are balanced, all conjugate space vectors and DC components are eliminated. mS = LmS iS + LmS 2 iS e j 4 + Lmfa i f (18) The first two terms of (18) are produced by currents in armature windings, i.e. armature reaction. The last term is similar to the mutual inductance term in conventional synchronous machines, which is the flux linkage resulting from field excitation. This is responsible for inducing internal electro-motive-force (emf). It is worth noting that the armature reaction of heteropolar inductor machine induces a 5th space harmonic which results in a double frequency flux linkage term LmS 2 iS e j 4 .because iS e j 4 = iS e j (e + 4S ) e j ( 0 + 40 ) and e is equal to 4S . On the 2 LmS 0 iS is 3 also induced in each phase by armature reaction, which has no contribution to terminal voltages. In a word, the invariant permeance component 0 contributes to the armature reaction flux linkage in line frequency while the fundamental permeance component 1 gives rise to a double line frequency flux variation with an offset. other hand, an invariant component of flux linkage
LS and LS 2 take leakage inductance in addition to magnetising inductances into account. In stator coordinate, the terminal voltage vector is expressed as d uS = iS RS + S (20) dt A resultant phase voltage waveform (Fig.4(a)) of 90 rotor mechanical degrees is simulated by FEA. The waveform is undesirable because it is greatly distorted by a second harmonic and a resulting second harmonic current will create additional heating in circuit and undesired torque ripple in machine, if the machine is loaded.
(a) (b) Fig.4. Normalised terminal voltage waveform of heteropolar inductor machine (a) before and (b) after modification
A possible method to eliminate second harmonic voltage is to force the second harmonic magnetising inductance LmS 2 to become zero. This is achieved by extending all coil spans (both armature and field) of the machine by 1/6. When the coil pitch becomes 7 stator slots, the term kdp 5 is reduced to almost zero. The modification is shown in Fig.5 and the terminal voltage waveform after modification is shown in Fig.4(b).
Fig.5. Modified winding configuration to suppress second harmonic
After the elimination of second harmonic, the flux linkage space vector is now similar to that of a conventional synchronous machine. S = LS iS + Lmfa i f (21) Unlike a conventional synchronous machine having a field current which is actually rotating in space, the field current
vector i f of a heteropolar inductor machine is the consequence

of the interaction between stationary field windings on stator and a rotating magnetic permeance wave in space. Although LS and Lmfa are analogous to self and mutual inductances in conventional machines, they have no close relations between each other since they are effected by different permeance components. The above space vectors are referenced to stator coordinate. If vector i f is considered to be the direct axis (d axis) in a rotating frame [10], the model is analogous to a conventional machine since i f and iS are rotating at a same angular speed e = 4S . Armature current space vector iS can be resolved into two orthogonal components aligned to direct and quadrature (q axis) axes to facilitate torque calculation. The relationship of vectors is shown in Fig.6.
The results for three field windings are j 3 2 j * mf 1 = N 2 N f i f r 1e 3 + 1 e 3 f i f r 0 6 16 j j 1 * kdp1 N S N f r 0 iS e 6 + iS (26) e 6 4 j j 1 * * kdp 5 N S N f r iS 1e 6 + iS 1 e 6 40 j j 3 2 * mf 2 = N 2 N f i f r 1e 3 + 1 e 3 f i f r 0 6 16 5 5 j j 1 * kdp1 N S N f r 0 iS e 6 + iS (27) e 6 4 j j 1 * * kdp 5 N S N f r iS 1e 2 + iS 1 e 2 40 j 2 3 2 * j mf 3 = N f i f r 0 N f i f r 1e + 1 e 6 16 j j 1 * kdp1 N S N f r 0 iS e 2 + iS e 2 (28) 4 j 5 j 5 1 * * k dp 5 N S N f r iS 1e 6 + iS 1 e 6 40 The field windings of a heteropolar inductor machine also form a three phase system in space indicated by (26-28), despite the fact that they are connect in series instead of star or delta in the case of armature windings. By denoting
Fig.6. Spacial and angular relationship of current vectors
In rotating frame, voltage and flux linkage space vectors are shown in (22) and (23) respectively. The superscript R denotes that the vectors are referenced to a rotating plane. d R R uS RS + SR + j 4S SR = iS (22) dt SR = LS iSR + Lmfa i R (23) f Torque equation is similar to that in conventional space vector theories for normal synchronous and induction machines [10, 11]. 2 Te = Im 4SR * iSR (24) 3 The factor of 4 is taking the number of pole pairs (i.e. the number of rotor saliencies) into account.
N2 f r 0 2 A simple result is obtained. mf = mf 1 + mf 2 + mf 3 = Lmf i f Lmf =
(29) (30)
Therefore no rotating space vector is generated by field windings and clearly the field series connected field windings are only linked by their own flux, therefore balanced armature currents have no contribution to field flux variation. Field leakage flux linkage also contributes to the total field flux linkage. Hence the total field flux linkage is expressed as in (31). f = Lf if (31) L f is the sum of magnetising inductance Lmf and leakage inductance. The field terminal voltage is expressed as d uF = i f R f + f (32) dt In summary, (22-24, 32) comprise the dynamic space vector equations for heteropolar inductor machines. III. TEST MACHINE EXPERIMENT
B. Field Equations The magnetising flux linkage for one of the field windings is calculated by integration over the field coil span, for example the magnetising flux linkage of field coil 1 is calculated as
mf 1 = 2 N f r
N f if + ma ( ) d 2 0
3
(25)
In steady state, differential terms in dynamic equations are equal to zero and they are rotating in same speed in space. The steady state equations are
(33) R R uS = iS RS + j 4S SR (34) uF = i f R f (35) 8 8 R Te = Im 4SR* iS = Lmfa i f iSq (36) 3 3 Test machine is experimented in lab in order to verify the theories. The test machine stator and rotor are shown in Fig.7. Stator windings are wound with a coil pitch of 7 stator slots in order to remove undesired second harmonic. The test rig is shown in Fig.8. A synchronous machine is directly coupled to the test machine and the armature windings are connected together. The synchronous machine and test machine are both driven by a DC motor in order to force the two machines to generate emf and current at a same electrical frequency. The torque angle of test machine is changed by altering the relative position between synchronous machine rotor and test machine rotor. The torque transducer is responsible of recording the torque and speed values.
SR = LS iSR + Lmfa i R f

The synchronous reactance of test machine is measured by short circuiting three phase windings and giving a certain field current. Since synchronous reactance is much greater than phase resistance at normal operational speed, it is assumed that the value is calculated by dividing open circuit voltage by short circuit current. Also, since in three phase balanced short circuit, the phase current lags internal emf by almost 90 degrees, the measured reactance value is therefore assumed to be unsaturated value. It is measured and calculated from the test machine that the synchronous reactance X S = 15.7 , slightly larger than theoretical estimation and FEA. It is because the theory does not include differential leakage reactance in calculation and 2D FEA is unable to simulate end turn leakage flux of test machine.
Fig.7. Rotor and stator of test machine
Fig.8. Test rig and relevant equipment for measurements
A. Test Machine Parameters The sizes and parameters of test machine are listed in Table 1.
Table 1. Test machine sizes and paraeters
Bore diameter Outer diameter Axial length Armature winding turns Field winding turns Minimum airgap length Angular speed 0 (simulated by FEA) 1 (simulated by FEA)
103 mm 170 mm 108 mm 150 turns/phase 150 turns/field winding 0.6 mm 52.3 rad/s (500 rpm) 1.296E-4 Wb/A 6.966E-5 Wb/A
B. Open Circuit Test Theoretically the peak value of internal emf per phase at open circuit may be calculate from the magnitude of vector uS by (22, 23) when iS is equal to zero. 2 4S Lmfa i f = 9 V 3 The factor of 2/3 in front of the equation takes into account the fact that the length of space vectors is 3/2 times the real phase value. The waveform of internal electro motive force (emf) is measured by oscilloscope when the test machine is fed by 1A field current and the armature terminals are open circuited. As shown in Fig.9, the waveform is close to a proper sine wave, except that a little slot harmonic is observed. This can be eliminated by skewing. The vertical and horizontal scales are 5V/div and 5ms/div respectively. The emf magnitude is about 9V. E=
Unsaturated armature reaction reactance of test machine at 500 rpm can be calculated from (9). X mS = 4S LmS 11.72 Leakage reactance per phase mainly consists of slot leakage reactance and end winding leakage reactance. Classical formulae [12] for leakage reactance are used to estimate the value of phase leakage reactance and an approximate value of 3.41 is obtained. The synchronous reactance is therefore the sum of magnetising and leakage reactances and equal to 15.13 . 2D FEA is also used to estimate the synchronous reactance and the value is 14.5 .
Fig.9. Open circuit emf when field current equals 1A
The open circuit magnetic curve (Fig.10) is measured to give an idea of saturation and finding proper operational point for test machine. Field current is increased from 0 to 7 A by an increment of 0.5 A.
Steady state theories are verified by experiments and transient experiments will be conducted in proceeding experiments in the future to verify the dynamic theories. IV. CONCLUSION
The space vector theory for heteropolar inductor machines is illustrated in this paper. Armature space vector differential equations and field dynamic equation are derived with important machine inductances estimated. Theory in steady state is verified by practical experiment.
Fig.10. Open circuit magnetic curve and airgap line of test machine at 500 rpm
REFERENCE [1] [2] J. H. Walker, "High-Frequency alternators," Electrical Engineers - Part II: Power Engineering, Journal of the Institution of, vol. 93, pp. 67-80, 1946. J. H. Walker, "The theory of the inductor alternator," Electrical Engineers - Part I: General, Journal of the Institution of, vol. 89, pp. 477-480, 1942. P. K. Dash and A. K. Das Gupta, "Design Aspects of Three-Phase Inductor Alternators," Power Apparatus and Systems, IEEE Transactions on, vol. PAS-88, pp. 1718-1724, 1969. A. Zulu, B. C. Mecrow, and M. Armstrong, "A Wound-Field Three-Phase Flux-Switching Synchronous Motor With All Excitation Sources on the Stator," Industry Applications, IEEE Transactions on, vol. 46, pp. 2363-2371, 2010. J. T. Chen, Z. Q. Zhu, S. Iwasaki, and R. Deodhar, "Low cost flux-switching brushless AC machines," in Vehicle Power and Propulsion Conference (VPPC), 2010 IEEE, 2010, pp. 1-6. C. Pollock and M. Wallace, "The flux switching motor, a DC motor without magnets or brushes," in Industry Applications Conference, 1999. ThirtyFourth IAS Annual Meeting. Conference Record of the 1999 IEEE, 1999, pp. 1980-1987 vol.3. R. Nair, "Inductor generators for alternative energy schemes," in Electrical Machines and Drives, 1993. Sixth International Conference on (Conf. Publ. No. 376), 1993, pp. 499-504. R. K. Lay, "The history and changing fortunes of the inductor alternator," Electronics and Power, vol. 14, pp. 484-487, 1968. D. J. Powell, G. W. Jewell, D. Howe, and K. Atallah, "Rotor topologies for a switched-reluctance machine for the 'more-electric' aircraft engine," Electric Power Applications, IEE Proceedings -, vol. 150, pp. 311318, 2003. P. K. Kovcs, Transient phenomena in electrical machines: Elsevier, 1984. W. Leonhard, Control of Electrical Drives: Springer, 2001. J. H. H. Kuhlmann, Design of Electrical Apparatus: John Wiley & Sons, 1950.
The air gap line has a slope of 9 V/A and the real magnetic curve diverges from air gap line at field current equals about 4 A. The magnetic curve converges to 50 V (peak) at 500 rpm. C. On Load Torque Test The machine is loaded by connecting the armature winding terminals to synchronous machine armature windings. The idea of measuring torque angle curve is to shift the relative rotor position between synchronous machine and heteropolar inductor machine while maintaining the values of armature current and test machine field current constant for each single test. The test machine field current is set to be 2 A, resulting in a peak internal emf of about 18 V. The armature current amplitude i ph is chosen to be 1.2 A. The torque values when test machine is under motoring mode are recorded according to different values of load angles. Load angle of test machine is varied from -90 degrees to +90 degrees by an increment of 10 degrees, as shown in Fig.11.
[3]
[4]
[5]
[6]
[7]
[8]
Fig.11 Fig.12
Fig.11. Variation of load angle Fig.12. Theoretical and practical torque values at different load angles
[9]
In theory, torque can be estimated from (36). 8 8 Te = Lmfa i f iSq = Lmfa i f iS sin (37) 3 3 The theoretical and practical values are compared in Fig.12. The torque-angle curve of a heteropolar inductor machine is similar to that of a non-salient pole synchronous machine, despite the fact that the rotor is of saliency.
[10] [11] [12]
Identification of Key Transmission Lines in Power Grid Using Modified K-core Decomposition
Zhang Xiangyu, Liu Feng*, Yao Rui, Zhang Xuemin, Mei Shengwei
State Key Laboratory of Power System Department of Electrical Engineering Tsinghua University, Beijing, P.R.China * lfeng@mail.tsinghua.edu.cn
AbstractPower grid today has been more interconnected than it used to be, resulting in frequent occurrence of cascading failure in the power system. According to historical data, many major cascading failures are caused by the outage of certain transmission lines. The transmission lines with potential to cause large scale cascading failure are called key transmission lines (KTL). To find out the KTL, this paper first proposes a correlation network which characterized the interaction between different lines in the system. Then, based on this new network, we identify the KTL in power system using the modified K-core decomposition, which has been amended to better suit the power systems inherit features. Finally, case studies utilizing cascading failure simulation on New England System and IEEE-118 system are conducted to substantiate the effectiveness and efficiency of the proposed approach. Keywordscascading failure, k-core decomposition, degree centrality, key transmission lines, correlation network
Zhang Zhenan, Li Xiaomeng

Electric Power Research Institute HAEPC Zhengzhou, Henan, P.R.China
topology metrics mentioned above [6]-[7]. Degree centrality is based on the assumption that the most connected nodes are crucial, this is reasonable since the robustness of a network closely related to the degree distribution [8]. However, recent study has showed that not every node with high degree is influential [9]. Beside the degree centrality, the concept of betweenness is also used to recognize key elements frequently. It assumes that the busiest node in the network demonstrates vital importance for it undertakes heavy information or power flow. This approach views the problem from a new perspective which is very inspiring. But when using in power system, some scholars point out that it assumes the power flows in the shortest path, which deviates from the reality [10]. Reference [11] contends that, among so many choices, the Kcore decomposition outperforms other approaches in searching the influential spreaders in complex networks [12]. K-core decomposition has been widely used in many disciplines; involved areas including telecommunication [13], Internet structure analysis [14], complex network visualization [15] and so on. In [9], the study of epidemic proliferation shows the effectiveness of locating the influential spreaders using K-core decomposition. Power system cascading failure shares many common with epidemic spreading but the application of K-core method in power grid should be considered meticulously. First, the key elements in power system are not invariable but changing according to the different operating conditions which makes the method that totally rely on system topology inappropriate; Second, during cascading failures, the transmission lines, which are the links between nodes, are more fragile and vulnerable than the buses, and thus it is the links but not the nodes should be the main study objects; Finally, unlike that of the highly complicated protein molecule structure, the average degree of the power grid does not increase with the augment of grid scale which results in a limited degree range [6]. According to the consideration above, a modified K-core decomposition is needed both to retain the merit it has in identifying vital spreaders and to accommodate with the power grid characteristics. Following part of the paper consists of five parts. In Part II, the switch of research object into a derivative network is introduced to better depict power system cascading failures. In Part III, an overview of classic K-core
I.
INTRODUCTION
The interconnection of power grid can optimize the allocation of resources; nevertheless, it will also facilitate the propagation of system failure. Some faults in the system merely have influence locally while others will cause extensive cascading failures. Many major blackouts, like the North America 2003 and the India 2012, are caused by initial transmission line faults [1]-[2]. As a result, distinguish lines by their ability of causing large scale cascading blackout is an essential work [3]-[4]. Thousand times of cascading failure simulation definitely will give the answer, but it is inefficient reasoning that the key transmission lines (KTL) vary with different operating conditions. Therefore, an appropriate approach which has balanced the accuracy and efficiency is needed. Methods for identifying key elements in complex network are summarized in [5]. Such approaches for undirected network include degree centrality, betweenness centrality, closeness centrality, eigenvector centrality and kcore decomposition while HITS and PageRank Algorithm are suitable for the directed networks. In this paper, what we focus is to find the key elements in power grid which will cause potential cascading failures. Power system is a typical complex network; therefore it can be analyzed by some classic
decomposition methods is briefly mentioned and the modified method is explained in detail. In Part IV, the cascading failure simulator used for verify the validity of modified K-core decomposition is explained. In Part V, New England System and IEEE 118-bus system are used for case study. Finally in Part VI, the conclusion is summarized. II.
BRANCH-BASED CORRELATION NETWORK
The matrix that reflects the correlation network is not symmetric reasoning that Pij and Pji are not necessarily equal. Therefore, the network is essentially a directed network because there are interaction Pij from node i to j and interaction Pji from j to i . In order to simplify the model, we define the weight between node i and j as follow:
Even though the K-core decomposition has outstanding ability for identifying key elements, however, in the Introduction, we have mentioned three aspects, including various operating conditions, difference of major study objects and innate property of low average degree, that forbid us from using this method directly to the power grid. Moreover, since the classic K-core decomposition regards only the system topology, it will inevitably allocate the generators, which sometimes locate at the periphery part of the system, into the trivial part. Nevertheless, as the power producer, generators actually have vital importance. This is another trait of the power grid which hinders the direct usage of K-core decomposition. Therefore, in order to tackle with the problems above, a novel correlation network which considers both the grid topology and power flow distribution is introduced in this part. What is new about this network is that the nodes of this network are corresponding transmission lines in the original power grid. During the operation of power grid, the N-1 contingencies should be checked. Utilizing the outcome of the routine inspection, a matrix of dimension n can be generated as follow, where n here stands for the total number of transmission lines in the power grid.
ij
Pij margin
j
Pji margin i
(2)
The reason we bring the margin i and margin j to the weight in (2) is because it is not only the power fluctuation that determines the impact one had on another, but also the margin of line being influenced should be considered. Thus, by the simplification above, the network becomes an undirected weighted network, one that considered both the grid topology as well as the power flow distribution. In retrospect of the problems we have mentioned earlier, they are not exist in the correlation network anymore since 1) this network is based on the transmission lines, which is what we mainly studied, and 2) this network is derived from N-1 check which already considered the original power flow distribution and 3) this network is better connected than the power grid itself therefore truly reflect the interaction between different elements. III. MODIFICATION OF K-CORE DECOMPOSITION
P P 1f 12 P 21 P2 f P31 P32 ... ... Pn1 Pn 2
P 13 P23 P3 f ... Pn 3
... P 1n ... P2 n ... P3n ... ... ... Pnf
(1)
A. A brief overview of classic K-core decomposition The K-core decomposition decomposes the network by successively pruning the nodes, classifies them into different layers. Each layers nodes have a Ks value which can represent the connectivity of this node to others in a better way than the degree centrality. Degree centrality shows advantage under the condition when the network is assortative. However, according to the calculation of assortativity coefficient [5]:
1
2 q
jk (e
j ,k
jk
q j qk ), r [1,1]
(3)
In the matrix above, the element Pij means the power fluctuation on line j caused by the outage of line i , while Pkf 0, k {1, 2,..., n} since it is not concerned. As a result, this matrix can reflect the mutual influence between any two different lines. Moreover, the value of all the elements in the matrix are determined by both the topology of the grid as well as the power flow that distributed on it, which are the two important factors we want to consider when identifying key lines. It is easy to understand that the square matrix is another form of a network, e.g. the node relationship matrix can aptly reflect the connection of nodes in the power grid. Thus, the matrix above is a form of certain network, which we defined as correlation network. In this new network, nodes represent all the transmission lines in the grid we studied while the connections between them are the measure of interaction. About the correlation network, below are more details.
2 in which qk is the excess degree distribution, q is the
variance of the excess degree distribution and e jk q j qk , j, k . The power grid has a negative r value, which is to say it is not assortative network, e.g. for IEEE-300 system, r 0.2206 [16]. In this case, the degree centrality will somewhat lose its reference value when analyzing power grid. In identifying the best spreader, K-core decomposition has better performance than other approaches such as the degree centrality and betweenness centrality, according to [11]. After decomposing the network, every node will own a Ks value, which determines its status in the system: the nearer the nodes to the core, the larger the Ks value is. Processing detail will not describe here but can be found in [9] and [12]. Although [9] have successfully verified the K-core decomposition in identifying the influential spreaders in the
Fig. 1. Demonstrative network for defect analysis.
network, the defect of the classic K-core decomposition still need to be considered. In Fig.1, after decomposing, the node A and node B will have the same Ks value, which means they are categorized as nodes with the same spreading ability. Nevertheless, this outcome is intuitively unreasonable, since even if we cut all the nodes connected to node B whose degree equal one, at this moment A and B are the same. With the excessive nodes connected to B than A, B should have better spreading ability yet K-core decomposition endows them the same Ks value. Upon this problem, An Zeng et al [11] put forward an approach called mixed degree decomposition (MDD) to better classify different layers of nodes. This work is enlightening, however, the tunable parameter it proposed does not change when dealing different nodes and the fixed value (0.7 in the paper) is not well-explained. Another weakness of the classic K-core decomposition is its applicability since it is aiming at analyzing the topology of a given network, yet no weight is taken into consideration. As a matter of fact, most of the networks in the real work are weighted, even in [9], the spreading of epidemic in human interaction network can introduce weight derived from something like the meeting frequency between individuals. Thus, to better classify the node, it is necessary to involve weight into the network; the necessity and method to consider weight into K-core decomposition is described in [17]. B. Modification of Classic K-core Decomposition To accommodate the innate property of the power grid, we have switched the study object to a new kind of correlate network. Still, besides the object, the decomposition approach itself should be modified for the reasons we mentioned above. In this paper, we modify the decomposing method in the following way: When conducting classic K-core decomposition, the degree of some inner layer nodes should be minus by 1 as long as the node in the outer layer which connects to it is removed. This is not totally reasonable since the connection between them has two meanings: their mutual interaction. The removal of out layer node A means we have finished dealing with it, but the inner layer node B it connected with is still waiting for process, then the influence B had on A should still be counted but not vanished with node A.
Fig. 2. Difference between processing nodes in classic and modified K-core decomposition: in classic K-core decomposition, when a node is removed, the degree of nodes connected to it will minus 1(Left); In modified method, even a node is removed, the nodes that once connect to it can keep the one direction weight.(Right)
As we mentioned above, when node j is removed, the influence node i had on j still stick with the degree of node i , as showed in Fig.2. Therefore, neither minus 1 like the classic K-core decomposition does, nor take 70% of the connection arbitrarily like the mix degree decomposition [11], we propose that the weighted degree of nodes i should minus by ji [0,1] , the portion of influence j had on i .
ji
IV.
Pji margin i ij
(4)
CASCADING FAILURE SIMULATOR
Cascading failure simulator is needed to verify the validity of the modified K-core decomposition approach. CFS in this paper resembles the OPA model: it is a DC power flow model which has computational efficiency and numerical stability and suitable for considerable times of simulation. Differences lie in that we do not consider the optimal power flow and the slow dynamics, like the upgrading of loads, generators and transmission lines. Program flow chart shows in Fig.3. In Parameter Preset, the stochastic parameters, such as the malfunction possibility of relay protection, and maximum simulation day number are set. Initial fault exists every day by randomly pick one line in the system as the trigger. Line Trip stage might cause line outage either because of hidden failure or out of overload reasons. During the simulation, the initial faults and power losses of each day are recorded which can be used in our later analysis. For example, if transmission line j acts like initial fault for N times, and the total power loss caused by j during the whole process of simulation are Pj _ total , then the average power loss caused by line j will be:
Pl _ av
Pj _ total N
(5)
The average power loss caused by different lines will be the gauge of their ability to trigger potential cascading failure. As for the cases we used, original power flow in them satisfies the N-1 check. Therefore, the N-1-1 contingency is caused only by hidden failure, namely malfunction of the relay protection; but following line outage could be triggered either by hidden failure or capacity overload. Parameters set in the
simulation are: relay protection malfunction probability P0 0.01 , total simulation day DAY 20000 for New England System.
Start Parameter Preset
Random Initial Fault
DC Power Flow Calculation
Line Trip (Hidden Failure/Relay Protection)
B. Case I. The New England system has 46 transmission lines, which map into the correlation network are 46 corresponding nodes. Using the data from N-1 contingency routine check, the link between different nodes can be established as proposed in Part II. Since we cannot think of a clear physical meaning of the betweenness of nodes in the correlation network at present, the following work mainly focus on the comparison between modified K-core decomposition and degree centrality. As for betweenness and other metrics, they will be tested in future work. Here, using both K-core and degree centrality methods to analyze the network, and compare the results of two approaches to see which is closer to the result of cascading failure simulation. To be specific, the top 7 transmission lines that identified by degree centrality and K-core decomposition as key transmission lines are listed in Table I for comparison. Moreover, the top 7 lines that triggered largest scale cascading failure in simulation are also put in the table. The average power loss data in the table is from the 20000-day data as we proposed in Part IV.
TABLE I.
Rank CFS
Line Outage Occurred?
N
Statistics about Load Loss Day = Day + 1
Y
Day < Dmax?
N
End
COMPARISON BETWEEN RESULTS FROM TWO APPROACHES

Pl_av (MW) Degree Value K-core Value
Fig. 3. Chart flow of cascading failure simulation program.
V.
CASE STUDY
In this paper, a method for identifying key transmission lines is proposed. Using the modified K-core decomposition to analyze the correlation network, which is derived from the real power grid, we can find out lines that might initiate large scale cascading failure. In this part, the New England System and IEEE 118-bus system are used to verify the approachs validity and accuracy. A. Verifying Procedure Conducting N-1 routine security check for the given power grid and harvest necessary data; Preprocessing acquired data and forming correlation network; K-core decomposition of correlation network using the modified method proposed above; Identifying the transmission lines that located at the inner most of the structure as key elements; meanwhile, utilizing degree centrality to identify key lines as well, which can be used for comparison. Thousand times of simulation using CFS above and record the average power loss caused by each line as initial faults; Compare the result of key transmission lines identification based on modified K-core decomposition and degree centrality with the most influential spreaders found by cascading failure simulation.
1 2 3 4 5 6 7
20 37 46 14 39 33 38
512.19 506.72 503.16 494.18 475.37 419.48 412.37
3 46 37 26 33 20 23
16.59 16.01 15.24 14.93 14.83 14.00 13.93
46 37 33 34 20 39 14
12.84 12.84 12.84 12.84 12.24 12.24 11.83
From Table I, it is obvious that the identification results of degree centrality and K-core decomposition are not the same. Considering the components of three sets, Fig.4 is used to demonstrate the accuracy of identification outcomes. Elements in the dark circle are transmission lines that cause large scale cascading failure in simulation; those in dash line circle are the key transmission lines identified by degree centrality approach while the ones in normal circle are lines predicted by the modified K-core decomposition method. 3, 23, 26
34
46, 37, 20, 33 39, 14 38
CFS
Degree
K-core
Fig. 4. Graph depicts identification and simulation result
From the result above we can see that the top 7 lines identified by modified K-core decomposition can largely correspond with the top 7 lines from the simulation, merely with one exception. Although Line 34 identified by decomposition approach does not appear in the top 7 severe lines in simulation, it ranked the 10th with an average power loss of 359.05MW, according to the simulation result. This is not severely deviated from the identification result reasoning that Line 34 does cause large cascading failure though it is not the top 7 largest. On the contrary, the degree centralitys result demonstrates a poorer correspondence to the reality, with three lines incorrectly identified. To be more specific, Line 3, with the highest degree among all the nodes, should cause large scale cascading failure according to degree centrality, yet it ranked the 16th in all 46 nodes. Whats more, the fourth largest degree belongs to Line 26, but it ranked the 31th with an average power loss of 163.77MW. All these incorrect identification demonstrates that the degree centrality sometimes cannot grasp the essence; however, fortunately, the modified K-core decomposition can aptly remedy this since by successively pruning the nodes, those with real potential to cause cascading failure can be identified while those with false appearance will be eliminated.
To look in depth this problem, it is necessary to explain the gist of the modified K-core decomposition here. Those has potential to cause large scale cascading failure are the transmission lines that is easily influenced by others and at the same time, has much impact on many others. When lines with these two attributions clustered together, their mutual interaction make the propagation of faults more easily and efficiently, thus become more influential spreaders of cascading failure. This is the essence of K-core here. Apparently, node 3, 23 and 26 do not belongs to the core even they have large or largest degree. Their degree reflects their influence as well as how easily they will be influenced; high degree but not in the core means they are nodes that are readily affected by others yet their influence to others is trivial. Therefore, even when they are failed and quite often, no following consequence occur and as a result, they cannot be the vital spreaders; this is the physical fundamental of the modified K-core decomposition. From the analysis above, the advantage modified K-core decomposition has over degree centrality in identifying key transmission lines stands out. Fig.5 shows the position of key lines in the average power loss distribution curve. Noting that it is a ranked curve, the number on the abscissa does not coordinate with the nodes number.
Average Power Loss(MW)
600 500 400 300 200 100 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 Average Power Loss Distribution CoreMember
Fig. 5. Modified K-core decomposition result and ranked CFS outcome for New England System: blocks represent core members.
C. Case II. To better illustrate the applicability of the modified K-core decomposition, the IEEE-118 system is also used for test. Since the standard IEEE-118 system does not have the capacity of each transmission lines. We simply allocate the value according to the original power flow as well as line voltage. The identification result is showed in Fig.6, all the elements from the inner part of the core structure have the ability to trigger large scale cascading failure. Moreover, the advantage K-core decomposition has over the degree centrality can also be found in this case. The fourth largest degree is Transmission Line 54, which is 32.02, however, the decomposition process trimmed its value to 5.07, a comparably low value. And the reality is that, after cascading failure simulation, the average power loss caused by Line 54 ranked the 145th out of the total 186 lines. This can also be
explained by the essence of modified K-core decomposition mentioned earlier. Case study on IEEE-118 system demonstrates the effectiveness of the modified K-core decomposition does not decrease with the increment of the grid scale. Those lines both have strong influence to others and are readily affected by others are classified into the core area. The strong interaction between the core members makes them capable for causing severe cascading outage.
Average Power Loss(MW)
70 60 50 40 30 20 10 0 1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106 113 120 127 134 141 148 155 162 169 176 183 Average Power Loss K-core
[3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15]
Fig. 6. Modified K-core decomposition result and ranked CFS outcome for IEEE-118 System: blocks represent core members.
VI.
DISCUSSION AND CONCLUSION

J. Johansson and H. Hassel. An approach for modelling interdependent infrastructures in the context of vulnerability analysis. Reliab. Eng. Syst. Saf. 95, 2010, pp. 13351344. J.Ugander, B.Karrer, L. Backstrom and C.Marlow. The anatomy of the Facebook social graph. 2011 XF.Wang, X.Li and GR. Chen. Network Science: An Introduction[M]. Beijing: Higher Education Press, 2012. R.Albert, I.Albert, and G.L.Nakarado. Structural vulnerability of the North American power grid. Physical Review E 69.2 (2004): 025103. A. B. M. Nasiruzzaman, H. R. Pota and M. A. Mahmud, "Application of centrality measures of complex network framework in power grid," in IECON 2011 - 37th Annual Conference on IEEE Industrial Electronics Society, 2011, pp. 4660-4665. M. Rosas-Casals, S. Valverde, and R. Sole, Topological vulnerability of the european power grid under erros and attacks, International Journal of Bifurcations and Chaos, vol. 17, no. 7, pp. 24652475, 2007. M. Kitsak, L. K. Gallos, S. Havlin, F. Liljeros, L. Muchnik, H. E. Stanley, and H. A. Makse, "Identification of influential spreaders in complex networks,", 2010, 6(11), pp. 888-893. E. Bompard, Di Wu and X. Fei, The Concept of Betweenness in the Analysis of Power Grid Vulnerability, in Complexity in Engineering, 2010. COMPENG '10., 2010, pp. 52-54. A. Zeng and C. Zhang, Ranking spreaders by decomposing complex networks, Physics Letters A, 2013. S.Carmi, S.Havlin, S.Kirkpatrick, Y.Shavitt, and E.Shir, A model of Internet topology using k-shell decomposition. Proceedings of the National Academy of Sciences, 2007, 104(27): pp. 11150-11154. D. Miorandi and F. De Pellegrini, K-shell decomposition for dynamic complex networks, in Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks (WiOpt), 2010 Proceedings of the 8th International Symposium on, 2010, pp. 488-496. S.Carmi, S.Havlin, S.Kirkpatrick, Y.Shavitt, and E.Shir, Medusa-new model of Internet topology using k-shell decomposition. ,2006. J.I.Alvarez-Hamelin, L.DallAsta, A.Barrat, and A.Vespignani, Large scale networks fingerprinting and visualization using the k-core decomposition. Advances in neural information processing systems 18 (2006): 41. JEC S nchez. A complex network approach to analyzing the structure and dynamics of power grids. Doctoral dissertation. The University of Vermont, 2009. A.Garas, F.Schweitzer, and S.Havlin. "A k-shell decomposition method for weighted networks." New Journal of Physics 14.8 (2012): 083030.
In this paper, first, the idea of creating a correlation network based on N-1 criterion is proposed. The original transmission lines in power grid become nodes in this special network, while the links between nodes reflect the interaction between transmission lines. Therefore, the problem of locating KTL in power grid turns into identifying key elements in such derivative network. Among so many metrics for finding key components in complex network, degree centrality is used quite often even though it has limitation. Next, to overcome the limitation, this paper modified the classic K-core decomposition both to retain the merits of this very method in locating key spreaders and to make it better accommodate to the power systems inherit characteristics. Further, the modified K-core decomposition is utilized to identify KTL in the New England power system and the result is compared with that harvested from degree centrality. By the comparison, a better correspondence to the result of cascading failure simulation shows the reliability of the outcome of the method we proposed. Finally, the verification on a larger IEEE-118 system demonstrates that this approachs effectiveness does not diminish with the augment of systems scale. Since in the approach we proposed, only the N-1 information is used, which means the formation considered merely the original power flow and N-1 data; the information loss is inevitable and thus it might results in some unidentified key transmission lines. Fortunately, the current approach can guarantee us that all elements in the core area are vital lines even there exist a small portion of key lines remain unidentified. Moreover, the accuracy of identification might be influenced by the systems original power flow, we assume that the method performs better in relatively heavier load level system yet this still need future investigation. REFERENCES
[1] A.Singh, and S. Aasma. Grid failure in Northern, Eastern and NorthEastern grid in 2012: Cause & its effect on economy of India An Review. , 2012. Final report on the August 14, 2003 blackout in the United States and Canadacauses and recommendations. Washington, D.C.: U.S. Dept. of Energy, 2004.
[16]
[17]
[2]
Smart Insular Grids: Opportunities and Challenges

Ozan Erdinc and Joo P. S. Catalo
University of Beira Interior, Covilha, Portugal, INESC-ID, Lisbon, Portugal, and IST, University of Lisbon, Lisbon, Portugal E-mail: oerdinc@yildiz.edu.tr; catalao@ubi.pt
AbstractThe ever increasing environmental concerns together with depletion risk of conventional fossil fuels that can threat the energy independence of energy importing countries have led to more interest in renewable energy sources of locally produced energy. Increase in such sources together with new types of integrated loads has required new concepts to operate bulk electricity power grid. Accordingly, a relatively new concept - the smart grid - has been widely publicized recently with huge investments of both developed and developing country governments in this area. The smart grid concept is currently in the test phase and insular grids have been generally considered as one of the most suitable structures to evaluate the applicability of "smart" solutions. Thus, the general structure and applicable smart grid based strategies are presented in this study to promote the overall operating performance of such systems. The opportunities as well as the possible challenges of smart solutions are discussed, and the future requirements are evaluated. Keywords-demand response; electric vehicles; energy storage; insular grids; renewable energy; smart grid.
Mehmet Uzunoglu and Ali Rfat Boynuegri

Yildiz Technical University, Istanbul, Turkey E-mail: uzunoglu@yildiz.edu.tr; alirifat@yildiz.edu.tr Fan et al. [4] provided a theoretical small-scale insular grid analysis in the smart grid concept for Pulau Ubin Island of Singapore with a discretely located solar energy production facility via photovoltaic (PV) panels, battery based energy storage systems, diesel generator based back-up units and several households. Jeju Island, the test bed determined by the Korean Government for smart grid applicability assessment, has been analyzed by Yoon et al. [5] and Kim et al. [6] from different perspectives. The EcoGrid EU funded project for the Danish Island of Bornholm has been presented by Jorgensen et al. in [7]. Moreover, several different demonstration projects evaluating the smart grid suitability within island based test beds from several points of view have been introduced to the literature recently. From this brief overview, it will be clear that the smart grid application is the concern of several countries with test demonstrations in several places of the world. This paper aims to give an overview of a "smartly" operated insular grid structure. The main parts of such a structure, which can even be evaluated as a microgrid, is demonstrated from production to end-users. General and particular smart grid solutions that can enhance the energy performance of the system, from production to consumption, are discussed. It is also to be noted that this study is also a part of an EU funded FP7 project, with acronym SiNGULAR, aiming to apply smart grid solutions to insular grids located in different parts of Europe [8]. The paper is organized as follows: Section II gives the general presentation of the main components of a smart insular grid structure. Afterwards, Section III includes the solutions applicable for such structures in the smart grid concept. Lastly, concluding remarks are presented in Section IV. II. COMPONENTS OF A SMART INSULAR GRID STRUCTURE The "smart" insular grid structure is composed of three levels namely production level, distribution level and end-user level, including different types of end users together with smart opportunities as shown in the block diagram given in Fig. 1. This section gives a brief overview of those hierarchical levels. A. Production Level The production level consists of different energy production options, including both conventional and new generation alternative/renewable energy systems. In addition, different large-scale storage options should be considered for ensuring the power balance between production and consumption for all conditions.
I.
INTRODUCTION
Due to the change in the former structure of the bulk electricity power grid with additional new generation facilities and load types, power system deregulation has become a widely regarded research area with new approaches and concepts. Among such new concepts for building the basis of the future electricity grid structure, the smart grid concept has recently drawn much attention from researchers and investors of both academic and industrial areas [1]. Smart grid is the term used for a fully automated power grid, monitoring and controlling every node of production, transmission, distribution and consumption with a two-way energy and information flow within all points of the grid structure. Increased penetration of non-dispatchable renewable energy sources and loads corresponding to electric vehicles (EVs) have led to a requirement of such data collection, monitoring and real-time control actions within the smart grid concept in order to reduce the necessity of large scale investments in new power plants [2], [3]. Application of smart grid technologies in insular areas is growing with construction of different test systems installed on different islands of several regions in the world. Even though the interconnected bulk power system structure is considered to be more appropriate in terms of stability, insular areas that can also create an important basis for possible islanding operation requirements are considered as one of the best test beds for the pre-evaluation of the smart grid concept before adopting it to wider areas such as cities with large populations, regions composed of several cities, or even to a whole country.
This work was supported by FEDER funds (European Union) through COMPETE and by Portuguese funds through FCT, under Projects FCOMP01-0124-FEDER-020282 (Ref. PTDC/EEA-EEL/118519/2010) and PEstOE/EEI/LA0021/2013. Also, the research leading to these results has received funding from the EU Seventh Framework Programme FP7/2007-2013 under grant agreement no. 309048.
Figure 1. Block diagram for an insular smart grid structure.
The formerly and mostly utilized way of energy production is by employing conventional types of energy conversion systems fuelled by petroleum, coal, natural gas, etc. A preconstructed conventional energy production system may serve as a part of the production level in smart insular grid structures. However, especially for insular grids where the source of such conventional fuels may be finite or unavailable and also significantly costly to supply from other regions with building a new infrastructure or using the existing infrastructure, the importance of alternative/renewable energy production facilities is growing rapidly, ensuring the chance of local source utilization [9]. Different alternative/renewable energy production facilities have been employed for remote power consumption centers like insular grids. Conventional diesel generators have been widely utilized, especially as back-up for these structures. Besides, several renewable energy production options exist nowadays that can be especially appropriate for insular grids. The wind and solar energy conversion systems via wind turbines (WTs) based wind farms and PVs based solar farms are the leading systems for this area, similar to the renewable energy facility installations all over the world [10]. Another alternative, namely tidal and wave power utilization, especially for insular areas located within oceans, seems promising as there is a great source of energy in wave movements in oceans and tidal actions are also usual for such lands.
In addition, the mature geothermal energy technology can be another well-employed option where there is a sufficient geothermal source. Furthermore, biomass energy production is also a topic of research for many areas that can be especially suitable for insular grids with rural areas, including sufficient usable wastes. On the other hand, according to the reports of the Department of Energy (DoE) in the USA, all generation and storage options should be considered in a smart grid structure [11]. Thus, the role of energy storage in new smart grid structures cannot be neglected. Recently, large-scale storage options such as compressed air or pumped water based storage systems have been utilized with a growing ratio. Such energy storage options can be utilized for storing hundreds of MWh of energy. However, the geographical dependence of such systems is a major drawback as a suitable underground storage space and an inclined land with multiple water storage options in different levels are necessary for compressed gas and pumped hydro systems, respectively. Large-scale new types of battery units are also considered as an option especially for smoothing the fluctuating power output of large power WTs. A 245 MWh Sodium Sulfur (NaS) battery system is available in Japan for ensuring dispatch of a 51 MW wind farm [12]. New generation lithium-ion batteries for EVs will also be a significant option of energy storage, particularly when the penetration of EVs on road transport reaches a considerable level.
As another storage option, the flywheel technology has already been employed in different areas, especially for rapid response to frequency deviations caused by the fluctuating nature of non-dispatchable renewable energy sources. As a possibly leading future option, the hydrogen storage systems composed of electrolyzer-large scale hydrogen tank-fuel cell (FC) combinations have been given importance in the research studies realized, especially on the cost reduction and safety improvement of the mentioned systems. The generated hydrogen can also be utilized directly for powering FC based hydrogen vehicles, which is also a topic of research for more than a decade. Lastly, the main grid connected to high-voltage main transmission system, generally with subwater electricity cables and a bulk power generation structure, also serves as a possible source of energy for insular grids. However, the effort to use more local energy production options will, in turn, reduce the utilization ratio of the bulk power system. B. Distribution and End-user Levels The mature transformer technology serves as the heart of the distribution level for insular grids. Distribution level includes many of such transformer units, each with a different power rating. Being generally connected to the production level via these transformer units, the end-user level is considered as the end of energy chain. The end-user level name in an insular grid structure was formerly called as consumption level due to the fact that the consumers are just consuming energy without the facility of producing power. However, as the applications of small scale grid tied renewable energy systems (roof top PVs, small WTs, pico-hydro systems for areas located near a small river, etc.) recently increased, the consumers have gained the chance of also selling energy back to the grid in low demand-high production periods. As an example, PV based power production is at the highest level at noon times when generally the consumption is at the lowest due to the fact that most people are out of their home during working hours. Thus, this is the reason that the name of this level has been changed from consumption to end-user level. The mentioned end-user level of an insular grid structure surely includes different kinds of end-users. These end-users can be listed under two main types: residential end-users and industrial end-users. Individual residential or industrial end-users have their own appliances/devices that can be classified under three main categories [13]: Thermostatically controllable appliances. Non-thermostatically controllable appliances. Non-controllable appliances. Controllable appliances provide the leading opportunity for smart grid operations. The thermostatically controllable devices include heating and ventilation air conditioners (HVACs), water heaters, etc., for residential end-users that can be switched from on to off position and vice versa and curtailed during peak periods of power demand, also considering the comfort settings of the residential owner. Besides, non-thermostatically controllable appliances can be controlled by positioning the appliance to work consuming a higher or lower power value without just totally opening or closing the appliance.
Washers, dryers, etc., for residential areas can be given as an example for such a kind of appliance. On the other hand, the non-controllable appliances are called as "non-controllable" due to the fact that these appliances, such as TVs, lights, etc. have a comfort lowering impact on consumers if controlled without the permission of the end-user. It is well known that there is a chance of automatic dimming for the lightening of an end-user unit, but this is considered in the concept of energy efficiency investments. As a new type of end-user appliance/load, EVs have recently gained more importance as electrification of the transport sector; as a major fossil fuel consumer, its a hot topic of environmental sustainability. EVs have a different structure with challenges/opportunities that should be examined in detail. As a load, the energy need of EVs can reach to the levels of new power plant installation requirements. The recommended charging level of a Chevy Volt as a small sized EV is 3.3 kW [14], which can even exceed the total installed power of many individual homes in an insular area. Thus, a special importance should be given to EV charging with smart solutions in the existing literature, especially with the consideration of high level EV penetration. On the contrary to the charging power challenge of EVs, the opportunity of using EV as a mobile storage unit with the vehicle-to-grid (V2G) concept may provide some opportunities, especially for shaping the demand without adding stress to transformers and accordingly to the bulk power system. Analysis of EVs is expected to have an important place in a better deregulation of electric power systems. III. "SMART" OPPORTUNITIES AND CHALLENGES Smart grid structures have the potential to offer many opportunities in all parts of the bulk power system. In insular grids, the main opportunity lies in two different points of view considering end-user level local regulation activities and the overall system operator strategies to efficiently and effectively manage the power flow within insular grid structures. These two interconnected points are discussed as follows. A. End-user Level Local Regulation Activities The end-user level activities aiming to support the smart grid opportunities mainly lies on demand response (DR) strategies employed in end-user local areas. DR is a term defined as Changes in electric usage by end-use customers from their normal consumption patterns in response to changes in the price of electricity over time, or to incentive payments designed to induce lower electricity use at times of high wholesale market prices or when system reliability is jeopardized by DoE and is composed of incentive based programs and price based programs (time-of-use, critical peak pricing, dynamic pricing, and day head pricing) [15]. The utilization of DR strategies is considered as mature for industrial customers, but this is a relatively new concept for employing in residential areas responsible for nearly 40% of energy consumption in the world [16]. There are many supporting devices/technologies for DR activities in residential areas. Especially, home energy management systems (HEMs) and smart meters have the leading role for effectively applying DR strategies.
Smart meters are new generation electronic meters that have the capability of two-way communications with the utility/system operator. For DR activities, smart meters can receive signals from the system operator, such as the maximum allowable level of power in a certain period (especially reducing the possible foreseen stress on the relevant transformer) or price signals determined in a dynamic way, and can share this information with HEMs. HEMs also receive information signals from smart appliances and/or smart plugs, if available, including state of the appliance, power consumption, etc. Also, the power production information of available renewable generation facilities is received by HEMs. Moreover, the state-of-energy value information of the EV after returning back to home from work is also transmitted to the HEMs. Evaluating this information from different parts of a smart end-user area, HEMs techno-economically decide the operation strategy for the on-board controllable appliances considering the comfort level and priority settings of the enduser, price of buying and selling power to grid, etc. It should also be mentioned that the communication infrastructure (called home area network HAN - for such small scale enduser areas) plays a crucial role in the proper operation of the mentioned management strategy. A block diagram of HEM operation is shown in Fig. 2. B. Overall System Operator Strategies The system operator in insular grids also has a vital role as a supervisor of the above mentioned DR activities and other system operating strategies. There are three main interconnected hierarchical parts of system operator role: forecasting, planning and operating, as shown in the system operator activity block diagram in Fig. 3.
First of all, forecasting part of the system operator activity provides a basis for all planning and operating activities. There are two main types of forecasting activities: Load forecasting and production forecasting (especially for non-dispatchable renewable energy conversion units). Load forecasting is mainly based on evaluating the end-user energy consumption habits differing for many types of enduser profiles. Together with the deployment of new load types like EVs that have different characteristics from conventional daily life end-user loads, load forecasting issue has become significantly more complex and subject to several recent literature studies [17], [18]. The examination of the daily habits for such new loads should be given a specific importance apart from the usual load types with a long term data collection of utilization habits. Load forecasting is an important topic for the system operator in order to effectively regulate the possible requirements of commissioning different power plants. On the other hand, production forecasting regarding renewable energy systems is recently another topic of interest for research studies [19], [20]. Apart from the more predictable nature of solar energy conversion systems having usual characteristics throughout the day (neglecting the instantly shading conditions during cloudy days), wind energy conversion systems are given a specific importance as the wind speed profile can be extremely variable even in minute scale. To overcome the possible fluctuating nature of these systems, especially in the planning phase, a forecasting strategy with a superior performance can be significantly helpful for the system operator. The planning part of the system operator strategy can be considered as the heart of system operator activity success. Evaluating the possible load variations and production facilities, the system operator provides a plan for a future operating horizon of the system (for instance, week ahead). Taking the possible load profile for the planning horizon into account, the system operator considers the best options to supply this load in the most economical way. These options include forecasted renewable facilities, conventional sources of energy and different kinds of storage options. The power quality of bulk power system connection point of the insular grid should also be considered in this planning part. The possible risks are all pre-evaluated for insufficient energy supply, overloading of transformers, etc., in the planning part. The definition of future electricity market prices is also realized in this part of the activity. Microgrid structures also represent a main topic of many recent literature studies, generally based on unit-commitment based scheduling at production level. Lastly, the operating part of the system operator activities is the observable results of the forecasting and planning performance of system operator activities. The system operator observes the system in real-time and takes the necessary regulations during the deviations from the planned performance in the operating part. The main grid connection point power quality issues should also be a prior concern of the system operator, such as fault conditions that may lead to disconnection of the microgrid from bulk power system and accordingly bringing upon islanded mode of operation.
Figure 2. Block diagram of a general HEM operation.
Figure 3. Block diagram of system operator activity in an insular smart grid structure.
The settings of protection equipments should also be adapted to such new operating conditions with smart adaptive control approaches, since using the settings (for instance, the current value setting of an overcurrent relay) in grid connected operation as the same in islanded operation mode may result in misoperation of protection equipments. Besides, within the operating part, there may be a conflict of automatic and man-handled operation of the system in the possible future penetration of smart grid technologies. The main idea of the smart grid leads to the "smart" automatic control of the overall system. Hence, the system operator should be a "smart" overall energy management center with embedded optimization algorithms in such a case. However, it is considered that getting all the control from a human may not be effective as expected and this issue may cause a conflict for regulating the balance between man-handled and automatic control ratio for sustaining the expected management performance. Here, the communication infrastructure (called wide area network WAN - for such a large-scale application area) is reasonably important for effective system operator performance, similar to end-user level local activities. It should be noted that building a cyber-secure infrastructure is a major challenge for such insular smart grid construction activities. Besides, a reasonable level of investment is also necessary for employing the other necessary components of a smart grid structure (smart meters, HEMs for each local end-user areas, monitoring devices for production and transformer units, etc.) apart from the communication infrastructure. Together with the adoptability improvement of the overall smart structure to new installations of production and consumption units, these may be considered as possible challenges to overcome for effective penetration of the smart grid concept. IV. CONCLUSIONS The recently growing importance of smart grid concepts has been attracting many researchers and also investors from academic, industrial and regulator areas. The implementation of smart grid technology still needs test bed based preevaluations. In this context, insular grids are widely recognized as a suitable test bed for applicability examination of this hot topic. A general overview of such "smart" insular grids was presented in this paper. The structure of insular electricity grids and areas where smart solutions can be employed were discussed. It is accepted that the demand side solutions are a new source of energy for the electricity regulator. However, the role of system operator performance is also significantly important, not less than the demand side local consumption decrement activities. Different conditions of system operation including fault events are also necessary to examine in order to have a better vision of the system operator performance. There are different areas of challenge, mostly in the terms of investment and technology development and at least a midterm period of system performance observation would be necessary before the bulk system totally upgrades to a smart grid. However, smart grid solutions seem to be attractive, especially for developing efficient microgrids like insular areas in the near future.
REFERENCES
[1] R. Moghe, F. C. Lambert, and D. Divan, "Smart stick-on sensors for the smart grid," IEEE Transactions on Smart Grid, vol. 3, pp. 241-252, 2012. K. Tanaka, K. Uchida, K. Ogimi, T. Goya, A. Yona, T. Senjyu, T. Funabashi, and C. H Kim, "Optimal operation by controllable loads based on smart grid topology considering insolation forecasted error," IEEE Transactions on Smart Grid, vol. 2, pp. 438-444, 2011. Y. Guo, M. Pan, and Y. Fang, "Optimal power management of residential customers in the smart grid," IEEE Transactions on Parallel and Distributed Systems, vol. 23, pp. 1593-1606, 2012. Y. Fan, V. Rimali, M. Tang, and C. Nayar, "Design and Implementation of stand-alone smart grid employing renewable energy sources on Pulau Ubin Island of Singapore," 2012 Asia-Pacific Symposium on Electromagnetic Compatibility (APEMC), pp. 441-444, Resorts World Sentosa, Singapore, 21-24 May 2012. D. H. Yoon, B. Y. Shin, C. S. Song, H. Lee, and G. Jang, "Study on Jeju Island power system considering multiple HVDC," 2010 IEEE Power and Energy Society General Meeting, Minneapolis, US, 25-29 July 2010. T. Kim, S. K. Park, and B. G. Lee, "What is appropriate strategy for smart grid business?: A Case Study of Test Bed in Korea," 5th International Conference on Ubiquitous Information Technologies and Applications (CUTE), Hainan, China, 16-18 December 2010. J. M. Jorgensen, S. H. Sorensen, and P. B. Eriksen, "EcoGrid EU - A prototype for European smart grids," 2011 IEEE Power and Energy Society General Meeting, Detroit, US, 24-29 July 2011. Smart and Sustainable Insular Electricity Grids Under Large-Scale Renewable Integration SiNGULAR FP7-EU Project [http://www.singular-fp7.eu/] M. Liserre, T. Sauter, and J. Y. Hung, "Future energy systems," IEEE Industrial Electronics Magazine, pp. 18-37, March 2010. Q. Fu, L. F. Montoya, A. Solanki, A. Nasiri, V. Bhavaraju, T. Abdallah, and D. C. Yu, "Microgrid generation capacity design with renewables and energy storage addressing power quality and surety," IEEE Transactions on Smart Grid, vol. 3, pp. 2019-2027, 2012. B. Fahimi, A. Kwasinski, A. Davoudi, R. S. Balog, and M. Kiani, "Charge it," IEEE PES Magazine, pp. 54-64, July/August 2011. B. Roberts, "Capturing grid power," IEEE Power&Energy Magazine, pp. 32-41, July/August 2009. P. Du, and N. Lu, "Appliance commitment for household load scheduling," IEEE Transactions on Smart Grid, vol. 2, pp. 411-419, 2011. M. Pipattanasomporn, M. Kuzlu, and S. Rahman, "An algorithm for intelligent home energy management and demand response analysis," IEEE Transactions on Smart Grid, Vol. 3, pp. 2166-2173, 2012. S. Borlease, "Smart grids: Infrastructure, technology and solutions," CRC Press, Boca Raton, 2013. K. J. Chua, S. K. Chou, W. M. Yang, and J. Yan, "Achieving better energy efficient air conditioning - A review of technologies and strategies," Applied Energy, vol. 104, pp. 87-104, 2013. L. Ghelardoni, A. Ghio, and D. Anguita, "Energy load forecasting using empirical mode decomposition and support vector regression," IEEE Transactions on Smart Grid, vol. 4, pp. 549-556, 2013. R. A. S. Fernandes, I. N. da Silva, and M. Oleskovics, "Load profile identification interface for consumer online monitoring purposes in smart grids", IEEE Transactions on Industrial Informatics, in press. J. P. S. Catalao, H. M. I. Pousinho, and V. M. F. Mendes, "Hybrid wavelet-PSO-ANFIS approach for short term wind power forecasting in Portugal," IEEE Transactions on Sustainable Energy, vol. 2, pp. 50-59, 2011. A. Botterud, Z. Zhou, J. Wang, J. Sumaili, H. Keko, J. Mendes, R. J. Bessa, and V. Miranda, "Demand dispatch and probabilistic wind power forecasting in unit commitment and economic dispatch: A case study of Illinois," IEEE Transactions on Sustainable Energy, vol. 4, pp. 250-261, 2013.
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9] [10]
[11] [12] [13]
[14]
[15] [16]
[17]
[18]
[19]
[20]
A New Power Line Communication Modem Design with Applications to Vast Solar Farm Management
Ahmed J. Abid, Ramzy S. Ali
Electrical Engineering Dept. University of Basrah Basrah, Iraq
Fawzi M. Al-Naima
Computer Engineering Dept. Al-Nahrain University Baghdad, Iraq
Z. Ghassemlooy, Zhiwei Gao

Engineering and Environment Faculty Northumbria University Newcastle upon Tyne, UK
AbstractThis paper presents a new power line communication (PLC) modem design which can control data flow with a new networking strategy to propagate signals for long distance without using extra cabling or signal repeaters. An investigation is also carried on the utilization of twin core power line cables in vast solar farm for controlling solar trackers and data acquisition purposes. Frequency responses of these power lines are analyzed and tested according to different coupling circuits. A simple design is proposed for PLC modem based on such channel characteristics, data flow control strategy, collision avoidance technique and error detection. Keywords- Coupling circuits; Data flow control; PLC; Solar farm management; Twin-core cable characteristics
follows. In Section II, the preliminary aspects of DC power line are introduced. Section III presents the channel calculations, and in section IV the design aspects of the modem are addressed. Section V describes the solar tracking management strategies. Section VI presents the result and discussions. The paper ends with the conclusion in Section VII. II. PRELIMINARY ASPECTS FOR DC PLC
This section presents the twin core cable characteristics and the signal coupling circuits. A. Channel characteristics: It is important to know the used channel characteristics before deciding on the used modulation technique, transmitted and received power, channel coding techniques, the used frequency and the available bandwidth. It is also important to know if the used channel has a frequency selective or a flat frequency response. The adopted channel is a power line cable consisting of twin cores, each with a radius (a) and the distance between the centers of the two cores is (d) as shown in the Fig. 1.
a d
I.
INTRODUCTION
The present revolution in communication systems, particularly stimulated by internet, offers the possibility of much greater monitoring and control for power system, leading to more effective, flexible and lower cost operation [1].Some designs took only channels into consideration [2], or only discussed modulation technique without studying the frequency response of the channel and coupler [3], or went with twisted pair cables but unavailable for high cross section cables[4].To reduce impulsive noises in channels a master controller was employed to control solar tracker running time. A number of costly techniques were presented and discussed in [5] and [6]. In the present paper, the contention-detection and contentionresolution procedures proposed in [7] are improved by adopting a strategy to control data flow based on channel sensing for multiple accesses under supervision of a master controller. The proposed modem is designed based on the frequency response of the twin wire power line with coupler. The channel and coupler have been simulated from 1kHz to 20MHz and the modulation technique and used frequencies have been selected based on the channel response. The system has the ability to control a large number of trackers in vast farm with high efficiency and accuracy. The system consists of a master controller which is programmed to control and acquire data from all the solar trackers via its slave controllers. A new networking protocol is proposed to control data flow and boosting the data for long distance. The remaining parts of the paper are organized as
Figure 1. Twin core power line cable
There are three important factors need to be calculated about this cable namely the capacitor per unit length, inductance per unit length and the characteristic impedance. The first two factors are important to expect the channel response and the last one is to achieve maximum power transfer through the channel. The main equations describing these factors are [8]: The capacitance per unit length: The inductance per unit length Where the approximation is valid for d a. The characteristic impedance (3) (1) (2)
Equations (1), (2) and (3) show that all these factors are proportional to two main dimensions namely (a & d) and there is a tradeoff between the inductance and capacitance of the channel based on these factors; i.e., the increment of distance between cables increases the inductance and reduces the capacitance and vice versa, and the same matter with core diameter. B. Coupling Circuit The coupling circuit performs the connection with the communication channel. It acts as a filter that passes the communication signals, while attenuating out-of-band frequencies. Its design mainly depends on the line characteristics, such as voltage, frequency, wiring style, etc. In some applications, the coupling circuit can be required to provide safety isolation and protection from high voltage disturbances as mentioned in [9]. 1) Capacitive Coupling: The capacitive coupling is the method used in most applications, since it requires direct parallel connection with the communication channel. The main coupling element is a capacitor that blocks the line voltage, while passing the high-frequency communication signals.Fig. 2 shows the capacitive circuit.
and Capacitance =1.4e-10F/m. Two pieces of cable have been used each 500 m long with three couplers, one for the master and two for the slaves. Fig. 4 and Fig. 5 show the circuit diagram and the frequency response for the power line, capacitive coupler and the signal trappers.
Figure 4. Capacitive coupling circuit
Figure 5. Capacitive frequency response `
The capacitive couplers with two different cables length have been searched for a good response as depicted in Table I.
TABLE. I SELECTIVE FREQUENCIES OF THE CAPACITIVE COUPLING Frequency (kHz) Gain (dB)
500m -17.1894 -9.33683 -6.48368 -3.7784 -1.59256 -1.92427 -7.19386 -7.3086 -3.86291 1000m -27.2834 -14.6116 -13.294 -10.1556 -7.60511 -4.86654 -4.50147 -4.77343 -11.7653
Figure 2. Capacitive coupler
2) Inductive Coupling: The inductive couplers are connected in series on the communication channel and inject a PLC current signal on the line. This is achieved through a specialized current transformer using appropriate high frequency ferrites. The transformer also provides galvanic isolation. Inductive couplers can be with or without connection to the communication cable as shown in Fig. 3.
f0 f1 f2 f3 f4 f5 f6 f7 f8
8.9 119.1 236 354 472 591 716 835 941
Moreover, Fig. 6 and Fig. 7 show the circuit diagram and the frequency response for the inductive coupler with the cable used.
Figure 3. Inductive coupler
Figure 6. Inductive coupling circuit
III.
CHANNEL CALCULATIONS
The system frequency response depends on three factors: the power line characteristics, the signal trappers and the used coupler. The channel characteristics and the effectiveness of each coupling circuit on the frequency response of the system have been simulated. The selected twin core copper cable has been used with the following features: cross section = 25mm2, PVC isolated, resistor = 4e-3 Ohm/ m, Inductance =5e-7 H/m
Figure 7. Inductive frequency response
The inductive couplers with two different cables length have been searched for a good response as depicted in Table II.
TABLE. II SELECTIVE FREQUENCIES FOR THE INDUCTIVE COUPLING Frequency (kHz) f0 660 f1 16595 Gain (dB)
500m 1000m
The modem should have its own channel sensing technique to detect the channel for the signal. Also, the system should have auto shut down the carrier frequencies if idle. Finally it is important to prevent the echo signal in transmitting mode.
-32.2167 -43.7027
-31.09 -83.1759
According to the channel response with frequency range from 10 kHz to 20 MHz at different cable's length (500 and 1000 m), the channel response is a selective and there is no flat frequency response range. Before selecting the carrier frequency it is important to observe that there are more items to be taken into consideration like signal trappers which allow the DC power and prevent RF signal. The impedances of such trappers are directly proportional to the signal frequency and coil inductance. Since the coil inductance in power line is a matter of cost and weight, it is therefore important to use a suitable high frequency to achieve the desired impedance. The modulation technique is also important because a narrow bandwidth is needed to achieve a flat response. IV. MODEM DESIGN ASPECTS
Figure 8 Cable equivalent circuit
A. Transmitter Circuitry: 1) Voltage Control Oscillator: The design adopts 74HCT4046A as a voltage control oscillator which has a centre frequency of up to 17 MHz at VCC = 4.5v. This VCO has been adjusted to generate the mentioned frequencies for logic "0" and "1" by selecting the right values for R1, R2 and CEXT. 2) Band Pass Filter: This filter is used to pass the mentioned frequencies only. 3) Power Amplifier: A class AB power amplifier has been used to increase the transmitted signal power. B. Receiver Circuitry: 1) Low Noise Amplifier: The first stage in the receiver is the low noise amplifier. This stage is important because the circuit sensitivity will depend on its capability to detect low signal to noise ratio which leads to increase the range of the transmitted signal in meter. 2) Parallel LC Resonator: This circuit is designed to resonate at logic "0" frequency to make the signal with two voltage levels. 3) Envelope Detector: This circuit is designed to separate the carrier frequencies and get the data signal back.The design used this technique to demodulate the RF signal which is simple and low cost demodulator. An ultra fast diode is used to pass the negative part of the signal followed by parallel RC circuit. It is important to select the appropriate RC time constant to detect the envelope of the required signal. An operational amplifier has been used to regenerate the signal by comparing the signal level with threshold level which is adjusted according to the received signal voltage. The RC time constant should be chosen according to the following equation: (4) Where fh is the highest carrier frequency and fm =9600 Hz. So as a midpoint 1/RC is as 50 kHz or RC =20 sec. 4) Channel sensing circuit:This circuit uses the same technique that has been used in the demodulation circuit. The
The channel frequency response shows that it is a narrow band frequency selective. Hence, it is required to adopt a narrow bandwidth modulation with carrier frequency within these resonated frequencies to achieve lowest attenuation on the receiver side which let the signal propagates for a long distance and cover the required vast area of the solar farm with no need of repeaters. According to these requirements and because the transmitted data in binary format, it follows that there are two nominated modulation techniques, namely the binary amplitude shift keying (BASK) and the binary frequency shift keying (BFSK). To avoid the data collision in the channel, the transceiver should have a channel sensing circuit which detects the carrier frequency in the channel. But that will be difficult in case of using BASK. Also, using a BASK with two voltage levels will still be more complex than BFSK from the implementation point of view because its need for more multiplier and adder circuits. Hence, the design will adopt BFSK with two frequencies one for logic "0" and the other for logic "1". The used frequencies for BFSK should be selected carefully according to the channel response. Table 1 suggests many frequencies could be used in case of capacitive coupling. The channel has been established as an equivalent circuit consisting of 14 LC pi sections, each with a coil of 0.106 m/m, 35.7H and a capacitor of 1nF. This is equivalent to a power line 25 mm2 in cross section and 1000m in length.According to the conducted laboratory test a good response least dependent on the channel length is achieved by using f2 = 236 kHz and f3= 354 kHz as depicted in Fig. 8.
circuit consist of an envelope detector and level comparator, RC time is selected here to detect the carrier only for sensing purposes ( ). 5) Flow control: There are two major matters in transmit and receive data in both directions, the first matter is to shut down the carrier wave in case if there is no data to transmit since TTL level did not turn off the voltage control oscillator (VCO) in logic "0". This is so important because as mentioned before the transceiver used collision avoidance. The other matter is to avoid the echo of the transmitted data to the transceiver itself. The schematic diagram in Fig. 9 shows the proposed PLC modem.
LCD displays time and date on the first row, latitude and declination angles on the second and tilt and polar angle on the third row [13]. The tracker shown in Fig. 11.B has been designed to track the sun in two axes for maximum radiation. The used slave controller is ATmega8 and the master controller is ATmega32.
Control unit CB
Solar Panel # N
Actuator1 Actuator2
Modem Control unit CB
Slave Controller # 1 Solar Panel # 1

Actuator1 Actuator2
Modem
Slave Controller # 2
GPS
Current Sensor
Modem
Master Controller
Legend Grid-Tie Inverter

Power Data PLC
Figure 10. Signal line diagram
Actuator 2 Actuator 1 Control unit
Figure 9. PLC modem
V.
SOLAR TRACKING FARM MANAGEMENT
The solar tracking farm consists of many trackers separated by a specific distance to track the sun in the sky for maximum solar irradiance. To control and monitor these trackers, a master and slave control system has been adopted. This control technique has many advantages since it has a centralized administration, gateway role for PLC network, management of the roles of each device, PLC and IP network hierarchy, easier network supervision as mentioned in [10]. The master controller calculates the sun location based on astronomical equation and the instance time, date and location. This master controller sends the sun location data periodically to all the slaves on the farm. At the same time it acquired the data from all these slaves, as the voltage, current and fault diagnostic code [11] and [12]. Fig. 10 shows the signal line diagram for the distributed units on the farm, shows the master and slave microcontrollers. Each slave controller manages two actuators to adjust the polar angle and the tilt angle of the solar panel. The control box in Fig. 11.A shows the slave controller which controls the two actuators of the tracker. It also displays all the data on 4x20 characters LCD. The photo shows that the
(A)
(B)
Figure 11.A-Slave controller, B- Dual Axis Solar tracker
A. Data Network Design 1) Data Structure: The message consists of the following four parts:head, control byte,message body and message ends. Message head contains preamble data, source & destination address and data length while the message body contains the main data. The control byte is having information about the message body content and the last part is the message end which contains the inner error detection checksum. 2) Data Flow Control a) Sharing the Message: The master sends all slaves a general message which has information about the sun location as shown in Fig. 12. This message has information about the tilt angle, polar angle, tracking type (1= vertical tracking, 2= horizontal tracking and 3= full tracking) [14].
Figure 12. Shared message
b) Dedicated Message:The master controller starts sending an addressable permission for each single tracker to send it data as shown in Fig. 13.A. When the slave tracker receives its own permission, it automatically replies a message containing the following data; panel voltage and current, fault code and acknowledgment of the general message as shown in Fig. 13.B.This acknowledgment gives the master controller an indication about the connection quality.If the general message reaches the slave correctly, the master controller will not send the data back again because it is already expired but it is used to evaluate the connection quality.
2) Flow Control: The microcontroller detects the channel sensing signal before transmitting any data.If this signal is low it means the channel is idle and the transmitter has a green light to transmit the data, if not the microcontroller will back off for a random time and detect the channel again according to N back off strategy. It is also programmed to set the "TxEN"signal high in transmitting mode, this signal enables the VCO output and disables the RX data to avoid echo signal.
Figure 13.Dedicated message
c) Soft controlled repeating: The master control memorizes the reliability of the connection of all the slaves in the farm. If the master control does not receive a reply to the dedication message or receiving one acknowledge for the shared data or the received data with error, the master controller starts to evaluate this salve and reduce its reliability with every time this happen and vice versa. Now the question is this, does this slave not receive the signal because it is so far from the slave? The answer of this question depends on the performance of the adjacent slaves. In case the answer is "Yes", such slave reaches a specific level of reliability and the master control starts sending the data to this slave in an indirect way.The master will ask a nearby slave to repeat the message for the unreachable slave, and then receives its reply to forward it to the master controller. Fig. 14 shows the soft repeating procedure.In Fig. 14.D, if the slave N doesn't answer slave N-1 a fault code will reply the master controller with control byte=0x08H. B. Microcontroller Software Duties The Software part of this design is a complementary part and its software duties can be concluded as: 1) Error control and detection:Two coding techniques are used, namely checksum for the inner data and CRC-16 for the outer data. The microcontroller calculates the checksum and the CRC-16 code before sending the data to the transmitter, This value is transmitted after the End of Packet for error detection purposes at the receiver.
Figure 14.Soft controlled repeater
3) Route Reliability estimation:One of the microcontroller important duties is to estimate the connection reliability level CRL between it and every slave in the field. In the initialization phase all slaves have the same CRL but this level could be an increase or decrease based on the quality of connection which depends on the detected error. 4) Fault detection: When the master controller receives any fault code from the slaves it automatically translates this code for the operator with ability to save some solution under user permission if the same problem happen in the future. 5) Trackers running time management:To avoid impulsive noise slave controller does not run the actuators unless it receives its own dedicated message. VI. RESULTS AND DISCUSSION
The present modem is simple, low cost and able to control the data transmission and reception for long distance without using extra repeaters. The adopted soft repeating technique can easily control the data flow in both directions. The used coding techniques for inner and outer message body can control the error of the transmitted data with high efficiency.
The main controller informs the user about the power supplied with each single tracker and the total power. In case of any fault in the trackers the master controller will inform the user by translating the fault code received from the slave. The master controller asks the user every time a fault will happen to save the treatment in case of the same fault occur in the future. The complete circuit has been simulated using Proteus simulation software. This simulation includes the power line, coupling circuit, modem and microcontroller type ATmega32 for master controller and ATmega8 for the slave controllers. The controller detects the channel sensing signal before transmitting data. If this signal is low then the TxEN signal will be set to enable the VCO out and disable the echo signal. Fig.15 shows the simulated input and output signals of the transmitter.
Voltage Tx
transmitting power or changing the circuit specifications. The proposed design offers a simple solution to control the solar trackers in a vast solar farm. It can monitor the supplied power by each single tracker in the field and having the ability to diagnose the fault in each of them. ACKNOWLEDGMENT Professor Fawzi Al-Naima acknowledges the financial support of the Institute of International Education (IIE). REFERENCES [1] J. Ekanayake, K. Liyanage, W. Jianzhong, A. Yokoyama, and N. Jenkins,
Smart Grid Technology and Applications. UK: John Wiley & Sons, 2012.
[2] M. Lienard, O. Carrion, V. Degardin, and P. Degauque, "Modeling and [3] E. Kabalc, Y. Kabalci, and I. Develi, "Modelling and analysis of a power
Analysis of In-Vehicle Power Line Communication Channels," IEEE Transactions On Vehicular Technology, vol. 57, no. 2, pp. 670-679, 2008. line communication system with QPSK modem," IEEE Electrical Power and Energy Systems, vol. 34, no. 1, pp. 1928, Jan 2012. Assessment of a PLC System for Spacecraft DC Differential Power Buses," IEEE Transactions On Electromagnetic Compatibility, vol. 53, no. 3, Aug. 2011.
RFOUT
[4] F. Grassi, S. Pignari, and J. Wolf, "Channel Characterization and EMC

Time
Figure 15. Transmitter signal
[5] P. Lopes, J. Pinto, and J. Gerald, "Dealing With Unknown Impedance and
The radio frequency signal shown in Fig. 15 is frequency shift keying between the two selecting frequencies which guarantees the maximum channel response. Fig. 16 shows the received signals at low noise amplifier input, resonator output and the envelope detector output.
Voltage
RFIN Resonator Output
Impulsive Noise in the Power-Line Communications Channel," IEEE Transactions On Power Delivery, vol. 28, no. 1, pp. 58-66, Jan. 2013. Noise Characterization of In-Vehicle Power Line," IEEE Transactions On Electromagnetic Compatibility, vol. 50, no. 4, pp. 861-868, Nov. 2008.
[6] V. Degardin, M. Lienard, P. Degauque, E. Simon, and P. Laly, "Impulsive [7] O. Amrani and A.R Rubin, "Contention Detection and Resolution for [8] J. Bird, Electrical Circuit Theory and Technology. Oxford, UK: Elsevier,
2003.
Multiple-Access Power-Line Communications," IEEE Transactions On Vehicular Technology, vol. 56, no. 6, pp. 879-887, Nov. 2007.
[9] F. Hendrik, L. Lutz, N. John, and S. Theo, Power Line Communications:

Theory and Applications for Narrowband and Broadband Communications over Power Lines. Southern Gate, UK: John Wiley and Sons, 2011. House, 2006.
Envelope Detector Output Time
[10] X. Carcelle, Power Line Communications in Practice. London, UK: Artech

Figure 16. Receiver signal
[11] F. Al-Naima, R. Ali, and A. Abid, "Design and implementation of a smart
The received radio frequency signal shows fixed amplitude in spite of the different frequencies and that came from the suitable selected frequencies. The resonated output has a sufficient drop in voltage which makes the digital signal regeneration more easy. VII. CONCLUSION In this paper, a complete solution for PLC modem has been proposed based on existing power line characteristics. The frequency responses of the channel and coupler have been simulated for a wide range of frequencies. The designed modem has advantages of simplicity, cost effectiveness and using new strategy to control the out of reach devices without increasing the
dual axis sun tracker based on astronomical equations," in The European Workshop on Renewable Energy Systems, Antalya, Turkey, 2012. system for a multi-mode solar tracking farm," in The European Workshop on Renewable Energy Systems, Antalya, Turkey, 2012. GPS and Astronomical Equations," in IT-DREPS Conference & Exhibition, Amman, Jordan, May 29-31, 2013, pp. 144-149. Design Based on Power line Communication," in REDEC, Beirut, Lebanon, 2012.
[12] F. Al-Naima, R. Ali, and A. Abid, "Design of a control and data acquisition [13] F. Al-Naima, R. Ali, and A. Abid, "Solar Tracking System:Design based on
[14] F. Al-Naima, R. Ali, and A. Abid, "Smart Home Energy Management:
Profit Based Unit Commitment Using Gbest Artificial Bee Colony Algorithm
Manisha Govardhan, Research scholar
Department of Electrical Engineering S.V. National Institute of Technology, Surat, India manishagovardhan@yahoo.com
Abstract Due to deregulation of power industries, generation companies strive to attain maximum profit without any compulsion of satisfying consumers demand. For generation companies, unit commitment is an optimizing task of scheduling units to acquire maximum profit with minimum operation cost. This paper proposes Gbest Artificial Bee Colony algorithm to solve profit based unit commitment (PBUC). The competence of proposed algorithm is endorsed on IEEE 26 unit test system. The simulation results confirm that PBUC has drawn more profit compared to traditional unit commitment. Keywords- Gbest artificial bee colony algorithm (GABC), Generation scheduling, Profit based unit commitment (PBUC), Unit commitment (UC) .
Ranjit Roy, Associate Professor

Department of Electrical Engineering S.V. National Institute of Technology, Surat, India rr@eed.svnit.ac.in
idea of Gbest Artificial Bee Colony algorithm is described in Section III. Simulation results are analysed in Section IV and Section V comprises conclusion. II. PBUC PROBLEM FORMULATION
I.
INTRODUCTION
Deregulation of power industries modernize earlier vertically integrated electric utilities into horizontally integrated electric utilities which comprises unbundling of generation (GENCOs), transmission (TRANSCOs) and distribution (DISCOs) companies. Deregulation of power system causes competition among the different GENCOs and allows end user entities to decide their supplier and procure energy at cheaper cost [1]. The motive of traditional UC is to obtain minimum cost while satisfying load demand and several fundamental constraints. While in deregulation environment, generation companies (GENCOs) impulse to gain maximum profit without any compulsion of accomplishing consumers demand is referred as profit based UC (PBUC). Many conventional and meta-heuristic techniques have been reported in literature to solve PBUC. These techniques include genetic algorithm (GA) [2], lagrangian relaxation (LR) [3,4], particle swarm optimization (PSO) [5], mixed integer programming (MIP) [6], new approach of muller method [7], differential evolution algorithm (DE) [8], artificial bee colony algorithm (ABC) [9], ant colony optimization (ACO) [10] and memetic algorithm [11]. This paper has implemented Gbest artificial bee colony algorithm [12] to solve PBUC. The outline of this paper is as follows: Section II epitomizes construction of PBUC with elementary constraints and basic
A. System objective PBUC is a crucial optimization problem of power system in deregulated energy market. The chief goal of PBUC is to maximize profit of generation companies (GENCOs) with appropriate allocation of committed units while satisfying several constraints. The objective function of PBUC can be characterized as follows: (1) max PF = RV TC where PF, RV and TC describes profit, revenue and total cost of a system. Revenue of any system in UC can be calculated as:
RV = Pi ( h ) * EP( h ) * ui ( h )
i =1 h =1
(2)
where EP(h) is an electricity price and ui(h) is an on /off status of generating unit i at hour h. While mathematical formulation of total cost can be expressed as [13]:
TC = [ FCi ( Pi ( h ) ) + STCi (1 ui ( h1) )]ui ( h )

i =1 h =1
(3)
where FCi (Pi(h)) and STCi are fuel cost and start-up cost of unit i at hour h, N is number of thermal units and H is total scheduled time horizon. The fuel cost function of a generating unit can be expressed in quadratic polynomial form as:
FCi (Pi (h) ) = ai + bi (Pi( h) ) + ci ( Pi (h) ) 2
(4)
where ai , bi and ci are cost coefficients of fuel cost of generating unit i. The start-up cost of unit i is defined as [14]:
STCi = i + i (1 e (Ti
off
/ i )
(5)
where is hot start-up cost, is cold start-up cost and cooling time constant of a unit. B. Constraints
is
III.
GBEST ARTIFICIAL BEE COLONY ALGORITHM
1) Power balance consrtaint: The addition of total power generation of all committed units should be always less than or equal to the load demand at that particular interval h. (6) N
This section covers the basic idea about the proposed gbest artificial bee colony (GABC) optimization algorithm employed to solve PBUC problem. The ABC focuses on a cluster of honeybees and their food finding habits. The bees are mainly classified into three groups namely employed bees, onlookers and scouts [15]. In ABC, bees fly to hunt food in multidimensional search space. Some bees seek food source depending on their prior experience and some find arbitrarily without using any experience. Employed bees convey their food information to the onlooker bees. The onlookers choose good food sources from those founded by employed bees and they further search food source nearby the selected food source. If there is no enhancement in the food source then scout bees fly and discover the new food source randomly without using experience. The algorithm steps are [12]: Initialization Initial population P= [P1, P2, P3,Pn] of randomly distributed solution is generated from the multi-dimensional search space, where n is the size of optimized parameters and P1, P2, P3,Pn are possible candidate solutions and each solution vector is specified as Pi= [P1,i P2,i P3,i Pj,i ...Pn,i]. Employed bee phase Update process is carried out for each candidate solution Pi. In this process a new food source vi is given by:
P
i=1
i ( h)
* ui(h) PL(h)
where Pi(h) is generated power output and PL (h) is load demand at hour h. 2) Generation limit constraint: Power generation of each generating unit must be within its maximum (Pimax) and minimum (Pimin) range. min max (7)
P i
P i P i
3) Minimum up/down time constraint The unit has to remain ON/OFF for predefined time period before any transition. Ti on MUTi Ti off MDTi (8)
where MUT and MDT are minimum up time and off minimum down time of a unit i respectively . Ti and on Ti are duration of unit i being continuously off and on respectively. 4) Spining reserve (SR) constraint: The sum of reserve and power generated should be lies min max between Pi and Pi of a generating unit.
vi j = Pi j + i j ( Pij Pkj ) + ij ( y j Pi j )
where
(12)
i j is a random number between -1 to 1, j {1,2n}
P i
min
P i (h) *ui ( h) + Ri (h) *ui (h) P i

i (h)
max
(9) (10)
and Pk signifies the another randomly selected solution from the population. i j is a random number between 0 to C and C is a positive constant. yj denotes the jth element of global best solution. After completion of this phase the best obtained solution is stored. Onlooker phase Each onlooker bee selects a food source according to the probability value
R
i =1
* ui ( h ) SR( h )
where Ri(h) is reserve of unit i and SR(h) is spinning reserve at hour h. 5) Ramp up / ramp down constraint: (11) P P RU
i(h) i ( h 1) i
pi = fiti / n fit n where fit indicates the
Pi ( h1) Pi ( h ) RDi
where RU and RD are the ramp up and ramp down limit of unit i. 6) Initial status Before starting scheduling of thermal units, initial condition of each unit must consider.
fitness value of the solution. All onlooker bees further hunts for the better food source in the neighborhood of the selected food source according to (12). Scout phase If the solution is not upgraded for certain trials, the scout bees generate a new random food source and repeat the search process.
IV.
RESULTS AND DISCUSSIONS
Simulation study carried out with IEEE 26 unit test system to solve PBUC using GABC. The parameters settings used to solve GABC are listed as follows: Population Number of trials Maximum iterations Number of onlooker bees 100 20 200 200
$41667 is gained in PBUC (i.e. $1064970-$1023303) as compared to traditional unit commitment which is graphically described in Figure 3.
8 7 6 x 10
4
UC PBUC
Profit ($)
5 4 3 2 1 0 2 4 6 8 10 12 14 16 18 20 22 24
The significant parameters of generating units and 24 hours load demand information are specified in Table.1 and Table.2 respectively [14,16]. The electricity price for 24 hour period extracted from [17] is shown in Figure.1. The simulation is performed for 20 trials and optimum solution is achieved. This paper implements traditional priority list (PL) method based on full load average production (FLAC) cost for appropriate on/off decision of generating units [18]. The units with lower FLAC cost turns on first for profit maximization and vice a versa. Initially GABC has been executed to solve traditional unit commitment. The generation scheduling of all units with associated cost and profit is displayed in Table.3. The convergence characteristic of this base case is given in Figure.2.
45 40
Time (hour)
Figure 3 Profit gained by UC and PBUC
V CONCLUSION Deregulation of power industry generates competitive environment for different GENCOs to achieve maximum profit while consumers are benefited of having different generation option to purchase energy at cheaper cost. This paper has solved profit based unit commitment (PBUC) using gbest artificial bee colony algorithm in deregulated environment. The simulation result confirms that the profit obtained per day by PBUC is $41667 higher than traditional UC. REFERENCES
[1] T. Venkatesan and M.Y. Sanavullah, SFLA approach to solve PBUC problem with emission limitation, Journal of Elect. Power and Energy Syst., Vol. 46, pp. 1-9, 2013. C.W. Richter and G.B. Sheble, A Profit-based Unit Commitment GA for the Competitive Environment, IEEE Trans. Power syst., Vol.15, pp. 715-721, 2000. P. Attaviriyanupap, H. Kita, E. Tanaka and J. Hasegawa, A hybrid LREP for Solving New Profit-based Unit Commitment Problem under Competitive Environment, IEEE Trans. Power Syst., Vol.18, pp. 229237, 2003. H.Y. Yamin and S.M. Shahidehpour, Unit Commitment using a Hybrid Model Between Lagrangian Relaxation and Genetic Algorithm in Competitive Electricity Markets, Journal of Elect. Power Syst. Res., Vol. 68, pp. 83-92, 2003. I. J. Raglend, C. Raghuveer, G. Rakesh Avinash, N.P. Padhy and D.P. Kothari, Solution to Profit-based Unit Commitment Problem Using Particle Swarm Optimization, Journal of Appl. Soft Compu., Vol.10, pp. 1247-1256, 2010. S. De La Torre, J.M. Arroyo, A.J. Conejo, and J. Contreras, Price Maker Self-scheduling in a Pool-based Electricity Market: A Mixed Integer LP Approach, IEEE Trans. Power Syst., Vol. 17, pp. 10371042, 2002. K. Chandram, N. Subrahmanyam, and M. Sydulu, New approach with muller method for profit based unit commitment, Power and Ener. Society General Meeting- Conversion and Delivery of Electrical Energy in 21st Century pp. 1-8, July 2008. C. Kumar and T. Alwarsamy, A Differential Evolution Algorithm with Population Diversity for Solving the Profit Based Unit Commitment, European Journal of Scientific Research , Vol.66, pp. 29-40, 2011.
35
EP ($)
30
25
20
[2]
15 0
10
Time (hour)
12
14
16
18
20
22
24
[3]
Figure 1 Hourly electricity price

18 16 14 x 10
6
[4]
Fuel cost
12 10 8 6 4 2 0 0 20 40 60 80 100 120 140 160 180 200
[5]
[6]
[7]
Number of iterations
Figure 2 Convergence characteristic of base case
[8]
The hourly generation scheduling of PBUC along with profit is displayed in Table 4. It is clear that an appreciable profit of
[9] C.C. Columbus and S.P. Simon, Profit based Unit Commitment: A Parallel ABC Approach using a Workstation Cluster, Journal of Compu. And Elect. Engg., Vol.38, pp.724-745, 2012. C.C. Columbus, K. Chandrasekaran and S.P. Simon, Nodal Ant Optimization for Solving Profit based Unit Commitment Problem for GENCOs, Journal of Appl. Soft Compu., Vol. 12, pp.-145-160, 2012. D.K. Dimitroulas and P.S. Georgilakis, A New Memetic Algorithm Approach for the Price based Unit Commitment Problem, Journal of Applied Energy, pp 4687-4699, 2011. Guopu Zhu and Sam Kwong, Gbest-guided Artificial Bee Colony Algorithm for Numerical Function Optimization, Int. Jour. of Applied Mathematics and Computation, Vol. 217, Issue 7, pp. 3166-3173, Dec. 2010. T. Senjyu, T. Miyagi and A. Saber, Emerging Solution of Large-Scale Unit Commitment Problem by Stochastic priority List, Journal of Electric Power Systems Research, pp. 283-292, 2006. [14] C. Wang and S.M. Shahidehpour, Effects of Ramp-rate limits on Unit Commitment and Economic Dispatch, IEEE Trans. Power Syst., vol.8, no.3, pp. 1341-1350, August 1993. [15] B. Basturk and D. Karaboga, An Artificial Bee Colony (ABC) Optimization, IEEE Swarm Algorithm for Numeric function Intelligence Symposium May 12-14, 2006, Indianapolis, Indiana, USA.
Journal of Appl. Energy, Vol.88, pp. 4687-4699, 2011.
[10]
[11]
[12]
[13]
[16] S.O. Orero and M.R. Irving, Large Scale Unit Commitment using a Hybrid Genetic Algorithm, Journal of Elect. Power and Energy Syst., vol.19, no.1, pp. 45-55, 1997. [17] D.S. Krischen, G. Strbac, P. Cumperayot and Dilemar de Paiva Mendes, Factoring the Elasticity of Demand in Electricity Prices, IEEE Trans. Power Syst., vol.15, no.2, pp. 612-617, May 2000. [18] A.J. Wood and B.F. Wollenberg, Power Generation, Operation and Control, New York: John Wiley & Sons, 1984
Unit 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
P (MW) 2.4 2.4 2.4 2.4 2.4 4.0 4.0 4.0 4.0 15.2 15.2 15.2 15.2 25.0 25.0 25.0 54.25 54.25 54.25 54.25 68.95 68.95 68.95 140.0 100.0 100.0
min
P (MW) 12.0 12.0 12.0 12.0 12.0 20.0 20.0 20.0 20.0 76.0 76.0 76.0 76.0 100.0 100.0 100.0 155.0 155.0 155.0 155.0 197.0 197.0 197.0 350.0 400.0 400.0
max
a ($/MW2) 0.02533 0.02649 0.02801 0.02842 0.02855 0.01199 0.01261 0.01359 0.01433 0.00876 0.00895 0.00910 0.00932 0.00623 0.00612 0.00598 0.00463 0.00473 0.00481 0.00487 0.00259 0.00260 0.00263 0.00153 0.00194 0.00195
($/MW) 25.5472 25.6753 25.8027 25.9318 26.0611 37.5510 37.6637 37.7770 37.8896 13.3272 13.3538 13.3805 13.4073 18.0000 18.1000 18.2000 10.6940 10.7154 10.7367 10.7583 23.0000 23.1000 23.2000 10.8616 7.4921 7.5031
Table 1 System parameters of generating units b c MUT MDT IS ($)

($) 24.3891 24.4110 24.6382 24.7605 24.8882 117.5551 118.1083 118.4576 118.8206 81.1364 81.2980 81.4681 81.6259 217.8952 218.3350 218.7752 142.7348 143.0288 143.3179 143.5972 259.1310 259.6490 260.1760 177.0575 310.0021 311.9102 (h) 0 0 0 0 0 0 0 0 0 3 3 3 3 4 4 4 5 5 5 5 5 5 5 8 8 8 (h) 0 0 0 0 0 0 0 0 0 -2 -2 -2 -2 -2 -2 -2 -3 -3 -3 -3 -4 -4 -4 -5 -5 -5 (h) -1 -1 -1 -1 -1 -1 -1 -1 -1 3 3 3 3 -3 -3 -3 5 5 5 5 -4 -4 -4 10 10 10 0 0 0 0 0 20 20 20 20 50 50 50 50 70 70 70 150 150 150 150 200 200 200 300 500 500
($) 0 0 0 0 0 20 20 20 20 50 50 50 50 70 70 70 150 150 150 150 200 200 200 200 500 500
1 1 1 1 1 2 2 2 2 3 3 3 3 4 4 4 6 6 6 6 8 8 8 8 10 10
RU (MW/h) 48.0 48.0 48.0 48.0 48.0 30.5 30.5 30.5 30.5 38.5 38.5 38.5 38.5 51.0 51.0 51.0 55.0 55.0 55.0 55.0 55.0 55.0 55.0 70.0 50.5 50.5
RD (MW/h) 60.0 60.0 60.0 60.0 60.0 70.0 70.0 70.0 70.0 80.0 80.0 80.0 80.0 74.0 74.0 74.0 78.0 78.0 78.0 78.0 99.0 99.0 99.0 120.0 100.0 100.0
Table 2 Hourly load demand

HOUR
1 1700 13 2590
2 1730 14 2550
3 1690 15 2620
4 1700 16 2650
5 1750 17 2550
6 1850 18 2530
7 2000 19 2500
8 2430 20 2550
9 2540 21 2600
10 2600 22 2480
11 2670 23 2200
12 2590 24 1840
DEMAND HOUR DEMAND
Table 3 Generation scheduling of traditional unit commitment

6 0 0 0 0 0 0 0 0 0 51.6 48.4 54.2 38.5 0 0 0 113.4 107.1 131.6 155.0 0 0 0 350.0 400.0 400.0 20724 86.8 34225 13414 24154 25476 72162 61446 74721 48000 58320 107950 98800 113475 110075 73713 261.0 818.0 349.4 0 0 74.3 23584 32026 35438 37354 38753 36287 400.0 400.0 400.0 400.0 400.0 400.0 400.0 36762 0 110075 73313 400.0 400.0 400.0 400.0 400.0 400.0 400.0 350.0 350.0 350.0 350.0 350.0 350.0 350.0 0 0 55.0 68.9 123.9 140.7 180.9 132.4 350.0 400.0 400.0 36030 85.4 108375 72259 0 55.0 106.7 138.9 138.1 124.6 108.5 68.9 0 55.0 110.0 165.0 68.9 123.9 100.2 155.0 151.4 155.0 155.0 155.0 155.0 155.0 143.4 155.0 155.0 145.2 108.5 86.9 350.0 400.0 400.0 37282 0 99560 62278 131.3 150.6 155.0 155.0 155.0 124.3 150.2 155.0 119.6 128.1 155.0 155.0 155.0 154.8 155.0 145.6 155.0 155.0 154.0 155.0 155.0 128.6 141.9 155.0 155.0 155.0 155.0 155.0 155.0 155.0 151.4 73.4 163.5 141.8 350.0 400.0 400.0 38013 0 100700 62686 0 51.0 100.0 100.0 100.0 32.9 0 44.6 84.3 96.4 51.0 100.0 74.6 100.0 100.0 98.7 97.2 86.1 100.0 73.8 51.0 100.0 100.0 100.0 100.0 100.0 100.0 73.0 87.6 100.0 76.5 77.5 55.7 155.0 114.2 155.0 155.0 108.4 148.9 109.4 350.0 400.0 400.0 36244 0 108375 72130 40.4 76.0 67.9 0 0 38.5 48.4 55.6 72.6 57.5 38.4 37.7 76.0 63.2 51.1 72.3 62.1 74.9 30.0 68.5 24.8 53.6 53.2 46.5 92.7 100.0 82.4 147.6 113.5 155.0 141.5 70.6 68.9 164.4 350.0 400.0 400.0 35748 0 75900 40152 36.8 75.35 74.5 76.0 76.0 57.9 68.3 61.9 76.0 76.0 76 75.7 67.9 76.0 17.9 56.2 76.0 70.9 66.9 72.0 55.3 76.0 76 67.3 0 0 0 0 0 0 0 0 0 0 0 0 0 76 25.2 52.7 63.0 73.4 44.9 100.0 155.0 102.2 155.0 142.5 92.7 111.4 155.4 350.0 400.0 400.0 31652 0 75000 43348 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 76.0 57.6 76.0 38.0 40.5 25.0 88.9 155.0 151.3 155.0 140.6 147.7 68.9 144.4 350.0 400.0 400.0 32746 0 76500 43753 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 10.4 0 0 0 0 0 0 0 0 0 0 0 0 0 11.3 0 0 0 0 0 0 0 0 4.5 9.8 7.2 13.5 18.6 0 0 34.5 74.3 71.0 43.2 44.5 73.0 97.0 145.1 155.0 155.0 155.0 86.7 123.7 116.1 350.0 400.0 400.0 35183 80.0 98800 63536 0 0 0 0 12 0 0 0 0 0 0 0 0 11.2 12.0 0 0 0 0 12 0 0 0 0 0 0 0 0 8.0 3.5 0 0 0 0 0 0 0 0 26.9 76.0 72.4 45.7 92.7 90.6 58.8 152.9 155.0 155.0 143.9 93.0 149.5 17.1 350.0 400.0 400.0 33691 0 74400 40709 0 0 0 0 12 0 0 0 0 0 0 0 0 10.9 10.2 0 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 0 0 0 0 0 0 0 0 0 65.4 59.1 39.0 49.3 57.3 58.2 98.2 97.3 98.3 102.7 155.0 76.5 93.1 0 350 400 400 27044 0 52800 25756 24 0 0 0 0 0 0 0 0 0 67.0 75.1 15.2.0 71.0 0 0 39.2 152.3 63.5 88.6 117.7 0 0 0 350 400 400 20259 0 34040 13780
Unit/ Hour U1
U2
U3
U4
U5
U6
U7
U8
U9
U10
40.3
56.9
56.0
39.3
53.1
U11
43.0
52.6
34.9
23.5
58.4
U12
38.5
U13
U14
U15
U16
U17
95.7
145.1
89.2
144.2
121.6
U18
148.9
70.9
94.3
141.4
121.0
U19
146.4
120.8
110.4
80.9
76.6
U20
78.5
133.5
155.0
120.4
130.5
U21
U22
U23
U24
350
350.0
350.0
350.0
350.0
U25
400
400.0
400.0
400.0
400.0
U26
400
400.0
400.0
400.0
400.0
Total 718775 1836.7 1743915 1023303
FC
18606
18983
18468
18517
19372
STC
81.6
RV
31450
32005
31265
31450
32375
PF
12844
13022
12796
12933
12921
Table 4 Generation scheduling of PBUC

5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 135 155 155 155 0 0 0 350 400 400 18848 0 32375 13527 13747 25216 27110 73618 62446 76952 75553 34225 48000 58320 107950 98800 113475 110075 173.6 181.0 741.0 0 355.4 0 0 20305 22603 30469 34332 35998 36523 34522 36042 74.3 110075 73958 400 400 400 400 400 400 400 400 400 400 400 400 400 400 400 400 350 350 350 350 350 350 350 350 350 400 400 30608 0 108375 77767 0 0 0 0 0 0 0 0 0 0 0 0 0 55 110 115.9 170.9 110.9 55 350 400 400 32810 376.1 99560 66374 0 0 192 192 197 197 197 186 194 193.1 165.9 155 155 131 155 155 155 155 155 154.7 154.4 155 155 107 155 148.1 155 155 155 140.2 152.9 154.1 88.24 178.5 165.4 110 350 400 400 34330 0 100700 66370 155 155 142 155 155 155 155 149.8 155 154.8 149.1 155 155 93.2 148 155 155 155 155 155 149.1 153 0 0 50.4 100 95.25 100 81.91 7.907 58.91 83.93 47.54 50.99 155 143.8 82.69 136.6 158.3 102.8 156 350 400 400 30382 0 108375 77993 0 0 74.1 92.5 100 100 100 93.98 94.45 59.57 99.9 88.72 0 0 97.2 100 100 100 99.85 100 67.11 61.05 100 100 0 38.5 69.7 76 76 0 0 38.5 41.79 38.01 76 47 76 95.56 69.38 59.18 142 154.7 137.6 154.8 118.2 3.801 179.7 350 400 400 32081 0 75900 43819 0 38.5 68.7 75.1 76 76 75.73 76 76 52.88 76 75.03 37.23 38.5 76 69.5 75.9 61.68 76 75.57 76 76 73.39 32.16 56.39 76 76 75.73 75.31 98.61 73.69 72.55 155 155 139.5 155 19.17 0 178.4 350 400 400 33177 0 75000 41823 38.5 76 26.5 65 76 76 74.09 76 76 76 70.12 46.8 75.93 76 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 76 72.66 31.42 76 100 100 72.07 155 155 155 152.3 74.17 0 180.4 350 400 400 30675 0 76500 45825 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 70.39 76 39.03 60.73 73.53 100 100 153.4 152.4 155 155 77.43 55 182.1 350 400 400 32153 244.2 98800 66403 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 76 76 63.17 19.33 79.49 29.77 26 138.9 155 122.9 129.5 122.7 110 181.1 350 400 400 29439 0 74400 44961 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 76 0 0 0 5.493 0 0 155 98.17 155 150.6 155.9 140.9 113 350 400 400 26975 0 52800 25825 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 110.41 129.46 89.77 72.598 172.64 101.15 13.975 350 400 400 21328 0 34040 12712 Total
Unit/ Hour U1
U2
U3
U4
U5
U6
U7
U8
U9
U10
U11
U12
U13
U14
U15
U16
U17
152
132
154.9
154.6
U18
143
153.8
155
135.8
U19
139
155
142.8
117.3
U20
114
139.2
87.36
142.4
U21
U22
U23
U24
350
350
350
350
U25
400
400
400
400
U26
400
400
400
400
FC
18221
18605
18131
18243
676800
STC
2145.36
RV
31450
32005
31265
31450
1743915
PF
13229
13400
13134
13207
1064970
Very Short Term Solar Irradiance Prediction for A Microgrid System in Taiwan based on Hybrid of Support Vector Regression and Grey Theory
Sryang Tera Sarena1), Kuo-Lung Lian1), TsaiHsiang Chen1), Tai-Di Huang1), ), Kuan-Sheng Tung1)
1)
National Taiwan University of Science and Technology Taipei, Taiwan
Ho
2)
Yung-Ruei Chang2), Y. D. Lee2), Yuan-Hsiang
2)
Institute of Nuclear Energy Research, Taoyuan County, Taiwan
Abstract In recent years, energy crisis becomes a global issue. The use of renewable energy in electricity generation has increased significantly. Natural resource such as solar energy is available in large amount, but it is unpredictable. Solar Irradiance Prediction (SIP) is very important to estimate Photovoltaic (PV) power generation. The generated power of PV will affect power dispatch, scheduling, or even the stability of a microgrid system. This paper proposes to use the Hybrid of Support Vector Regression (SVR) and Grey Theory Models for very short term (VST) SIP (in the range of minutes). The proposed model has been validated with the actual measured data and will be implemented in the Energy Management System (EMS) of a microgrid system in Taiwan. Comparing to the models based on either the Grey theory or the SVR, the proposed method yields higher accuracy. Keywords- Grey Theory; Solar Irradiance Prediction; Support Vector Regression
A microgrid system is a small power system consisting of distributed generators, storage system and loads. In order to properly manage the power flow of the microgrid, the energy management system (EMS) needs to predict the power produced by the PV system. In this paper, a prediction model for solar irradiance is proposed. The model is to predict the solar irradiance in a very short term (VST) in the range of minutes. The capabilities of Artificial Neural Networks (ANNs) [1] in non-linear mapping have attracted attentions in the field of time series predicting [2-6]. However, the ANNs take considerable time in data training. More conventional regression methods such as Autoregressive Moving Average (ARMA) [7, 8] or Autoregressive Integrated Moving Average (ARIMA) [8] have been used in several occasions [9-14]. In some cases, the results are even better than those of the ANNs [11]. Nevertheless, the conventional methods require large sample of data in constructing the model. For VST SIP, only one kind of input data is used. Since the interval is small, the changes between values from time to time are not as large as those in a long time interval. The prediction model based on one variable of input data can be modeled by the Grey Theory Model (GM) [15]. The models are constructed from uncertain systems with small amount of data [16] and have been used in many applications such as time series prediction of PV power generation [17], gold price [18], power demand [19], and wind speed [20], as well as block motion estimation [21] and motor drive controller [22]. Nonetheless, there are not enough linear factors involved in GM [16]. Support vector regression (SVR) [23, 24] has been widely applied to solve nonlinear predicting problems by a linear learning machine in many fields [25-31]. It has very good learning performance and generalization ability [32] with less learning time. Still, in some case, the support vectors
I.
INTRODUCTION
In recent years, energy crisis has become a global issue, and there is a strong growth in solar power generation development that requires prediction of solar irradiance to estimate the amount of the solar energy that a photovoltaic system can provide. Although the solar irradiance is relatively constant outside the earth's atmosphere, local climate influences can cause wide variations in available irradiance on the earths surface from site to site. In addition, the relative motion of the sun with respect to the earth will cause surfaces with different orientations to intercept different amounts of solar energy. This phenomenon makes solar irradiance very challenging. Solar Irradiance Prediction (SIP) is very important to estimate photovoltaic (PV) power generation. The generated power of PV will affect power dispatch, scheduling, or even the stability of a microgrid system.
This project is supported by Institute of Nuclear Energy Research, Taiwan (Contract Number: NL1010270)
are sensitive to some certain samples and the result may be distorted if the particular samples are polluted by noise and outliers [33]. The main purpose of this work is to develop algorithms of predicting and modeling of Solar Irradiance that is suitable for photovoltaic system of a microgrid constructed by Institute of Nuclear Energy Research (INER). From the irradiance prediction, solar power produced by the high concentrating PV (HCPV) system can be forecasted. This paper proposes to use the Hybrid of Support Vector Regression and Grey Theory Model (SVR-GM) for VST SIP (in the range of minutes). The accuracy of the prediction is evaluated with the actual data measured and other forecasting methods. II. GREY THEORY AND SUPPORT VECTOR REGRESSION MODEL
Step 4: The GM(1,1) model can be constructed by establishing a first order differential equation for as: and its solution is given by: 1 1 , (8) , (7)
where, 1 denotes the prediction of x at time point (k+1) and the coefficient , T can be obtained by the Ordinary Least Squares (OLS) method: , where 2 3 1 1 1 2 , 3 ,,
T T T T
(9)
A. Grey Theory The Grey Theory [15] is based on the Grey Model, GM(n, m), where n is the order of the grey difference equation and m is the order of the variables. The Grey Model, GM (1, 1) is one of the most frequently used grey forecasting model [17]. The number of the sample data of GM (1, 1) must be equal to four or more. In addition, the data must be taken at equal intervals and in consecutive order without skipping any data. The constructing process of GM (1, 1) model is as follows: Step 1: Denote the original data sequence by
1 , 2 , 3 ,, , 4 , (1)
(10)
(11)
Step 5: Applying the inverse AGO (IAGO) [19], we can have the predicted value of the primitive data at time (k+1)
1 1 1 .
where n is the size of sample data. Step 2: The Accumulated Generating Operation (AGO) [2] sequence, x(1) can be found by (2), (3) and (4). 1 , where 1 and , 2,3, , . 1 , 2 , 3 ,,
(12)
B. Support Vector Regression SVR is a regression method from the class of Support Vector Machine [23]. This method is (2) characterized by the maximal margin algorithm, where a non-linear function is learned by a linear learning machine in a kernel-induced feature space while the (3) capacity of the system is controlled by parameters that do not depend on the dimensionality of the space [24]. (4) Here, we use a function called -insensitive loss function. It is a loss function that ignores errors that are within a certain distance of the true value. Suppose we have a training data: , ,, , , where represents input space. By some nonlinear mapping , the training data is mapped into some feature space, in which a linear function is defined as in (13). , , , (13)
Step 3: The generated mean sequence of x(1), called can be found by (5) and (6) , (5) 1 , 2 ,, is the mean value of the adjacent data, , 2,3, , .
(6)
where,
Here, we predict the solar irradiance in every five minute. SVR-GM is applied in VST SIP. As shown in (14) Fig. 2. In this method, we combine GM and SVR model. minimize Essentially, we compare the error from each method and , whichever yields the best result will give more impact in , subjected to the next prediction. Our proposed SVR-GM is described , 0 as follows: The constant 0 determines the trade-off between the flatness and the amount up to which Step 1: We use two weights in SVR-GM to get more or deviations larger than are tolerated. less impact from GM and SVR, as: By using Lagrange multiplier techniques and , , , the optimization problem of (14) yields the following dual optimization problem. maximize subject to , 0, , 0 , 1 2 1 2 1. (17)
In -SVR, the goal is to find a function that has most deviation from the actually obtained targets for all the training data, and at the same time, is as flat as possible [30]. The meaning of flatness in the case of (13) is seeking a small [34]. One way to ensure this is by requiring:
The meteorological data have been collected since 2009 by the INER and were sampled for every minute. The data for VST SIP is sampled at every five minute. IV. THE PROPOSED MODEL
We get these parameters from several trials. Step 2: for iteration = 1, prediction is made with the GM model. GM prediction is the main prediction at this (15) stage because training is still needed for SVR model. , (18) where P is the main prediction and GM is the GM prediction. Step 3: for iteration = 2, the prediction is based on (16) 1 2 , (19) where SVR is the SVR prediction. Here, more weight is given directly to SVR prediction due to SVRs superiority than GM in previous trials. Step 4: for iteration > 2, if the previous error of SVR prediction is less than the previous error of GM prediction then use (19);otherwise, use (20). 2 1 (20) Details of SVR-GM model are shown in Fig. 2. V. EXPARIMENTAL VALIDATION ,
The decision function takes the form of
The function , corresponds to a dot product in some feature space. III. DATA COLLECTION The microgrid system under study is shown in Fig. 1. It is the first outdoor microgrid test bed, established by INER in Taiwan. The system consists of 21 units of 1.5 kW HCPV, one unit of 65 kW microturbine, and 60 kWh lithium iron phosphate battery.
In this study, the forecasting performance is evaluated in terms of forecasting error, defined as the Mean Absolute Percentage Error (MAPE) which is defined in (21), 100% , (21) is
Figure 1. The Microgrid system of INER
where is the measured solar irradiance data and the forecasted solar irradiance value.
Figure 2. The GM-SVR Model
This numerical example contains actual data and results of VST SIP of three consecutive days (May 29, 2009 until May 31, 2009) using GM, SVR and proposed model. Table 1 shows the values of MAPE when the value of W1 is varied from 0.6 to 1. It is observed that the best result is yielded by setting WI to be 0.949. This will result in 10.1028% for MAPE.
TABLE 1. WEIGHT TRIAL W1 0.6 0.7 0.8 0.9 0.94 0.948 0.9485 0.949 MAPE (%)
GM SVR Proposed Method
0.9495 0.95 0.955 0.96 0.97 0.98 1
11.575 11.575 11.575 11.575 11.575 11.575 11.575
11.216 11.216 11.216 11.216 11.216 11.216 11.216
10.1029 10.103 10.103 10.1052 10.1073 10.1098 10.117
The forecasting performance in this study is also defined in term of Absolute Error (AE), which is: , (22)
11.575 11.575 11.575 11.575 11.575 11.575 11.575 11.575
11.216 11.216 11.216 11.216 11.216 11.216 11.216 11.216
10.68 10.446 10.249 10.128 10.1068 10.1032 10.103 10.1028
where is the measured solar irradiance data and is the forecasted solar irradiance value. In Fig. 3, the AEs of models based on the grey theory, SVR, and proposed methods are compared. It has been noted that the proposed method yield the highest accuracy.
0.6
[9]
GM SVR
0.5
Proposed Method
[10]
IIrradiance [kW/m2
0.4
[11]
0.3
[12]
0.2
0.1
[13]
29
30
31
Date Figure 3. The AE of Five-minutes-ahead Solar Irradiance Prediction results (2009/May)
[14]
[15] [16] [17]
VI.
CONCLUSIONS
The estimation of photovoltaic (PV) power generation is a basic requirement in a microgrid system for the operation planning. PV power generation estimation requires accurate SIP. The use of SVR-GM is proposed for the VST SIP. The accuracy of the proposed methods has been validated by the measured data and other well-known prediction methods such as the grey theory-based and SVR-based models. The proposed method yields better results, and will be implemented in the EMS of the microgrid of INER. REFERENCES
[1] M.T. Hagan and M.B. Menhaj, Training feedforward networks with the Marquardt algorithm, IEEE Transactions on Neural Networks, vol. 5, no. 6, 1994. L. El Chaar, L. A. Lamont and M. Karkoub, Modeling climate parameters for renewable energy applications in the UAE using neural networks, Proc. CIGRE/IEEE PES Joint Symposium, 2009. D. T. Mirikitani and N. Nikolaev, Recursive Bayesian recurrent neural networks for time-series modeling, IEEE Transactions on Neural Networks, vol. 21, no. 2, 2010. W. Yan, Toward automatic time-series forecasting using neural networks, IEEE Transactions on Neural Networks and Learning Systems, vol. 23, no. 7, 2012. J. Pucheta, C. Rodriguez Rivero, M. Herrera, C. Salas and V. Sauchelli, Rainfall forecasting using sub sampling nonparametric methods, IEEE Latin America Transactions, vol. 11, no. 1, 2013. W. C. Yeh, New parameter-free simplified swarm optimization for artificial neural network training and its application in the prediction of time series, IEEE Transactions on Neural Networks and Learning Systems, vol. 24, no. 4, 2013. J. A. Cadzow, ARMA modeling of time series, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. PAMI-4, no. 2, 1982. H. Lutkepohl and M. Kratzig, Applied Time Series Econometrics, Cambridge: Cambridge University Press, 1892.
[18]
[19]
[20]
[21]
[2]
[22]
[3]
[23] [24]
[4]
[25]
[5]
[6]
[26]
[27]
[7]
[8]
[28]
A. J. Conejo, M. A. Plazas, R. Espnola, and A. B. Molina, Dayahead electricity price forecasting using the wavelet transform and ARIMA models, IEEE Transactions on Power Systems, vol. 20, no. 2, 2005. J.L. Torres, A. Garca, M. De Blas, A. De Francisco, Forecast of hourly average wind speed with ARMA models in Navarre (Spain), Solar Energy, vol. 79, 2005. G. Reikard, Predicting solar radiation at high resolutions: a comparison of time series forecasts, Solar Energy, vol. 83, 2009. A. Mauludiyanto, G. Hendrantoro, M. H. Purnomo, T. Ramadhany, and A. Matsushima, ARIMA modeling of tropical rain attenuation on a short 28-ghz terrestrial link, IEEE Antennas and Wireless Propagation Letters, vol. 9, 2010. A. P. Lanari B, P. Poignet and C. Geny, Pathological tremor and voluntary motion modeling and online estimation for active compensation, IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 19, no. 2, 2011. J. Liu, Z. Deng, Self-tuning weighted measurement fusion Wiener filter for autoregressive moving average signals with coloured noise and its convergence analysis, IET Control Theory and Applications, 2012. J.L. Deng, Introduction to grey system theory, The Journal of Grey System, vol. 1, 1989. S. Liu and Y. Lin, Grey Systems Theory and Applications, Berlin: Springer, 2010. Y.Z Li, R. Luan and J.C. Niu, Forecast of power generation for grid-connected photovoltaic system based on grey model and Markov chain, Proc. IEEE Conference on Industrial Electronics and Applications, 2008. E. Kayacan, B. Ulutas and O. Kaynak, Grey system theorybased models in time series prediction, Expert Systems with Applications, vol. 37, pp. 17841789, 2010. C. Hsu and C. Chen, Applications of improved grey prediction model for power demand forecasting, Energy Conversion and Management, vol. 44, pp. 2241-2249, 2003. C.Y. Huang, Y.W. Liu, W.C. Tzeng and P.Y. Wang, Short term wind speed predictions by using the grey prediction model based forecast method, Proc. IEEE Green Technologies Conference, 2011. J.M.Jou, P.Y.Chen, and J.M.Sun, The gray prediction search algorithm for block motion estimation, IEEE Transactions On Circuits and Systems For Video Technology, vol. 9, no. 6, 1999. H.K. Chiang and C.H. Tseng, Integral variable structure controller with grey prediction for synchronous reluctance motor drive, IEE Proc.-Electr. Power Appl., vol. 151, no. 3, 2004. V. N. Vapnik, Statistical Learning Theory, New York: Wiley, 1998. N. Cristianini and J. Shawe-Taylor, An Introduction to Support Vector Machines : and Other Kernel-Based Learning Methods, New York : Cambridge University Press, 2000. J. D. Martn-Guerrero, G. Camps-Valls, E. Soria-Olivas, A. J. Serrano-Lpez,J. J. Prez-Ruixo, and N. V. Jimnez-Torresu, Dosage individualization of erythropoietin using a profiledependent support vector regression, IEEE Transactions on Biomedical Engineering, vol. 50, no. 10, 2003 C. H. Wu, J. M. Ho, and D. T. Lee, Travel-time prediction with support vector regression, IEEE Transactions on Intelligent Transportation Systems, vol. 5, no. 4, 2004. A. Shilton, D. T. H. Lai, and M. Palaniswami, A division algebraic framework for multidimensional support vector regression, IEEE Transactions on Systems, Man, and CyberneticsPart B: Cybernetics, vol. 40, no. 2, 2010. E. E. Elattar, J. Goulermas and Q. H. Wu, Electric load forecasting based on locally weighted support vector regression, IEEE Transactions On Systems, Man, And CyberneticsPart C: Applications And Reviews, vol. 40, no. 4, 2010.
[29] O. A. Omitaomu, M. K. Jeong, and A. B. Badiru, Online support vector regression with varying parameters for timedependent data, IEEE Transactions on Systems, Man, and CyberneticsPart A: Systems and Humans, vol. 41, no. 1, 2011. [30] J. X. Liu, J. Li and Y. J. Tan, An Empirical Assessment on The Robustness of Support Vector Regression with Different Kernels, Proc. Fourth International Conference on Machine Learning and Cybernetics, 2005. [31] C. C. Chang and C. J. Lin, LIBSVM: a library for support vector machines", ACM Transactions on Intelligent Systems and Technology, 2011. [32] C. Jin, Software reliability prediction based on support vector regression using a hybrid genetic algorithm and simulated annealing algorithm, IET Software, 2011. [33] Z. Kou, J. Xu, X. Zhang and L. Ji, An improved support vector machine using class-median vectors, Proc. International Conference on Neural Information Processing, 2001. [34] A. J. Smola and B. Scholkopf, "A tutorial on support vector regression," in Statistics and Computing, vol. 14, Netherland: Kluwer Academic Publishers, 2004, pp. 199222.
Intelligent reserve management with high penetration of wind power

Liu Jin /Yu Jilai /Liu Zhuo
Electrical Engineering Department Harbin Institute of Technology Harbin, China
AbstractThe rapid variation of wind power output is severe where wind power penetration is high. In urgent rapid ramp event period, the rapid ramp event is liable to cause second blow to the power system. In order to face the challenge from high wind power penetration, an approach for realizing the intelligent spinning reserve management is proposed in this paper. The intelligent management not only forecasts the traces in variation process, but also searches for future hidden trouble from the forecasting errors, forecasts the evolution trend, hence the look ahead capability for the worst extremity is enhanced obviously. Hence the intelligent management has the ability to adapt the change from normal state to urgency situation, and to mine the potential of system reserve contribution (SRC). The different reserve demands can be satisfied in different time intervals. Moreover, the approach can realize the target coordination among guarding against the impacts from rapid ramp events, and can enable minimize the additional cost of urgent optimization comparing with conventional optimization. All above-mentioned advantages can create favorable conditions for more renewable energy (especially wind energy) into power grids. Keywords- spinning reserve; wind power; rapid ramp events; intelligent management; distribution feeders
Liu Guangyi
China Electric Power Research Institute Beijing, China ramp rate are found out, not only the commitment and decommitment of peaking units could be implemented up against difficulties caused by rapid ramp events, but also the real time dispatch system may be adjusted so as to enhance the capability for coping with the worsen errors of rapid ramps. In the current running thermal unit commitment framework, it is possible to mine the potential of spinning reserve with high wind power penetration. In other words, there are feasible approaches to increase the system spinning reserve without starting any generating unit [4]-[6]. From the perspective of intelligent management, it is important to manage intelligently the balance of active and reactive power between supply-side and demand-side in power systems [7] [8]. The intelligent management not only traces the variation process, but also seeks for future hidden trouble forecasts the evolution trend. In urgent rapid ramp event period, the system operators are high-strung, they need to know about the worst possible future. Hence it is necessary to enhance the look ahead capability for the worst ready, such as preparation for the second blow out of the early appearance of the latter rapid ramp event. The optimization is carried out under the intelligent spinning reserve management. It realize the targets of fossil energy saving, emission reduction and guarding against the second blow by all means. On one hand, mining the spinning reserve potential by dint of adaptable optimization among the thermal generating units, on the other hand, striving for the supports from demand response, neighboring grids, and pumped storage stations for creating favorable conditions to guard against the second blow. The intelligent management for spinning reserve has added a lot of intelligent functions to the real-time dispatch. The real-time dispatch is an objective application of conventional energy management system (C-EMS), however the C-EMS does not provide supporting platform for intelligent functions. Hence the intelligent management for spinning reserve is difficult to be realized in power grid control center. For working out this problem, we proposed to establish a new objective multi-agent system (MAS) named
I.
INTRODUCTION
Increasing wind energy penetration levels and the resulting variation in wind energy production are becoming major factors in power system operation. Rapid ramping events result from changes in wind power or high speed cut-out events and pose a challenge to the efficient and reliable operation of the power grid [1]. When the system net load rapidly ramps up during peak load period or ramps down during valleyload period, the sudden variations in wind power and the net load forecasting errors could cause noticeable power mismatches. Rapid ramping events in wind energy production will bring the net load forecasting significant difficulties resulting in forecast error worsening. There are effective approaches to schedule spinning reserve optimally [2] [3]. When the uncertainties ranges for balancing capacity, ramping capacity, ramp duration and
This paper was supported by research grants from China National Natural Science Foundation (51077027), and National High Technology Research and Development Program of China(863 Program)( 2011AA05A105), and Science and Technology Research Foundation of SGCC(DZB17201200260).
extended real time dispatch (ERTD) on the multi-agent based distributed computing platform [9] for new generation of EMS (N-EMS) [10]. In ordinary state the ERTD implement conventional real time dispatch functions, when the system is in the urgent rapid ramp event period, the ERTD will implement the intelligent spinning reserve management functions This is our main contribution with respect to previous works such as [4-6]. Besides, the authors of this paper put forward the functional design of the intelligent slave agents of two function MASs (forecasting MAS and optimizing MAS). II. ADAPTABLE OPTIMIZATION The adaptable optimization that is actualized under intelligent spinning reserve management has obvious adaptability. The adaptability is represented in following two aspects. On one hand, for the purpose of realizing the goals of fossil energy saving, emission reduction and guarding against the second blow, different optimizations are executed for different thermal unit subset so as to transfer outputs from a thermal unit subset to another subset flexibly. On another hand, the optimizations can be adapted to time variation of the SRC management state during the course of the urgent rapid ramp event period A. Two subsets of the running thermal units From the viewpoint of spinning reserve management, the set of the running thermal units may be divided into two subsets: subset Srich and subset Slack [4-6]. Srich is the subset whose units are rich in spinning reserve. Slack. is the subset whose units are lacking in spinning reserve. The reserve contribution RC i of thermal unit i belonging to subset Srich in a time interval can be expressed as (1) RC i = Tri , unit i S rich T is the duration of a time interval (5 minutes) ri is its ramp up (or down) rate. Tri equals to its full speed output increase (or decrease) during a time interval of real time dispatch. The reserve contribution RC j of thermal unit j belonging to the subset Slack cannot be expressed as (2) or (3), because these units are lacking in spinning reserve, but it can be expressed as
SRCk = SRCk =
k iSrich
Tri +
Tri +
k jSlack
k jSlack
( Pjkmax Pjk )
(4)
k iSrich
( Pjk Pjkmin )
(5)
RCj = Pj max Pj , RCj = Pj Pj min ,
k j Slack k j Slack
(2) (3)
In above formulas, the first item on right hand equals to the total amount of spinning reserves which can be taken out from the running thermal units belonging to subset Skrich during time interval k. The second item equals to the total amount of spinning reserves which can be taken out from the running thermal units belonging to subset Sklack during time interval k . During peak load period, up reserve contributions of the units belonging to subset Slack can be increased if their generations are decreased, at the same time if generations of the units belonging to subset Srich are increased and the increase may not reduce their up reserve contributions. Over here it is evident that there is possibility for mining fairish augment potential of system reserve contributions (SRC) by shifting outputs from subset Srich to subset Slack , i.e. the unit outputs of subset Slack decrease by SHFT and the unit outputs of subset Srich increase by the same amount. SHFT is the amount of output shifts from subset Slack to subset Srich. The generation decrements in subset Slack are compensated by MW increments from subset Srich such that the SRC is enhanced and the load demand requirement during urgent rapid ramp event situation may be satisfied. The output shifts SHFT will influence the results of conventional optimization for real time dispatch. Objective function of the optimization consists of the costs of fossil fuel and the costs for pollutant emission with respect to the whole set of running thermal units. The objective performances of the running thermal units belonging to subset Slack are better than that of the running thermal units belonging to subset Srich. In order to minimize the additional cost, an approach by name two subset optimization was proposed [4-6]. In the approach, two optimizations are implemented for subset Srich and subset Slack respectively. The objective functions and the inequality constrains are similar to each other, but the equality constrains about the output shift are different obviously. The equality constrain about SHFTk for subset Slack (6) Pik = Pik SHFTk
k iSlack k iSlack
Pjmax, Pjmin are the upper and lower output limit of unit j respectively. System reserve contribution (SRC) is defined as the amount of spinning reserves which can be taken out from the whole set of running thermal units to meet various requirements during a time interval (5 minutes) of real time dispatch. Thus the system reserve contribution of time interval k can be expressed as follows.
The equality constrain about SHFTk for subset Srich + SHFTk Pjk = Pjk
k jSrich k jSrich
(7)
where,
k iSlack
and Pik
k jSrich
are the total output of the Pjk
subset Slack and the subset Srich determined in conventional optimization for time interval k respectively SHFTk is the shifted quantity of outputs from the subset Slack to subset Srich in time interval k. The amount SHFTk of output shifts has a time-variation feature. The states and evolution tendencies of various time
intervals are different, hence their output shifts are different each other. B. Three management state of time intervals On the management of system spinning reserve, the management state of the time intervals during urgent rapid ramp event period may be divided into following three states: 1) SRC enhancing state 2) SRC retaining state 3) SRC discharging state Suppose PLkmax is the maximum net load requirement in look-ahead horizon of time interval k, which is the forecasted net load of a real time dispatch interval whose value is the largest one within the look-ahead horizon. The system reserve contribution SRCk is desired to be sufficient enough. If the system reserve contribution SRC k is not sufficient enough, that is (8) SRCk < P k P k
L max L
management state during the course of the urgent rapid ramp event period. It is helpful to realize the goals of fossil energy saving, emission reduction and guarding against the second blow. When situation is extremely urgent, probably only depend mining the inner potential among the running thermal units is not enough, at this time it is possile to strive for assistances from demand response with the help of electricity market [11]. The concrete way of doing can be expressed as follows. III. APPLICATION OF MAS TECHLOLOGY Multi-Agent System (MAS) has a flexible autonomy feature; hence it is a promising technology in implementation of the platform proposed by reference [9]. As mentioned in the introduction, the opening feature of the platform enables the intelligent management agents to be admitted easily to the new generation of EMS (N-EMS). In order to make the intelligent spinning reserve management can be realized in power grid control center, had better to carry out MAS technology application. N-EMS consists of many application objectives, the conventional real-time optimal dispatch may be an application objective of them. Each of the application objectives may be divided into several functions. These functions are the basic elements of N-EMS. A. Tri-layered MAS architecture in N-EMS A tri-layered MAS architecture is designed by reference [9] for N-EMS. Function MASs reside in bottom layer; they integrate the foundation of the whole N-EMS. Each function has its individual function MAS in bottom layer. The function MASs contains three kinds of basic elements: master agent, slave agent and client agent. Master agents are responsible for distributing calculation task among multiple computers and then collecting and compiling the results from each computer. These agents may be invoked periodically, or driven by contingencies in power system or by a request from client agent. Slave agents are deployed in computers of the cluster, and perform the calculations requested by master agent. These agents complete the assigned task, and then send the results back to master agent. In Fig. 1 the slave agents S1 perform the conventional calculations, and the slave agents S2 perform the calculations about intelligent management. Client agents are deployed in man-machine interaction (MMI) workstations which are responsible for MMI. Individual function MASs in bottom layer communicates and cooperates with each other through blackboard to implement an objective. Assume the ERTD (extended real time dispatch) may be such an objective. In middle layer, there are many objectives MASs. Suppose DR (demand response), ERTD, SSA (steady-state security analysis) etc. were such objective MASs. All the individual objective MASs in middle layer communicate and cooperate
where P is the forecasted net load of interval k, then time interval k must resides in SRC enhancing state. If the task of the SRC enhancing was implemented entirely in former time interval (e.g. interval k-1 or k-2, etc), but from the latest forecasting results the rapid ramp would not occur in interval k, then the interval resides in SRC retaining state. If the task of the SRC enhancing was implemented entirely or the enhanced SRC was retained in former time interval and from the latest forecasting results the rapid ramp would occur in interval k, then interval k resides in SRC discharging state. Owing to timing error of short term forecast, it is possible that the following cases would appear 1) a rapid ramp occurred in interval k which is appointed to reside in SRC retaining state according to forecasting results; 2) The rapid ramp did not occur in interval k which is appointed to reside in SRC discharging state according to forecasting results. In case 1), it is not terrible that the primary frequency regulation will withstand the impact of the rapid ramp successfully because there is a sufficient system reserve contribution retained in time interval k. In case 2), it is not serious also, if the forecasted rapid ramp did not come, the running generating units would not discharge system reserve contribution by the light of nature. If only sufficient system reserve contribution was retained, dont be afraid of the second blow brought by the latter rapid ramp whose coming is earlier than forecasted. Anyhow by dint of the intelligent spinning reserve management, different optimizations are executed for different thermal unit subset so as to transfer output from a thermal unit subset to another subset flexibly, and the optimizations can be adapted to time variation of the
k
L
1, W3 = 0,
SYSTEM MAS (top layer)
( N o u ta g e 1) [( N u p 1) ( N d o w n 1)] O th e r
(11)
New Generation of EMS

blackboard
OBJECTIVE MAS
(middle layer)
It cannot say that as long as the power system resides in one (or two) of the three cases, then the present time is in urgent rapid ramp event period. The sufficient and necessary condition of the rapid ramp event urgency state : CRRE = 1, where,
SSA
DR
ERTD blackboard
1, CR RE = 0,
( C 1 = 1) ( C 2 = 1) O th e r
(12) (13)
FUNCTION MAS bottom layer
M C
C1 can be describe by
S2
S1
S2
1, C1 = 0,
( W 1 = 1) (W 2 = 1) ( W 3 = 1) O th e r
S1
FORECAST
OPTIMIZATION
C2 can be describe by
1, C2 = 0, S R C k < P Lkm a x P Lk O th e r
(14)
Fig.1. Objective MAS ERTD in N-EMS
with each other through blackboard to implement entire NEMS function. B. Slave Agent S2 of Forecast MAS As mentioned above, in Fig.1, the slave agents S1 perform the conventional calculations, and the slave agents S2 perform the calculations about intelligent management. The slave agents S2 of Forecast MAS will perform the intelligent management calculations for cognizance whether the present time interval is in the urgent rapid ramp event period; hence it is a cognitive agent, involves considerable logic calculations and needs to symbolize its environments. The urgent rapid ramp event periods appear mainly in following three cases In case 1, the operating point of power system resides in peak load section of the net load curve, and there are upward ramps on the section. In case 2, the operating point of power system resides in valleyload section of the net load curve, and there are downward ramps on the section. Case 3 appears after some contingency happened; meanwhile there exist rapid ramps of the net load curve. Let W1, W2 and W3 express the circumstance symbols of case 1, case 2, and case 3 respectively. Nup is the number of upward rapid ramps in the look ahead horizon, Ndown is the number of downward rapid ramps in the look ahead horizon, Noutage is the number of outages; THpeak is the threshold value of peak loadTHvalley is the threshold value of valley load, then
1, ( PL THpeak ) ( N up 1) W1 = 0, Other
1, W2 = 0, ( PL THvalley ) ( N down 1) Other
By dint of (12), the slave agents S2 of forecast MAS can cognize whether the present time is in the urgent rapid ramp event period. The intelligent management function of the objective MAS ERTD will be start up and the slave agents S2 of optimization MAS will be awaken as soon as the cognizance is confirmed, heretofore the objective MAS ERTD only implemented still the function of conventional real-time dispatch. C. Slave Agent S2 of optimization MAS In Fig.1, the slave agent S2 of Optimization MAS performs three functions: 1) the logic calculations for cognizance in what management state does each time interval reside ; 2) the adaptable optimization; 3) The intelligent decision of striving for assistances from demand response. The logic calculations for cognizance about management states may be carried out as follows. Let SRCEnhance, SRCRetain and SRCDischarge express the management state symbols of SRC enhancing state, SRC retaining state and SRC discharging state respectively. j expresses the number of the time interval in which PL max k will appear. PL max k is the forecasted net load of a real time dispatch interval whose value is the largest one within the look-ahead horizon of time interval k. V k 1 is the symbol of SRC enhancing (or retaining) state in the time interval( k-1). The sufficient and necessary condition for that the time interval k resides in SRC enhancing state is
SRCEnhance = 1
where
1, S R C En hance = 0,
(15)
(9) (10)
k ( j > k ) (C 2 = 1) (V k 1 = 0) O ther
The sufficient and necessary condition for that the time interval k resides in SRC retaining state is (16) SRCRetain = 1 where
1, S R C R etain = 0,
( j > k ) (V O ther
k 1
= 1)
interval h, then the calculating results may be seen in Table .

TABLE II. CALCULATING RESULTS OF REAL TIME DISPATCH DURING URGENT RAPID RAMP EVENT WITHOUT FORECASTING ERRORS
Time interva l h h+1 h+2 h+3 h+4 h+5 h+6 h+7 SUM
The sufficient and necessary condition for that the time interval k resides in SRC discharging state is
SRCDischarge = 1 where
(17)
(MW) 27120 27213 27456 28012 28526 28399 27975 27893 222594
PL K
SRC (MW)
1055.0 1327.346 1315.397 518.6022 460.122 465.0 465.0 465.0
RClack (MW) 590.0 562.346 550.397 53.6022 220.122 0.0 0.0 0.0
RCrich (MW) 465.0 765.0 765.0 465.0 240.0 465.0 465.0 465.0
SHF T
(MW)
Cost (yuan)
311454.6 312960.5 316173.6 312665.1 329484.9 327773.6 322149.2 321076.3 2553737.8
Add. Cost (yuan)

252.4688 593.3125 732.531 0.0 0.0 0.0 0.0 0.0 1578.3123
1, ( j = k ) (V k 1 = 1) SDischarge = 0, Other As mentioned in Section , the main contents of the adaptable optimizations are the two subset optimizations for different SRC management states. In reference [6], the two subset optimizations were illustrated in detail. The intelligent decision of striving for assistances from demand response can be realized with the help of formula (18)
C DR ( C RRE = 1) ( f ave - min < T H f ) 1, (18) = 0 , Other = 1 express that the demand response mechanism
590 890 690 0.0 0.0 0.0 0.0 0.0
TABLE III. CALCULATING RESULTS OF REAL TIME DISPATCH DURING URGENT RAPID RAMP EVENT WITH FORECASTING ERRORS
Time interval h h+1 h+2 h+3 h+4 h+5 h+6 h+7 SUM
(MW) 27120 27213 27456 28012 27894 27975 28526 28398 222594
PL K
SRC (MW)
1055.0 1327.346 1315.397 518.6022 714.995 694.996 429.873 465.0
RClack (MW) 590.0 562.346 550.397 53.6022 249.995 229.996 189,873 0.0
RCrich (MW) 465.0 765.0 765.0 465.0 465.0 465.0 240.0 465.0
SHFT
(MW)
Cost (yuan)
311454.6 312960.5 316173.6 312665.1 321297.0 322351.4 329483.6 327760.2 2554146.0
Add. Cost (yuan)

252.4688 593.3125 732.531 0.0 207.5938 201.47 0.0 0.0 1987.3761
where C DR has been decided to initiate for providing tertiary reverve service. f ave - min is the average frequency deviation in one
minute. T H f is the threshold value of f ave - min . In smart grid, some distribution feeders own distributed energy resources (DER), they can provide considerable tertiary reserve service through demand response. IV. SIMPLIFIED NUMERICAL EXAMPLE
Suppose the parameters of running thermal units in a power system are shown in Table I.
TABLE I.
Type A B C D E
total Capacity
(MW)
590 890 690 0.0 250 230 0.0 0.0
Suppose there are forecasting errors from time interval h+3 to time interval h+7 in the look ahead horizon of time interval h, as shown in Fig. 2. In this case, the calculating results may be seen in Table III .
PARAMETERS OF RUNNING GENERATING UNITS

Number
j k l n o
ai
0.00979 0.01293 0.00227 0.00239 0.00168
bi
7.884 6.373 9.616 9.385 8.979
ci
531.3 514.5 373.1 368.4 340.8
300 300 600 600 1000 29100
15 16 10 8 9 58
Ramp rate (MW/m) 3 3 6 6 10
e c a b d
These running thermal units take on the net load. In the interest of the energy saving and emission reduction, in Table I, ai, bi, ci are the coefficients of the cost function for a certain thermal unit i, in which f i ( Pi ) is the fossil fuel cost (including emission tax) of thermal unit i during a time interval of real time dispatch (5 minutes). f i ( Pi ) = a i Pi 2 + bi Pi + c i
interval h
Fig. 2. Forecasting errors in the look ahead horizon of time interval h
In Table and Table III, SRC denotes system reserve contribution; RClack denotes reserve contribution of subset SlackRCrich denotes reserve contribution of subset Srich . In the look-ahead horizon for real time interval h, there are 2 ramps observable. If the actual net load is very close to the forecasted results for the look ahead horizon of time
In Table , from the time interval h to the time interval h+7, the system comes through two rapid ramps. The total cost in above 8 time intervals is 2554146.0 Yuan; the total additional cost for output shifting between subsets in above 8 time intervals is 1987.3761 Yuan. The percentage of additional cost is 0.07781%, as compared with other measures (such as starting peak units), it is extremely small.
In Table III , the system comes through two rapid ramps also, but the latter rapid ramp is advanced to time interval h+4(segment ij) due to timing error of short term forecast. Owing to the time interval h, time interval h+1, and time interval h+2 reside in SRC enhancing state, the system reserve contribution has been enhanced sufficiently, and hence it can outlive two blows coming from two succession rapid ramps in time interval h+ 3 and time interval h+ 4 respectively. Although the system reserve contribution enhancement can support the power system to endure the second blow, but it is not necessary to increase the additional cost for output shifting between subsets. Actually in Table III , the total additional cost for output shifting between subsets in above 8 time intervals is 11578.3123 Yuan, less than that in Table .
[6]
[7]
[8]
[9]
[10]
When the latter rapid ramp is advanced to time interval h+4(segment ij), in front of time interval h+4 the net load will begin descanting, hence time interval h+4 and time interval h+5 need not reside in SRC retaining state anymore. In other words, in time interval h+4 and time interval h+5, the power system need not shift outputs between subsets anymore so that the total cost is smaller. V. CONCLUSION It is dangerous if the succedent rapid ramp event occurs early when the strength of power system damnified by previous rapid ramp event is not restored as yet, so that the look ahead capability for the worst extremity must be enhanced. Mining the spinning reserve potential between subset Srich (whose units are rich in spinning reserve) and subset Slack. (Whose units are lacking in spinning reserve), the adaptable optimization can realize the targets of fossil energy saving, emission reduction and guarding against the second blow in urgent RAPID RAMP EVENT period. It has very small additional cost, can create favorable conditions for more and more renewable energy into power systems. The opening feature of the distributed computing platform based on MAS enable the intelligent management agents to be admitted easily to the new generation of EMS, hence the research on multi-agent system (MAS) applications for the approach proposed is quite useful to make the intelligent spinning reserve management can be realized in power grid control center. REFERENCES
[1] C. W. Potter, Potential benefits of a dedicated probabilistic rapid ramp event forecast tool, IEEE / PES, Power System Conference and Exposition, pp1-5, 2009. H.Wu, H. B. Gooi. Optimal Scheduling of Spinning Reserve with Ramp Constraints, Proc. IEEE. Power Eng.Soc. Winter Meeting, vol. 2, pp. 785-790, 1999. H. B. Gooi, D. P. Mendes, K. R. W. Bell et al. Optimal Scheduling of Spinning Reserve, IEEE Transaction on Power Systems,Vol. 14,No. 4, November 1999 L. Chang-yi, L. Jin, L. Zhuo, Studies on enhancing power system response ability for climbing steep peak load, Automation of Electric Power System, Vol. 26 No.2 January 2002, China, pp.34-37 L. Jin, P. Yi, L. Chang-yi, Y. Ji-lai, Coordination Optimal Dispatching in Process of Climbing Peak, Proceedings of the CSEE, Vol. 26 No. 2 January 2006 pp. 36-40
[11]
L. Wei, L. Jin, L. Zhuo, Enhancing Operational Flexibility with Significant Wind Energy Penetration, The annual Electrical Power and Energy Conference (EPEC 2011), Winnipeg, MB, Canada, Oct. 2011. Z. Xudong, L. Jianming, S. Hao et al. Orderly Consumption and Intelligent Demand-side Response Management System under Smart Grid, Power and Energy Engineering Conference(APPEEC) 2010 Asia-Pacific pp.1-4 M. Z. C. Wanik, I. Erlish and A. Mohemed. Intelligent Management of Distributed Generators Reactive Power for Loss Minimization and Voltage Control, 2010 15th IEEE Mediterranean Electrical Conference, pp 685-690 Z. Boming, Z. Chuanlin, W. Wenchuan. A multi-agent based distributed computing platform for new generation of EMS, IEEE / PES, Power System Conference and Exposition, pp1-7, 2009, pp. 1-7 Z. Boming, S. Hongbin, W. Wenchuan. A new generation of EMS implemented in Chinese electric power control centers, Proc. 2008 IEEE PES General Meeting, July 20-24 2008, pp. 1-3 F. D. Galiana, F. Bouffard, J. M. Arroyo, and J. F. Restrepo, Scheduling and pricing of coupled energy and primary, secondary, and tertiary reserves, Proc. of the IEEE, vol. 93, pp. 19701983, Nov. 2005.
[2]
[3]
[4]
[5]
Techno-economic Optimization of Hybrid PV/Wind/Battery/Diesel System Considering Effects of Wind Turbine Parameters on Optimal Wind Turbine Selection Using Genetic Algorithm
Abdullrahman A. Al-Shammaa
Department of Electrical Engineering King Saud University Riyadh, Saudi Arabia ashammaa@ksu.edu.sa
Abstract This paper presents an optimum sizing methodology to optimize the hybrid energy system (HES) configuration based on Genetic Algorithm (GA). The proposed optimization model has been applied to evaluate the techno-economic prospective of the HES to meet the load demand of remote village AlSulaymania. Al-Sulaymania located in the northern part of Saudi Arabia near Arar. The optimum configuration is not achieved only by selecting combination with lowest cost but also by finding a suitable renewable energy fraction that satisfying load demand requirements with zero rejected load. Furthermore, this study investigated the effects renewable energy fraction and wind turbine parameters on optimum wind turbine selection. The simulation results indicated that the optimum wind turbine selection is not affected only by the wind turbine speed parameters or by the wind turbine rated power but also by the desire renewable energy fraction. It found that the wind turbine rated speed has a significant effects on optimum wind turbine selection and the wind turbine that has the lowest cut-in speed and lowest rated speed is the one has the lowest cost of energy. Whereas, the wind turbine rated power has no consistent effects on optimal wind turbine selection. The proposed HES was found to have the lowest cost of energy amongst the other HES systems with 0.1396 $/kWh at 65% renewable energy fraction. The PV/Bat/DG hybrid system is only feasible when wind resource is very limited and solar energy density is high. On other hand WT/Bat/DG hybrid system is only feasible at high wind speed and low solar energy density. Keywords- Hybrid energy system; renewable energy fraction; Genetic algorithm; Cost of energy; Annualized system cost.
Khaled E. Addoweesh
Department of Electrical Engineering King Saud University Riyadh, Saudi Arabia khaled@ksu.edu.sa increase in global temperature include a rise in sea levels and a change in the amount precipitation, as well a probable expansion of subtropical deserts. Consequently, it is necessary to find alternative energy sources to overcome the continuously energy consumption increasing in the last decade while minimize the negative impacts on the environment. The renewable energy resources such as solar and wind energy are being widely used to reduce fuel consumption for electrical power generation either in grid-connected or standalone mode of operation [3]. Depending on the costumers load demand requirements and the availability of renewable energy resources, sometime two sources or more combined together to form HES such as PV/wind turbine/diesel generator system employing battery bank as secondary energy storage. This kind of HES attenuates single sources fluctuations and reduce storage system requirement which in turn reduce overall system investments [4], [5]. Additionally, integrating different energy resources improves system reliability and efficiency and decreases system cost. However, in comparison HES with single energy systems, HES appear to be more complicated due to the use of two different energy resources jointed together. Moreover, the nonlinear characteristics of the different HES components and high number of parameters that have to be considered for the optimum design, and the fact that the optimum system design and control strategy of the system are mutually dependent, makes the optimum design HES complicated. Consequently, optimum sizing methodology is essential to utilize the renewable energy resources efficiently and cost-effectively. The optimum sizing process can assist to guarantee the lowest investment cost with an adequate and efficient use of renewable resources. The hybrid energy systems have been studied comprehensively [6-13]. However, these studies did not consider the effects of wind turbine parameters such as WT speed characteristics and WT rated power on electricity cost. Additionally, few researchers have compared the performance of various HES at a common
I.
INTRODUCTION
The Earth's average surface temperature rose by 0.74 0.18 C over the period 19062005 due to global warming phenomenon [1]. Even though this small rise in earth temperature may not sound like much it has that huge influence, even a half degree increase in the earth temperature can affect the weather and planet operation. The effects of an
location [11]. In order to find the optimal sizing of renewable energy components, fifteen types of WT from different manufacturers with different characteristics were used. Motivated by reducing our dependence on fossil fuel and addressing the shortcoming of previous research into this topic, this paper presents a complete optimization methodology for optimally sizing of HES composes of wind turbines, diesel generator with battery banks as energy storage system using GA. The proposed system intended to supply a remote area located in the northern part of Saudi Arabia. The optimum configuration is not only achieved by selecting combinations with lowest cost, but also by finding a suitable renewable energy fraction that satisfies load demand requirements with zero Loss of Power Supply Probability (LPSP). Four parameters were included in the optimization method; number of PV array, number of wind turbine, wind turbine type and numbers of battery bank. Ease of Use II. RESEARCH METHODOLOGY
GT (1) PPV = PR f PV 1 + P T c T c ,STC GT ,STC Where, PR is the rated power of the PV; fPV is the PV derating factor; GT,STC is the incident radiation at standard conditions; GT is the incident radiation on the tilted PV array; P is the temperature coefficient of power; Tc is the PV temperature; Tc,STC is the PV temperature under standard conditions.
Wind Turbine System Model Choosing a suitable model is very important for wind turbine output power simulation. There are several existing models for estimating wind turbine output power, such as linear, quadratic model and model based on Weibull parameter. In this study, the wind turbine output power is approximated by a quadratic model [15], which can be expressed as:
A. Modeling of the Hybrid Energy System Components A HES modeling is a crucial step before any optimal sizing process. The block diagram of proposed HES is shown in Fig. 1. A General methodology for modeling PV, wind turbine, battery bank and diesel generator is described below.
0 V < Vc , V > V f 2 2 V Vc PW T (V ) = Pr 2 ............Vc V V f 2 Vr Vc Vr V V f 1
(2)
Battery Bank Model Batteries are essential component in the HES since solar and wind resources are not consistently available due to the weather conditions and the time of the day. Hence, battery is sized to meet the mismatch energy. Battery sizing process depends on many factors such as maximum depth of discharge, temperature correction, and nominal battery capacity. Depending on the energy generated by PV and wind turbine and the load requirements, the state of charge (SOC) of battery can be calculated from the following equations [9]: E (t ) SOC (t ) = SOC (t 1)(1 ) + EGA (t ) L bat (4) inv Where, SOC(t) and SOC(t-1) are the battery bank state of charge at time t and t-1; is hourly self discharging rate; EGA is the total energy generated; EL is the load demand; inv and batt are the efficiency of inverter and battery. Diesel Generator Model Diesel generator is required to supply load when renewable energy is not sufficient. The fuel cost of the DG can be calculated as follow [7]:
CDG = CF
t =8760 t =1
F (t )
(5)
Figure 1: Schematic diagram of HES.
Where; F(t) is the hourly fuel consumption and it can be calculated as follow [9]: F ( t ) = APDG ( t ) + BPR (6) Where, PR is the DG rated power, PDG(t) is power generated (kW), CF is the fuel cost per liter and A=0.246 l/kWh and B=0.0845 l/kWh are the fuel curve coefficient [7].
System Constraints Model System Reliability Model In this study, power system reliability is expressed in terms of LPSP [5] which is defined as the probability that an insufficient energy results when the HES is unable to supply the load. This method can be summarized as follows:
PV System Model The hourly output energy of the PV array is given by the following equation [14]. The PV energy is calculated using solar radiation, ambient temperature and manufacturing data.
If EL is lower than WT energy generated (PWT), then surplus energy is used to charge the battery banks via AC/DC and charge controller. If battery banks capacity (CBatt) reaches it maximum level (SOCmax), then abundant energy will be dissipated. If EL is higher than PWT, the deficit energy will be supplied by PV arrays. If both PV arrays and WT failed to fulfill the load requirement, the battery banks will satisfy the deficient energy if battery SOC(t) is higher than minimum level SOCmin. If EL is higher than energy generated from PV and WT and if battery SOC(t) is equal to SOCmin, diesel generators will be started to supply load as follows: If deficient energy is lower than minimum diesel generator operating point (PDGmin), then diesel generator will be operated at its minimum level and surplus will be dumped. While, if deficient energy is higher than PDGmin and lower than diesel generators rated power (PDGrated), deficient energy will be supplied by diesel generator. Otherwise, if deficient energy is higher than PDGrated, diesel generator will be operated at rated power and the rest of deficient energy will supplied by battery. The loss of power supply probability can be defined by [5]:
t =T
The Ccap is given by: Ccap = CPV + CWT + CBatt + CDG
(10)
Where, CPV, CWT, CBat and CDG are the capital cost of PV array, wind turbine, battery bank and DG. The annualized replacement cost can be expressed by: Carep = Crep SSF i, Yrep Where, Crep is the replacement cost of the component, SSF is the sink fund factor, Yrep is the component lifetime.
B. Site Selection The second stage of this study involved the selection of a specific isolated community upon which to base the modeling process, data profile and load demand. The selected remote community was to act as a case study for other communities in a similar isolation. Al-Sulaymania village (Arar, Saudi Arabia) was chosen as it satisfied the criteria of non-connection to the electrical grid, being a small and centralized community, having moderate wind and solar radiation conditions, having information on community living requirements. This village has the basic life requirements, such as schools, mosques, clinic center, water pump, and police center as shown in Fig 2. The existing plant consists of three DG each of 456 kW connected in parallel to a common buss and connects to step up transformer rated of 1250 kVA as shown in Fig. 3.
(13)
(7) T The PFT is defined as the time that the load is not satisfied when the energy generated from energy sources is insufficient and the battery bank is exhausted.
Renewable Energy Fraction Model The REF is defined as the fraction of the energy delivered to the load that originated from renewable sources and it can be calculated using the following equation [14]: EL , DG REF = 1 (8) 100 EL , served Where; EL,DG is load served by diesel generator. Pure renewable system corresponding to REF=100%, while pure diesel system corresponding to REF=0%. So, excluding these boundaries, the remaining values correspond to HES. System Economics Model The economical approach, according to the concept of Cost of Energy (COE), is developed to be the objective function of the system cost analysis in this study and can be expressed as [14]: ASC (9) COE = EL , served The annualized system cost (ASC) is composed of the annualized capital cost Cacap, the annualized replacement cost Carep and the annualized maintenance cost Camain. ASC = Cacap + Carep + Camain (10)
LPSP =
t =0
PowerFailureTime( PFT )
Figure 2: Village Community.
Figure 3: Al-Sulaymania power plant.
The annualized capital cost of each component is given by:

Cacap = Ccap CRF i, Y proj
(11)
Where, Ccap is the initial capital cost; CRF is the capital recovery factor, Yproj is period lifetime.
C. Computer Model Development The third stage of this study used the aforementioned collected data in order to produce a computer based model for the analysis of energy requirements of the village community. At any given hour interval, wind speed, solar radiation and power required by the community were determined. The flow chart of the optimization model is shown in Fig. 4. In proposed configuration, the optimum configuration includes an optimum number of photovoltaic array, number of battery bank, number of wind turbine and wind turbine type.
appropriate combination of WT type, number of WT, PV array power and number of battery banks to optimally match the load demand requirement.
A. Effect of WT Parameters on optimal WT Selection In order to reach a suitable COE value at a proper REF and to verify the effects of REF on optimal WT selection, the impact of REF variation on COE for the fifteen WTs are investigated. According to Fig. 7, it is interesting to note that the priority WT sequence is not affected only by the WT speed parameters or by the WT rated power but also by the desired REF. It can be clearly seen that the WT rated speed has a significant effects on priority sequence. Whereas, the WT rated power hasnt a consistent effects on optimal WT sequence. Also, at high REF, comparison between HES combinations that contain WTs with equal cut-in speed which include WT1, WT2, WT3, WT4, WT6, WT7, WT8, WT9 and WT12, the HES composed of WT with the lowest rated speed is the one that has the lowest COE. The highest COE value is recorded amongst HES composed of WT12 which has higher rated speed and all the remaining HES have much lower figures as the WT rated speed decreases. Whereas COE values of HES combinations that include any of WT5, WT10, WT11, WT13, WT14 and WT15 are much higher than those having lower cutin speed, with an exceptions of those contain WT5 and WT14.
Figure 4: Flow Chart of the Optimal Sizing Model using GA.
D. System Specification Specific data relating to HES components was necessary in order to accurately calculate the performance of the HES. The technical characteristic of the fifteen different types of WTs are given in Table 5. The capital and O&M cost of the HES components are shown in Table 6.
Table 1: Technical Specifications of the Wind Turbines Figure 5: Impact of REF on wind turbines optimum sequence. Table 3: Optimal sizing based on 65% of REF and 0.1$/liter diesel price.
Table 2: The Costs and Lifetime of the System Components
III.
Genetic Algorithm based matlab code has been developed to find out the optimal sizing HES consists of PV, WT, DG and battery bank as storage system intended to supply AlSulaymania village. The diesel generator rated power is sized to meet the peak load demand. The study investigates an
The COE of HES contain WT5 is slightly lower than those composed of WT8, WT9 and HES compose of WT14 is somewhat lower than those compose of WT12. That resulted from lower rated speed of WT5 as compared to WT8 or WT9 whereas WT14 has much lower rated speed in compared with WT12. On the contrary, for REF less than 75 %, the HES combination with lowest cumulative rated power is the optimum one, with an exception of WT3 and WT4. Moreover,
the COE value of HES combination that contain WTs having different rated power and slight different speed characteristics such as WT9, WT14 or WT10, WT12; the optimal wind turbine is the one that has the highest rated power.
B. Optimal Sizing Results It can be depicted from Fig. 8 that the COE is not linearly dependent on REF, since at high REF a little raise in REF will results in high increase in COE value. Moreover, at REF the required storage system is higher and the surplus energy is also high. Therefore, the proposed HES with 65% of renewable energy fraction seem to be the suitable REF in term of COE value and surplus energy.
the village under study which is located approximately 50 km to the nearest distribution network, the COE of HES is cheaper than an electricity grid extension, when the COE drown from grid varies between 0.06 and 0.1334 $/kWh.
Figure 7: The Breakeven Grid Extension Distance for the Optimized HES. Figure 6: Renewable energy fraction impact on cost of energy
Furthermore, it can be also depicted from Fig. 8 and Table 8 that the optimum WT to be chosen is Lagerwey (WT1) as it result in minimum COE of 0.13964 $/kWh and minimum excess energy only 13.7% of the energy generated. The program results are validated with HOMER. The results of the optimal sizing for HES are shown in the Table 10.
C. Comparison with Grid Extension Traditionally, standalone HES are installed in remote location that are far from the electricity grid. Farther away from the grid, the HES are optimal. Nearer to the grid, grid extension is optimal. In order to determine the distance from the grid, which results in an equal COE for both HES and grid power, the following equation is used [20]:
ASC = Cacap + Carep + Camain
(10)
Where COEgrid is the COE from the grid ($/kWh), ALD is the annual load demand (kWh) and Cext is the cost of grid extension ($/km). Fig. 10 shows the breakeven distance that makes electricity from decentralized HES cost effective over that from a grid source for the same range of input component costs. It is clear that the BEGED for this hypothetical system in this location for load demand of 1124.9 MWh varies from 3.85 to 232.2 km depending on COE from grid and cost of grid extension. If the load centre is more than 232.2 km away from the grid, the COE of HES is cheaper than an electricity grid extension, when the COE drown from grid is 0.01337 $/kWh and cost of grid extension is 10000 $/km. Whereas, if the load centre is more than 77.62 km away from the grid, the HES is economically preferable, when the COE drown from grid is 0.01337 $/kWh and cost of grid extension is 30000 $/km. For
D. Sensitivity Analysis Results The results of the analysis are reasonably sensitive to the assumptions about the uncertain parameters of wind speed and solar radiation as these parameters strongly affect configuration of the system. Thus uncertainty with wind speed and solar radiation considered as a sensitivity variable because only one year data have been used in the analysis. The optimization results are represented in graphical forms represents optimal system types at certain wind speed and certain solar radiation for various PV costs. This permits identification of system types and components size for various locations around area under study. The systems shown in Fig. 11 reflects the optimization results based on PV cost of 2290 $/kW and diesel price of 0.1 $/L. Two important observations can be depict is that PV/Battery/diesel hybrid system comes into the picture only when wind resource is very limited and solar energy density is very high. On other hand Wind/Battery/diesel hybrid system is only feasible at high wind speed and low solar energy density. With expectation that PV cost is decreasing in near future, the remarks in Fig. 12 are made with lower cost level. The systems shown reflects the optimization results based on PV cost of 900 $/kW and diesel price of 0.1 $/L. It can be noticed that for wind speed higher than 4.89 m/s the PV/Wind/Battery/diesel hybrid system is feasible over whole range of solar radiation. The cost of energy of such system varies between 0.1136 $/kWh to 0.1511 $/kWh. Whereas at 4.39 m/s, the PV/Battery/diesel hybrid system becomes economically feasible. The COE of such system is much higher and varies between 0.1310 $/kWh and 0.1523 $/kWh. Selection of an optimum system type is explored from different approach in Fig. 13, where the solar energy density and PV
cost is considered fixed. Wind speed and fuel price are taken as sensitivity parameters. According to the graph, the DG system is the cost-effective solution for wind speed less than or equal 5.89 m/s and diesel price less than or equal 0.1 $/L. While for diesel price of 0.2 $/L, the diesel system is only the economical solution for wind speed less than or equal 4.89 m/s.
system is only feasible at high wind speed and low solar energy density. The DG system is the economical solution for wind speed less than or equal 5.89m/s and diesel price less than or equal 0.1 $/L. While for diesel price of 0.2 $/L, the DG system is only the economical solution for wind speed less than or equal 4.89 m/s. The PV/Bat/DG hybrid system is economically acceptable only at very low wind speed and diesel price higher than 0.3 $/L. Whereas WT/Bat/DG hybrid system is the economical solution at high wind resources and over whole range of diesel prices for the village. ACKNOWLEDGMENT
Figure 8: Optimal System Type for PV cost=2290$/kW and REF=65%.
This work was financially supported by the National Plan for Science and Technology (NPST) program, King Saud University; Project Number: 09 ENE 741-02. REFERENCES
Figure 9: Optimal System Type for PV cost=900$/kW and REF=65%. [1] Elhadidy M.A. Shaahid S.M., "Technical and economic assessment of grid-independent hybrid photovoltaic-diesel-battery power systems for commericial loads in desert environments," Renewable and Sustainable Energy Review, vol. 11, no. 8, pp. 1794-1810, 2007 [2] Whole Global Greenhouse Warming Internet. (2010) Global Greenhouse Warming. [Online] [3] Balachandra P., Ravindranath N.H. Deepak P.K., "Grid-connected versus stand-alone energy systems for decentralized powerA review of literature," Renewable and Sustainable Energy Reviews, vol. 13, p. 20412050, 2009 [4] Chengzhi L, Zhongshi Li, Lin Lu, Hongxing Y.W., "Current status of research on optimum sizing of stand-alone hybrid solarwind power generation systems," Applied Energy, vol. 87, p. 380389, 2010 [5] Lu L, Zhou W Yang HX, "A novel optimization sizing model for hybrid solarwind power generation system," Solar energy, vol. 81, no. 1, p. 7684, 2007. [6] Nayar C.V. Ashari M., "An optimum dispach strategey using set points for a photovoltaic (PV)-diesel-battery hybrid power system," Solar Energy, vol. 66, pp. 1-9, 1999 [7] Hayashia D., Yonaa A., Urasakia N., Funabashib T. Senjyua T., "Optimal configuration of power generating systems in isolated island with renewable energy," Renewable Energy, vol. 32, p. 19171933, 2007 [8] Al-Hadhrami L.M. Rehman S., "Study of a solar PVedieselebattery hybrid power system for a remotely located," Energy, vol. 35 , pp. 4986-4995, 2010 [9] Zhang L., Barakat G. Belfkira R., "Optimal sizing study of hybrid wind/PV/diesel power generation unit," Solar Energy, vol. 36, pp. 100-110, 2011 [10] Zhang L., Barakat G. Belfkira R., "Optimal sizing study of hybrid wind/PV/diesel power generation unit," Solar Energy, vol. 36, pp. 100-110, 2011 [11] Campbell H., Sanguinetti C. Hessami M.A, "A feasibility study of hybrid wind power systems for remote communities," Energy Policy, vol. 39, pp. 877-886, 2011. [12] Nehrir M.H., Wang C. Nelson D.B., "Unit sizing and cost analysis of stand-alone hybrid wind/PV/fuel cell power generation systems," Renewable Energy, vol. 31, pp. 1641-56, 2006 [13] Yang H., Shen H., Liao X. Ai B., "Computer-aided design of PV/wind hybrid system," Renewable energy, vol. 28, pp. 1491-1512, 2003. [14] National Renewable Energy Laboratory (NREL). Homer-Analysis of micropower system options. [Online]. http://analysis.nrel.gov/homer/default.asp [15] Johnson G.L., Wind energy systems. USA: Prentice Hall, 1985 [16] Solar buzz Retail Pricing Environment. (2012, March) Solarbuzz. [Online]. www.Solarbuzz.com [17] TRADINGECONOMICS. (2012, January) Saudi Arabia Interest Rate. [Online]. www.TRADINGECONOMICS.COM [18] Mara Isabel Blanco, "The economics of wind energy," Renewable and Sustainable Energy Reviews, vol. 13, p. 13721382, 2009
Figure 9: Optimal System Type for PV cost=2290$/kW and REF=65%.
IV.
CONCLUSION
A methodology of optimum sizing of a standalone HES consisting of PV arrays, wind turbines, battery banks and diesel generator using GA is presented in this paper. This optimization methodology takes in consideration the effects of wind turbine parameters on optimal wind turbine selection. The technical models are represented based on LPSP technique and renewable energy fraction whereas economic system model is developed based on annualized system cost. According to the simulation results related to the case studied in this study, it can be concluded that: The COE is not linearly dependent on REF, since at high renewable energy fraction a little increase in renewable energy fraction will results in high jump in COE value. Moreover, at high renewable energy fraction the required storage system is higher and the surplus energy is also high. The optimum wind turbine selection is not affected only by the wind turbine speed parameters or by the wind turbine rated power but also by the desire renewable energy fraction. The wind turbine rated speed has a significant effect on optimum wind turbine selection whereas, the wind turbine rated power hasnt a consistent effects on optimal selection. The PV/Bat/DG hybrid system comes into the picture only when wind resource is very limited and solar energy density is very high. On other hand WT/Bat/DG hybrid
Renewables and Plug in Electric Vehicles Modeling on Electricity and Gas Infrastructures Scheduling in Presence of Responsive Demand
Mahmoud-Reza Haghifam Electrical Engineering Islamic Azad University (IAU), South Tehran Branch
Tehran, Iran Senior Member, IEEE haghifam@modares.ac.ir
Samaneh Pazouki Electrical Engineering Islamic Azad University (IAU), South Tehran Branch Tehran,Iran Student Member, IEEE samanehpazouki@gmail.com
Samira Pazouki Electrical Engineering Islamic Azad University (IAU), Qazvin Branch Tehran,Iran Student Member, IEEE samirapazouki@gmail.com
AbstractGreenhouse gases emission due to fossil fuels and growing energy needs are developing progressive advances of Distributed Generation technologies. Combined Heat and Power (CHP), Renewable Resources, Plug in Electric Vehicles (PEVs) as energy storages and Demand Response (DR) are great examples of the technologies. The advances are subject to efficiency enhancement, emission lower, reliability enhancement and economic improvement in smart grid. In this paper, Energy Hub (EH) model is flexibility extended as a strong solution for robust operation of multi carrier energy systems which are integrated to the technologies in smart distribution network. In other words, in this paper, Renewable (wind and biomass), PEVs and demand shifting of DR are mathematically modeled and comprehensively combined to gas and electricity infrastructures under EH approach. Then they are utilized to serve hub required demands. GAMS; high level algebraic modeling software is employed for Mixed Integer Linear Programming (MILP) model of proposed approach to endorse when and how much of which technology should be optimally operated to achieve minimum operation costs. Finally, impact of the technologies on hub operation costs and imported electricity and gas from grid is evaluated. Keywords-Smart Grid, Multi Carrier Energy Systems, Renewables, PHEV, Demand Response
I. INTRODUCTION Smart grid transfers unidirectional electrical power system from fossil fuels generation plants to bidirectional environment. The grid provides opportunity for customers to participate electricity management or sell their additional electricity to the grid to receive revenue. Energy hub is expanded as a robust solution for optimal operation of multi carrier energy infrastructures; gas and electricity [1-3]. The hub has strong potential for integrating distributed generations such as; renewable, energy storages and demand participation to reduce operation costs in response to different required demands such as electricity and heat [4-5]. Impact of renewable such as Photo Voltaic (PV) on optimization problem in energy hub approach is seen in [6-7]. Energy storage technologies such as battery, flywheel, and compressed air energy storage are rapidly developingcommercial and their appliances are broadly expanding valuable for renewable fluctuations. Energy storages are
modeled and formulated in order to discharge their batteries to smooth wind fluctuations [8-11]. Demand Response is defined electricity usage pattern change via consumer to follow spot electricity price or receiving predesigned incentive to reduce electricity in high price time or while system reliability imperils. Demand response programs are separated to subsections; incentive based DR and time based rate which is entirely discussed in [12]. DR could be more sufficient and more dispatch able to operate with intelligent communication technologies which is adequately implemented to coordinate of renewable resources oscillations to inhabit transmission system expansion investment. Heat load management of demand side is evaluated in response to reduce operation costs in [13-15]. Gas and electricity infrastructures are optimally operated under energy hub approach by (Mixed Integer Nonlinear Programming) MINLP model of MATLAB [3] and in (Mixed Integer Linear Programming) MILP model of GAMS [16]. This paper is aimed to utilize gas and electricity infrastructures which are comprehensively integrated to smart grid technologies; wind, biomass, PEVs and DR in response to emission lower and operation costs under energy hub approach as follow: proposed energy hub is introduced in section II. Proposed hub is scheduled in section III. Session IV discusses simulation results. Finally, conclusion is debated in session V. II. PROPOSE ENERGY HUB Proposed energy hub (Fig.1) would be robustly scheduled with hub central unit via intelligent information technologies in order to minimize operation costs to satisfy hub demands which are advanced by smart grid important drivers. Complex attributes of integrated renewable resources with electricity and gas infrastructure of conventional grid raises the need for technologies and techniques which complement the features. Demand response and energy storage have many qualities that make them well suited for this role in smart grid scene. Energy hub approach is strongly facilities the complex interaction of the technologies to minimize energy operation costs. Hub receives grid electricity and wind power to supply hub electricity demand. PEVs as energy storages are also utilized to provide electricity demand. Hub has strong potential to sell
Figure 1. Proposed energy hub
Hub electricity demand is supplied (2a) by wind , imported electricity for transformer from power through transformer efficiency . It could be network supplied by CHP through imported network gas for CHP . Demand could be shifted down and its efficiency in in high price time and in high demand or shifted up low price time and in low demand. PEVs would be charged , and discharged , in required times by hub control and information technologies in required times. PEVs arrive, park and departure time is read from parking by hub central control unit and they are asked driver when they want to leave parking. Total electrical storage of PEVs which are available in PEVs parking are collected to decide when and how much to charge or discharge. Wind power is modeled in , wind (2b). shows wind speed. When wind reaches to turbine starts to produce power till rated power of wind turbine , when wind increases till , wind turbine produces , wind turbine will rated power.When wind rises more than be off. , , are related to wind turbine characteristics [17]. Hub has potential to sell additional electricity to the grid. is provided by (2c), (2d); heat Hub heat demand and its efficiency , CHP power exchanger power through its efficiency , boiler power through its efficiency and discharge power of thermal storage . Required power for absorption chiller is reduced from heat exchanger power and additional produced heat is saved in thermal storage . Produced gas power from biomasses and their gasification efficiency and network gas are used for CHP and boiler in (2e). is provided (2f) by absorption Hub cooling demand chiller through its efficiency . , is constrained by its remained Parking power power in PEVs batteries while arriving in parking , 1 , , and discharge , power and its loss charge power with its efficiency in (3a). Thermal storages power are restricted by its state of charge and discharge power and 1 , its charge loss power and its efficiency in (3b).
+
additional electricity to the grid and to receive revenue. Two different biomasses with gasification technologies produce gas power. Hub receives biomass gases and network gas to supply its CHP and boiler to satisfy hub heat demand. Heat exchanger is used to convert high temperature heat to normal heat. Thermal storage is employed to store additional heat and it is used in required times. Absorption chiller uses high temperature heat to produce hub required cooling demands. Hub has potential to sell additional heat to the grid. III. PROPOSED ENERGY HUB SCHEDULING Energy hub is economically scheduled based on minimum operation costs. Hub sometimes receives revenue from selling power to the grid. Objective function (1) is related to purchase and gas from network for or selling electricity and boiler with electricity and gas CHP price. Biomass from type one is purchased with its price 1 and biomass from type two 2 with its price is also sold to grid with grid heat 2. Additional heat is operated benefit price . Demand response based on its operation cost . Hence, the hub could be powerfully scheduled in order to operate the advanced technologies for minimum operation costs: Minimizing;
-
0,
,
,
(2a)
2b
, 0,
(2c) (2d) (2e)
2
, , 1 , , . ,
(2f)
(3a)
(1)
1 +
(3b)
TABLE I. I Proposed Energy Hub Parameters and PEVs Parameters
0.98 0.9
N,H N1 N2 N3 N4 N5 N6 N7 N8 N9 N10
0.35 0.4
H01 H02
0.8 0.8
H03
0.95 0.9
H04
0.93 0.8
H05
0.9 0.9
H06
0.03 0.03
H07 H08 8 6 20 2 6 4 12 +
30 0 500
H H09 2 + + + + + + + +
500 1000
H10 + + + + + + + + +
900 1000
H11 3 + + + + + + + + + H12 + + + + + + + + + +
500 700
N,H N1 N2 N3 N4 N5 N6 N7 N8 N9 N10
0.1 0.1
H13 + + + + + + + + + + H14 + + + + + 28 30 + + +
8 10
H15 + + + + + + + +
13.2 13.2
H16 + + 30 25 25 + + +
400 10
H17 25 25 H18
4 22
H19
0.032 0.01745
H20
0.0776 10
H21 H22
1000
200 800
H24
300
H23
+ + +
+ + +
+ + +
30 30 +
20
14 Wind Power (kW) 12 10 8 6 4 2 0
Wind Power Demands (kWh)
1400 1200 1000 800 600 400 200 0 Le(H) Lh(H) Lc(H)
1 2 3 4 5 6 7 8 9 101112 131415 161718 192021 222324 Time (H)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Time (H)
Figure 2. Wind turbine power (Pew(H)) at 24 hours a day
Figure 3. Electricty, heat and cooling demands at 24 hours a day

20 Price(Cent/kWh) 15 10 5 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Time(H) e(H) g(H) SHCT(H)
Demand response constraints are formula ated in (4a)-(4d). Total electricity demands which are shifted d up should be equal to total electricity demands which are shifted down in d response (4a). Shifted up and shifted down power of demand is limited between load participation fact tor of shift up LPFshup and load participation factor of shif ft down LPFshdo in (4b), (4c) in sequence. Binary variabl les of shift up and shift down powers are applied to prevent acting shift down and shift up pow wer at the same time in (4d). PEVs powers of parking constraints are limited in (5a)age are restricted (5d). PEVs powers in parking as energy stora power in (5a). between minimum and maximum Charge and discharge power of parking sh hould be limited between min and maximum power with their r charge and discharge efficiencies in (5b) and (5c c) in sequence. Binary variables of charge and discharge powers ower at the same are used to prevent charge and discharge po time in (5d). Thermal storage is limited between its in n minimum and and discharge maximum power in (6a), charge powers of thermal storage through h charge and discharge efficiencies of the storage in n (6b) and (6d). Binary variables of charge and discharge e are used to prevent charge and discharge performance at the same time. Biomass1, biomass2, grid gas, grid electricity and pipeline heat are respectively restricted in (7a), (7b), (7c), (7d) and (7e). , boiler , CHP , ab bsorption chiller Transformer and heat exchanger sizes also cause constraints c in
Figure 4. Electricty (e(H)), gas (g(H)) and shift (SHCT(H)) costs at 24 urs a day hou
(8a), (8b), (8c), (8d) and (8e) for importing electricity, gas and heat from network. DR Constraint: 0 0 0 Parking Constraint: N, H 0 , 0 0 0 , , 1 (4a) * *
(4b) (4c) (4d) (5a) (5b)
, , , 1
(5c) (5d) (6a) (6b) (6c) (6d)
Thermal storage constraint: 0 H S 0 0 0 1
TABLE II. Electricity load suppliers in two different cases (case1, case 4)
+ -
time T 1,4 t1 t2 t3 t4 t5 t6 t7 t8 t9 t10 t11 t12 t13 t14 t15 t16 t17 t18 t19 t20 t21 t22 t23 t24 600 470 500 420 400 400 420 600 600 500 600 700 700 700 800 800 800 800 1000 1200 1200 1200 1000 800
400 400 400 400 400 0 0 0 0 0.3 0.3 0.3 0.3 0.3 0 0 0 77.8 77.8 77.8 77.8 400 400 400
TABLE III. Heat load suppliers in two different cases (case1, case 4)
,
Time PEVsStorage T 1 4 t1 t2 t3 t4 t5 t6 t7 t8 t9 t10 t11 t12 t13 t14 t15 t16 t17 t18 t19 t20 t21 t22 t23 t24 700 700 700 700 700 700 550 550 420 420 700 700 700 420 420 420 550 550 550 550 620 620 550 550 Load Sold Heat 1 29.3 29.3 29.3 29.3 29.3 29.3 337.7 337.7 393.8 393.8 29.3 29.3 29.3 393.8 393.8 393.8 337.7 337.7 337.7 4 29.3 29.3 29.3 29.3 29.3 29.3 337.7 337.7 393.8 393.8 29.3 29.3 29.3 393.8 393.8 393.8 337.7 337.7 337.7
Chiller 1,4 222.2 222.2 222.2 222.2 222.2 222.2 55.5 55.5 133.3 133.3 222.2 222.2 222.2 133.3 133.3 133.3 55.5 55.5 55.5 55.5 166.6 166.6 55.5 55.5 Storage 1,4
Load Wind 1
Electricity Load Suppliers Electricity CHP DR 4

265.3 119.3 153.0 63.2 40.8 448.9 471.4 612.2 612.2 509.8 611.9 713.9 713.9 713.9 816.3 816.3 816.3 695.0 838.9 768.6 900 571.9 714.2 489.7 248.8 119.4 119.4 248.8 120.1 119.4
Heat Load Suppliers Boiler CHP 1 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 389.0 758.5 758.5 1100 1100 4 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 1100 389.0 757.7 758.5 1100 1100 1 4
1,4
60 47 50 42 40 40 42
20.4 103.0 -69.3 159.1 150.1 204.0 241.0 740.3 389.7 342.3 536.0 643.5 647.6 606.8 710.7 682.6 701.8 450.1 650.5 523.7 670.3 349.4 491.8 288.3
30 7.7 226.7 91.9 13.4 4.0
-41 100 120 120 120 100 80
58.0 1.4 93.0 134.3 29.4 30.0 67.7
284.3 136.5 136.5
284.3 137.3 136.5
337.7 337.7
337.7 337.7
20.0
Imported Electricty from Network at Every Case

1000 Imported Network Electricity (kW) Case 1 Case 2 Case 3 600 400 200 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Case 4
Pipeline Constraint:
0 0 0 0 0
800
(7a) (7b) (7c) (7d) (7e) (8a) (8b) (8c) (8d) (8e) IV. SIMULATION RESULTS
Converter Constraints:
-200 -400
Time (H)
Figure 5. Comparison imported electricty from network for different cases (case1, 2, 3 and 4)
Simulation is run under CPLEX solver of GAMS software in less than 0.01 second. Required information of proposed energy hub such as converters efficiency and their restrictions, storages efficiency and their constraints, load participation factors, network constraints, wind turbine characteristics, heat benefit, biomasses price and etc which are given in Table I. Arriving, park and departure times of ten PEVs in parking are registered by hub central unit in Table I. Initial state of charge or remained energy of vehicles batteries are registered by hub central control unit. Their required energy and departure times of vehicles owner are also asked and registered by hub control unit.
Wind speed is depicted in Figure 2. Hub required demands; electricity, heat and cooling demands are drawn in Figure 3. Electricity and gas prices and demand shifting operation costs are shown in Figure 4. Four different cases of combination of distributed energy resources to the hub are introduced, and their effects are evaluated on hub operation costs. Energy hub is integrated to renewable (wind and biomasses), energy storages (PEVs and thermal storages), demand shifting program of demand response in case 1. Energy hub is integrated to renewable and energy storages in case 2. Energy hub is integrated to energy storages and demand response in case 3. Energy hub is integrated to renewable and demand response in case 4.
Hub Operation Costs (cent/kWh)
250000 200000 150000 100000 50000 0 Case 1 Case 2 Case 3 Case 4
Hub Operation Costs at Every Case e
185334.1 187997 220803 235150.2
Figure 6. Hub operation costs in differen nt case
Case 1 (renewable, storages and DR R) and case 4 (renewable and DR) are economically scheduled for supplying electricity and heat demand in Ta able II and III in sequence. They are evaluated for impact of renewable r (wind and biomass), energy storages (PEVs and th hermal storages) and demand shifting program of deman nd response on electricity and gas infrastructures (Figure e 5). Figure 5 reveals that how much electricity should be purchased from electric network at every case. As it is employed from Figure 5, case 1 (integration of renewable, storages s and DR to the hub) purchases less electricity than other o cases from electric grid. Case 2 (renewable, storages) purchases more electricity than case 2 from grid, and it purchases less electricity than case 3 and case 4 from grid. Case C 3 (storages and DR) purchases more electricity than cas se 1 and case 2, and it imports less electricity than case 4 fro om grid. Case 4 (renewable and DR) purchases more electr ricity than other cases from grid. It is observed from Figure 5, 5 less electricity is purchased from grid by utilizing smart gr rid technologies. It shows importance of combination of the technologies on reducing hub operation costs through purchasing p less electricity from electric network. Table II and Table III reveal and compare hub economical scheduling of case 1 (wind, biomass, b PEVs, thermal storage and DR) and case 4 (renewa able and DR) to supply required electricity and heat demands s in sequence. Comparison of case 1 and case 4 in Tab ble II shows that hub would rather supply its electricity dema and with wind and with shifting demands in high h pricetimes. (Positive mark of shifting demands show ws that demand could be shifted up in low demands and in off peak times. Negative mark of shifting demands show ws that demand could be shifted down in peak demand ds and in high electricity prices). CHP is utilized in ca ase 1 as well as case 4 in high electricity demands. Importan nt change occurs in importing electricity from network in two cases. (Positive marks of introduce importing electricity from network in required times. Negative mar rks of show selling additional produced electricity to grid). Case 1 (renewable, storages and DR) imports less electricity than case 4 (renewable and DR) because PE EVs storages in
parking provide some part of electricity demand. Therefore, hub operation cos sts are less than case 4. Table III shows that the ere are not sensible changes in utilizing the technologies for f supplying heat demands at every case. On the other han nds, hub purchases network gas for CHP and boiler of case 1 (renewable, storages and DR) as well as case 4 (renewable e and DR). Also, hub sells equal additional produced heat to the network in case 1 and case i heat demands by boiler 4. Hub prefers to supply its because hub has potentia al to sell its additional heat to the heat pipeline to recei ive revenue and decrease its operation costs. It is seen th hat CHP is employed when electricity and heat demand ds are high and hence boiler is utilized less than other time es to provide heat. Network gas power is imported till ma aximum gas network capacity because gas price is less than n electricity prices. Impact of integrating dis stributed energy resources to the hub is compared in Figure 6. 6 It is observed from Figure 6, hub operation costs is sensib bly decreased by combination of the technologies to the hub in case 1 (renewable, storages and DR). Case 2 (renewab ble and storages) declines hub operation costs less than case c 1 and more than case 3 (storages and DR) and case 4 (renewable and DR). Case 3 (storages and DR) decrease es hub operation cost less than case 1 (renewable, storages and a DR) and case 2 (renewable, storages) and more than case e 4 (renewable and DR). Case 4 reduces hub operation costs s less than case 1 (renewable, storages and DR), case 2 (re enewable and storages) and case 3 (storages and DR). As it is seen from Table II in comparison with Table III, there are just sensible chan nges on purchasing electricity from network in every cas se. Also, Figure 5 shows that comparison of purchasing electricity from network for different cases. As it is seen from Figu ure 6, hub operation costs are directly affected by purchas sing electricty from network in Figure 5 and Table II accord ding to technologies sizes. Also, Figure 6 accordingly show ws impact of PEVs as energy storage, renewable energy especially wind power and f DR on hub operation costs to demand shifting program of satisfy a commercial load in 24 hours a day. V. CONCLUSION In this paper, an energ gy hub, which is combined to renewable (wind, biomass ses), energy storages (PEVs parking as energy storage an nd thermal storage) and demand shifting program of deman nd response, is mathematically modeled and economically scheduled. The hub supplies a commercial load with electri icity, heat and cooling demands. Heat exchangers, absorption n chiller are also used to convert high temperature heat to usable heat and cooling hub ation is utilized for converting required demands. Gasifica biomass to natural gas for su upplying CHP and boiler. Hub is operated and scheduled to o achieve minimum operation costs. Results are evaluated d and compared in comparison with four different cases. Comparison C of different cases demonstrates that hub opera ation costs are directly affected
by purchasing electricity from network. Result shows combination of renewable (wind and biomass), energy storages (PEVs and thermal storages) and demand response (demand shifting program) declines purchasing network electricity carriers. Therefore, hub operation costs will decrease by integration of combination of smart grid technologies through purchasing less electricity from network. Results also approve that gas energy carrier is purchased as equal as in different cases. Also, results debate PEVs as energy storages and wind as renewable energy has more effective than thermal storage, biomasses and DR programs on hub operation costs. REFRENCES
M. Geidl, G. Koeppel, P. Favre-Perrod, B. Klockl, G. Andersson, K. Frohlich, "Energy hubs for the future," IEEE Trans. Power and Energy Magazine, vol. 5, no.1, pp. 24-30, 2007. [2] M.Geidl, Integrated modeling and optimization of multi-carrier energy systems, PhD Thesis, ETH Zurich, 2007. [3] M. Geidl, G. Andersson, "Optimal power flow of multiple energy carriers," IEEE Trans. Power Systems, vol. 22, no.1, pp. 145-155, 2007. [4] L. Carradore and F. Bignucolo, Distributed multi-generation and application of the energy hub concept in future networks, in Proc. 2008 IEEE Universities Power Engineering (UPEC) Conf., pp. 1-5. [5] L. Forbes, S. J. Galloway, G. W. Ault, An approach for modellling a decentralized energy, in Proc. 2010 IEEE Universities Power Engineering (UPEC) Conf., pp. 1-5. [6] M. Arnold, G. Andersson, Modeling and optimization of renewable; applying the energy hub approach, in Proc. 2010 IEEE Power and Energy Society General Meeting Conf., pp. 1-8. [7] M. Schuzle, L. Friedrich, M. Gautschi, Modeling and optimization of renewable: applying the energy hub approach, in Proc. 2008 IEEE Sustainable Energy Technology Conf., pp. 83-88. [8] M. Geidl, G. Andersson, "Optimal coupling of energy infrastructure," in Proc. 2007 IEEE Power Tech Conf., pp. 1398-1403. [9] A. Parisio, C. D. Vecchio, and A. Vaccaro, "A robust optimization to energy hub management," Elsevier, Electrical Power and Energy Systems, vol. 42, pp. 98-104, 2012. [10] M. D. Galus, R. La. Fauci, G. Andersson, Investigating PHEV wind balancing capabilities using heuristics and model predictive control, in 2010 IEEE Power and Energy Society General Meeting Conf, pp. 1-8. [11] M. D. Galus, G. Andersson, Power system considerations of plug-in hybrid electric vehicles based on a multi energy carrier model,in Proc. 2009 IEEE Power and Energy Society General Meeting Conf., pp. 1-8. [1]
[12] FERC, Staff Report, Assessment of demand response and advanced metering, www. FERC. Gov, Aug. 2006. [13] P. Ahcin, M. Sikic, "Simulating demand response and energy storage in energy distribution systems," in Proc, 2010 IEEE Power System Technology Conf., Hangzhou, China, pp. 1-7. [14] F. Kienzle, P. Ahcin, G. Andersson, "Valuing investment in multi energy conversion, storage and demand side management systems under uncertainty," IEEE Trans. Sustainable Energy, vol. 2, pp. 194202,2011. [15] M. Houwing, R. R. Negenborn, B. D. Schutter, "Demand response with micro CHP systems," IEEE Trans. Proceeding in IEEE, vol. 99, NO.1, pp. 200-213, 2011. [16] A. Sheikhi, A. M. Ranjbar, and H. Oraee, "Financial analysis and optimal size and operation for a multi carrier energy systems," Elsevier. Energy and Builduing, vol. 48, pp. 71-78, 2012. [17] P. Giorsetto and K. F. Utsurogi, Development of a new procedure for reliability modeling of wind turbine generators, IEEE Trans. Power Apparatuses and Systems, vol. 102, No. 1, January 1983.
Mahmoud-Reza Haghifam (IEEE M95SM06) was born in Iran in 1967. He received the B.Sc., M.Sc. and Ph.D. degrees in Electrical Engineering in1989, 1992 and 1995, respectively. He joined Tarbiat Modares University as assistant Prof. in 1995. He is now a Full Professor at the Department of Electrical Engineering, Tarbiat Modares University (TMU) as well as Islamic Azad University (South Tehran Branch), Tehran, Iran. Also he is a research Fellow of Alexander Von Humboldt in Germany. He has been awarded by DAAD and AvH in 2001, 2006, 2009 and 2011 for research stays in German universities. He was as visiting Prof. in University of Calgary, Canada in 2003. His main research interests are Power system Planning, Power System Reliability, Electric Distribution
System and Smart Networks. He is a Senior Member of IEEE and an IEEE Iran Section Industrial Relationship Officer.
Samaneh Pazouki (IEEE S12) was born in Tehran, Iran. She received the B.Sc. degree in Electronic Engineering from Islamic Azad University (IAU), Garmsar Branch, Iran. She is currently pursuing the M.Sc. degree in Electrical Engineering at Islamic Azad University (IAU), South Tehran Branch, Tehran, Iran. Her main research interest is Power System Operation, Power System Planning, Power System Reliability, Electric Distribution System, Plug-in Electric Vehicles and Smart Networks. She is a Student Member of IEEE. Samira Pazouki (IEEE S13) was born in Iran. She is currently pursuing the B.Sc. degree in Control Engineering at Islamic Azad University (IAU), Qazvin Branch, Iran. Her main research interest is Power System Control, Plug-in Electric Vehicles, Wind, Photo Voltaeic and Smart Networks. She is a Student Member of IEEE.
Feasibility Study Of An Intentional Islanding Operation With A New Adaptive Load Shedding
Hasmaini Mohamad , Nofri NY Dahlan
Faculty of Electrical Engineering, University of Technology MARA(UiTM), Shah Alam, 40450,Selangor, Malaysia.
Hazlie Mokhlis,Mazaher Karimi

Department of Electrical Engineering, Faculty of Engineering, University of Malaya,50603 Kuala Lumpur, Malaysia
Ab Halim Abu Bakar

University of Malaya Power Energy Dedicated Advanced Centre (UMPEDAC), Level 4, Wisma R&D, 59990, Kuala Lumpur, Malaysia.
Abstract - Intentional islanding operation has recently been considered by some utilities to effectively use the available renewable energy generation. In this paper, an intentional islanding operation of hydro Distributed Generation (DG) is proposed. Governor and Automatic Voltage Regulation (AVR) are properly developed such that the DG exhibits a good dynamic response once a part of the network is islanded. A new adaptive load shedding scheme is also proposed to address the issue of overload or insufficient generation after the system is islanded. The scheme is based on a combination of adaptive and intelligent techniques. The disturbance magnitude of response based and event based is treated differently: 1) event based is based on the power imbalance 2) response based is based on the swing equation which relies on the frequency and Rate of Change of Frequency measurement. To investigate the effectiveness of the controller and the load shedding scheme, dynamic simulation on an intentional islanding operation is performed on an existing Malaysia network interconnected with mini hydro generation. The simulation results demonstrate that the island is successfully established with the frequency and voltage response during the islanding operation are within the acceptable limit. Index Terms -- Islanding Operation, Load Shedding, Smart Grid, Hydro Generation, Distributed Generation
There are many publications that have discussed on the planned/intentional islanding operations. The operation is feasible provided that a great attention is given towards several technical issues of islanding. Amongst the issues are power quality, out of phase synchronism and protection system. To address the issues of power quality and out of phase synchronism, an appropriate controller need to be developed for the DG to operate in two operation modes i.e.grid connected and islanding [1].The complexity of the controller design would vary with the type of DG (rotating and inverter type). Of significance is to ensure that the frequency and the voltage response of the island is dynamically stable and within their acceptable limit. Unlike the interconnected grid system, the frequency response of an isolated and islanded system could be more severe when subjected to a transient disturbance. The severity greatly varies with the size of the isolated system or in other words the total system inertia. The smaller system inertia tends to cause a quick frequency drop. It also has a limited spinning reserve to cover for the shortfall of the generation. Thus, these characteristics have justified the requirement of particular approaches in designing the adaptive load shedding scheme in the isolated system [2-4]. For the islanded distribution system, not much work has been carried out. The developed load shedding schemes so far have focused on the approaches for obtaining the optimal load shedding in the islanded system. Among the strategies are the scheme based on the frequency and df/dt information, customers willingness to pay and load histories [5] and the best time to shed the loads [6]. This paper proposes an intentional islanding strategy which including DG controller and a new adaptive load shedding scheme for an islanded operation. The proposed load shedding scheme will provide a solution for two scenarios in an islanded distribution system i.e. the moment the island disconnected from the grid and during islanded system. The strategy utilizes a combination of adaptive and
I. INTRODUCTION When considering a high penetration of Distributed Generation (DG), the decision to disconnect DG when islanding occurs is not appropriate. The utility shall fully utilize the DG to continue supplying power to the load in the island system. However, without a proper control of an islanding operation, the idea to implement islanding could create risks and hazards to the island and grid. A smart distribution system so called Smart Grid comprising monitoring, advanced control and communications capabilities can facilitate automation control for a stable and seamless islanding operation.
intelligent UFLS scheme adopting the event-based and response-based method as proposed in [7] to tackle those scenarios respectively. It would be able to intelligently and optimally shed the load in accordance with the load priority. Dynamic simulation is carried out considering existing Malaysia distribution network that consists of two small units of hydro generation connected in parallel to the grid. The study is simulated using the PSCAD/EMTDC simulation tool. The effectiveness of the proposed load shedding scheme in the islanded system is investigated and discussed. In order for the islanded system to have a seamless operation and a successful load shedding implementation, the distribution grid is assumed to be facilitated with the smart grid technologies. II. INTENTIONAL ISLANDING STRATEGY In advance operation planning, a good control strategy and complete information on the system network are amongst the essential requirement to perform intentional islanding operation. Figure 1 shows the concept of the proposed strategy. The islanding scheme consists of a Main/intelligent Islanding Controller (MIC) located in Distribution Control Centre (DCC) which is able to communicate with the governor and excitation controller of the DG/DGs unit and the utility grid to perform the desired operation in the islanded system. The MIC receives/ transmits the signal from/to grid system and the islanded system via a good and reliable communication link. The MIC is designed to intelligently control the whole islanding operation in response to the opening of particular circuit breakers. As a result, the DG is isolated from the grid and islanding mode of operation is subsequently activated for continuously supply power to a particular island configuration/area. Any on-going islanding activities will be monitored at the DCC so that the operator could alert the maintenance personals the current status of the islanded system and to cease the islanding operation during emergency state. To support the planned islanding operation, this study has adopted the guideline on distribution DG islanding developed by BC Hydro research group [8]. The focus of the guideline is on the essential considerations, technical issues and operations and safety aspects required for the intentional islanding operation of DG. It highlighted various considerations in every perspective of islanding operation issues of which are very useful to be adopted for any microgrid/islanding operation. A. Generator Controller A key success of the islanding operation is relying on the DGs governor that specifically performs the speed and active power control of the DG. In Malaysia, most of the plants utilized hydraulic mechanical governor, but when considering islanded system, a better approach is to use governor with PID controller. The governor applied in the study is as illustrated in Figure 2.
Figure 1: Control strategy for an islanded operation
The controller is designed in two modes of operation; grid connected and islanded. For grid connected operation, fixed power control mode is adopted whereas for islanded operation, isochronous mode is applied. The fixed power mode is represented by the predetermine load reference value whilst the isochronous mode is represented by the PID controller. The PID controller helps to maintain the stability of the islanded system. It is used to bring the speed/frequency and the voltage phasor as close as possible to their reference value, thus facilitate the synchronization process. During transient response, the system frequency for the hydro turbine application is controlled to be within the limit of 47.5Hz52Hz; otherwise the over/under-frequency protection will trigger and trip off the DG to avoid the risk of out of phase reclosure.
Figure 2: Speed-active power (Pf) control block diagram using PID governor III. PROPOSED LOAD SHEDDING SCHEME FOR ISLANDING OPERATION The proposed load shedding scheme introduces a new approach of UFLS Scheme for an islanded system. In general, the scheme uses rate of change of frequency (ROCOF) to determine the amount of load to be shed. The load shedding is performed according to the load priority. The loads are classified into 3 categories; vital, semi-vital and non-vital. The non-vital load will be the first priority to be shed, followed by the semi-vital and lastly the vital load. Figure 3
illustrates the overall concept of load shedding scheme for an islanded distribution network connected with Mini Hydro power plants.
M a in G rid
1563.4kW) are connected together in the network. As indicated in Figure 4, a Normal Off Point (NOP) for 11kV subsystem is located at the intersection of two load feeders. During normal operation, the required total power output to be delivered by the DG units to the distribution network is 2.8MW. From the load flow analysis using the PSSE software, a total of 2.5MW is absorbed by the loads in the network (including distribution losses) and the remaining is exported to the grid system.
Island ed Distrib ution Ne tw ork

Load Shed C ontroller M odule
D istrib u te d G e n eratio n DG1 Frequency C alculator M odule DG2
Figure 3: Concept of proposed load shedding scheme for an islanded network The scheme consists of two main modules; (a) Frequency Calculator (FCM) and (b) Load Shed Controller (LSCM). The FCM will continuously monitor the frequency of the DG and send the real-time frequency value to the LSCM via the communication link. Real-time measurement unit and Remote Circuit Breaker (RCB) are facilitated at each of the load feeder. The system state variable measurement (i.e. active power, frequency and voltage) and the breaker status are monitored by the LSCM. The system information is required by the LSCM to determine or estimate the amount of power imbalance that will be used to identify the right amount of load to be shed. Consequently, the LSCM will initiate load shedding instruction to the RCB to disconnect the corresponding load feeder. IV. TESTING ON DISTRIBUTION NETWORK In this study, the proposed load shedding scheme is incorporated into the proposed intentional islanding operation. A simulation model is developed using the test system as shown in Figure 4. The two mini hydro generation units rated 2MVA are operated at 3.3kV voltage level. Both DGs are synchronous generators which are equipped with a governor, a hydraulic turbine with all the necessary valves for control of water flows and an excitation controller. Two parallel units of 2MVA generator transformer are connected to the DGs to stepping up the voltage level to 11kV. The distribution network is connected to the transmission grid via two parallel step up transformer (11kV/132kV) rated 30MVA. A number of load feeders (total peak load: 2327.4kW, light load:
The system is islanded when there is a loss of main at the grid substation feeder, RCB1. A new bus section that interconnects two transformer feeders (bus 1106 and 1105) is proposed to bypass power generation from DG to a part of the load that being isolated by the NOP. This is also to allow a bigger islanding coverage thus optimize the available generation. If the bus section is left opened, the load located at the right side of the network will be disconnected and thus loss their power supply throughout the islanding operation. The bus section is set to normally off during normal operation and normally on upon the activation of the first islanding configuration. This islanding operation would require load shedding when one of the DG is forced or tested to trip off either immediately after system islanded or during islanding operation. This study keeps the existing load data and uses the proposed controller as of discussed previously. The load shedding scheme is integrated and controlled in the MIC. Furthermore, instead of measuring at each of load feeder, the load is measured at several main feeders. This is due to reduce the measurement points, and hence reduce the overall cost to install measurement units. Those points with the amount of load to be shed are tabulated in Table 1. The location of measurement points is as depicted in Figure 4. As highlighted in Table I, the load is ranked based on the load category: non-vital, semi-vital and vital, in which independent on the amount of load. The amount of load on each of the feeder is varied within the range of peak to light load. The variation is to indicate the load profile of the island system. It is also used to investigate the effectiveness of the proposed load shedding scheme. Apart from peak load, there are three other load variations as presented in Table 1. The total amount of load is reduced from 2.327MW (peak load) to 2.16MW, 1.942MW and 1.687MW respectively. These amounts of load are selected considering the island could overload in the event of one DG is tripped off. Note that light load scenario is not considered in this study since the total amount of load (1.56MW) is within the generation capacity, thus will not trigger the load shedding scheme.
Figure 4: Load measurements points in the test system TABLE I. LOAD RANKING TABLE
Load Ranked 1 2 3 4 5 6 7 8 9 10 11 Total Bus Number Peak Load P (MW) 0.022 0.046 0.326 0.088 0.053 0.132 0.095 0.132 0.132 0.577 0.723 2.327 Q (MVAR) 0.014 0.028 0.2 0.055 0.033 0.082 0.059 0.082 0.082 0.358 0.448 1.441 Intermediate Load 1 P (MW) 0.021 0.042 0.302 0.082 0.049 0.123 0.088 0.123 0.123 0.536 0.671 2.16 Q (MVAR) 0.013 0.026 0.186 0.051 0.031 0.076 0.055 0.076 0.076 0.332 0.416 1.338 Intermediate Load 2 P (MW) 0.019 0.038 0.272 0.074 0.044 0.110 0.079 0.110 0.110 0.482 0.604 1.942 Q (MVAR) 0.012 0.024 0.167 0.046 0.028 0.068 0.049 0.068 0.068 0.299 0.374 1.203 Intermediate Load3 P (MW) 0.016 0.033 0.236 0.064 0.039 0.096 0.069 0.096 0.096 0.418 0.524 1.687 Q (MVAR) 0.01 0.020 0.145 0.039 0.024 0.059 0.043 0.059 0.059 0.259 0.325 1.042 Load Category
1026,1047 1013 1059,1057, 1154, 1050 1055 1012 1060 1079 1056 1058 1046,1018, 1019,1020,1168 1157,1004,1141, 1151,1064,1029, 1010,1039
Non-vital Non-vital Non-vital Non-vital Non-vital Semi-vital Semi-vital Semi-vital Semi-vital Vital Vital
A.
Case Study
In order to test the proposed scheme, an intentional islanding is simulated on the developed model of test system network. Prior to islanding, a total power of 2.8 MW is dispatched by the generators. Islanding operation is simulated and activated with the opening of incoming grid substation breaker, RCB1 at time t=10 seconds (s). Several events are simulated to justify the effectiveness of the proposed load shedding scheme when integrated with the intentional islanding operation. The scheme should be able to cope with different range of power imbalance between load and generation following the events. The events are simulated
on different loading conditions (as recorded in Table I), which are peak and three ranges of medium load. The tested events are summarized in Table II. TABLE II. CASE STUDIES
Case studies Case I Case II Descriptions System islanded and one of DG is tripped off during islanding operation. System islanded with different range of load condition.
1)
Case I
ranked 1 until 9 (based on the look up table as in Table I). The load shedding value is summarized in Table III. As can be observed in Figure 5, following the tripped, the active power dispatched for DG1 for peak and medium scenario increase to 1.36MW and 1.28MW respectively. Meanwhile, the system frequency of the peak and medium load drops to 48.77Hz and 48.81Hz respectively, resulting from the load shedding. Without the LSCM, the frequency might drastically drop and eventually become unstable. It should be noted that the frequency gradually recovers to nominal value but with some overshoot with a peak value of 50.6 Hz. The rise could be due to the step response of the developed PID controller. It is also could be due to the load is not optimally shed considering the reduced load measurement points introduced in this case study. More measurements points mean more load will be appropriately ranked. This would result in optimal load shedding. Most importantly, the frequency transient response for both events are within the acceptable frequency limit of 47.5Hz and 52Hz [9]. Thus, this proves that the proposed load shedding scheme has successfully sustained the islanding operation by shedding the load according to its priority and load disturbances. TABLE III. LOAD SHEDDING BY LSCM
In this case, two scenarios are considered, of which system islanded with peak and medium load with a total amount of load is 2.32MW and 2.16MW respectively. In order to observe the response of the load shedding scheme, one of the DG (DG2) is tripped off during islanding operation at simulation time, t = 45 seconds. The simulation result is as shown in Figure 5. The simulation results present comparison responses of DG1 serving two different amounts of load.
Active Power(MW)
2 1.5 1 0.5 0 0 52 51 50 49 48 0 1.08 1.06 1.04 1.02 1 0.98 0.96 0 10 20 30 10 20 30
DG1 PEAK DG1 MEDIUM DG1 PEAK DG1 MEDIUM DG2
10
20
30
Time(s)
40
50
60
70
80
Frequency(Hz)
Time(s)
40
50
60
70
80
Voltage(pu)
DG1 PEAK DG1 MEDIUM
Load Case Peak Load Medium Load
Power Imbalance (MW) 1.16 1.08
Load Shed (MW) 1.026 0.953
Time(s)
40
50
60
70
80
Figure 5: Simulation results for Case I. Following islanding, the system frequency of peak and medium load rises to 51.06Hz and 51.76Hz respectively and then slowly recover to 50Hz. The high frequency of medium load as compared to peak load is due to the large power imbalance resulting from system islanded. This can be observed from the reduced amount of active power dispatched by the generator immediately after system islanded. The power dispatched for medium load is reduced more than the peak load. It shall be noted that there is no overload in the system during this time, thus no load shedding is required (LSCM is not initiated). A different scenario can be observed when DG2 is tripped off at t= 45s. Following the event, the speed of the DG1 starts to decelerate as a result of losing half of the island generation (1.16MW and 1.08MW out of total peak and medium load respectively). Each DG has a provision of spinning reserve to be used during emergency state. However, for this case, the spinning reserve is only allocated with a total of 0.2MW for each DG. Considering the spinning reserve, the LSCM initiates its event based strategy and quickly shed a total of 1.026 MW and 0.953MW of the peak and medium load from
As for the voltage, following islanding, the voltage level drops from 1.04p.u to 0.98p.u and finally restore to 1.0 p.u. However, for the second event, the voltage steeply increases to 1.06p.u of which is acceptable when it is resulting from a large disturbance [9]. 2. Case II
In this case, intentional islanding operations consider two different total amount of active load which are 1.94MW and 1.68MW respectively. These two medium loading ranges are introduced into the system under study for investigating the response of the proposed load shedding scheme with different loading scenario. Different consideration is needed when total amount of load in the test system is less than 2.1MW. The power imbalance is huge such that if the normal islanding operation is carried out, it could lead to overshoot in frequency of the generator to more than 52Hz. This will initiate the over frequency protection and eventually cease the islanding operation. To prevent this, DG2 is forced to trip off following system islanded. Thus, DG1 is left to supply the whole island. Considering a spinning reserve of 0.2MW for each DG, DG1 is only capable to supply maximum power of 1.6MW to the island. In this case, it is obviously required load shedding to stabilize the islanding operation. The
simulation results comparing two responses of DG1 in different loading condition are as shown in Figure 6.
Active Power(MW)
2 1.5 1 0.5 0 -0.5 0 20 40 60 80
DG1 1.94MW DG1 1.68MW DG2
deviation of voltage from 1.055p.u to 1.01p.u is caused by a large transition of reactive power from grid connected to islanded system. The value is depending on the change of reactive power dispatched by the DG in which relies on the dispatched power during grid connected and the reactive load in the island. V. CONCLUSION
Time(s) Frequency (Hz)

50.5 50 49.5 49 48.5 48 47.5 0 1.06 10 20 30
DG1 1.94MW DG1 1.68MW
Time(s)
40
50
60
70
80
1.04 1.02 1 0.98 0 10 20 30
DG1 1.94MW DG1 1.68MW
Time(s)
40
50
60
70
80
The concept of intentional islanding operation is presented in this paper. The proposed intentional operation has incorporated a new approach of UFLS scheme to solve the stability and under-frequency issue for an islanded system in a distribution network. Two strategies based on the response based and event based have been proposed. The effectiveness and robustness of the scheme have been investigated based on two scenarios of event based. It is observed that the developed scheme managed to immediately perform the load shedding in one step. Simulation results show that the optimal amount of load can be shed according to the well-determined load prioritization. This proves that with a proper DG control and load shedding scheme dedicated for the island, a successful islanding operation can be achieved. The power quality of the island can be sustained thus allowing the island to persist. ACKNOWLEDGEMENT This work was supported by the Malaysian Government and University of Malaya, Kuala Lumpur under HIR/MOHE research Grant (Grant Code: D000004-16001) and the University of Technology MARA, Malaysia. REFERENCES
[1] F. Katiraei, C. Abbey, S. Tang, and M. Gauthier, "Planned islanding on rural feeders ; utility perspective," in Power and Energy Society General Meeting - Conversion and Delivery of Electrical Energy in the 21st Century, 2008 IEEE, 2008, pp. 1-6. T. Tomsic, G. Verbic, and F. Gubina, "Revision of the underfrequency load-shedding scheme of the Slovenian power system," Electric Power Systems Research, vol. 77, pp. 494-500, 2007. J. G. Thompson and B. Fox, "Adaptive load shedding for isolated power systems," Generation, Transmission and Distribution, IEE Proceedings-, vol. 141, pp. 491-496, 1994. C. Concordia, L. H. Fink, and G. Policies, "Load shedding on an isolated system," Power Systems, IEEE Transactions on, vol. 10, pp. 1467-1472, 1995. P. Mahat, C. Zhe, and B. Bak-Jensen, "Underfrequency Load Shedding for an Islanded Distribution System With Distributed Generators," Power Delivery, IEEE Transactions on, vol. 25, pp. 911-918, 2010. E. E. Aponte and J. K. Nelson, "Time optimal load shedding for distributed power systems," Power Systems, IEEE Transactions on, vol. 21, pp. 269-277, 2006. H. Seyedi and M. Sanaye-Pasand, "New centralised adaptive load-shedding algorithms to mitigate power system blackouts," Generation, Transmission & Distribution, IET, vol. 3, pp. 99-114, 2008. BCHydro. (2006, Distribution power Generator Islanding Guideline Available: www.bchydro.com S. Tenaga, "The Malaysian Distribution Code," 2010.
Voltage(pu)
Figure 6: Simulation results for Case II. As can be observed in Figure 6, the active power dispatched by DG2 drastically drop to zero as a result of shutting down the DG following islanding. The system frequency begins to decrease in response to an excess of load in the island. The LSCM start to initiate its event based strategy to prevent further decrement of frequency. By taking into account the reserved generation, the power imbalance is determined. The power imbalance and total load shed for medium load 2 and 3 is presented in Table IV. TABLE IV. LOAD SHEDDING BY LSCM
Load Case Medium Load 2 - 1.94MW Medium Load 3 - 1.68MW Power Imbalance (MW) 0.34 0.087 Load Shed (MW) 0.329 0.049 [4] [2]
[3]
In this case, 0.329MW (load ranked 1 to 3) and 0.049MW (load ranked 1 to 2) of load is shed from medium load 2 and 3 respectively. As a result of load shedding, the frequency drops to 48.33Hz and 48.10Hz respectively and manage to recover to its nominal. However, a small overshoot in frequency with a value of 50.2Hz and 50.3 Hz is noticed for both respective load conditions. The frequency response however has demonstrated the effectiveness of load shedding scheme integrated with the proposed MIC to ensure the stability of the intentional islanding operation. Meanwhile, the voltage spikes to a value of 1.055p.u following islanding and then recovers to 1.01p.u during steady state. The large
[5]
[6] [7]
[8] [9]
Flexible DG Interface Using Repetitive Control

Amr I. Said, Student Member, IEEE, Ehab F. El-Saadany, Senior Member, IEEE
Electrical and Computer Engineering Department University of Waterloo Waterloo, Canada
AbstractDistributed generation (DG) units are widely spread in most electrical power systems. Such large extensive presence attracted many researchers to optimally exploit the available DG units in order to enhance the overall performance of the main power system grid. In this work a repetitive control-based technique is applied to operate the DG unit as an active power supplier and power quality enhancer. The enhancement provided to power quality is summarized in filtering load current harmonics and supporting reactive power sharing as well. The main advantage of the repetitive control is simplicity. Current filtering, and active and reactive powers sharing control are done directly in the abc frame. There is no need for the dq, and its inverse, transformation. However, the applied technique shows a high quality performance under normal conditions and fast, robust performance under sudden load changing. Keywords- Active power filter (APF), distributed generation (DG), repetitive control.
Figure 1 the. Typical DG unit interfacing via electronic converter
I.
INTRODUCTION
Electrical distribution networks undergo severe stresses due to increase in nonlinear loads. That results in high reactive and harmonic components in systems currents. Consequently, AC power lines losses, load side voltage variations, and sensitive equipments malfunctioning increase too. The common Active Power Filter (APF) [1], [2] presents an effective solution. Now APF is mature enough in industry and extensively spread commercially despite its inverter overhead cost. Our world is in a very dangerous threat due to the global warming. The main reason for the global warming is the use of fossil fuel in energy production. So, nations have committed reduction of fossil fuel stations and utilization of green energy sources instead. Wind energy, photovoltaic cells, and fuel cells are very promising candidates. The main concern that must be pointed out here is that the power generated via green sources is formed as a DC in one of the generating phases. The topology of the electrical power grid has been completely changed with increasing the green sources penetration. A green source (DG unit) is connected with the main grid via electronic converters as shown in Figure 1. Generally speaking the insertion of the electronic converter is considered as an unreasonable cost from the economical perspective [3]. To relieve that economic burden, the electrical converters should be utilized in the most efficient way. A remarkable notice is that the DG interfacing and the active power filter have the same construction. That inspired Kuo and Liang [4] to develop a dual operation PV system. The
proposed technique aims at exploiting the DG converter to operate as active power filter during rainy days and to do its normal function, as an active supply, in sunny days. Another approach proposed by [5] is to make use of agentbased communication as well as local measurement to provide nonactive power in addition to active power supply in order to compensate distorted currents. The authors of [6], [7] implemented the idea of dual operation of DG inverters to control the source active power and handle power quality issues improving system balance, filtering harmonics and supporting reactive power. Authors in [8] proposed repetitive control to handle the dual operation of DG inverters. The repetitive control has eliminated the dq transformation required to control the inverter. That results in a faster operating mechanism without additional H\W required for dq transformation process. Although the proposed technique shows a good performance in filtering current, and controlling DG active power and maintaining unity power factor load; it has two main shortfalls. First, it lacks the flexibility to control reactive power shared by the DG. In other words, the control loop aims to totally compensate the reactive power required by the load, so the load is always seen as unity power factor load from the grid source perspective. Even if this operation is preferable from the source side, it puts additional stresses over the DG inverters. The second drawback is the moderate speed response after a sudden change in loading. In this work, a modification for the control loops is illustrated in order to solve the mentioned draw backs. The proposed control loops will mainly fulfill the same targets
with a wider capability of reactive power sharing control with faster transient response. This work is organized as follows: Section II gives an overview of the control structure as overall. Detailed information on the current control loop is given in Section III, whereas Section IV 0 covers the voltage control loop and active power sharing. Results of the achieved work are discussed in Section V. Finally, the work is concluded in Section VI.
internal model of i via the first couple of poles at fundamental frequency. That guarantees zero steady state error; i.e., lim ( ) 0, where ( ) is the error as a function in time. The analysis of control loop performance against disturbance is investigated in (2) that relates the grid current and the disturbance - the load current in our case.
( ) II. CONTROL STRUCTURE A complete control structure is shown in Figure 2. The main structure can be divided into two main loops, outer and inner. The purpose of the outer loop is to assign the required for the main grid current sinusoidal reference waveform ( : , , ). is composed of three components: fundamental component of the load current, DG active, and reactive current sharing (1).The fundamental component of load current ( ) is directly extracted from load current via Phase Locked Loop(PLL), whereas the reactive power sharing ( ) is obtained by a separate open loop shown in Fig.3. The is scaled with the knowledge of reference reactive power system voltage- and then filtered via low pass filter to get is the ( ) . The final reactive current waveform multiplication of I ( ) and a waveform leads the v by 90 as illustrated in Figure 3. The last component, active current sharing i , will be extensively discussed in Section IV 0 ( ) ( ) ( ) ( ) (1)
( )
( )
(2)
It can be noticed that the transfer function of (2) has zero gain for any sinusoidal inputs with frequencies integer multiples of ( ). Based on Fourier analysis, the load current can be represented by a summation of sinusoidal waveforms that have integer multiples of fundamental frequency . This interprets the success of repetitive control in rejecting load fundamental and harmonics disturbances.
( )
The inner loop is responsible for handling the actual grid current , i , in order to match the required reference current ,i . This is the most challenging part, as we have a sinusoidal reference current to be tracked. It is a common practice to use dq transformation to transform the sinusoidal reference to a constant one so common PI controller can be employed[9]. However, the repetitive control can take over tracking the reference sinusoidal wave forms directly without any dq conversion.
III.
Figure 2 Main control structure CURRENT CONTROL LOOP

( )
The current control loop is shown in Figure 4. The reference i is compared with the actual grid current i . The error is passed to a repetitive controller and then to the inverter system (G (s) ). It is worth mentioning that the load current , , is considered a disturbance in this analysis. What we are concerned about in this control loop are performance and stability. The performance is judged by the value of the steady state error and the ability to reject disturbance. The repetitive controller can be simplified as shown in Figure 5. Neglecting the effect of ( ), which is a single order low pass filter, it is obvious that the repetitive controller resembles an infinite number of poles coincide with frequencies of integer multiples of ( ). By assigning , where is the system angular frequency (50 Hz in our case), we create an
Figure 3 Reactive power control loop
( )
( )
( )
( )
log (Ms)
( )
So
( )
Magnitude (db)
Kq
Figure 4 Current control loop

()
()
()
()
()
-log (Ms)
() 1
1 ( )
( )
Figure 5
Simplified current control loop

Ws Frequency (Hz)
The current loop stability is investigated via the loop characteristic equation (3). 1 G (s) K e
G ( ) L
( )
(3)
Figure 6
the Closed loop criterion to guarantee stability

() ()
Where: G (s)
()
()
( )
()
Defining the sensitivity transfer function S : As : S (4). 1 S K e

T
Figure 7 Voltage control loop

G ( )
The closed loop characteristic equation can be rewritten as 0

T
G (s)
KI
(6)
(4) 1 (small gain The voltage loop characteristic equation is given by (s) CV s
V K
The loop is stable iff S K e theorem). However, S K e

T
V KI
(7)
With the undamped natural frequency S K 1 (5)

V KI CV
(8)
and he damping ratio is governed by The last inequality in (5) is easier to be traced to verify the . stability condition through the design of Defining and : Where, ( ) 1, system stability can be guaranteed if criterion is illustrated in Figure 6. IV. ( , | ( )| ) . This
V KP CV
(9)
VOLTAGE CONTROL LOOP
The main purpose of the voltage control loop is to keep the voltage on the DC capacitor constant. The outline of the voltage control loop is shown in Figure 7, where based on the proof provided in [8]. For simplicity the controller is a PI controller. ( ) applied in that system
The active current i (t) is a multiplication of the PI controller loop output, I ( ) , and a unity sinusoidal waveform in phase with the grid voltage as illustrated in Figure 2. So, a smooth I ( ) is required in order to reduce higher harmonics in the reference current i (t) (1). If the bandwidth of the voltage loop is set to a value much lower than the grid supply frequency, the disturbance effects of PL (t) and PDG (t) will be substantially attenuated by the voltage control loop. If the bandwidth of the voltage loop is set to 1/n times the supply frequency, f , the integral K I becomes
( ) CV V
KI
(10)
The proportional gain K can be calculated to follow unity power damping ratio as follows:
CV V
Currents (A)
30 20 10 0 -10 -20
Grid Current Load Current
(11)
However, the voltage loop will slowly respond to any sudden changes in the load value due to the low bandwidth. That is why the direct feed-forward signal from load is proposed in this work. The roll of the feed-forward configuration is to fasten the overall system response after load changing.
-30 -40 1.24 1.25 1.26 1.27 1.28 Time(sec) 1.29 1.3 1.31
Figure 8 Operating the DG unit as an active power filter
V.
RESULTS
20 Currents (A) 0
For the purpose of illustrating the main outcome of this work a Matlab Simulink was employed. Figure 8 shows the performance of the proposed system in filtering a square waveform load current, whereas Figure 9 shows the filtration of the same load current in addition to active power sharing from the DG unit. Harmonics of load and grid currents in the former case are shown in Figure 10. It is clear how the proposed configuration succeeds in reducing the harmonic components from 24.85% - in load current- to 1.07% -in main grid current. To illustrate the effectiveness of the proposed configuration in controlling the DG reactive power sharing, three different cases are provided. In the different cases the load is highly inductive (PF=0.7 lagging) but with different reactive power sharing from the DG side. In case 1, shown in Figure 11, the DG just contributes in the active power sharing. Case 2, shown in Figure 12, the DG takes over all the load reactive power. Case 3 which is a pure theoretical case, shown in Figure 13, the DG provides reactive power over the load requirements. Such flexibility in reactive power compensation is missed in.[8]. The system transient response is tested by applying a sudden 50% load reduction. Figure 14 and Figure 15 show the main grid current and capacitor voltage transients after the load change without the load feed-forward signal. The grid current takes around 20 cycles to cope with the load change, whereas the DC capacitor voltages increases from 650V to 720V before reaching its initial steady state value again. The same situation is shown in Figure 16 and Figure 17 after adding the feedforward current signal loop. However, the grid current copes with the change of load current after 2.5 cycles approximately, whereas the DC capacitor voltage rises to 661V before a fast return to its initial value of 650V. This is another aspect where the proposed system outperform the proposed one in [8].
-20
-40 1.51 1.52 1.53 1.54 1.55 Time(sec) 1.56 1.57 1.58 1.59
Figure 9 Operating the DG unit as an active power and active power supply
1 Ratio of Fundamental 0.8 0.6 0.4 0.2 0
4 5 6 Harmonic Order
10
Figure 10 FFT for load and grid currents waveforms
20 Currents (A) 10 0 -10 -20 -30 0.26
Supply Voltage Grid Current Load Current
0.28
0.3
0.32 0.34 Time(sec)
0.36
0.38
Figure 11 Load current, grid current and voltage case1
Supply Voltage Grid Current Load Current Currents (A) 40 20 0 -20 -40 -20 0.46 0.47 0.48 0.49 0.5 0.51 Time(sec) 0.52 0.53 0.54 1.96 1.98 2 2.02 2.04 2.06 Time(sec) 2.08 2.1 2.12 Grid Current Load Current
20
Currents (A)
10
-10
Figure 12 Load current, grid current and voltage case 2
Figure 16 Load and grid currents after sudden change in load current with feed-forward load signal
700
20 10 Currents (A) 0 -10 -20
Capacitor Voltage (volt)
Supply Voltage Grid Current Load Current
690
680
670
660
650
0.26
0.28
0.3 0.32 Time(sec)
0.34
0.36
1.6 1.8 2 2.2 2.4 2.6 Time(sec) 2.8 3 3.2
Figure 13 Load current, grid current and voltage case 3
Figure 17 DC Capacitor voltage after sudden change in load current with feedforward load signal
30 20 Currents (A) 10 0 -10 -20 -30 -40 2
Figure 1. Figure 2. Figure 3. Figure 4. Figure 5.

2.1 2.2 2.3 2.4 2.5 Time(sec) 2.6 2.7 2.8 2.9 3
VI.
CONCLUSION
Figure 14 Grid currents after sudden change in load current without feedforward load signal
In this work a system is proposed for controlling the DG unit to serve multipurpose functions. On one hand the proposed configuration handle the DG active power. On the other hand, it plays a significant role in enhancing the power quality of the main grid. The power quality aspects tackled, in this work, are the harmonic filtration and power quality improvement. As illustrated through the simulation results, the proposed system succeeded to fulfill the active power control and the power quality enhancement. This work outperforms the configuration proposed in literature in two main points. The first is the flexibility in controlling the reactive power sharing of the DG unit. According to the proposed configuration the DG reactive power can be directly set to a reference KVAR to control the degree of reactive power compensation. The Second is the fast transient response after the load changing. This control design depends on a direct feed-forward from the load to fulfill this purpose.
720 Capacitor Voltage (volt) 700 680 660 640 620 600 2
ACKNOWLEDGMENT
2.5 Time(sec) 3 3.5
Figure 15 DC Capacitor voltage after sudden change in load current without feed-forward load signal
Authors would like to express profound gratitude to Mrs. Jane Russwurm for the generous time she offered to enhance the writing quality of this paper.
REFERENCES
[1] K. Tsang and W. Chan, "Design of single phase active power filter using analogue cascade controller," IEE Proc. Electr. Power Appl., vol. 153, no. 5, 2006. S. Gonzalez, R. Carcia-Retegui and M. Benedetti, "Harmonic computation technique situable for active power filters," IEEE trans. Ind. Electron., vol. 54, no. 5, 2006. F. Blaabejerg, R. Teodorescu and M. Liserre, "Overview of control and grid synchronization for distributed power generation systems," IEEE Trans. Ind. Electron., vol. 53, no. 5, 2006. Y.-C. Kuo, T.-J. Liang and J.-F. Chen, "A high-efficiency single-phase three-wire photovoltaic energy conversion system". K. Macken, K. Vanthournout and J. Van deneybus, "Distributed control
of renewable generation units with integrated active filter," IEEE Trans. Power Electron., vol. 19, no. 5, 2004. [6] M. Cirrincions and G. Vitale, "A single-phase DG generation unit with shunt active power filter capability by adaptive neura filtering," IEEE Trans. Ind. Electron., vol. 55, no. 5, 2008. X. Tang, K. Tsang and W. Chan, "A power quality compensator with DG interface capability using repetive control," IEEE transactions on energy conversion, vol. 27, 2012. H. Akagi, Instantaneous power theory and applications to power conditioning, IEEE press, John Wily and sons, 2007. A. Yazdani and R. Iravani, Voltage sourced converters in power systems, John Wily and sons, 2010.
[7]
[2]
[8] [9]
[3]
[4] [5]
Solution to Economic Emission Load Dispatch Problem Using Modified Artificial Bee Colony Algorithm
H.T. Jadhav Electrical Engineering Department Rajarambapu Institute of Technology Islampur, India. htj@ritindia.edu Shubham Raj Electrical Engineering Department SV National Institute of Technology, Surat, India shubhamraj108@gmail.com Ranjit Roy Electrical Engineering Department SV National Institute of Technology, Surat, India rr@svnit.eed.ac.in
Abstract This paper presents an application of modified artificial bee colony algorithm (MABC) to determine the optimized solution of economic and emission load dispatch (EELD) problem. The EELD problem is formulated as a biobjective problem by taking minimization of fuel cost and emission levels as objectives. In order to convert a biobjective problem into a single objective function weighing factor is used. Effectiveness of the MABC algorithm is verified by applying it on five standard test systems and the outcomes are compared with the latest reported literatures. It is proved from the results that MABC algorithm is more powerful than other algorithms. Keywords- economic and emission load dispatch; modified artificial bee colony algorithm; price penalty factor; weighing factor
employed optimization techniques is presented. In section IV result and discussion of different test cases is provided. Finally, conclusions are drawn in section V. II. EELD PROBLEM FORMULATION
The EELD problem seeks the best generation schedule for the generating plants to supply the required demand plus transmission losses with minimum production cost and emission. The objective function and constraints for the EELD problem are as formulated below. A. Classical EELD problem In this paper the objective function which represents total cost of operation (TC) consists of two independent objectives namely fuel cost (FC) and emission (E). The bi-objective EELD problem is expressed in single-objective form as in (1) using a price penalty factor ( h ) [3]. Minimize; FEELD = w1 FC + w2 hE subject to; 1. Power balance constraint, (1)
I.
INTRODUCTION
A considerable segment of the worlds power plants are using fossil fuels like natural gas, coal, oil as principal resource for production of electricity. It is becoming most important to make use of existing resources consciously and supply electricity at lowest rate. Economic dispatch (ED) is a key task in power system operation and planning. The primary objective of classical ED is to allocate the amount of power produced by generating units to among the loads and limit total operating cost while satisfying all equality and inequality constraints of the system. As a result of global warming, the environmental pollution is becoming an alarming aspect to the world. Therefore the classical ED problem is modified to economic emission load dispatch (EELD) to produce power in cost effective manner and also with minimum pollution [1]. Artificial bee colony (ABC) algorithm invented by Karaboga has been proved to be more effective than some conventional biological-inspired algorithms like genetic algorithm (GA), differential evolution (DE) and particle swarm optimization (PSO). But, ABC is superior at exploration and poor at exploitation. This paper presents a recently developed optimization method, where ABC algorithm is modified to guide the search of candidate solution towards the global optima [2]. The significant contributions of this paper are as follows: Modified artificial bee colony algorithm (MABC) is implemented to solve EELD problem and applied to five standard test systems consisting of 6, 10, 11, 14 and 40 thermal units. The corresponding results are compared with the methods available in recent literatures. Rest of the paper is arranged as follows: Section II provides the EELD problem formulation. In section III an overview of
P = P
i i =1
+ PL
(2)
2. 3.
Inequality constraint,
Pimin Pi Pimax
Ramp rate limit,
(3) (4)
Pi Pi ini URi Pi ini Pi DRi
where F C : fuel cost of ith thermal unit; E :emission in kg/h; w and w are weighing factors such that
1 2 w + w = 1 ; h is price penalty factor; Pi: output power of 1 2 ith generating unit in MW; n: number of generating units in
the system; PD :total load demand in MW; PL : total transmission loss of the system in MW; Pim ax and Pim in are the maximum and minimum generation confines of the actual power output of ith unit respectively; U R i and D R i are ramp up and ramp down limits of ith unit, respectively; Pi in i is the power output of ith unit at previous hour in MW/h.
B. Fuel cost functios Two different types of fuel cost curves are considered in simulation studies namely quadratic cost curve and quadratic cost curve with valve point loading effect and given in (6) and (7) respectively. In a practical power system, the power plants with several valves in steam turbines are available. To maintain the active power equilibrium these valves are opened and closed in sequence. The valve point loading effect adds ripples in the objective function and makes it extremely nonlinear.
In ABC algorithm, first half of the colony is occupied by the employed bees and the second half includes the onlooker bees. For each food source there is only one employed bee assigned i.e. the number of employed bees is equal to the number of food sources surrounding the hive. The scout starts its search cycle once the employed bee has exhausted its food source position. The colony consists of a set of possible solutions x i ( i = 1, 2, , eb ) , where i indicates the size of population and e b is the number of employed bees. The preference of selecting a food source pi is carried out probabilistically and the equation is given by,
pi = fiti k fitk
Nf
FCi ( Pi ) = (ai + bi Pi + ci Pi 2 )
i =1
(5)
F C i ( Pi ) = ( a i + bi Pi + c i Pi 2 ) + ei sin ( f i ( Pi m in Pi )) (6)
where a i , b , ci i
(9)
ei and f i are cost coefficients of ith unit.
where
fiti is fitness value of the of ith solution and
C. Prohibited operating zones Thermal power generating units may have certain POZs where operation is constrained due to physical limits of the power plant components, steam valve, vibration in shaft bearing, etc. These zones generate discontinuities in cost curve and the cost function is modeled as follows.
L Pim in Pi Pi ,1 L Pi = Pi U , k 1 Pi Pi , k , k = 2, zi PU P P i i , m ax i , zi
N f is the number of food sources. Once the onlooker bee

chooses a food source, it is further modified by the following equation,
( xik x jk ) where k {1, 2,...e } , j {1, 2,....d } and b

v ik = x ik + ik
(10) d are the
(7)
dimension of the problem
ik
is a random number between
are the lower and upper limits of where Pi ,Lk , Pi U ,k prohibited operating zone of unit i, prohibited operating zones in unit i.
th
zi is the number of
D. Emission function: The overall fuel emission of environmental pollutants such as green house gases, SOx and NOx caused by fossil fuel fired thermal power plants can be expressed as [4], (8) E i ( Pi ) = 10 2 ( i + i Pi + i Pi 2 ) + i exp ( i Pi )
where i , i , i
[1, 1]. If the modified food source has a better or an equal nectar amount than the previous one, it will replace older one and becomes a new member of the population, that is, a greedy selection method is applied between new and old food sources. If the new food source cannot be further modified after a certain number of trail limits, the food source is discarded and consequent bee becomes a scout bee. Trial limit is the essential control factor of the ABC algorithm. The scout bee randomly discovers a new food source according to,
xi , j = xmin, j + r ( xmax, j xmin, j )
(11)
i and i
are the emission coefficients of
where r is a random number between [0, 1], xmin, j and xmax, j are the minimum and maximum limits of the
ith unit.
III. OVERVIEW OF OPTIMIZATION TECHNIQUES EMPLOYED
A. Artificial Bee Colony Algorithm

In 2005, Karaboga [5] introduced an innovative metaheuristic optimization technique called the artificial bee colony (ABC) algorithm. The family of artificial bee consists of employed, onlooker and scout bees. Each employed bee deals with just one food source (solution) at a time, which they are presently exploiting or are employed at. They hold information concerning this particular food source, its track and distance from the hive. Then, by a way of specific type of dance called waggle dance, employed bees share this data with the bees waiting in the hive. The bees waiting in the hives dance area are called onlooker bees.
ith solution respectively. The optimization algorithm has four control parameters namely; limit cycle value, number of employed and onlooker bees and maximum number of cycles (MCN).
j t h parameters of the of
B. Modified Artificial Bee colony Algorithm

Encouraged by PSO [6], which, in order to improve the exploitation capability, takes advantage of knowledge of the global best solution and to direct the search of candidate solutions in feasible region, equation (10) is modified as:
v ik = x ik + ik
where C is a random number between [0, 2]. According to (12), new candidate solution can be driven towards the global
( xik x jk ) + C (G b est
so l
ik
(12)
Gbest sol ; therefore it can boost the exploitation capability of ABC algorithm. The value of C should be
best solution suitably set to balance exploration and exploitation capability. IV. IMPLEMENTATION OF EELD BY MABC ALGORITHM The variables in optimization process are active power outputs of thermal units in MW. The step by step procedure to implement EELD problem using MABC algorithm is given below. Step 1: Specify the cost and emission coefficients, valve-point coefficients, prohibited operating zones, ramp rate limits and active power limits for each thermal unit. Specify parameters of the MABC algorithm such as ( N e ), ( N o ), trial counter limit and C. Step 2: Initial population is generated arbitrarily from the multi-dimensional search space,
Step 5: All onlookers chooses a food source position of employed bee with a probability Pi using (9). A particular food source with better quality attracts more onlookers. Step 6: Once the onlooker bee chooses a food source position of employed bee, it is further modified to obtain even better food source position by using (12). Again greedy selection is applied to retain better solution results and discard substandard solution. Step 7: If the solution is not improved for certain number of trial limits, the scout bee produces a new food source according to (13) Step 8: Remember the best solution obtained so far. Increment the iteration count. Step 9: If the stopping criterion is not meet, go to step 4. The stopping criterion in this case is the maximum number of cycles (MNC). V.
X = X 1, X 2,........ X m , where m is the size of the population and X 1, X 2 ,........ X m

are
X i = [ Pi1 ,......PiN ] is subjected to iterative search processes
candidate
solutions.
Each
solution
vector
of the employed bees, the onlooker bees and the scout bees. The index N represents the number of thermal units and uniformly between their upper and lower values as per (13).
i [1, m ] . All variables given by vector X i are distributed

Pij = p + ij Pmax, j P min, j min, j j = {1, 2,........ N}
(13)
For given scheduling horizon with N thermal units, an initial population is generated as shown in (14):
P 11 P 21 X = : : P m1
P 12 P 22 : : P m2
.. . . . .. . . . : : : . : : : . .. .. . ..
. : P mN
P 1N P 2N :
(14)
In order to demonstrate the effectiveness and achievability of MABC algorithm for economic emission load dispatch problem five test systems are taken into consideration. The algorithm has been applied to 6, 10, 11, 14 and 40unit test systems. The proposed method was implemented in MATLAB software. In each simulation study, the algorithm was run for 50 independent trials. The best results obtained for each test case are presented below. A. Six unit system This test system comprises six generating units with quadratic fuel cost curve. The cost coefficients and the emission coefficients are taken from [7]. The total load demand to be met is 1000 MW. The results of six unit test system with MABC algorithm are compared with other algorithms reported in previous literatures. The best compromising solution is given Table I while Table II gives comparison with -iteration [7], recursive approach [7], particle swarm optimization (PSO) [7], differential evolution (DE) [7], genetic algorithm (GA) [ 8], simplified recursive [7] and gravitational search algorithm (GSA)[8]. It can be seen that among all methods MABC gives lower value of fuel cost and emission simultaneously.
TABLE I. EELD PROBLEM SOLUTION USING MABC ALGORITHM FOR SIX UNIT TEST SYSTEM (TOTAL DEMAND=1000 MW). Unit 1 2 3 4 5 6 Fuel cost ($/h) Emission (kg/h) MABC 82.1907 78.9586 167.6496 162.6398 250.1576 258.4005 51249.0000 827.0461
Step 3: The objective function value of each food source position is evaluated using (1). Step 4: Each employed bee produces a new candidate food source in the surrounding area of its current position using equation (12). If this value exceeds its limit, the parameter is set to its limit value. The modified food source position is tested for all technical constraints. Then the fitness value of each candidate food source is evaluated. If the new fitness value is better than the old one, the newly found food source position is remembered; otherwise, the old one is retained.
TABLE II. COMPARISON OF MABC ALGORITHM WITH OTHER METHODS FOR SIX UNIT TEST SYSTEM (TOTAL DEMAND=1000 MW). Methods -iteration [7] Recursive approach [7] PSO [7] DE [7] Simplified recursive [7] GA [8] GSA [8] MABC Fuel cost ($/h) 51264.6 51264.5 51269.6 51264.6 51264.6 51262.31 51255.7880 51249.0000 Emission (kg/h) 828.720 828.715 828.863 828.715 828.715 827.2612 827.1380 827.0461
B. Ten unit system

In this case the test system consists of ten generating units with valve loading effects. The total load demand is set at 2000 MW. Cost coefficients and the emission coefficients of thermal generating units, and the loss coefficient matrix required to calculate the network losses are taken from [9].
TABLE III. COMPARISON OF BEST COST AND BEST EMISSION RESULTS FOR TEN UNIT SYSTEM (TOTAL DEMAND=2000 MW) Unit DE [9] Economic Emission dispatch dispatch 55.0000 55.0000 79.8063 80.0000 106.8253 80.5924 102.8307 81.0233 82.2418 160.0000 80.4352 240.0000 300.0000 292.7434 340.0000 299.1214 470.0000 394.5147 469.8975 398.6383 111500.0000 116400.0000 MABC Economic Emission dispatch dispatch 55.0000 55.0000 79.9851 80.0000 108.8087 81.8505 101.0186 82.2131 81.6073 160.0000 80.7539 240.0000 300.0000 291.4903 339.8778 296.1792 470.0000 395.8016 470.0000 399.0752 111500.0000 116400.0000
Table III indicates the results for best economic dispatch and best emission dispatch for ten unit test system with total load demand of 2000 MW. The results obtained by MABC algorithm are compared with those obtained by DE method. From table III it is obvious that MABC performs much better than DE method. As stated in table, the fuel cost obtained for best economic and best emission dispatch by DE and MABC algorithms are 111500.00 $/h and 116400.00 $/h respectively. However the emission level obtained by DE algorithm for the first case is 4581.00 kg/h and that by MABC algorithm is 4579.4054 kg/h. In second case i.e., best emission dispatch, the emission level mentioned by DE is 3923.40 kg/h but according to the schedule given by author actual emission level comes out to be 3932.40 kg/h. The emission level obtained by MABC algorithm is 3930.7606 kg/h which proves that the results achieved by MABC algorithm are superior to other algorithms reported in previous literatures. Fig. 1 shows the convergence characteristic of ten unit test system for best economic dispatch with load demand 2000 MW.
1 2 3 4 5 6 7 8 9 10 Fuel cost ($/h) Emission 4581.00 3923.40a 4579.4054 3930.7606 (kg/h) a According to schedule given by author emission is found to be 3932.40 kg/h.
1.16 1.15 T o ta l c o s t ($ /h ) 1.14 1.13 1.12 1.11 0 x 10
5
C. Eleven unit test system In this test case, eleven thermal generating units are considered. The system data like quadratic cost function coefficients, emission coefficients are taken from [7]. The total load is set to 2500 MW. Table IV shows the best compromising results obtained for this case by MABC algorithm and its comparison with other methods.
TABLE IV. COMPARISON OF THE BEST COMPROMISING SOLUTIONS FOR ELEVEN UNIT TEST SYSTEM (TOTAL DEMAND=2500 MW) Unit 1 2 3 4 5 6 7 8 9 10 11 Fuel cost ($/h) Emission (kg/h) Simplified recursive [7] 139.672 112.781 145.802 221.527 136.774 218.578 140.261 345.046 329.484 363.645 346.430 12,424.94 2003.300 GSA [8] 138.9382 110.2728 147.9728 221.1072 137.7986 217.9015 141.3801 349.6497 327.3178 363.4766 344.1847 12422.6626 2002.9499 MABC 131.3255 109.3195 154.6564 216.8766 177.1431 200.4872 171.2090 366.5573 307.5986 352.0699 312.7559 12418.0000 1988.4000
20
40
60
80
100 120 Iterations
140
160
180
200
As it is clear from the results that the fuel cost and emission level obtained by MABC algorithm is 12418.00 $/h and 1988.4 kg/h respectively which is much lesser than that obtained by simplified recursive and GSA method. The values of fuel cost and emission obtained by simplified recursive method are 12424.94 $/h and 2003.300 kg/h respectively and those by GSA are 12422.6626 $/h and 2002.9499 kg/h respectively.
Figure 1. Convergence characteristic of ten unit test system for best economic dispatch solution at load demand 2000 MW.
D. Fourteen unit test system In this example, fourteen thermal generating units are taken into consideration. The cost function and emission function coefficients and transmission loss coefficients are
taken from [10]. The total load demand for 14 unit system is assumed to be 2000 MW.
TABLE V. BEST ECONOMIC DISPATCH AND BEST EMISSION DISPATCH FOR 14 UNIT (TOTAL DEMAND TEST SYSTEM = 2000 MW) TribeMDE [11] Economic Emission dispatch dispatch 329.5196 189.5197 299.5997 150.0000 94.7998 130.0000 119.7331 130.0000 249.7331 162.5091 184.8666 270.407 135.0000 199.8422 109.8666 177.9192 124.7331 162.0000 137.1997 160.0000 57.9996 80.0000 80.0000 80.0000 62.3999 85.0000 52.3999 55.0000 37.7589 9180.9474 5069.2919 32.1972 10798.0159 2887.8203 MABC Economic Emission dispatch dispatch 239.2474 192.798200 150.0000 150.000000 85.0484 130.000000 130.0000 130.000000 196.3868 163.551400 239.8505 281.244200 215.2179 201.900400 188.3051 164.596500 162.0000 162.000000 156.8617 159.958200 74.7503 80.000000 77.4856 80.000000 82.8149 85.000000 37.8026 54.999800 35.7769 9146.8000 3341.2000 36.048800 10667.0000 2909.6000
Unit 1 2 3 4 5 6 7 8 9 10 11 12 13 14 Total loss (MW) Fuel cost ($/h) Emission (kg/h)
best emission dispatch obtained by GBAC algorithm and their comparison with other algorithms for 40unit test system with a load demand of 10500 MW. The results of MABC method are compared with Th PSO algorithm. The best fuel cost and the corresponding emission reported by Th PSO algorithm are 121412.5704 $/h and 359902.1885 kg/h respectively. Also, the best emission dispatch and the corresponding economic dispatch reported by ThPSO algorithm are 176682.264 kg/h and 129995.2713 $/h respectively. The best fuel cost and the emission obtained by MABC algorithm are 121412.5561 $/h and 359901.4409 kg/h respectively. The fuel cost evaluated by MABC algorithm is lesser than that obtained by Th PSO algorithm. As it is obvious from the results that, the best emission dispatch by MABC algorithm is 176679.4239 kg/h which is superior to methods reported in [11]. Also the corresponding fuel cost is 129990.00 $/h and it is less than that obtained by Th PSO algorithm. Fig. 2 shows the convergence characteristics of best economic while Fig. 3 shows the convergence characteristics of total cost of operation (TC) for best emission dispatch for this test system when the total load is 10500 MW.
TABLE VI. COMPARISON OF BEST COST AND BEST EMISSION OBTAINED BY GBAC ALGORITHM FOR 40UNIT TEST SYSTEM (TOTAL DEMAND =10500 MW) Th PSO [4] Economic Emission dispatch dispatch 110.80182 114.0000 110.80003 114.0000 97.399913 120.0000 179.7331 169.368007 87.799785 97.0000 140.0000 124.25741 259.59965 299.711395 284.59966 297.914857 284.59965 297.260101 130.0000 130.0000 94.0000 298.410142 94.0000 298.026011 214.75979 433.557636 394.27937 421.728407 394.27937 422.779653 394.27937 422.779654 489.27937 439.412855 489.27937 439.402886 511.27937 439.412857 511.27937 439.412855 523.27937 439.446399 523.28065 439.44640003 523.27937 439.7720652 523.27937 439.7720652 523.27937 440.1117645 523.27937 440.11I7645 10.0000 28.99370448 10.0000 28.99370193 10.0000 28.99370196 87.799891 96.99999999 190.0000 172.3319030 190.0000 172.3319033 190.0000 172.3319036 164.80149 200.0000 194.39276 200.0000 MABC Economic dispatch 110.7998 110.8001 97.3999 179.7332 87.7999 140.0000 259.5997 284.5997 284.5996 130.0000 94.0000 94.0000 214.7598 394.2794 394.2793 394.2794 489.2794 489.2793 511.2794 511.2795 523.2795 523.2792 523.2793 523.2794 523.2794 523.2795 10.0000 10.0000 10.0000 87.8001 190.0000 190.0000 190.0000 164.8000 199.9998
Unit 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35
Table V highlights the results obtained by MABC algorithm for 14 unit test system and its comparison with TribeMDE algorithm [11]. As it can be seen from the results that, the best economic dispatch schedule obtained by TribeMDE algorithm is 9180.9474 $/h which is more than that obtained by MABC method. Also the corresponding losses and emission level reported by TribeMDE algorithm are 5069.2919 kg/h and 37.75894 MW respectively; which are higher to a great extent than the results obtained by MABC method. The optimum economic dispatch estimated by MABC method is 9146.8 $/h.Furthermore the emission and losses obtained by MABC algorithm are 3341.2000 kg/h and 35.7769 MW respectively. The fuel cost and emission level reported in [11] are 10798.0159 $/h and 2887.8203 kg/h respectively. However the actual transmission loss calculated using B-matrix data employed in [10] is more than the reported value of 32.19721 MW. On the other hand, the emission obtained by MABC algorithm is 2909.6 kg/h and the corresponding fuel cost and losses are 10667.0000 $/h and 36.0488 MW respectively.
E. Forty unit test system

In this section the feasibility of MABC has been investigated on large scale test system considering 40-units with valve point effects, ramp rate limits and POZs. The load demand of 10500 MW is set. The input data for this case is taken from [9]. Table VI gives the results of best economic dispatch and
Emission dispatch 114.0000 114.0000 120.0000 169.3675 97.0000 124.2570 299.7105 297.9140 297.2592 130.0000 298.4093 298.0251 433.5564 421.7268 422.7781 422.7781 439.4117 439.4017 439.4117 439.4117 439.4453 439.4453 439.7709 439.7709 440.1106 440.1106 28.9931 28.9931 28.9931 97.0000 172.3314 172.3314 172.3314 200.0000 200.0000
36 37 38 39 40 Fuel cost ($/h) Emission (kg/h) 200.0000 110.0000 110.0000 110.0000 511.27937 121412.5704 359902.1885 200.0000 100.8383774 100.8383768 100.8383772 439.4128536 129995.2713 176682.2646 194.3974 110.0000 110.0000 110.0000 511.2792 121412.5561 359901.4409 200.0000 100.8380 100.8380 100.8380 439.4117 129990.0000 176679.4239
1.5
x 10
1.45
handle the equality and non-equality constraints. To demonstrate the efficiency of the MABC algorithm, it is implemented on different standard test case systems like 6, 10, 11, 14 and 40 unit. For 6 and 11 unit test case systems best compromising solution obtained by MABC algorithm and their comparison with other methods is provided. For 10, 14 and 40 test cases, best economic and best emission dispatch schedule obtained by MABC is presented. The results obtained by MABC algorithm are compared with those obtained by other methods reported in previous literatures. It is shown that the MABC algorithm performs better than other algorithms reported in recent literature. Moreover the MABC optimization technique has fast convergence speed and hence requires less computational time. REFERENCES
[1] F. Yao, Z. Y. Dong, K. Meng, Z. Xu, Ho-Ching Iu, Quantuminspired particle swarm optimization for power system operations considering wind power uncertainty and carbon tax in Australia, Unpublished. [2] G. Zhu, S. Kwong, Gbest-guided artificial bee colony algorithm for numerical function optimization, Applied math and comp 2010, vol. 217, pp. 31663173. [3] S. Mondal, A. Bhattacharya, S. H. nee Dey, Multi-objective economic emission load dispatch solution using gravitational search algorithm and considering wind power penetration, Electr. Power and Energy Syst 2013, vol. 44, pp. 282292. [4] T. Niknam, H. D. Mojarrad, Multiobjective economic/emission dispatch by multiobjective -particle swarm optimisation, IET Gener, Trans & Distri. 2012, vol. 6, pp. 363377. [5] D. Karaboga, B. Basturk, On the performance of artificial bee colony (ABC) algorithm, Applied Soft Computing 2008, vol. 8, pp. 687 697. [6] A. Mahor, V. Prasad, S. Rangnekar, Economic dispatch using particle swarm optimization: A review, Renewable and Sustainable Energy Reviews 2009, vol. 13, pp. 21342141. [7] R. Balamurugan, S. Subramanian, A Simplified recursive approach to combined economic emission dispatch, Electr Power Comp and Syst 2008, vol. 36, pp. 1727. [8] U. Gven, Y. Snmez, S. Duman, N. Yrkerend, Combined economic and emission dispatch solution using gravitational search algorithm, Scientia Iranica D. Article in press. [9] M. Basu, Economic environmental dispatch using multi-objective differential evolution, Applied Soft Computing 2011, vol. 11, pp. 28452853. [10] K. K. Mandal, N. Chakraborty, Effect of control parameters on differential evolution based combined economic emission dispatch with valve-point loading and transmission loss, Int J Emerg Electr Power Syst 2008, vol. 9, pp. 118. [11] T. Niknam, H. D. Mojarrad, B. B. Firouzi, A new optimization algorithm for multi-objective Economic/Emission Dispatch, Electr. Power and Energy Syst 2013, vol. 46, pp. 283293.
Total cost ($/hr)
1.4
1.35
1.3
1.25
1.2 0
100
200 300 Iterations
400
500
Figure 2. Convergence characteristics of best economic dispatch for 40 unit test system with total demand 10500 MW
x 10
5
6 5.5
T o ta l c o s t ($ /h )
5 4.5 4 3.5 0
20
40
60
80
100 Iterations
120
140
160
180
200
Figure 3. Convergence characteristics of best emission dispatch for 40 unit test system with total demand 10500 MW
VI.
CONCLUSIONS
In this article, the modified artificial bee colony (MABC) algorithm is used to solve the economic emission load dispatch (EELD) problem. MABC algorithm can effectively
Design and Implementation of Current based Vector Control Model of Brushless Doubly Fed Induction Generator
F. Khazaeli Moghaddam, H. Gorginpour, A. Hajbabaei, S. ouni, and H. Oraee
Electrical Engineering Department, Sharif University of Technology (SUT), Tehran, Iran
AbstractThis paper is aimed at proposing a current based vector control model of the brushless doubly fed induction generator, modelling the presented control method, as well as implementing the proposed algorithm by DSP. In order to achieve the purpose, by presenting a detailed coupled circuit model of BDFIG, the vector model and then the current based vector control algorithm of the mentioned machine are acquired. The way of independent control of torque and power, and also the structure of speed controller amongst the proposed control model are discussed. Additionally, the concepts behind the proposed structure of the speed control system and the way of determining the model parameters are explained. Then the general model is simulated in Matlab/Simulink environment. In continue, in order to get the simulation results in the real drive system, an efficient BDFIGs driver with the purpose of attaining the acquired results of simulation is built. The implementation process, including the design and selection of the driver components, and the implementation of the algorithm on DSP is presented with details. The evaluation of implementation results reveals that the designed controller, based on the proposed algorithm, meets drive requirements sufficiently with an appropriate dynamic performance and a stable operation in the duration of working in generator mode.
a kind of coupling well-known as cross coupling. Cross coupling is a kind of coupling occurs between the electric fields of the stators control and power windings. Regarding the inequality of pole numbers in power winding (PW) in comparison with control winding (CW), these two cannot have a direct coupling. In synchronous mode, BDFIG produces a torque which is controlled by the load angle, like a synchronous generator. The division of power between the two windings occurs with the ratio of , in which 1 and 2 are the electrical angular velocity of the power and control windings respectively. Hence, for a little deviation from synchronous speed, a variable speed system based on a converter with fractional ratings of the total power is required to be used. It has been proved that in synchronous mode of operation, the power factor correction of the machine is easily possible through handling the BDFIGs slip power; moreover, the power factor of the PW can be lead [4]. In order to achieve the operational purposes of this machine, it is necessary to use an efficient control method. Regarding the exclusive performance of vector control method and the numerous advantages by which are achieved, this method is widely used in the design of electrical drives [5]. The other methods, in comparison with VC, have some deficiencies. The other commonly used methods are based on direct power control or direct torque control. DTC methods use estimated torque and flux; therefore, as an indirect result of changes in voltage, changes in flux bring about, this method operates inefficiently. The performance of DPC considerably depends on the accuracy of the sensors; hence, it does not offer a robust method of control. Despite the complexity of vector control method, it has a lot of preferences over these methods. The PI controllers, used in VC based structures, increase the controllers robustness against the internal dynamics of the system which is not modelled [6]. Additionally, the controller is more robust against changes in the system parameters and the measurement noises. Lately, several papers have proposed various types of vector control method and their respective advantages [7-9]. In [10], the authors present a vector control method; however, does not concern about testing and implementing the proposed algorithm in a real drive system. Various VC structures presented in papers have a difference in selecting both control inputs and control references. One major difference between these methods is the reference angle by which the transformation between dq and abc frames occurs. In this paper, the theory of vector control of BDFIG based on current control, considering the whole drive system
Keywords-Brushless doubly fed induction generator; Current based vector control algorithm; simulation model; Drive system; Design and implementation I. INTRODUCTION Brushless doubly fed induction generators (BDFIGs) have the potential to be the next generation of wind generators. BDFIGs have some privileges in comparison with DFIG which can make them popular amongst the medium and high power wind system technologies. Omitting brushes results in more strong body and less need of repairs, which makes BDFIG a more reliable structure. Another immense advantage is the lower rate of the gearbox used in the drive train system which leads to the less cost of gear box and lower mechanical losses. BDFIG technology needs a fractional rated power converter to be a variable speed turbine [1]. This is the other achievement that considerably decreases the cost and loss of converter in comparison with full-converter based IGs. These advantages make BDFIG an efficient system for the next generation of wind turbines. The different structures of BDFIGs such as the typical standard BDFIG are compared and depicted in [2-3]. During the normal operation, BDFIG operates in synchronous mode, in which the stator and rotor windings have
requirements is presented. The coupled circuit model and consequently, the vector model of BDFIGs in an efficient reference frame are proposed. Based on the vector model of BDFIG, the efficient vector control model of the machine is introduced. In study on vector control structure, the aim is determining the relations between the selected control inputs and the both electromechanical torque and electric power. It will be shown that using suitable control inputs and control reference frame, this relation is a pure gain which is a good achievement for the method of control. According to the proposed method, the speed control loop will be designed. The paper is organized as follows: First, through the presentation of BDFIGs detailed model, its current based vector control strategy is established. Second, the proposed algorithm is simulated via MATLAB. The drive system is designed and then the control algorithm is implemented by DSP (TMS320F2812), and the results of implementation are checked by the simulation results. The whole process of design and implementation is explained in detail. Eventually, in the last section, the success in implementing BDFIGs driver will be examined and the paper will be concluded. II. COUPLED CIRCUIT MODEL OF BDFIG Consider a system containing n windings. The currents and terminal voltages of the windings are respectively shown with and . Using the combination of faradays law and ohms low, the equation (1) is concluded: (1) where R is the resistance matrix and is the linkage flux of each circuit. This model is coupled winding model, namely coupled circuit model. In rotational reference frame, the model depends on rotor speed and can be written as: (2) The electric torque produced by machine is: (3) The coupled circuit model of BDFIG, first time, was proposed in [11]. In BDFIG, regarding the fact that rotor voltage is zero, expressing v and i in terms of the parameters of the stator PW and CW, and the rotor is simply acquired from equation (4). The parameters marked by index are vectors. , (4)
0 0 0 0 0 0 0 0 0
0 0 0
0 0
(5)
0 0 0 0 0
Regarding equation(3), the torque can be written as follows:
1 2
1 2
1 1 2 2 1 1 2 2
(6) In here, it is required to remind the mechanical differential equation of the system: (7) where J is the summation of the generator and load inertias, and Tl is the mechanical loads torque applied to BDFIG. Here, it is assumed that the friction is equal to zero. By comprising (5) and (7), and considering the relation between angular velocity of rotor (r) with the changes of rotor position (r), the state space for the dynamic analysis of BDFIG will be:
0 i i i dt 0 1 J A (8)
where A is:
0 0 A 0 0 0 0 0 0 0 0 0 0
It is notable that the voltages of vs1 and vs2 are written for the four wired system. Designing the structure of BDFMs is in such a way that the mutual inductance between the power and control windings is equal to zero. The mutual inductance matrix, M12=M12T, is the function of rotor angle, whereas, the mutual inductance matrixes Ms1 Ms2 and Mr are constant regarding the fact that their windings are standstill in each other views. Therefore, by applying equation (2) we will have:
It should be mentioned that BDFM is inherently a nonlinear and time variant system. The dependency of system equations on rotor angle r makes it complex to control because r is changing during normal machine operation. Furthermore, i is a quadratic equation in terms of torque.
The resulted equations from the coupled circuit model will be rewritten in appropriate reference frame so that forms vector model of the machine. In other words, by means of selecting a reference frame rotating with synchronous speed and oriented properly, the model can be controlled by linear regulators. III.
VECTOR CONTROL OF BDFIG
A) BDFIG Vector Control Structure As mentioned before, vector model of BDFIG results in a linear system for it. Hence, to control the system, linear controllers can be used. Fig.1 depicts the proposed control strategy. As it can be seen, in this figure, the three-phase inverter which is connected to the control winding (CW) is utilized in current control mode. A hysteresis controller, in each moment, commands the state of switches to fire efficiently each switch in inverters structure.
m,ref
iqs2 ,ref
is2 ,ref
is2
s2 ,ref
I ds2 ,ref
ids2 ,ref
Fig.1.BDFIG proposed current based vector control structure
In this equation, Is2,ref = Ids2,ref + jIqs2,ref , is current vector of the control winding. In fact, the above equation is the inverse vector transformation of the three phase parameters; although, the reference angle is not arbitrary and for an appropriate operation of vector control structure it should properly be determined. Determination of the referenced angle is the most important part of the vector control method, because it affects the whole sections of vector control model. Also, it is the most sensitive part, because a small measurement error can influence the system operation. As earlier discussed, the proper value for this angle made the machine linear and time invariant. The most appropriate reference angle value is selected so that, firstly, the relation between the torque and the control inputs gets linear without any compensating term in the control structure and, secondly, this relation does not be depended on the machine parameters. In most of the utilized reference angles for the control of electrical machines, often the first constraint is satisfied; however, these angles are usually based on the vector model of machine, therefore, the machines parameters are appeared in calculations. Using such a method, error in the operation of controller is usually unavoidable. In here, in order to satisfy the two aforementioned requirements, the reference angle is considered as the following: P 1+P 2 s 2,ref = 1dt (10) m 2 In order for the vector control method based on the reference angle defined in (10) operates properly, synchronization between the power winding and control winding parameters is necessary. Without the synchronization, the system cannot operate desirably. It can easily be proved that by the use of the reference angle in (10), BDFM always operates in synchronous mode. B) BDFIGs Electromechanical Torque Control Algorithm In this section, the torque control algorithm based on the current of CW as the Control Parameter is proposed. To achieve this, the relation between the torque and q-component of current is derived. Using the synchronous reference frame and the mentioned reference angle, the electromechanical torque can be written as the following equation:
Te = GIqs2 + Td
The operation of hysteresis controller on each bridge of inverter is in such a way that if the current of each phase is more than the reference value, the controller will turn on the downward switch in the same leg to reduce the current value, and if the current the same phase is less than its reference value, the controller will turn on the upward switch to increase the current value. The robustness against changes in machine parameters and the fast dynamic of response are the main advantages of the proposed structure. Regardless of the structure, the current control loop should operate properly from the viewpoints of stability and dynamic performance. In this case, it can be claimed that the current control input can be in access in each moment. It should be noted when the control inputs are the currents, the speed control loop will present a better performance in comparison with the case that control inputs are the voltages. As depicted in Fig.2, the reference current input is determined by a dq-abc transformation on currents, using the suitable reference angle. The way of computing the current references, using the dq current components of control winding and the reference angle is shown in equation (9). t 2 4 j s 2 , ref j j 3 3 i s 2,ref = Re1 e e I s 2, ref e (9)
(11)
in which: 9 9 9 27 G = p1Ms1r D1D2 + p2Ms2r D1 D2ds1+ D2Ms1r D1 2 2 L L

D1 = L N M NN
(12) (13) (14)
D2 = M s1r LN M NN
r1 ()
M st 2r
As it can be seen, in the above equations, G is a pure gain which relates the torque to the input control parameter (q component of control winding current). Besides the controllable section of the torque equation which can be independently controlled by iqs2, the torque has another part which is uncontrollable. Td is called the uncontrollable part of torque and is considered as a disturbance input in the design and analysis of BDFIG control system. In here, the main
achievement is deriving a pure gain relation between the control input and the torque using the proposed control method. This relation results in the possibility of utilizing a linear controller to control the rotor speed.
n =
=
GKK J
(16) (17)
(1 + K P ) 2
GK JK I
C) BDFIGs Reactive Power Control Algorithm Throughout this section, the relation between the reactive power and the d-component of control windings current will be determined. Through writing the power windings current vector in terms of the control winding current, the rotor flux and the power winding flux, and using them in the reactive power equation, and separating the rotor flux in terms of r1 and r 2 , the reactive power will be written as follows:
Q 1 = 9 3 (Ms1r (LN MNN)1Vqs1r1 Ms1r (LN MNN)1Mst2rVqs1Ids2) + Qd 2L 2
K is selected regarding the stability margin of inner control loop. G, in vector control model, is continuously changing during the system operation. For example, the changes of terminal voltage result in the change of G. The inner control loop is designed with a high stability margin so as ensure that the changes of operating conditions do not cause instability. E) Simulation Results of The Proposed Algorithm According to the discussions about BDFIG drive system during the previous sections, the system is simulated via Simulink and shown in Fig.3. As it can be seen, the system (15) maintains the following subsystems:
BDFIGs model consisting of: BDFIGs mathematical model; the unit transferring power and control windings voltages from abc to dq; the circuit calculating the linkage flux; the circuit computing the current vectors from the linkage fluxes; the unit modelling the mechanical equations, using current and voltage vectors; torque estimation block; rotor position and speed estimators. Inverters model consisting of: the mathematical model of inverter; voltages phase calculation block. Inverters command circuit consisting of: switching command circuit; hysteresis current controller; current transformer unit from dq to abc. Speed controller unit. As we know, the drive system should operate in such a way that during the period on which BDFIG supplies nominal power, the voltage and the frequency of power point coupling stay constant. Whatever this requirement get achieved by more independency of machine parameters and load changes, and higher dynamic of response, it will be much worthier. Regarding the proposed structure of the control system and based on the simulation results, in order to the power windings voltage and frequency stay constant, the characteristics of the voltage amplitude and frequency of control winding versus rotor speed must be as demonstrated in the Figures 4 and 5. IV. DRIVE SYSTEM IMLEMENTATION AND BUILDING In this section, the purpose is implementing and developing BDFIGs driver. To achieve this, the following steps are taken.
It can be seen that the reactive power is also consisted of controllable and uncontrollable parts. Qd is the uncontrollable part of the equation. The equation (15), with the state variable r1 , results a linear system in which r1 is state vector, Ids2 is the control input, and Vqs1 & Qd are disturbance inputs. Hence, the relation between the power windings reactive power and
Ids2 is linear, even though the relation is not a pure gain.

According to the aforementioned explanations, both the torque and the power of BDFIG can be independently controlled based on the current components of CW.
D) BDFM Speed Controller The operation of speed controller on speed error, obtained by the subtraction of reference speed from the actual speed, will produce the reference value of controller input parameter iqs2 which by regulating machines torque, makes the actual speed adjusted to the reference speed. In the proposed method, as discussed earlier, there is a linear relation between the input control and torque which is a pure gain; hence, the desired speed control loop is designed by a linear controller. Fig.2 shows the proposed BDFIG speed control based on the proposed vector control algorithm.
Td Tl
Te
m, ref
KP +
KI
Iqs2,ref
1 Js + B
Fig.2.Speed control loop
In this figure, the speed controller, vector control structure, and shaft speed dynamic model are demonstrated. As it can be seen, speed controller produces the reference control input, and the control input modify BDFMs torque so that the desired speed get achieved afterward. The structure considered for the speed controller consists of an inner control loop and an outer control loop in which the inner control loop is the gain of k and the outer control loop is a PI controller. Hence, for the designed speed controller the three independent parameters K, KP and KI should be determined. The mentioned speed loop has the natural frequency and damping factor as below:
A) Drive Systems Infrastructure The devices used for building BDFIGs driver are listed as:
Inverter: inverter is operating as the machine side converter. According to the project requirements, the inverter is built with the following specifications: dc link voltage 350 volt, output current 10 A, six isolated supplies for driving the six switches, the capability of controling fan and bypassing the charging resistance of dc link capacitors, holding off the switches by -5 volt when there is no firing command, auxiliary power supply through the SQPS system, over current and short circuit protection by fuse.
DSP processor: in fact, by implementation of the control algorithm by DSP, the suitable signals to set the inverter switching signals are produced. in here, the DSP
charactrized by TMS320F2812 is used to do the driver implementation test.
Fig.3.Simulated model of BDFIG and its control system based on the proposed vector control algorithm
120
100
80
60
40
20
0 400
450
500
550
600
650
Fig.4. Voltage amplitude of control winding vs. speed in nominal operation

15
Rotational Speed (rpm)
Control Winding Frequency (Hz)
10
signal waveform. The amount of bandwidth is usually considered equal to 20% frequency of first harmonic component intended to be firstly eliminated by the filter. The resonant frequency of the filter branch, should be selected in such a way that the resonant frequency become much more than the dangerous frequencies like (5th, 7th, 11th, 13th, 17th, 19th and 23th). The ratio is mostly considered equal to 10. Using the three equations provided by the three mentioned criteria, the value of filter components is achieved as follows: R= 0.06 ohm L= 4.74 H and C= 1300 F DC link capacitor: consisting of two paralleled 470F and 600v electrolytic capacitors. The other devices include: voltage and current sensors, SQPS power supply, and the connectors such as ICDs, etc.
Control Winding Voltage Amplitude (volt)
-5
-10 400
450
500
550
600
650
Fig.5. Frequency of control winding vs. speed in nominal operation
Diode bridge: diode bridge operates as the grid side converter to play the role of power supply for the inverter. The used bridge is specified by: the nominal values of 4 A for its current and 600 V for its voltage. Shunt high-pass filter: In order to Select the appropriate values for R, L and C in design of filter, three criteria including: filter cut-off frequency, bandwidth, and the filter branch resonant frequency are considered [12]. The filter cut-off frequency is practically considered as a quarter of the first remarkable harmonic frequency component in the
B) Implementation of Control Algorithm on DSP In order to use DSP facilities, efficient software associated with the DSP should be utilized. The software conveys the instructions to DSP through Programming codes. The software used for Texas processors is Code Composer Studio. This code is comprised so that the following sequence of implementation levels gets provided: set of the PWM generator, measurement of the required current and voltage samples, creating the closed-loop current control structure, sampling the position by encoder and estimating the rotor speed, creating the structure of a closed-loop speed control.
V.
EVALUATION OF THE IMPLEMENTATION RESULTS
The implemented prototype of BDFIGs driver designed based on the previous descriptions is depicted in Fig.6. To illustrate the desirable operation of the implemented drive system, several produced inverter signals recorded by Fluke are depicted as below. The different waveforms are achieved through changing a potentiometer which sets the sample of speed applied to DSP. For example, at the rotor speed 600 rpm,
in or rder of fixing the terminal voltage and frequency f of power p wind ding to have a constant pow wer transfer, the t recorded output o of Fluke, F which represents th he control win ndings voltag ge, is depicted in Fig.7. . In order to achieve a the simulation resu ults as depicted in figure es 4 and 5, fo or the differen nt amount of f rotor spee ed, the Flukes signal is re ecorded. To evaluate the sy ystem oper ration, the ma ain componen nt of control windings vo oltage shou uld be measur red; hence, Fo ourier transfo ormation is us sed to deter rmine the dif fferent freque ency harmonic c components s. For the recorded r signa al in Fig.7, the e frequency sp pectrum is dep picted in Fig.8. To reach h the drive purposes, p DSP P sends the si ignals gene erated by the e implementa ation of the algorithm, to o the inverter to do the e switching process. p When n the speed is 600 rpm, , the amplitud de of the main frequency component of co ontrol wind dings voltage e and frequenc cy are 71.579 volt and 9.92 28 Hz respe ectively.
Fig.6. The designed and implemented d BDFIGs driver r accompanied by y the required labora atory equipment
Fig7. Fluke F recorded sig gnal of the contro ol winding voltag ge and frequency for the speed of 6 600 rpm
Fig.8. The frequency f spectru um of the control windings voltag ge waveform reco orded by Fluke (r= 600 rpm)
T The results demonstrated d in TABLEI, are achieved via doin ng the same process for the different rotor speeds s. By comp mparing the res sults of TABL LEI with the simulation re esults, the capability c of BDFIGs driver in follow wing the simu ulation
resul lts will be clarified. Fig.9 shows a com mparison bet tween simu ulation and imp plementation results. As A it can be se een, in these figures, the results achieve ed by the implementatio on of the driv ve system foll low the simul lation
results precisely. Hence, the implementation of the proposed algorithm is properly done. In fact, in here, by changing the rotor speed, to achieve maximum power point tracking in a BDFIG based wind turbine, the amplitude and frequency of control windings voltage is changing so that the terminal amplitude and frequency of power windings voltage stay constant. This is the main controlling purpose of a variable speed system like the wind system. The output voltage of the inverter is comprised of pulses which its fundamental frequency component is controlled through the written code implemented on DSP. As it can be seen, the control of BDFIG for the different conditions of rotor speed and loading, continuously, precisely and with a high dynamic is achieved.
TABLE I. FLUKE RECORDED SIGNALS OF CONTROL WINDING
performance drive system for BDFIG. By this representation, the control of the mentioned generator, based on the proposed algorithm versus the various amounts of rotor speed in reaction to the different environmental conditions for the purpose of tracking the maximum power required in a variable speed system, is achieved.
ACKNOWLEDGMENT
Authors would like to thank Dr. Farhad Barati from department of electrical engineering at University of South Carolina, Columbia, USA, for his scientific supports during performing a part of the project. REFERENCES
[1] [2] [3] [4] [5] [6] R. Spe, A. K. Wallace, and H. K. Lauw, Performance simulation of brushless doubly-fed adjustable speed drives, IEEE Industry Applications Society Annual Meeting, San Diego, CA, 1989. R. Li, A. K. Wallace, and R. Spe, Two-axis model development of cage-rotor brushless doubly-fed machines, IEEE Transactions on Energy Conversion, Vol. 6, no. 3, pp. 453-460, 1991. Ehsan Abdi, Modeling and Instrumentation of Brushless Doubly-Fed Machines, Ph.D. dissertation, University of Cambridge, 2006. F. Barati, S. Shao, E. Abdi, H. Oraee, and R. McMahon, Synchronous operation control of the brushless doubly-fed machine, IEEE International Symposium on Industrial Electronics, ISIE2010, July 2010. D. W. Novotny, and T. A. Lipo, Vector control and dynamics of AC drives, Oxford Press, 1996. Andres E. Leon, Juan Manuel Mauricio, and Jorge A. Solsona, Fault Ride-Through Enhancement of DFIG-Based Wind Generation Considering Unbalanced and Distorted Conditions, IEEE Transactions on Energy Conversion, Vol. 27, no. 3, SEPTEMBER 2012. S. Shao, E. Abdi, F. Barati, and R. A. McMahon, Stator-flux-oriented vector control for brushless doubly-fed induction generator, IEEE Transaction on Industrial Electronics, pp. 4220 4228, Oct. 2009. S. Shao, E. Abdi, and R. A. McMahon, Vector control of the brushless doubly-fed machine for wind power generation, In Proc. IEEE ICSET, pp. 322327, Nov. 2008. Protsenko, K., and Dewei Xu Modeling and Control of Brushless Doubly-Fed Induction Generators in Wind Energy Applications, IEEE Transactions on Power Electronics, Page(s): 1191 - 1197, 2008. F. Barati, H. Oraee, E. Abdi, S. Shao, and R. McMahon, The brushless doubly-fed machine vector model in the rotor flux oriented reference frame, In Proc. 34th IEEE IECON, pp. 14151420, 2008. A. K. Wallace, R. Spe, and H. K. Lauw, Dynamic modelling of brushless doubly-fed machines, Conference record of the IEEE Industry Applications Society Annual Meeting, San Diego, CA, 1989. F. Khazaeli Moghaddam, A. Ameli, H. Askarian, and N. Farokhnia, Nonlinear Control of STATCOM in Presence of Filter Based on Indirect Control of DC Link Voltage, IEEE 18th EPDC conference, 2013.
Speed (rpm) 600 580 560 540 520

120
Voltage amplitude (volt) 71.579 54.594 38.823 24.264 18.198
Frequency (Hz) 9.928 7.949 5.911 3.939 1.851

Simulation Results Experimental Results
Control Winding Voltage Amplitude (volt)
100
80
60
[7]
40
20
[8]
450 500 550 600 650
0 400
[9] [10] [11]
Control Winding Frequency- Main Component (Hz)
15
(a)
Simulation Results Experimental Results
10
[12]
0
-5
-10 400
450
500
550
600
650
(b) Fig.9. The comparison between drive implementation results and the simulation results: (a)voltage amplitude, (b)frequency.
VI. CONCLUSION A detailed current based vector control model of BDFIG is investigated in this paper. Afterwards, the control of torque and power, and the design of speed control loop is proposed. The proposed model is simulated, and then the implementation process of BDFIGs driver, accomplished on the basis of the results of the simulation model, is discussed. The evaluation of the driver performance supports the capability of the proposed control algorithm and the success in developing a high
A Novel Energy Management System using Ant Colony Optimization for Micro-grids
Ayman Esmat
Electrical and Control Engineering Dept. Arab Academy for Science and Technology Cairo-Egypt aaesmat@gmail.com
Amr Magdy
Electrical Power and Machines Dept. Ain Shams University Cairo-Egypt amr.magdy@eng.asu.edu .eg
Walid ElKhattam
Electrical Power and Machines Dept. Ain Shams University Cairo-Egypt walid_el_khattam@hotm ail.com
Ahmed M. ElBakly
Basic Science Dept. Arab Academy for Science and Technology Cairo-Egypt aelbakly1964@yahoo.co m
AbstractMicro-grids have spread in many distribution systems worldwide. They offer safe and reliable operation for their consumers by managing the micro-grids power generation and trading with the main grid. Furthermore, micro-grids can help in integrating and promoting for Renewable Energy Sources (RES) and reducing the environmental impacts of traditional centralized generation. This paper proposes a novel Energy Management System (EMS) in micro-grids using Ant Colony Optimization (ACO) technique to solve the generation dispatch problem. A combined cost optimization scheme is investigated to minimize both operational cost and emission levels while satisfying the micro-grids load demand. Furthermore, the proposed EMS is used to evaluate promoting RES implementation in micro-grids despite of their high capital cost using the combined economic emission dispatch problem. The proposed EMS was implemented using MatLab and tested on two case studies with and without RES-WindTurbine (WT). The obtained results from the proposed technique are compared with those calculated using two other Techniques; Lagrange and Gradient to evaluate the proposed method. The outcomes are evaluated and discussed. Finally, conclusions are reported. Index Terms- Ant Colony Optimization, combined economic emission dispatch problem, energy management system, microgrid.
I.
INTRODUCTION
Micro-grids are modern small scale versions of the centralized electricity system, designed to supply electricity for small communities such as villages or commercial areas like industrial factories. They were meant to achieve specific goals such as reliability, and carbon emission and operational cost reduction. Micro-grids have emerged in the electric industry increasingly attracting consumers [1]. During the utility's disturbances, micro-grids can isolate themselves and operate to improve customer service, with minimum or no load disturbance, while being energized by various Distributed Generation (DG) technologies such as micro-turbines, fuel cells, solar and wind power generators [2]. One of the problems associated with the operation of power systems/micro-grids is economic dispatch of existing generation/DG units within their territories. Economic dispatch is the assignment of the optimal power output for the generating units in a power system while attaining the minimum operating cost possible and satisfying the load demand in a reliable manner. Short term scheduling for
generating units with the ability to adapt to any system changes or constraints can eventually save a large amount of cost. In order to make micro-grids a more feasible and viable alternative to conventional power systems, new management techniques are required to manage the operations of microgrids. These techniques must accommodate the dynamic behavior and the multi-constraints environment of the microgrids. Several classical techniques were previously used to address this problem such as homogeneous Linear Programming Algorithm [3]. Modern computational intelligence techniques addressed the economic dispatch problem such as Particle-Swarm-Based-Simulated Annealing [4]. In [5-8], the economic dispatch problems using the Ant Colony technique were evaluated and the results showed better cost savings, effectiveness, and flexibility to solve this problem under study. In [9], the combined economic and emission problems with transmission losses and conventional generation was discussed. Three methods were examined and compared; Genetic Algorithm, Classical Ant Colony, and Lambda techniques. Results showed the superiority of the Classical Ant Colony algorithm in minimizing the cost and the thermal emissions. In [10], Ant Colony and Particle Swarm Optimization were compared to solve the classical economic dispatch problem considering RES. Results were discussed and the advantages of the ACO are highlighted. A similar optimization problem in [9] was examined with Evolutionary Programming and Lagrange Techniques in [11] and [12] respectively. In [13], the combined economic emission dispatch problem was evaluated for a mix of conventional and RES using Gradient method. In this paper, a novel EMS for micro-grids is proposed to solve the combined economic emission dispatch problem using ACO taking into consideration both conventional and RES-WT generation. In Section II, the ACO technique and its enhancement for the proposed EMS are illustrated. The proposed EMS for micro-grids formulation is explained. The combined economic emission dispatch cost function and its constraints are provided in Section III. In Section IV, two case studies were carried out. The first case was used to validate the ACO technique. The second case was implemented to evaluate promotion of RES-WT in micro-grids using the proposed EMS to minimize generation cost of the whole system including the WT investment cost. The obtained results
are shown and discussed. Finally, conclusions are discussed in Section IV. II. ANT COLONY OPTIMIZATION TECHNIQUE
Where; is the total fuel cost, N is the number of generators, is the real output generation of the generator, and a, b, c generator. are the cost coefficients of the - Emission Function The total emission of atmospheric pollutants caused by operation of fossil fueled generators can be expressed as follows: 2 is the total fuel cost and , , are the cost Where; generator. coefficients of the Therefore, the combined economic emission dispatch problem is formulated as a single optimization problem as follows:
A. Basic Prinicipal In real life, ants are capable of finding the shortest path to a food source from their nest using their pheromones. Pheromones are chemical substances deposited from ants bodies on the ground leaving a trail to a favorable path to the food source for each ant. These favorable paths are used by other members of the ant colony depending on the intensity of these pheromones in succeeding visits to the food source. ACO uses this phenomenon and applies it to solve real life optimization problems. B. ACO Algorithm for EMS The proposed energy management system was designed using the Ant Colony technique due to its effectiveness and accuracy. Some features were added in the proposed technique to enhance its results [6] and decrease its computational time, and they can be explained as follows: Priority listing of generating units: A ranking priority based list is proposed in the algorithm. This list increases the performance of obtaining new solutions by ranking the effect of the generating units. It creates new solutions and decreases the search time. Zooming feature: This feature helps in reducing the search space by zooming on the optimum solution every iteration. The reduction process is based on the best solution found by the technique in the previous iteration. Moreover, it helps in finding more accurate results for optimal solution. PROPOSED ENERGY MANAGEMENT SYSTEM FOR MICRO-GRIDS Energy Management refers to an efficient and effective use of energy to minimize cost, improve energy efficiency, and reduce greenhouse gas emissions by making use of RES and reducing the impacts of blackouts or interruptions in energy supplies. The proposed EMS addresses the combined economic emission dispatch problem in a typical micro-grid by providing hourly active power generation set-points while satisfying the load demand. It aims at minimizing operating cost and emissions. The proposed EMS was designed using ACO technique in the MatLab environment. A. The optimization cost function: The combined economic dispatch problem is addressed as a single optimization problem with respect to fuel cost and emissions function as follows: Fuel Cost Function + Emission Function - Fuel Cost Function The generator fuel cost function is represented as a quadratic equation and can be expressed as follows: 1 III.
(3) Where; is the price penalty factor which is the ratio between the maximum fuel cost and maximum emission for each generator [11]. 4 B. The Constraints: The combined economic dispatch problem is constrained by power balance and inequality constraints of a micro-grid. - Power Balance Constraint Total power generation must satisfy the load demand plus the power transmission losses at any given time. The power balance equation is illustrated as follows 5 Where; is the total load demand
- Inequality Constraint Total power generation for each unit lies between minimum and maximum limits. The inequality constraint can be described as follows: 6 IV. SIMULATION RESULTS In this section, two case studies were carried out. The first case is used to verify the proposed EMS based ACO technique. On the other hand, the second case study evaluates the proposed EMS to promote for RE-WT investment in Micro-Grids. The micro-grid under study is assumed to be working in the island mode. Thus, there is power exchange with the main grid (selling/buying).
A. ACO Technique Verification In order to verify the effectiveness of the proposed EMS based ACO technique, a case study was carried out on a fiveconventional generator system given in [12]. The generation power limits and fuel and emission coefficients, shown in Tables I to III, and four load demand scenarios were used. The combined economic emission optimization problem was minimized using (1) to (6). Tables IV and V show a comparison of the results obtained from the proposed ACO and those from Lagranges method reported in [12].
TABLE I. G1 G2 G3 G4 G5 TABLE II. G1 G2 G3 G4 G5 TABLE III. G1 G2 G3 G4 G5 GENERATORS POWER MAX-MIN LIMITS Max Power (MW) 75 125 275 250 300 Min Power (MW) 10 20 30 40 50
TABLE V. COMPARISON BETWEEN TOTAL COST OF OPERATION BETWEEN PROPOSED ACO TECHNIQUE AND LAGRANGE'S TECHNIQUE Ft ($*103) Pd (MW) Scenario-1: 250 MW Scenario-2: 300 MW Scenario-3: 350 MW Scenario-4: 400 MW Lagrange's Technique [12] 1,723.34 2,482.06 3,378.81 4,413.57 ACO Technique 1,427.60 2,024.00 2,840.80 3,865.40 Cost Saving (%) 17.16% 18.45% 15.92% 12.42%
It is observed from Table V that the proposed ACO technique reported better results and minimum cost at all four demand load scenarios under study that vary from 12.5 to 18.5%. B. Proposed EMS-ACO Technique for Promoting RE-WT in Micro-Grids In [13], a renewable based micro-grid was modeled and an optimization process was carried out using the Gradient method technique with respect to the total generation cost of the whole system without/with RE-WT. A comparison was made using the same coefficients and constraints with the ACO technique proposed to include REWT into consideration. A case study was carried out to satisfy a daily load profile shown in Table VI. The micro-grid model is assumed to have three conventional generators (two synchronous generators and one Combined Heat and Power (CHP) generator) and a non-dispatchable WT generator. The generators power limits and cost coefficients are described in Tables VII and VIII. A daily WT power generation output based on a given wind speed is shown in Figure 1.
TABLE VI. Time (Hrs) Load (MW) Time (Hrs) Load (MW) Time (Hrs) Load (MW) 1 140 9 210 17 170 LOAD DEMAND OVER 24 HOURS PERIOD 2 150 10 230 18 185 3 155 11 240 19 200 4 160 12 250 20 240 5 165 13 240 21 225 6 170 14 220 22 190 7 175 15 200 23 160 8 180 16 180 24 145
GENERATORS FUEL COST COEFFICIENTS a ($/hr) 0.008 0.003 0.0012 0.001 0.0015 b ($/hr) 2 1.8 2.1 2 1.8 c ($/hr) 25 60 100 120 40
GENERATORS FUEL EMISSION COEFFICIENTS (kg/hr) (kg/hr) (kg/hr) 0.018 0.015 0.0105 0.008 0.012 -0.805 -0.555 -1.355 -0.6 -0.555 80 50 60 45 30
TABLE IV. COMPARISON BETWEEN GENERATORS DISPATCHED POWER USING PROPOSED ACO TECHNIQUE AND LAGRANGE'S TECHNIQUE Pd (MW) G1
Lagrange ACO Lagrange ACO Lagrange ACO Lagrange ACO Lagrange ACO
250 49.9 74.1 50.1 43.3 49.8 30.1 49.9 44.7 50.1 57.8
300 59.9 75.0 60.1 63.2 59.8 33.2 59.9 44.7 60.1 83.9
350 59.9 74.9 70.1 78.4 69.8 36.4 59.9 47.0 70.1 113.2
400 79.9 74.7 80.1 97.2 79.8 41.3 79.9 54.0 80.1 132.9
TABLE VII.
GENERATORS MAX-MIN POWER LIMITS Max Power (MW) Min Power (MW) 37 40 50
G2
G3
G1 G2 G3 TABLE VIII.
150 160 190
G4
GENERATORS FUEL COST COEFFICIENTS a ($/hr) b ($/hr) 21 20.16 20.4 c ($/hr) 0.024 0.029 0.021
G5
G1 G2 G3
1530 992 600
is the operation and maintenance cost per power ($/KW), unit generated energy ($/KW). The obtained results, Table X, show the proposed ACO technique saved 0.75% (no emission cost) and 4.56% (considering emission cost) over the Gradient method reported in [13] of total cost. In the meanwhile, comparing the ACO results in Tables IX and X, installing RE-WT in the micro-grid saved 2.73% of the total system cost despite of the high WT investment cost taking into consideration the emission cost of conventional generation. V.
Fig. 1 A Daily WT Power Generation
CONCLUSIONS
Two scenarios were carried out: - Scenario-1: Three fueled generators: The proposed EMS with ACO technique was implemented to minimize the combined economic emission dispatch problem, equation (3). The obtained results were obtained for a daily load demand and it was compared with the results obtained by the Gradient method and reported in [13]. Table IX shows the generation dispatch of the three generators and the total operating cost without and with emissions cost. Results show that the proposed ACO technique saved 0.53% (no emission cost) and 3.98% (considering emission cost) over the Gradient method in total cost. Moreover, for the gradient method, the reported dispatch power are taken from [13] without taking the emission impact into consideration. Then, the emission cost is calculated based on the optimized dispatched generation without emission consideration. However, the proposed ACO technique, dispatch problem used equation (1) or (3) for no emission or emission cost consideration respectively. That is why the dispatched generation output is different. Scenario-2: Three fueled generators and one WT generator: According to the worldwide practice for RE generation/transmission, RE has the priority to be transmitted as soon as it is generated over the fossil fuel generated electricity. Therefore, the WT generated power can be considered a negative load and used to reduce the load demand values. The rest of the load is supplied from the conventional generators in the system where the economic dispatch problem is used. The proposed EMS -ACO is carried out to solve the combined economic emission dispatch problem taking into consideration the RE-WT investment cost. The aim is to minimize the total generation cost of the system. Cost function for the WT generator considers the investment cost of the equipment and the operation and maintenance cost. The cost function can be described as follows: (6) 1 1 7 -
This paper presents a novel energy management system for a micro-grid with and without RES-WT to solve the combined economic emission dispatch problem. The proposed EMS was designed using the Ant Colony technique and tested on two different case studies. In the first case, a micro-grid with conventional generation, the proposed ACO technique was compared with Lagrange's method to verify the proposed EMS-ACO technique. The cost saving reaches up to 18.5% over the results obtained from the Lagrange's method with respect to the combined economic and emission dispatch problem. In the second case, micro-grid without/with RE-WT, the proposed ACO technique was compared with Gradient method. Results show that the proposed EMS-ACO technique saved 3.98% over the Gradient method taking emission cost into consideration without RE-WT in the micro-grid. However, considering RE-WT, the ACO saved 4.56% over the Gradient method considering the emission cost as well. In the meanwhile, results show that investing in RE-WT has saved 2.73% of the total generation cost in the system taking into consideration the WT investment cost inspite of its high capital investment. In a nut shell, the proposed EMS-ACO technique offered cost savings for the combined economic emission dispatch problem in all cases allowing more efficient management on the given micro-grids. Furthermore, it showed that investing in WT can save the system running cost of the conventional generation in the micro-grid which leads to promote WT investments in micro-grid. REFERENCES
[1] S. Chowdhury, S.P. Chowdhury and P. Crossley, "Microgrids and Active Distribution Networks,". London, United Kingdom: The Institution of Engineering and Technology, 2009. Konstantinos Angelopoulos, "Integration of Distributed Generation in Low Voltage Networks: Power Quality and Economics," MSc. Thesis, Dept. Mech. Eng., University of Strathclyde, Glasgow, Scotland, 2004. Rabih A. Jabr, Alun H. Coonick and Brian J. Cory, A Homogeneous Linear Programming algorithm for the security Constrained Economic Dispatch Problem," IEEE Transactions on Power Systems, vol. 15, no.3, August 2000. Chun-xia Dou, Xing-bei Jia, Zhi-qian Bo, FangZhao and Dong-le Liu, "Optimal Management of MicroGrid based on modified particle swarm optimization algorithm," Power and Energy Engineering Conference, Asia-Pacific, March 2011. T. Niknam, A.M. Ranjbar, A.R. Shirani and A. Ostadi, "A new approach based on ant colony algorithm to distribution management system with regard to dispersed generation," 18th International Conference on Electricity Distribution, Turin, Italy, June 2005. Saravuth Pothiya, Issarachai Ngamroo and Waree Kongprawechnon, "Ant Colony Optimization for economic dispatch problem with non-
[2]
[3]
[4]
[5]
Where; is the wind generation, is the NPV coefficient, r is the interest rate assumed to be 9%, N is the investment lifetime assumed to be 20 years, is the investment cost per installed
[6]
smooth cost functions," International Journal of Electrical Power & Energy Systems, volume 32, pp. 478487, 2010. [7] Yun-He Hou, Yao-Wu Wu, Li-Juan Lu and Xin-Yin Xiong, "Generalized Ant Colony Optimization for Economic Dispatch of Power systems," International Conference on Power System Technology Proceedings, October 2002. [8] Enrico Carpaneto, Gianfranco Chicco "Optimal Operational Planning of large distribution systems with ant colony search," 15th Power Systems Computation Conference, Liege, Belgium, August 2005. [9] R.Bharathi, M.Jagdeesh Kumar, D.Sunitha and S.Premalatha, "Optimization of combined Economic and emission dispatch problem," Power Engineering Conference, Singapore, 2007. [10] C.M. Colson, M.H. Nehrir and S. A. Pourmousavi, "Towards Real-Time Microgrid power management using computational intelligence methods,"Power and Energy Society General Meeting, Minneapolis, USA, July 2010. [11] P. Venkatesh, R. Gnanadass, and Narayana Prasad Padhy, "Comparison and Application of evolutionary programming techniques to combined economic emission dispatch with line flow constraints," IEEE Transactions on Power Systems, vol. 18, no. 2, May 2003. [12] S. Krishnamurthy and R. Tzoneva, "Multi Objective Dispatch problem with valve point effect loading using cost and emssion criterion," International Journal of Computer and Electrical Engineering, vol. 4, no. 5, October 2012. [13] Noel Augustine, Sindhu Suresh, Prajakta Moghe and Kashif Sheikh, "Economic Dispatch for a microgrid considering renewable energy cost functions," Innovative Smart Grid Technologies, Washington, DC, USA, Jan 2012.
APPENDIX TABLE IX. SCENARIO-1: COMPARISON BETWEEN TOTAL GENERATION COST USING GRADIENT TECHNIQUE AND PROPOSED ACO TECHNIQUE (NO WIND) No Emission Gradient Method Pd (MW) Generation (MW) [13] G1 140 150 155 160 165 170 175 180 210 230 240 250 240 220 200 180 170 185 200 240 225 190 160 145 7 15 19 24 28 32 36 40 53 82 91 99 91 74 59 40 32 45 59 91 78 49 24 11 G2 11 19 24 28 32 36 40 45 57 87 95 103 95 78 64 45 36 49 64 95 82 53 28 15 G3 122 115 112 109 105 102 98 95 100 61 54 47 54 68 77 95 102 92 77 54 64 88 109 119 Total Gen. cost ($/hr) [13] 6298 6483 6579 6677 6778 6881 6986 7094 7795 8300 8569 8848 8569 8040 7548 7094 6881 7024 7548 8569 8168 7316 6677 6387 177,105 ACO Technique Generation (MW) G1 37 41 42 44 46 48 49 51 61 68 71 74 71 64 58 51 48 53 58 71 66 54 44 39 G2 45 48 50 51 52 54 55 57 65 71 73 76 73 68 62 57 54 58 62 73 69 59 51 47 G3 57 61 63 65 67 69 71 72 84 92 96 99 96 88 80 72 69 74 80 96 90 76 65 59 Total Gen. cost ($/hr) 6152 6380 6495 6611 6727 6843 6959 7076 7787 8269 8512 8757 8512 8027 7549 7076 6843 7194 7549 8512 8148 7312 6611 6266 176,166 Gradient Method Generation (MW) [13] G1 7 15 19 24 28 32 36 40 53 82 91 99 91 74 59 40 32 45 59 91 78 49 24 11 G2 11 19 24 28 32 36 40 45 57 87 95 103 95 78 64 45 36 49 64 95 82 53 28 15 G3 122 115 112 109 105 102 98 95 100 61 54 47 54 68 77 95 102 92 77 54 64 88 109 119 Total Gen. cost ($/hr) 8498 8541 8593 8667 8761 8875 9011 9167 10525 11868 12657 13529 12657 11162 10008 9167 8875 9344 10008 12657 11505 9541 8661 8509 240,783 Emission ACO Technique Generation (MW) G1 37 40 41 42 44 45 46 47 56 61 64 66 64 59 53 47 45 49 53 64 60 50 42 38 G2 44 47 49 51 52 54 55 58 67 72 75 80 75 69 63 58 54 58 63 75 71 60 51 46 G3 59 63 65 67 69 71 73 75 88 96 101 104 101 92 84 75 71 78 84 101 94 80 67 61 Total Gen. cost ($/hr) 7349 7755 7963 8175 8390 8608 8830 9056 10482 11502 12033 12578 12033 10985 9993 9056 8608 9285 9993 12033 11242 9517 8175 7550 231,191
TABLE X. SCENARIO-2: COMPARISON BETWEEN TOTAL GENERATION COST USING GRADIENT TECHNIQUE AND PROPOSED ACO TECHNIQUE (WITH WIND) No Emission Pd (MW) Generation (MW) [13] G1 140 150 155 160 165 170 175 180 210 230 240 250 240 220 200 180 170 185 200 240 225 190 160 145 5 8 12 10 22 28 24 18 48 52 80 84 79 54 63 29 29 43 59 91 78 49 23 11 Gradient Method Total Gen. cost ($/hr) G2 G3 [13] 10 12 16 14 26 32 28 22 52 55 84 88 83 57 70 33 33 47 64 95 82 53 27 15 123 121 118 120 110 105 108 113 89 105 62 60 64 99 59 104 104 93 76 54 65 88 109 119 6297 6474 6565 6650 6759 6865 6940 7013 7662 8164 8450 8662 8437 7967 7509 7044 6870 7195 7544 8567 8167 7314 6674 6389 176,179 ACO Technique Generation (MW) G1 37 38 39 39 43 46 44 42 54 62 67 68 66 61 55 46 46 52 57 71 66 54 44 39 G2 45 46 47 46 50 52 51 49 59 66 70 71 69 65 60 53 53 57 62 73 69 59 51 47 G3 56 58 59 58 64 67 65 62 76 85 91 92 90 84 77 67 67 74 80 95 90 76 64 59 Total Gen. cost ($/hr) 6143 6335 6445 6520 6686 6815 6875 6925 7657 8151 8425 8628 8415 7960 7497 6997 6823 7183 7544 8511 8147 7310 6605 6263 174,856 Gradient Method Generation (MW) [13] G1 5 8 12 10 22 28 24 18 48 52 80 84 79 54 63 29 29 43 59 91 78 49 23 11 G2 10 12 16 14 26 32 28 22 52 55 84 88 83 57 70 33 33 47 64 95 82 53 27 15 G3 123 121 118 120 110 105 108 113 89 105 62 60 64 99 59 104 104 93 76 54 65 88 109 119 Total Gen. cost ($/hr) 8499 8499 8513 8504 8631 8763 8673 8575 9505 10663 11643 11970 11551 10504 9732 8788 8794 9275 9977 12643 11493 9528 8649 8506 235,639 Emission ACO Technique Generation (MW) G1 37 37 38 38 42 44 42 41 50 56 61 62 60 56 51 44 44 49 53 64 60 50 42 38 G2 44 45 46 45 50 52 51 49 60 67 71 73 71 67 60 52 53 58 63 75 71 60 50 46 G3 58 59 61 60 66 69 67 64 79 89 95 97 95 87 80 70 70 77 83 10 0 94 80 67 61 Total Gen. cost ($/hr) 7311 7558 7742 7775 8207 8480 8446 8363 9842 10901 11580 11899 11527 10646 9744 8686 8518 9232 9970 12027 11237 9508 8148 7537 224,884
Power Quality Measurement and Evaluation of a Wind Farm Connected to Distribution Grid
Mustafa Sekkeli , O. Fatih Kececioglu
Dept. of Electrical and Electronics Engineering Kahramanmaras Sutcu Imam University Kahramanmaras, Turkey msekkeli@ksu.edu.tr , fkececioglu@ksu.edu.tr
Ceyhun Yildiz
Electrical Power Resources Survey and Development Ankara, Turkey cyildiz@yegm.gov.tr
Hakan Acikgoz
Dept. of Electrical Science Kilis Vocational High School Kilis, TURKEY hakanacikgoz@kilis.edu.tr
Abstract Wind power can bring new challenges when it is connected to the power grid. Generated power from wind energy system is always fluctuating due to the fluctuations in the wind. This paper shows a study on Power Quality (PQ) analysis of wind turbines installed in Hatay re-gion and has been working for three years. Power quality parameters such as voltage, current, active, reactive and appearent power and harmonics are measured, analyzed and evaluated taking into consideration IEEE 519-1992 standards. These parameters are continously meas-ured for three monts. The recorded parameters are voltage and current rms values of all single phase for every 10 power frequency cycles, average values of current and voltage harmonics every 3 sec, active, reactive, and apparent power values and the power factor every second. Keywords- Wind farm, power quality, Harmonic
I.
INTRODUCTION
supply [1,11]. Although there are many parameters affecting on power quality, harmonics are one of the major power quality problems in power systems. Harmonic disturbances are a phenomenon associated with the distortion of the fundamental sine wave and are produced by nonlinearity of electrical equipment. Harmonics causes increased currents, power losses and possible destructive overheating in equipment [1,12-14]. There are power quality standards that define the maximum allowable limit of distortion in voltage and current waveforms of the power supply. Many standards, guidelines and recommendations including IEEE standard 519-1992 and IEC 61000 series of standards have come into effect in this regard [1, 15]. Also, some indices like Total Harmonic Distortion (THD) and Total Demand Distortion (TDD), used respectively for voltage and current harmonics [15]. In this study, an investigation is carried out to measure and evaluate the power quality in a wind farm installed at Hatay region. II. POWER QUALITY INDICES UNDER HARMONIC
DISTORTION
The negative effects of fossil fuels have forced researchers finding new and renewable enegy sources that are both environmentally more suitable and renewable [1-4]. Wind energy is a renawable energy sources and has shown very fast development in the whole World. Wind energy, in special, have received huge interest by both private investors and governments. Integration of wind power into power systems presents many new challenges. Like conven-tional power plants, wind power plants must provide the power quality required to ensure the stability and reliability of the power system it is connected to and to satisfy the customers connected to the same grid [5-9]. Although many operational aspects affect wind power plant operation, this paper, focuses on power quality [1,10]. This narrower definition of power quality (PQ) allows to focus on supply waveform problems, its frequency and magnitude. Power quality has received a great deal of attention recently, since the increased use of power electronic devices which are non-linear loads drawing nonlinear currents from the power
Harmonic indices have been developed to assess the service quality of a power system with respect to the harmonic distortion levels. These indices are measures of the effective value of a waveform and can be applied to both the current and the voltage [1,11]. The IEEE-519 doc-ument has set limits on the level of allowable harmonics [15]. Several indices are available for harmonic analysis; however, the two most commonly used are the total harmonic distortion (THD) and the total demand distortion (TDD). Mathematical formulations of (THD) and (TDD) for voltage and current are given in equations (1), (2) and (3) respectively [1,15]:
THDV =
V
h=2
2 h
V1
(1)
THDI =
I
h=2
2 h
IV. (2)
RESULT AND DISCUSSION
I1
Where V1 and I1 are the RMS value of the fundamental, and Vh and Ih are the RMS value of the h-order harmonic component. Total Demand Distortion (TDD): Is the total harmonic cur-rent distortion defined by the ratio of the RMS value of the sum of the individual harmonic amplitudes to the maximum or rated demand load current IL as shown in the following expression:
In this study, electrical parameters of wind farm such as voltage, current, frequency, active, reactive, appearent power, power factor and harmonic distortions are given and analized as follows. These values are measured and obtained for the time periods of three monts. The recorded parameters are voltage and current rms values of all singal phase for every 10 power frequency cycles, average values of current and voltage harmonics every 3 sec, active, reactive, and apparent power values and the power factor every second. Waveforms of the voltage and current measured from wind farm feeder while all wind tur-bines (12 wind turbines) were working are given in the Fig.2 and Fig.3 respectively. Voltage measurement is performed at High voltage level of 154 kV substations for single phase-neutral. High voltage ranges in Turkey for three phases are 154 and 380 kV and single phase 90 and 220 kv 10% respectively. As seen from true-rms single phase voltage variation in 154 kV side in Fig.7, voltage level does not exceed the limit through measurement period. According to the Fig.7 although a slight ripple of voltage waveform, it is seen that nominal volt-age amplitudes are nearly constant at about 90 kV.
TDDI =
I
h=2
2 h
IL
III. DESCRIPTION OF WIND FARMS
(3)
Belen wind farm is installed at Hatay province of Turkey in mediterranien region. Wind farm has been installed with a total capacity of 36 MW and constituted by 12 wind turbine. Rotor diameters and hub heights of wind turbines are 90 m. and 80 m. respectively. The nominal power of each tubines are 3 MW manufactured by Vestas V90. So far wind farm has been effectively worked and produced electrical energy. The picture of wind farm derived from Google earth is shown Fig.1.
Figure 2. Voltage values and waveform of wind turbine output
Figure 1. Picture of wind farm derived from Google Earth
In wind turbines of installed wind farm, three-phase wound rotor induction generator is used. Each wind turbine generator has been connected to its own 0.69/34.5 kV star-delta connected transformer. The neutral point of the transformer is grounded to diminish the third harmonic voltages. These generators are connected to 154 kV power lines via 154/ 34.5 kV, 50 MVA transformers. The substation also ensures that the electric power generated from wind is de-livered to the transmission line at constant voltage level of 154 kV and 50 Hz.
Figure 3. Current values and waveform of the wind turbine output
Similarly, Waveform of the current are also very ripple depending on generated power at wand farm. Voltages and currents are illustrated as single phase. Waveforms of active, reactive and apparent power measured on wind farm feeder while all wind turbines were generating active power, are
shown in Fig.4. These 12 turbines are identical and assumed to be running under same shaft torque. The power parameters have been measured and recorded at every 10 min on wind farm feeder.
Figure 4. Waveforms of the active, reactive and appearent power measured at wind farm feeder.
Measurement results show that the real power has been generated between 0 and 20 000 kW while the reactive power varies between 0 and 2500 kVAr. Changes in wind speed often result in wind turbine active and reactive power fluctuations. The output real power depends on the wind speed on wind energy conversion system.
Reactive power compensation systems have been installed on the wind farm in order to minimize reactive power generated from wind turbines. Power factor values measured from the feeder after compensation are given in Fig.5. As shown in Fig.5, power factor are almost close to 1 value desired set point by means of the compensation system.
Figure 5. Power factor values measured on wind farm feeder
Output frequency at wind farm feeder is very important and it has to be constant at 50 Hz which is nominal value at turkey distribution system. As seen on Fig.6, nominal frequency is constant at 50 Hz. Voltage and current harmonics are evaluated on wind farm. Measured total harmonic distortion (THD) of voltage and current in wind farm feeder is shown in Fig.7.and 8 respectively. This measured values show that the THD is usually less than the limit specified by IEEE Std. 519-1992 for industrial harmonics loads [1,15]. Voltage and current distortion limits are given in Table.1 and 2 according to IEEE Std. 519-1992 respectively [15].
The power quality data from measurement are processed in accordance with IEEE 519-1992 [15]. As seen from Fig.12, in evaluation of voltage harmonics, THD of voltage for Wind Farm exceed the limits defined in IEEE 519-1992 in Table. 1. In Table.2, standard limits are specified from acceptable current harmonic limits defined by IEEE 519-1992 [15]. By com-paring limit values in Table 2 to measured values, limit of total current distortion is not exceeded in Wind Farm as shown in Fig.13.
Figure 6. Nominal frequency values measured at wind farm feeder.
TABLE I.
VOLTAGE DISTORTION LIMITS FOR GENERAL DISTRIBUTION SYSTEM.
Voltage Distortion Limits Bus Voltage at PCC 69 kV and below 69.001 kV through 161 kV 161.001 kV and above TABLE II. Individual Voltage Distortion THD (%) 3.0 1.5 1.0 Total Voltage Distortion THD (%) 5.0 2.5 1.5
CURRENT DISTORTION LIMITS FOR GENERAL DISTRIBUTION SYSTEM AT MEDIUM AND HIGH VOLTAGE
Current Distortion Limits for General Distribution Systems (120 V Through 69 kV) Maximum Harmonic Current Distortion in Percent of IL Individual Harmonic Order (Odd Harmonics) Isc/Il <11 11<h<17 17<h<23 23<h<35 <20* 4.0 2.0 1.5 0.6 20<50 7.0 3.5 2.5 1.0 50<100 10.0 4.5 4.0 1.5 100<1000 12.0 5.5 5.0 2.0 >1000 15.0 7.0 6.0 2.5
35<h 0.3 0.5 0.7 1.0 1.4
TDD 5.0 8.0 12.0 15.0 20.0
Figure 7. THD of voltage for wind farm on output feeder for three phase
Figure 8. THD of current for wind farm on output feeder for three phases
V.
CONCLUSION
In this paper, the power quality issues of wind plants at the point of common coupling with the HV transmission network are investigated. Voltage, current, frequency, active, reactive and apparent power, power factor and harmonics have been analyzed and evaluated based on IEEE 519-1992. The investigation shows that the harmonic values of wind plants have a slightly negative impact on the medium level of transmission network while the other parameters are mainly good. This analysis on real measurements is performed on total output of wind farms having 12 turbines. ACKNOWLEDGMENT Authors would like to thanks BELEN Wind farm, its own and manager and Wind Power Monitoring and Forecast Center (RITM) for their helps, collaboration and supports in our study. REFERENCES
[1] [2] C. Sankaran, Power Quality, Crc press, New York, 2002. M.M. Ertay and H.T. Duru, Harmonic analysis and passive power filter design for a medium voltage power system according to IEEE standards, Energ. Educ. Sci. Tech-A, vol. 29(2), pp. 1253-1262, 2012. F. Alessandro, Measuring electric power quality: Problems and perspectives, Measurement, vol. 41, pp. 121-129, 2008. A.P. Stavros, and S. Fritz, Power-Quality Measurements in an Autonomous Island Grid With High Wind Penetration, IEEE T Power Deliver, vol. 21(1), pp. 218-224, 2006. M. Sekkeli, Investigation of Power Quality Impact on the Textile Factories, Electronic Journal of Textile Technologies, vol. 3(3), pp. 111, 2009. M. Sekkeli and A.S. Yilmaz, Bir ta krma tesisinde g kalitesi seviyesinin lm ve deerlendirilmesi, Pamukkale University Journal of Engineering Science, vol, 15(3), pp. 317-323, 2009. M. Sekkeli and N. Tarkan, Development of a novel method for optimal use of a newly designed reactive power control relay, Int J Elec Power, vol. 44, pp. 736-742, 2013. A. Tascikaraoglu, M. Uzunoglu, B. Vural and O. Erdinc, Power quality assessment of wind turbines and comparison with conventional legal regulations: A case study in Turkey, Appl. Energ., vol. 88, pp. 1864-72, 2011. M. Imal, M. Sekkeli, C. Yildiz and O.F. Kececioglu, Wind energy potential estimation and evaluation of electricity generation in Kahramanmaras, Turkey, Energ. Educ. Sci. Tech-A, vol. 30(1), pp. 565-576, 2012. A. Kusiak and H. Zheng, Optimization of wind turbine energy and power factor with an evolutionary computation algorithm, Energy, vol. 35, pp. 1324-32, 2010. C. Kocatepe, A. Inan, O. Arikan, R. Yumurtaci, B. Kekezoglu, M. Baysal, A. Bozkurt and Y. Akkaya, Power quality assessment of gridconnected wind farms considering regulations in Turkey, Renew. Sust. Energ. Rev., vol. 13, pp. 2553-61, 2009. O.S. Mutlu, E. Akpnar and A. Balkc, Power quality analysis of wind farm connected to Alacat substation in Turkey, Renew. Energ., vol. 34 pp. 1312-18, 2009. L. Hocine and M. Mounira, Effect of nonlinear energy on wind farm generators connected to a distribution grid, Energy, vol. 36, pp. 325561, 2011. Z. Chen and F. Blaabjerg, Wind farm - A power source in future power systems, Renew. Sust. Energ. Rev., vol. 13, pp. 12881300, 2009. IEEE Recommended Practices and Requirements for Harmonic Control in Electrical Power Systems. IEEE Standard 519-1992, USA, 1992.
[3] [4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14] [15]
Short Term Economical Scheduling in an Energy Hub by Renewable and Demand Response
Samaneh Pazouki Department of Electrical Engineering Islamic Azad University (IAU), South Tehran Branch Tehran, Iran Student Member, IEEE samanehpazouki@gmail.com Mahmoud-Reza Haghifam Department of Electrical Engineering Tarbiat Modares University (TMU) Tehran, Iran Senior Member, IEEE haghifam@modares.ir Samira Pazouki Department of Electrical Engineering Islamic Azad University (IAU), Qazvin Branch Tehran,Iran Student Member, IEEE samirapazouki@gmail.com
Abstract Energy hub is a super node in electrical power system which has strong potential to receive varying energy carriers such as electricity and natural gas in hub input and as a minimum cost solution determines when and how much of which carrier should be purchased and stored in order to persuade required demands. Energy hub concept is not constricted to specific carriers and sizes; it can be extended for different approaches in response to different energy needs. In this paper, we present optimal scheduling of an extended energy hub for 24 hours under smart grid key drivers; renewable energy, heat and electrical storages, demand response in deregulated market. Numerical results through CPLEX (high performance linear programming solver) of GAMS software approve that proposed approach based on minimum cost solution can indicate when and how much of which carrier should be purchased, stored, interrupted or supplied by renewable resources to satisfy a commercial load. Impact of smart grid technologies and a comprehensive sensitive analysis on hub parameters; prices (electricity, gas, demand response and heat benefit prices), pipelines (electricity, gas and heat), technologies capacity (transformer, boiler, CHP, wind turbine, electrical and heat storages), wind speed and load participation factor is also evaluated on hub operation costs. Index TermsSmart Grid, Energy Hub, Mathematical Model, Optimal Operation, Wind, Energy storage, Demand response
A. Problem Outline Effective coordination of multi generation systems under smart grid technologies; i.e. renewable, storages, demand response with conventional grid originate novel challenges to develop predicting and scheduling tools to operate demand and supply balancing in deregulated market [2], [3]. B. Energy Hub as a Proposed Solution Integrated energy systems have been considered recently as Micro Grid[4], Hybrid Energy Hub[5], the latest approach originates in VOFEN (Vision of Future Energy Network) project [6] which defines Energy Hub as a super node in electrical power system to receives varying energy carriers such as gas and electricity, then strongly schedules when and how much of which carrier should be purchased and stored to provide hub required demands. Energy hub simplifies multiple systems optimization. (See [7], [8]). Model could be flexibility extended; demand response advance together with potential of storage technologies allows integrating variable generation with the fluctuations in hub in Big buildings, Industrial plants, Bounded geographic areas (country [9]), Island systems (electric vehicle [10]) C. Related Works Reference [1] operates and plans combined cooling, heat and power system (CCHP) plant under energy hub approach. Reference [11], [12] investigate impact of renewable generation on an optimization problem. Reference [13] balances hub wind power fluctuating with Plug in hybrid electric vehicle (PHEVs) as an energy storage; Vehicle to Grid approach (V2G). Reference [14] simulates incorporation of demand response (heat load management) with potentiality to shift load from high tariff time to low price time as energy storage for multi-energy network. Reference [15] is flexibly optimized as well as Reference [14] via proposed volatile market uncertain. Reference [16] analyses energy costs with intelligent price-based control concept (demand response) of
I.
INTRODUCTION
VER the past two decades, progressive of technology in distributed generation is leading to organize smart grid which is an umbrella for distributed generation and multi carrier systems in response to emission lower, efficiency enhancement, reliability improvement and cost reduction forward future energy needs. Combined heat and power plant (CHP) is a great example of the technologies which has the ability to convert one energy carrier to different energy carriers in order to establish multi generation systems [1] which allow arriving to the significant benefits. In addition, need for implementation of storages and demand response is raised to renewable especially variable generation to provide flexibility for supply and demand balancing [2].
Figure 1. Proposed energy hub in this paper for supply a commercial load with CHP, Boiler, Transformer Converters, Electrical and Heat Storages, Wind Turbine and Demand Response Program
micro CHP (below 15kw capacity) for household via model predictive control strategy, where demand response extremely reduces household cost of real-time pricing fluctuating about 1%-14%. Reference [17] employs the approach for smart network management by applying curtailment program of demand response. Reference [18] surveys demand response and energy storage impacts on domestic energy costs in residential area which is supplied within photo voltaic. D. Objective of the Paper This paper is organized to achieve economic optimal operation of the integrated energy resources for 24 hours based on energy hub solution as follow; Problem is drawn and modeled under proposed hub in section II. Section III mathematically optimizes presented energy hub by proposed objective function. Section IV investigates hub scheduling and sensitive analysis of hub parameters in details. Finally, conclusion is debated in section V. II. PROBLEM MODELING UNDER ENERGY HUB APPROACH Problem is drawn under energy hub approach in this part: A. Proposed Energy Hub for Problem Outline Energy hub is observed consist of in-output, direct connection, converters and storages from power system. A.1 Input and Output Energy Carriers Hub can receive variety of different grid energy carriers in its input. In this paper, below grid carriers are received: Grid electricity energy carrier Network natural gas energy carrier Excess produced powers are sold to grid in this light: Electricity demand Heat demand A.2 Direct Connections Direct connection transfers input grid energy carriers to hub output without any change in its nature such as hydraulic pressure, electric voltage in overhead lines, gas pipelines. No direct connection is used in this paper for proposed solution. A.3 Converters Converters change input carriers quality via e.g. fuel cell,
combustion engine, compressor, power electronic, etc to another form of output carriers. Employed converters: Transformer: for converting electricity Wind turbine: to provide electricity via wind power CHP: for heat, electricity produce from natural gas Boiler: to vary natural gas to heat energy carriers Note: Wind converter is neglected due to its high efficiency. A.4 Renewable Energy Due to global warming, Renewable Portfolio Standard (RPS) was adopted for many states to get a fraction of their electricity of the supplier from renewable energy by specified date [3]. Variability as variable generation output fluctuations and Uncertainty as less predictable magnitude and time [2] are outlined by North American Electric Reliability Corporation (NERC). This outlook observes: Wind power A.5 Energy Storage Storages thermally, electrically and chemically save energy and delivers to grid in required times. They set up to compensate stochastic feature of primary energy such as wind, wave and save energy. Electricity could be stored by super capacitor, super conductive, battery, PHEV [13], hydro reservoir, compressed air storage [2]. Used storage: Electrical storage Thermal storage A.6 Demand Response Demand response (DR) is described electricity consumption pattern change from normal usage in response to spot market of electricity price changes over time via smart technology [19]. DR is categorized to two sections (See [20] for comprehensive overview). In this paper, DR is modeled: Curtail able /Interruptible load service Certifiable is defined as an energy interruption by hub owner via smart technologies for max benefit, technical and economical results instead of utilizing generation unit. A.7 Proposed Energy Hub Proposed hub (Figure 1) consists of three converters (CHP, Boiler and Transformer) with two storages (Electrical and Heat) which supply electricity and heat loads from grid electricity and gas carriers. Wind power preferably supplies hub output. DR by curtailment potential acts in hub. Hub has potential to sell excess produced electricity and heat to grid. B. Problem Modeling under Energy Hub Approach In this part, problem via approach is formulated as follow: B.1 Converters Converter changes input renewable and non renewable carriers to different carriers with changing in its quality. (C) is defined coupling factor between input (P) and output (L) energy flow in time instant (H) (1). The energy carriers are introduced by Greek letters such as , ,.. The input powers are limited by minimum and maximum capacity restriction (2). If an input carrier enters to N converters, dispatch factor () defines how much P splits into K converters [7] (3). , corresponds to input and dispatched power of converter
K. Dispatch factor constraints are restricted in (4). In linear programming total power is dispatched in (5). Here, linear programming is employed for more speed [21]. (1) (2) . = (H) 3 1 , , =1 (4) 0 = (5) B.2 Energy Storage and its Dynamic model of energy storage constraint in response to primary renewable fluctuant and hub costs reduction is defined in (6)-(9). Storage content constraint , charge and discharge power of storages through its charge and discharge efficiencies and minimum and maximum factor of maximum capacity of storage is formulated in (7)(9). shows loss factor of storage. 1 (6) (7) (8) (9) B.3 Demand Response Curtailment program can directly curtail load via advance technologies in (10) and its constraints in (11). would be equal one for flexible load and it takes zero for fixed load. (10) 0 (11) III. SHORT TERM SCHEDULING IN ENERGY HUB
additional electricity is produced. Curtailment program of demand response supplies electricity demand. Heat demand is supplied (16) by CHP produced power and its efficiency, boiler produced power and its and it is . Heat storage is charged efficiency in required times. Additional produced discharged heat is sold to grid. Electrical storage is constrained (17) by remained power in storage 1 , its charge power , its and its loss power through its loss discharge power and its charge . Electrical storage factor is restricted in (19)-(21). and discharge power Heat storage is constrained (18) by remained 1 , its charge power , its power in storage and its loss power through its loss discharge power and its charge and . Heat storage factor is restricted in (22)-(24). discharge power Curtailed electrical demand is limited in (25). Imported network electricity is limited by transformer capacity CAPT (26). Imported gas from network is restricted by CHP capacity CAPC (27) and boiler capacity CAPB (28). Minimizing: OF= 24 1 24 1
0, , , 0,
+ 24 1
- (13)
(14)
Subject to: A.Objective function of Energy Hub Objective function of energy hub is described on minimizing costs by purchasing power or selling power with related carrier price (12). OF= (12) B.Objective function of Proposed Energy Hub Objective function (13) of proposed hub is economically scheduled based on purchasing electricity by grid and by , purchasing network gas for CHP its price and gas price and purchasing network gas for boiler , curtailed demand and its price . gas price Objective function constraints are also presented in this part. Electricity demand is supplied (13) by network electricity power through transformer and its efficiency . CHP produces electricity power through its gas to electricity . Wind turbine mathematically efficiency modeled in (14) and produces electricity power through its converter efficiency which is neglected in this context due its high efficiency. Electrical storage is discharged to produce electricity power and it is charged when
(15)
(16) = = 0 0 0 0 0 0 0 1 + 1 + (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) (27) (28)
Electricty and Heat Required Demands

1400 1200
TABLE I. 0.98 0.35 0.4 0.9 0.9 0.9 0.9 0.9 0.07
PROPOSED HUB PARAMETERS 0.03 3 0.03 3 200 200 1300 1300 500 0.9 800 10 22 4 400 0.01745 0.0776 0.032 1000 800
Electricty Heat
Demand (kWh)
1000 800 600 400 200 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Time (H)
LPF
CAPT
Y X ZZ CAPB CAPC
Figure 2. Electricity and heat hub required d demands

Electricty, Gas and DR prices
20 18 16 14 12 10 8 6 4 2 0
Electricty Gas DR
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Time (H)
Figure 3. Electricity, gas and curtailed dem mand prices

12
Wind Speed
Wind Speed
Wind Speed (m/sec)
10 8 6 4 2 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 1 18 19 20 21 22 23 24 Time (H)
Figure 4. Wind speed at 24 hour a d day
IV.
NUMERICAL RESULTS
Hub required electricity and heat demand ds are depicted in Figure 2. Electricity, gas and curtailed el lectricity demand prices are illustrated in Figure 3. Wind spe eeds are shown in Figure 4 at 24 hours a day. Proposed hu ub parameters are denoted in Table I. Simulation results com mpares importance of integrating renewable, energy storag ge and demand response parameters on hub operation cots in n Figure 5. Hub is economically scheduled under the techno ologies based on changing electricity price and wind speed at 24 hours a day in order to investigate when and how much of w which technology should be operated to satisfy electricity dem mand in Table II, III and heat demand in Table IV,V in response to hub operation costs. Finally, sensitive analysis w with 10% increase is applied on prices (electricity, gas, dema and response and heat benefit), pipeline capacities (electricit ty, gas and heat), technology sizes (transformer, boiler, CH HP, wind turbine, electrical storage and heat storage), wind d speed and load participation factor of curtailment deman nd in Table VI. Results show that which change is more imp portant than other changes to reduce hub operation costs. Figure 5 shows hub is operated with hig gh operation costs without smart grid technologies. Integrating wind turbine to
the hub decrease hub operation co osts till 45%. Integrating wind and electrical storage to the hub b decreases hub operation costs till 46%. Integrating wind, electrical storage and demand response to the hub decline hub operation costs till 53%. Hub operation costs are redu uced by the technologies, and it depends on technologies capacities. c Whatever the technologies sizes increase hub oper ration costs improve. Study of Table II and III (increa asing electricity prices till 10%) shows that when electrici ity price increases hub purchases electricity power with high h price. Gas power is purchased from network in peak demand d times in order to decline hub operation costs. When CHP C is committed, boiler is employed less and less heat is sold d to the grid. Heat storage is charged when heat demand is low. Boiler produces additional heat and it is discharged when CHP is committed and boiler is employed less (when n heat is produced less). When wind power is more, less elec ctricity is purchased from grid or sometimes additional electr ricity is sold to the grid when wind is high. Electrical st torage is charged when electricity price is low and it is dis scharged when electricity price is high to decrease opera ation costs. Interrupting electricity demand most depends on n load participation factor and it less related to interruptin ng price at every hour. Increasing DR interrupts electricity demands d more. Study of Table IV and V show ws that when wind speed increases, less electricity is purchase ed from grid or sometimes additional produced electricity by hu ub is sold to the grid. CHP could be utilized less in peak demand d times. Boiler is employed for producing heat, and th herefore additional heat is sold to the heat pipeline to achieve revenue and decline hub rged when heat demand is operation costs. Heat storage is char low and it is discharged when heat demand d is high. Electrical storage is charged when electricity price p and demand are low and it is discharged when electricity price and demand are high. Interrupting electricity deman nd most depends on load participation factor and it is less rel lated to interrupting price at every hour. Increasing dema and causes interruptible electricity demands occur more. Table VI shows that hub opera ation costs decrease when the technologies are integrated to the grid. Integration of wind, electrical storage and demand d response (case 4 or main case) to the hub decrease hub opera ation costs more than case 1 (hub without wind, electrical storage and demand response), case 2 (hub with wind), case c 3(hub with wind and electrical storage). Sensitive analysi is is investigated on all of hub parameter to study the effect of hub parameters on costs.
Price (cent/kWh)
TABLE II.
SUPPLYING ELECTRICTY DEMAND WITH INCREASING 10% ELECTRICITY PRICE
TABLE IV.
SPPLYING ELECTRICTY DEMAND WITH INCREASING 10% WIND SPEED
Supplying Electricity Demand by 10% Increase in Electricity Price
Supplying Electricity Demand by 10% Increase in Wind Speed
Hours
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Hours 600 470 500 420 400 400 420 600 600 500 600 700 700 700 800 800 800 800 1000 1200 1200 1200 1000 800 93.7 -29.6 -1.1 3.4 140.8 226.5 198.2 545.2 68.9 246.8 286.4 381.3 381.3 270.2 387.3 503.5 659.8 181.6 760.2 185.8 299.7 345.8 508.3 265.3 188.8 188.8 188.8 188.8 188.8 188.8 188.8 188.8 493.8 188.8 188.8 188.8 188.8 500 437.5 188.8 188.8 268.2 188.8 1300 1300 1226.3 240 240 400 400 400 347.8 347.8 347.8 137.5 137.5 137.5 137.5 211.2 211.2 211.2 211.2 211.2 211.2 211.2 298.8 298.8 298.8 347.8 347.8 347.8 400 42 32.9 35 29.4 28 28 29.4 42 42 35 42 49 49 49 56 56 56 56 70 84 84 84 70 56
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
26.7 180 180 7.3 180 180 19.4
600 470 500 420 400 400 420 600 600 500 600 700 700 700 800 800 800 800 1000 1200 1200 1200 1000 800
26.7 180 173.2 180 180 19.4
93.7 -29.6 -1.1 -49.7 87.6 279.7 147.1 494.1 17.8 195.7 216.0 310.9 310.9 199.8 317. 433.1 589.5 89.0 667.6 343.2 643.3 663.1 455.1 265.3
188.8 188.8 188.8 188.8 188.8 188.8 188.8 188.8 493.8 188.8 188.8 188.8 188.8 500 437.5 188.8 188.8 268.2 188.8 600 188.8 188.8 240 240
400 400 400 400 400 400 187.6 187.6 187.6 187.6 280.1 280.1 280.1 280.1 280.1 280.1 280.1 389.6 389.6 389.6 400 400 400 400
26.7 180 180 7.3 180 180 19.4
42 32.9 35 29.4 28 28 29.4 42 42 35 42 49 49 49 56 56 56 56 70 84 84 84 70 56
26.7 180 173.2 180 180 19.4
TABLE III.
SUPPLYUNG HEAT DEMAND WITH INCREASING 10% ELECTRICITY PRICE CHANGE
TABLE V.
SUPPLYING HEAT DEMAND WITH INCREASING 10% WIND SPEED CHANGE

Supplying Heat Demand by 10% Increase in Wind Speed
Supplying Heat Demand by 10% Increase in Electricity Price
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
700 700 700 700 700 700 550 550 420 420 700 700 700 420 420 420 550 550 550 550 620 620 550 550
375.5 375.5 375.5 375.5 375.5 375.5 500 500 500 500 500 448.9 375.5 500 500 500 500 500 500 500 500 500 500 500
1111.1 1111.1 1111.1 1111.1 1111.1 1111.1 1111.1 1111.1 806.1 1111.1 1111.1 1111.1 1111.1 800 862.4 1111.1 1111.1 1031.7 1111.1
73.6 1060 1060
188.8 188.8 188.8 188.8 188.8 188.8 188.8 188.8 493.8 188.8 188.8 188.8 188.8 500 437.5 188.8 188.8 268.2 188.8 1300 1300 1226.3 240 240
1 2 3 4 5 6
25.5 25.5 3.0 155.5 124.4 73.3
7 8 9 10 11 12 13 14 15 16
31.2 155.5 25.5 14.1 25.5 30 100 63.1
17 18 19 20 21 22 23 24
700 700 700 700 700 700 550 550 420 420 700 700 700 420 420 420 550 550 550 550 620 620 550 550
375.5 375.5 375.5 375.5 375.5 375.5 500 500 500 500 500 448.9 375.5 500 500 500 500 500 500 500 474.9 455.5 500 500
1111.1 1111.1 1111.1 1111.1 1111.1 1111.1 1111.1 1111.1 806.1 1111.1 1111.1 1111.1 1111.1 800 862.4 1111.1 1111.1 1031.7 1111.1 700 1111.1 1111.1 1060 1060
188.8 188.8 188.8 188.8 188.8 188.8 188.8 188.8 493.8 188.8 188.8 188.8 188.8 500 437.5 188.8 188.8 268.2 188.8 600 188.8 188.8 240 240
25.5 25.5 3.0 155.5 124.4 73.3
31.2 155.5 25.5 14.1 25.5 180 19.4
Sensitive analysis (Table VI) on electricity, gas, curtailed and heat benefit prices of the hub with wind, electrical storage and demand response show that when prices increase, hub operation costs also increase, except heat benefit price (case 8) which its price increase hub operation costs decrease because hub receives more revenue from grid. Increasing electricity (case 9), gas (case 10) and heat pipelines (case 11) capacities decrease hub operation costs especially in case 10 (increasing gas pipeline) because network gas is provides
energy carriers with low prices for the hub. Increasing transformer (case 12), boiler (case 13), CHP (case 14), wind turbine (case 15), electrical storage (case 16) and heat storage (case 17) decline hub operation costs. Wind turbine capacity is more effective than other technologies capacities on hub operation costs due to it free energy. Increasing wind speed (case 18) shows importance of rising wind speed on operation costs. Case 19 shows importance of increasing load participation factor on decreasing hub operation costs.
TAB BLE VI.

Case No 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 200000 180000 160000 140000 120000 100000 80000 60000 40000 20000 0 Technologies W ES + + + + + + + + + + + + + + + + + + Prices DR e g DR R HB
Sensitive Analysis on Hub Parameters

Cases Pipelines T B C OF Sizes W ES HS Others Wspeed LPF 181404.8 99129.7 96835.5 84570.8 93574.7 96920.8 84944.7 69968.7 84570.8 69230.0 82816.7 84570.8 81468.6 84570.8 76547.9 84340.7 84485.1 67198.2 83345.0
EH + + + + + + + + + + + + + + + + + + +
+ + + + + + + + + + + + + + + + +
+ + + + + + + + + + + + + + + +
10% 10% 10% % 10% 10% 10% 10% 10% 10% 10% 10% 10% 10% % 10% 10%
[4]
[5] [6] [7]

EH EH,W 99129.7 EH,W,ES S 96835.5 EH,W,ES,DR 84570.8
Costs
[8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21]
181404.8
Figure 5. Hub operation costs at different cases (case 1, 2, 3 and 4)
V.
CONCLUSION
was integrated to In this paper, an energy hub which w renewable, electrical storage and curtailm ment program of demand response was mathematically modeled and omically operated formulated. Proposed energy hub was econo to achieve minimum operation costs un nder smart grid technologies key drivers; renewable (wind d), energy storage and demand response program. Impact o of integrating the technologies on hub operation costs was inv vestigated. Results show importance of wind as renewable energy on hub operation costs. Sensitive analysis on hub parameters; prices, pipelines capacities, sizes, wind speed and load participation factor was compared. Results show imp portance of heat benefit price, gas pipeline, wind turbine size e and wind speed on hub operation costs. Proposed hub with h wind, electrical storage and curtailment program was scheduled under electricity price and wind speed changes in response to when and how much of which technology shoul ld be operated to achieve minimum operation costs. REFERENCES
[1] [2] [3] al analysis and optimal A. Sheikhi, A. M. Ranjbar, and H. Oraee, "Financia size and operation for a multi carrier energy systems s," Elsevier. Energy and Builduing, vol. 48, pp. 71-78, 2012. ts of emerging flexible NERC, Special Report, "Potential reliability impact resources, "http://www.nerc.com//, August. 2010 ns. Power and A. Ipackchi, F. Albuyeh, "Grid of future," IEEE Tran Energy Magazine, vol. 7, no.2, pp. 52-62, 2009.
R. H. Lasseter, "Micro grid," in Proc. 2002 IEEE Power Engineering R. i functional energy bus and its Frik, P.Favre-Perrod, "Proposal for a multi interlink with generation and consumption n", High voltage Laboratory, ETH Zurich, 2004. d, B. Klockl, G. Andersson, K. M. Geidl, G. Koeppel, P. Favre-Perrod Frohlich, "Energy hubs for the future," IEEE Trans. Power and Energy Magazine, vol. 5, no.1, pp. 24-30, 2007. M. Geidl, G. Andersson, "Optimal power flow of multiple energy carriers," , pp. 145-155, 2007. IEEE Trans. Power Systems, vol. 22, no.1, M. Geidl, G. Andersson, "Optimal coup pling of energy infrastructure," in Proc. 2007 IEEE Power Tech Conf., pp. 13 398-1403. M. Salimi, H. Ghasemi, S. Vaez-zadeh, " Irans electricity and natural gas networks modeling by means of energ gy hubs," in Proc. 2012 IEEE Environment and Electrical Engineering Conf. C , pp. 648-654. M. D. Galus, G. Andersson, Power system m considerations of plug-in hybrid electric vehicles based on a multi energy carrier c model, in Proc. 2009 IEEE Power and Energy Society General Meetin ng Conf., pp. 1-8. M. Arnold, G. Andersson, Modeling and optimization of renewable; oc. 2010 IEEE Power and Energy applying the energy hub approach, in Pro Society General Meeting Conf., pp. 1-8. M. Schuzle, L. Friedrich, M. Gautschi, Modeling and optimization of a in Proc. 2008 IEEE renewable: applying the energy hub approach, Sustainable Energy Technology Conf., pp. 83-88. son, Investigating PHEV wind M. D. Galus, R. La. Fauci, G. Anderss balancing capabilities using heuristics an nd model predictive control, in 2010 IEEE Power and Energy Society Gen neral Meeting Conf, pp. 1-8. P. Ahcin, M. Sikic, "Simulating demand d response and energy storage in energy distribution systems," in Proc. 2010 IEEE Power System Technology Conf., Hangzhou, China, pp. 1-7. aluing investment in multi energy F. Kienzle, P. Ahcin, G. Andersson, "Va conversion, storage and demand sid de management systems under uncertainty," IEEE Trans. Sustainable Ene ergy, vol. 2, pp. 194-202,2011. M. Houwing, R. R. Negenborn, B. D. Schutter, S "Demand response with micro CHP systems," IEEE Trans. Procee eding in IEEE, vol. 99, NO.1, pp. 200-213, 2011. Introducing a new framework for A. Ghasemi, M. Hojjat, M. H. Javadi, "I management of future distribution networ rks using potentials of energy hub ," in Proc. 2012 IEEE Smart Grids Conf., pp. p 1-7. F. Adamek, "Demand response and en nergy storage for a cost optimal residential energy supply with renewable e generation," Ph.D. dissertation, http://www.eeh.ee.ethz.ch/psl/research/vof fen.html, 2011. FERC, Staff Report, Assessment of demand response and advanced metering, www. FERC. Gov, Aug. 2006. r assessment2007-2016, NERC, Special Report, "Long-term reliability http://www.nerc.com/files/, Octobr.2007 A. Parisio, C. D. Vecchio, and A. Vaccaro, "A robust optimization to energy hub management," Elsevier, Electrical Power and Energy Systems, vol. 42, pp. 98-104, 2012. A. Shahmohammadi, M. M. Dalvand, M. S. Ghazizade, A. Saliminia l linear optimization with energy "Energy hubs structural and operational storage elements," in Proc. 2011 IEEE E Power and Energy Conversion Systems Conf., pp. 1-6.
Operation Costs (cent/kWh)
A Study of Hybrid Excitation Axial Flux Motor Topology for Electric Vehicle Traction.
Pelizari, A.
LMAG EPUSP So Paulo, Brasil ademir.pelizari@usp.br
Chabu, I.E.
LMAG EPUSP So Paulo, Brasil ichabu@pea.usp.br
AbstractThis paper presents a study of a non-conventional topology of axial flux machine for electric vehicle application using a hybrid excitation brushless system. In this type of motor both a high power and also a high torque densities are required, due the fact that the space is reduced in the vehicle as well in order to provide an excellent torque response. Hence the behavior of both mechanical and electromagnetics amounts of the machine were investigated. Keywords- electric traction motor; electric vehicle application; simulation of electric motors; axial flux machine; traction inwheel motors.
In (1) ROUT is the external radius of the disc in [m], the constant A is the rms linear current density, i.e., Am / 2 in [A.turns/m], B1AVG the fundamental air gap average flux density in (T), the factor Kd is the RINT / ROUT ratio of the disc. The constant Am is the peak value of the linear current density and it can be calculated as:
Am = 2 . N1. IA ROUT RIN . (RIN + ( ) 2 m1 .
(2)
I.
INTRODUCTION
The electric motors with double excitation have been developed since 50s. One of the main advantages of these machines is due the fact that it can operate at constant power region under flux weakening [1,2] which is an important feature especially in electric vehicle traction applications [3,4]. One of the problems to develop this type of motor is to achieve a good performance and a good range control of the excitation system due the high levels of flux density in some parts of the machine magnetic circuit [5]. Thereby the main purpose of this paper is a study of a non-conventional topology of Hybrid Excitation Axial Flux Machine (HEAFM) with double stator in order to make its operation at constant power region possible and also to make sure the requisites of vehicles electric traction is achieved. The research was focused on magnetic and mechanical aspects of the machine based on analytical method and pre-established values of torque and speed of the vehicle.
Developed Torque [N.m]
In (2) m1 is the number of phases of the stator, N1 the number of turns of the winding stator and IA the phase current of the stator winding. The air gap average flux density in (1) can be obtained in terms of the air gap maximum flux density Bg , as follows in (2)
B1AVG =
2 . Bg . sin . 2
(3)
In (3) the term is the relationship between the coil pitch and the pole pitch. Since the average flux density already has been defined, the developed torque can be evaluated as function of both the ranging of the linear current density and the average flux densities. The figure 1 depicts such situation:
Developed Torque [N.m] = f [Rout, Am]
160
140
A= 20000 A= 30000
120 A= 40000 100
II. THE DOUBLE SIDE AXIAL FLUX MOTOR TORQUE EQUATION In order to predict the torque performance of the electric machine, the influence of dimensions and magnetic variables were investigated. The developed torque equation of the double side axial flux motor [6] can be calculated as follow:
80
60
40
20
0.1
0.11
Td = 2 . . B1AVG . A . [ Kd - Kd ] . (ROUT )
3
0.12 0.13 External Radius (Rout) [m]
0.14
0.15
0.16
(1)
Figure 1 Developed torque as function of Disc External Radius.
This work was supported by CNPQ (Conselho Nacional de Desenvolvimento Cientfico e Tecnolgico, in Portuguese; English: National Council for Scientific and Technological Development)
III. ARMATURE DESIGN. The sizing of the armature was based on analytical methods [4,5,6] using the following data Power Rated (PR): 10 (kW) Input Voltage Supply (V1): 440 (V) Speed (NS): 600 (rpm) Frequency (f): 120 (Hz) Efficiency (): 0.9 Input Voltage / Airgap Voltage Ratio (): 0.9 Coil pitch / Pole pitch Ratio ( ): 0.637 Peak Magnetic Loading (AM): 38000 Max. Air gap Flux Density (B1MAX) [11]: 0.65 (T). These values have been used to design the electric motor at the constant torque region until 600 (rpm) [7,8]. At the weakening region (figure 2), the motor speed reach out 1200 (rpm) obtained through the excitation system control.
Rout
POLE =
B1AVG . dsp
(6)
Rin
In (6) dsp is the differential area per pole, calculated as:
dsp =
2 . . R . dr p
(7)
Thus, replacing (7) in (6) results
POLE = . B1MAX
(DOUT)2 - (DIN)2 8.pp
(8)
Since Kd is a dimension factor [6] it can be determined as

Kd = (RIN) (DIN) = (ROUT) (DOUT) = 1 3
(9)
Substituting Kd factor in (8), after some mathematical treatment results
POLE = . B1MAX .
. DOUT 2 (1 Kd 2 ) 8 . pp
(10)
Flux per Pole = f (pp, Dout)
Figure 2 Load requirements.
Flux per Pole [mWb]
0.8 0.6 0.4 0.2 0 0 2 4 6 8 10 0.1 0.2 0.3 0.5 0.4
According to rated characteristics, more specifically the frequency and the synchronous speed [9], the number of poles can be evaluated as:
p= 2 . f . 60 NS
(4)
External Diameter [m]
Pair of Poles
In (4) f is the rated frequency in (Hz), Ns the synchronous speed in (rpm) [12,13]. The next step was the average flux density B1AVG calculation as function of peak value of air gap flux density. Hence, the fundamental flux density per pole can be calculated as follow in (5):
p
Figure 3 Relationship among POLE , DOUT and pair of poles.
The figure 3 illustrates the dependence of the flux per pole as function of DOUT and pair of poles. Since KD is the dimension factor as function of Kd it can be calculated as
1 KD = [1 + Kd].[1-Kd 2 ] 8
B1AVG = B1MAX . sin (p) d

p
0
(5)
(11)
The magnetic flux per pole is determined in (6), based on average flux density per pole:
In this manner, the external diameter of the disc of axial flux motor [4] can be determined as follows
IV. HYBRID EXCITATION DESIGN.

DOUT = . PN . KD. K1. ns. B1MAX . AM . .cos
2
External Diameter of the Disc = f (Bmax [T], Am [A.turns/m])
(12)
0.5 0.4
0.3 0.2
0.1 1 2 3 4 x 10
4
0.1 0.2 0.3 0.4 5 6 7 0.8 0.7 0.5 0.6
Magnetic Loading [A.turns/m]
Maximum Flux Density [T]
Figure 4 Behavior of DOUT = f (B1MAX (T), AM (A.turns/m).
The hybrid excitation is composed by two systems [14,15,16]: one system consists of two pairs of coils allocated in the armature (see figure 9 and 10) in which produce 0.325 (T) controlled by a d.c. source of 100 [V]. The other system is the permanent magnet system composed by 24 distributed permanent magnets fixed to the rotor in which produces a maximum air gap flux density value of 0.325 (T) as well [7,11]. The main concern is to avoid that high levels of saturation occurs at the ferromagnetic material, for instance, teeth, yokes or even in the cover since not only one but two stators provide magnetic flux through the rotor. Therefore a maximum air gap flux density value of 0.65 (T) was considered, with both the excitation systems operating (electric excitation and permanent magnet excitation). Referring to the figure 8, the magnetic flux produced by the electric excitation establishes two flux paths through the rotor. One of them passes through the bottom yoke and the other one by the top yoke.
The relationship between DOUT, B1 and AM can be seen in figure (4). Substituting the rated values in (12) results
DOUT = 0,313 ( m )
And the internal diameter consequently is
External Diameter of the Disc [m]
DIN =
DOUT 3
DIN = 0,181 ( m )
(13)
Therefore, from (10) the flux per pole becomes
POLE = 1.778 m ( Wb )
Hence the number of turns of the stator winding was calculated as
Figure 5 Magnetic flux paths on the top and on the bottom of the rotor.
N1 =
. V1 . 2 . f . K1. POLE
= 250 [turns]
(14) Since the air gap flux density in non-sinusoidal, it is more convenient to assume an average flux density B1AVG at the air gap using a form factor KF
B1AVG = KF i B1MAX (16)
Since the prototype is a double side stator machine, the both stator windings were connected in series wye (Y) and that way the electric current of the armature can be calculated by
IA =
PN m1 . 2 . (VL/ 3) . COS .
= 14,5 [A]
(15)
From the average flux density in (16) the flux per pole can be determined which SP is the area of the pole
POLE = B1AVG i SP
And the developed torque is
Td = 145.76 [Nm]
(17)
Once the flux per pole is produced by two stators, in order to avoid magnetic saturation the flux density at the bottom of the pole as function of a minimum area (figure 6 and figure 7) becomes:
BMIN_AREA = 2 . PLO . KLEAKAGE SP _ MIN
(18)
BOTT_YOKE = POLE_PARTIAL . KLEAKAGE .
2.p (19) 2
Hence the flux density at the bottom yoke of the rotor becomes
BBOTT_YOKE =
BOTT_YOKE LBY . LR . KSTACK
(20)
The flux density at the top yoke of the rotor is

BTOP_YOKE = TOP_YOKE LTY . LR . KSTACK
(21)
Figure 6 Cut view of the minimum area of the pole.
Figure 9 View of the slotted armature.
And flux density at the teeth of the armature (figure 9) can be calculated as
BTEETH = POLE NSLOTS STEETH . KF . . KSTACK p
(22)
In (22), the term STEETH represents the area of the teeth, KSTACK is the stacking factor. And finally the flux density at the yoke of the armature was evaluated as
Figure 7 Flux path at the minimum area of the pole.
In (18) KLEAKAGE is the leakage factor.
BYOKE
POLE 2 = SYOKE . KSTACK
(23)
V. PROPOSED TOPOLOGY. The details of the proposed axial machine topology can be seen in figure 10 in a frontal view. The electric excitation is represented by 2 pairs of fixed coils in opposite sides of the rotor.
Figure 8 Variables of the rotor.
Referring to the figure 8 the variables LPOLE_PITCH, LPOLE, LTY, LBY, LR and LG were changed in order to determine acceptable values of flux density in all parts of the machine [17,18]. The flux per pole at the bottom yoke of the rotor can be evaluated by
Figure 10 Axial topology frontal view.
The figure 11 illustrates the proposed topology under analysis in an exploded view. In this figure it can be seen with more details the electric excitation system (in green color and red color) in opposite sides of the rotor [10]. The three phase winding is a consequent-pole concentrated coils type.
INTERNAL DC EXCITATION COIL (NORTH POLE) THREEPHASE WINDING
The dimensions of the rotor and the armature were determined based on the acceptable flux densities in all parts of the machine. In summary, the flux densities calculated are
BMIN_AREA = 1.9 (T) BBOTT_YOKE = 0.405 (T) BTOP_YOKE = 0.27 (T) BTEETH = 0.98 (T)
BYOKE = 0.56 (T)
ROTOR
INTERNAL DC EXCITATION COIL (SOUTH POLE)
SLOTTED ARMATURE
The results showed that there was not magnetic saturation in any part of the machine, in spite of the high level of flux density at the minimum area of the pole.
PERMANENT MAGNET
VII. CONCLUSION. Through the general torque equation (1) and the external diameter equation (12) was possible to determine the linear current density as well as the main dimensions of the armature disc. On the rotor, it was necessary to vary the main dimensions such as area of the pole, the pole pitch and the rotor length to obtain acceptable levels of flux densities. The concern in this type of design is preventing in that way, a critical condition which is magnetic saturation since the saturation point used in this project is 2.1 (T).
Figure 11 Proposed axial topology.
VI.
RESULTS.
Using average flux densities for double excitation system as well as the permanent magnet excitation system, the torque values in both regions, i.e. constant torque and constant power were obtained. According to figure 12, the lower the magnitude of dc excitation voltage, the greater the speed of the vehicle. The limit of flux weakening occurs when the voltage source is set to zero and consequently the resultant flux is about the half of the total average flux density, i.e., the relationship between the speed is approximately of 2:1.
REFERENCES
[1] [2]
[3]
[4]
[5]
[6] [7] [8] [9] Figure 12 Torque results.
Hendershot, J. R.; Miller, T. J. E., Design of brushless permanent-magnet motors, 1994: Clarendon press. Caricchi, F.; Crescimbini, F., Axial-flux permanent-magnet machine with water-cooled ironless stator, proc. Ieee Power Tech Conf., pp. 98-103, 1995. Sulaiman, E. ; Kosaka, T. ; Matsui, N., Design optimization of 12Slot-10 Pole hybrid excitation flux switching synchronous machine with 0.4kg permanent magnet for hybrid electric vehicles, IEEE (ECCE), 8th International Conference on, pp. 1913 1920, 2011. Profumo, F.; Zhang, Z.; Tencoti, A., Axial flux machines drives: a new viable solution for electric cars, IEEE trans. Ind. Electron., vol. 44, pp. 39, 1997. Hoang, E.; Lecrivain, M.; Hlioui, S.; Gabsi, M., Hybrid excitation synchronous permanent magnets synchronous machines optimally designed for hybrid and full electrical vehicle, IEEE (ECCE), 8th International Conference on, pp. 153 160, 2011. Gieras, J.F., R. Wang, and M. J. Kamper, Axial flux permanent magnet brushless machines, 2004. Husain, I., Electric and hybrid vehicles design fundamentals, CRC press, 2003. Lowry, J.; Larminie, J., Electric vehicle technology explained, John Wiley and Sons, 2003. Emadi, A.; Ehsani, M.; Miller, J., Vehicular electric power systems: land, sea, air and space vehicles , CRC press, 2004.
[10] Miller, J.M., Propulsion systems for hybrid vehicles, iet power and energy series 45, 2004. [11] Chabu, I.E.; Silva, V.C.; Nabeta, S.I.; Afonso, M.A.M.; Cardoso, J.R.,Axial flux concentration technique applied to the design of permanent magnet motors: theoretical aspects and their numerical and experimental validation Electric machines and drives, IEEE, 2005. [12] Kamper, M.; Wang, R., Analysis and performance evaluation of axial flux air-cored stator permanent magnet machine with concentrated coils, ieee electric machines & drives conference, 2007. [13] Chen, A.; Nilssen, R.; Nysveen, A., Performance comparisons among radial flux, multistage axial flux, and three-phase transverse-flux PM machines for downhole applications, IEEE Trans. Ind. Appl., vol. 46, n 2, pp.779-789, 2010. [14] Wang, S.; Xia, Y.; Wang, X., State of the art of hybrid excitation permanent magnet synchronous machines, Electrical Machines and Systems (ICEMS), International Conference on, pp.1004 1009, 2010. [15] Xia, Y. ; Wang, S. ; Ma, M. ; Hao L.; Qiu A. ; Huang, S.; Basic principles of hybrid excitation PM synchronous generator utilizing harmonic for excitation Electrical Machines and Systems (ICEMS), International Conference on, pp. 1010 1013, 2010. [16] Kefsi, L. ; Touzani, Y. ; Gabsi, M., Hybrid Excitation Synchronous Motor control with a new flux weakening strategy, VPPC, IEEE, p.p. 1 5, 2010. [17] Kong, L.; Wen, X.; Fan, T.; Zhu L., The field excitation adjustment characteristic of the parallel magnetic circuit hybrid excitation machine, Electrical Machines and Systems (ICEMS), International Conference on, pp. 1676 1680, 2010. [18] Hoang, E.; Lecrivain, M.; Hlioui, S.; Gabsi, M., Hybrid excitation synchronous permanent magnets synchronous machines optimally designed for hybrid and full electrical vehicle, IEEE (ECCE), 8th International Conference on, pp. 153 160, 2011.
DC/DC Converters for High Power Application: A Survey

A. Lachichi
ABB Corporate Research Centre SE-721 78 Vsters, Sweden amel.lachichi@se.abb.com
Abstract Wind energy collection/transmission systems are briefly reviewed, and the trend towards the use of a medium voltage DC collecting grid is identified by illustrating the market penetration of wind energy. The required technological change brings new challenges especially for the design of DC/DC converter topologies. The paper reviews topologies that have mainly been applied in transmission application and design issues to satisfy the main grid connection requirements are summarized. KeywordsHigh power application; multilevel converters; renewable energies DC/DC converter;
in using multi-converters structure is noticeable. Some literature discusses as well multilevel DC/DC converters, though those latter will meet more balance problems than that in AC/DC converters due to not only the asymmetric DC output voltage but because fewer redundant switching states remain available. This paper is organized as follow: In section II, an overview of the current development of wind energy power conversion system is presented in order to demonstrate the state of the art and to illustrate the trend towards the arrangement using a medium voltage DC collecting grid. Possible power conversion systems are briefly presented in order to justify the choice of DC medium voltage levels as those latter are not yet standardized. In the last section, DC/DC converter topologies that could interface a medium voltage and a high voltage DC grid are presented. II. OFFSHORE WIND FARM OVERVIEW The variable speed wind turbine technology using permanent magnet synchronous generators (PMSG) and full scale power converters appears to be the most successful configuration for the near future [1]. Today there are wind turbines ranging from a few watts applications up to an expected nominal power of 10 MW for large scale off-shore installations. As the technology matures, it is worth noting that wind farms will grow in size. The average size of projects being planned by developers increases to up to 600 MW [3], and hypothetical wind farms rated at 1 GW are already being investigated by academia. In the current state of the art, wind farms collecting grids are built in a medium voltage AC (10-33kV) as depicted in Figure 1 (a). However, DC collection at medium voltage level is considered to be more efficient with regard to power cabling, and is presented as being the future trend for offshore wind energy conversion systems (Figure 1 (b)). Although no technical-economic justification is made, the DC medium voltage level is set similar to the AC voltage level chosen in the conventional AC collecting concept for comparison purposes as exemplified in [4]. It is worth noting that various approaches for collecting the power in a medium voltage DC grid have been investigated and compared in term of costs of investment, operation and maintenance. In most cases, a
I.
INTRODUCTION
Being promoted by government-level policy, and established by commercial generation projects, renewable energies such as wind and PV have experienced one of the most significant growth in percentage of 20 % each year compared to other energies sources [1]. At the same time, despite the maturity of AC collection systems in term of protection for instance, there has been a renewed interest, both technically and economically, in DC collection and transmission of those renewable energies. Various approaches for collecting the power in a medium voltage DC grid have been investigated and compared with the traditional AC collecting grid in term of costs of investment, operation and maintenance. High power DC grids are a completely new technology and the required equipment for designing a high power medium to high voltage DC system are today not available. Most of the academic research works published so far tackle design issues at both system and components levels where, in most cases, a sensitivity analysis is carried out aiming to determine factors which most likely would affect this economic assessment and could favor the choice of a medium DC collecting grid [2]. However, the evolution toward the use of a medium voltage DC grid in many applications is a major technological change. One of the main challenges is the design of the DC/DC converter. The most common DC/DC converters encountered in low power applications cannot be extended to high power transmission application due to poor performances those converters exhibit. Other approaches are adopted and a trend
sensitivity analysis is as well carried out aiming to determine factors which most likely would affect this economic assessment and could favor the choice of a DC collecting grid. Clearly, the benefit of having a MVDC collecting grid is evident [2].
the ratio between the input and the output voltage is not only determined by the transformer ratio but also by the control of the power. The converter is in a soft switching mode under all load conditions if the converter operates with a neutral transformer ratio of 1. This helps avoiding poor efficiencies at partial load conditions which is common in wind farm. In order to carry out a comparative analysis between the dual full bridge converter with ZCS capability, and the series resonant converter, different semiconductors technology are used at the MV side of each converter. It is concluded that better performances regarding the efficiency is achieved with the series resonant converter, reaching an efficiency of 98.6 %.
(a)
(b) Figure 1. (a) Conventional power conversion system for offshore wind farms. AC collection and DC transmission. (b) Possible power conversion system. DC collection / DC transmission grid.
Figure 2. DC/DC structure formed by parallel/series connection of individual building block.
III.
DC/DC CONVERTER TOPOLOGIES
A wide variety of DC/DC converters are reported in the literature. However, even though they are mature for low power applications, few can be applied in high power transmission application due to low efficiency those topologies can exhibit. However, three options have been proposed and are reviewed in the following section. A. Multi-converter approach The multi-converter approach is one of the most predominant structures encountered in this kind of application. The building block of the structure is connected in parallel at the input, and is connected in series at the output (Figure 2). The most encountered building blocks are either the Dual Active Bridge converter or the Series Resonant Converter in either its two, three or n-phases configuration (Figure 3). Earlier research published in [5] demonstrated that the 3-phase dual active DC/DC converter presents interesting characteristics, such as lower filter ratings requirements compared to its single-phase counterpart, which makes the topology more suitable in high power application. The control of the power is realized by phase shifting the switching signals between the inverter and the rectifier of the topology. Besides,
Figure 3. Building block for the multi-converter approach. (a) Dual active bridge. (b) Series resonant converter.
To handle the high current from the medium voltage grid, it is proposed to use the multilevel cell as a building block for the topology as shown in Figure 4. The current flowing through each leg can be equally shared especially when the inductances are coupled [6]. Switches with lower current ratings can be chosen. It is worth pointing out that this cell is as well known in the literature as the interphase transformer
and it has been introduced as being the dual of the three-level flying capacitor converter [7]. Another benefit of this arrangement is that the effective switching frequency of the converter is increased making the filtering easier. On the high voltage side, any AC/DC converter arrangement is realized. In [8], the cell is used as the building block for a sevenlevel active neutral point converter with reduced number of components. The topology is shown in Figure 5. Note that reducing the number of switches in the converter is based on the reduction of the redundant switching state, and the switches that would be removed must not be involved in the creation of a voltage level. In this topology, switches connected to the DC bus bar have a high current rating which implies paralleling many switches to handle the high current.
and is shown in Figure 4. It is demonstrated that the voltage ratio can reach higher values without extreme duty cycle compared to the boost converter and can operate as well in bipolar mode.
Figure 4. Multilevel DC/DC converter without AC intermediate link.
Figure 4. Three-phase multilevel converter using a multilevel cell.
Front-to-front connection of multilevel converters is the second solution encountered in high power DC/DC converters interfacing medium to high voltage DC-grid (Figure 5). Moreover, it is possible to connect different multilevel converter technologies from each side of the DC grid in order to satisfy design requirements such as high current capability from the medium voltage side of the topology. Besides, the intermediate AC transformer can be optimized and designed at frequencies other than 50 Hz in order to increase the power density of the topology.
Figure 5. Front to front connection of different multilevel converter
Figure 5. Parallel active neutral point clamped converter with multilevel cell.
B. Multilevel approach Multilevel DC/DC converters without an intermediate AC link are derived from known DC/AC multilevel converters [9, 10]. However, the diode-clamp DC/DC converter has fewer redundant switching states and hence has limitation in DC/DC applications while the flying-capacitor converter can be used in different DC application due to the fact that the voltage balancing can be easily achieved as exemplified in [10]. A self balancing multilevel boost converter is proposed in [11, 12]
As exemplified in [13, 14], two single-phase modular multilevel converters connected in front-to-front configuration through a medium frequency transformer is presented as shown in Figure 6. Independently from the number of seriesconnected cells in each arm, both modular multilevel converters are controlled in a way to create only 2 levels AC output voltage with variable amplitude. This operating mode allows reducing the size of the cells capacitor.
REFERENCES
[1] [2] R. Teodorescu, et al, Grid Converters for Photovoltaic and Wind Power Systems, John Wiley and Sons, Ltd, ISBN 978-0-470-05751-3, 2011. P. Bresesti et al, HVDC Connection of Offshore Wind Farms to the Transmission System, IEEE Tran ON Ener. Conv. Vol. 22, No. 1, pp 37/43, 2007. List of Off-shore farms, available online: www.thewindpower.net. C. Meyer, Key Components for Future Offshore DC Grids, PhD thesis, Univ. of Aachen, ISBN 978-3-8322-6571-7, 2007. R. De Doncker et al, A three-phase soft switched high power density dc/dc converter for high power application, Tran. in Ind. App., Vol. 27, No. 1, pp. 63/73, 1991. R. Hausmann, et al, Three-Phase DCAC Converter Using Four-State Switching Cell, IEEE Tran. On Pow. Elec., Vol. 26, No. 7, pp. 1857/67, 2011. T. Meynard, et al, Parallel MultiCell Converters for High Current: Design of InterCell Transformers, 2010. N. Bouhalli, Etude et intgration de convertisseurs multicellulaires parallles entrelacs et magntiquement coupls, PhD thesis, INPT France, 2009. F. Zhang et al Study of the Multilevel Converters in DC-DC Applications, IEEE proceed. PESC 2004. F. Zhang et al, A New Design Method for High Efficiency DC-DC Converters with Flying Capacitor Technology, 2008. J. C. Rosas-Caro et al, Novel DC-DC Multilevel Boost Converter, J.C. Rosas-Caro et al, A DCDC multilevel boost converter, IET Power Electronics, Vol. 3, Iss. 1, pp. 129/37, 2010 S. Kenzelmann et al, A versatile dc/dc Converter Based on MMC for Energy Collection and Distribution, IET 2011. S. Kenzelmann et al, A Versatile DC-DC Converter for Energy Collection and Distribution using the Modular Multilevel Converter, EPE 2011
[3] [4] [5]
[6] Figure 6. Front-to-front connected M2LC.
IV.
CONCLUSION AND FUTURE TREND
[7] [8]
Wide panoplies of DC/DC converter topologies can be found in the literature and they are mature in low power application. However not all can be extended and applied in high power transmission application as they exhibit poor performances. In order to increase the amount of the power processed from the DC/DC conversion stage, several options such as multi- converter approach, use of multilevel converters, and the use of inductive coupling cells are used to build up the DC/DC converter topology. Multi- converter topologies using either the dual active bridge converter or the series resonant converter as building block are the state of arts nowadays. Multilevel DC/DC converters (without an intermediate AC link) seem to be a straightforward solution to be used in high power transmission application. However, the main drawback remains the capacitors voltage balancing not only because of the asymmetric DC output voltage but because fewer redundant switching states are available. A self-balancing multilevel boost converter has been proposed and can be used for instance to maintain the balancing of a diode-clamped converter. The front-to-front modular multilevel converter is the third solution encountered in high power DC/DC converters interfacing medium to high voltage DC-grid. It is worth mentioning that due to expected increasing levels of wind power, the front-to-front modular multilevel converter will be predominant as the M2LC is becoming mature in HVDC application.
[9] [10] [11] [12] [13] [14]
A Non-Intrusive Load Monitoring System Based on A Cascaded Method

K. L. Lian, K. S. Tung
Department of Electrical Engineering National Taiwan University of Science and Technology Taipei, Taiwan ryanlian@mail.ntust.edu.tw
AbstractNon-intrusive load monitoring (NILM) refers to a method of detecting the current energy consumptions of a house or a building using a single set of sensors at the utility service entry (USE). The voltage of USE in Taiwan can vary within 5%. NILM, which uses active (P) and reactive powers (Q) as the load signatures may suffer from very low detection rates. In this paper, we proposed a new NILM method based on cascaded strategy. The proposed algorithm consists of three stages. In the first stage, the conventional PQ method will be employed. In the second stage, a linearized admittance method will be used to update the fundamental components of the current signals in the database due the voltage variations at USE. Finally, in the third stage, the updated current spectra from the second stage will be used for identifying the loads passed down from the first stage. It is shown via experiments that the proposed method yield better detection rate than the conventional PQ method. Keywords- Non-intrusive load monitoring; voltage variations; linearized admittance; cascaded stages
Y. C. Su
Department of Research and Development Chicony Inc New Taipei City, Taiwan hello0716@hotmail.com However, such a method is only suitable for appliances characterized by significant spikes in power draw [5]. Leeb et al have used the concept of spectral envelope (SE) [8] for developing a NILM system. The SE is a vector of the first several coefficients of the short-time fast Fourier transform (FFT) of the transient current signal. Although this method can detect numerous appliances including the variable loads, the method requires excessive training for each appliance before classification and monitoring can be performed [5]. S. Patel et al. have proposed to monitor electric noise in a socket for transient signal [9]. However, the computational expense of capturing and analyzing transient noise can be excessive. Moreover, as suggested in [10], the appliance signature may depend on the household electrical wiring so that an appliance connected to a different socket may not be correctly detected. H. Y. Lam et al. in [11] have proposed to employ I-V curves for characterizing different loads. However, such an algorithm has not yet been successfully applied for a NILM system. There have been efforts to apply artificial intelligent methods such as artificial neural networks (NN) and fuzzy system are apply for NILM such as those presented in [12-14], [17-18]. However, these methods suffer from the fact that they require a long time for training in order to get satisfactory results, and provide little insights on the physical correlations between load characteristics or signatures, and detection accuracies. In [6], the authors have proposed to combine several of the existing algorithms such as active and reactive power (P-Q) method, eigenvalue method and instantaneous power waveform to form the committee decision mechanism. However, the problem of this method is that identification of a load will fail if a tie is reached by the committee. In this paper, a new NILM method based on the cascaded strategy is proposed. In the first stage, ten loads with the maximum likelihood will be first selected based on the decision rules of the first sage. Note that the number, 10 was randomly chosen. Then, they will be passed to a second stage to account for the voltage variation (discussed in details in section ). Finally, in the third stage, only one of the ten loads will be identified as the appliance in use. Unlike [6], the proposed method will never have a tie decision, and is shown The rest of the paper is organized as follows. Section II
I.
INTRODUCTION
Non-intrusive load monitoring (NILM) refers to a method of detecting the current energy consumptions of a house or a building using a single set of sensors at the utility service entry (USE) [1]. The idea of the NILM was developed in the 1980s at MIT by Hart [2]. His developed system uses the variations of real and reactive power to detect the status of loads in the P-Q plane [2]. However, several problems associated with this method are as follows: 1. The voltage variations of the utility may result in overlapping of loads in the P-Q plane [3]. 2. Different loads may consume the same active and reactive powers. 3. The method will fail to identify appliances that have nondiscrete changes in power [4]. Over the past two decades, there have been many other methods proposed for NILM. References [5] [6] and [16] provide a comprehensive list of available methods. Some of the most prominent methods and those not covered in [5] and [6] are briefly described as follows. Cole and Albicki [7] made use of edges and slopes as appliance features, in addition to the steady-state power draw.
This work was supported in part by the Ministry of Education, Republic of China (Taiwan).
describes the system setup. In Section III, the proposed method is described, and the experimental results are presented in Section IV. Finally, a conclusion is given in Section V. II. SYSTEM SETUP Fig. 1 shows the schematic diagram of the NILM system under study. This prototype system monitors the USE to a circuit breaker panel, which provides electrical hookup to the loads. The AC voltage is 110 V and 60Hz from the utility company. The loads are single-phased, and their power ratings are listed in TABLE I.
Fig. 2 The overall actual setup
III.
PROPOSED METHOD
The proposed NILM method consists of three cascaded stages. The overall flow diagram of the proposed NILM method is as shown in Fig. 3. The descriptions for each block in Fig. 3 are delineated as follows.
Fig. 3 Flow diagram of the proposed method Fig. 1 The schematic diagram of the NILM system under study TABLE I. APPLICANCE POWER RATINGS
ID 1 2 3 4 5
Name Lamp Electric kettle Hair dryer TV Vacuum cleaner
high low
Power (W) 21 500 750 200 113 1100
As indicated in Fig. 1, the measured voltage and current signals at the entry point are sent to a dedicated PC via an A/D card for display and analysis. Fig. 2 shows the overall actual setup.
A. Database Database plays an essential role in the NILM system as it is needed for load identification. In the proposed scheme, we will store three characteristics of the appliances. They are the PQ values, linearized admittances at the nominal condition, and harmonic spectra. Although there are five loads, there are a total of 47 combinations because the hair dryer can operate at the high speed and low speed mode. All the three features will be recorded for each of these combinations. The following briefly describes each of the three features. 1) PQ values All the PQ values recorded for database are measured at the nominal condition. That is the utility voltage at USE is at 110V, and the frequency is at 60 Hz. As mentioned in Section I, one problem of the PQ method is that it cannot deal with the situation when the utility voltage is varied. The utility voltages in Taiwan in general are varied within 5%. Therefore, one way of handling it is to record the values of PQ for each voltage variation in step of 1% as done in [13]. This requires a total of 517 (11 47) different values of P and Q. Such a method will require big efforts for data recording. To counteract the problem of the voltage variations, we proposed
to estimate the admittances of the loads and save them in the database as well. 2) Admittance Since most of the appliances are nonlinear in nature, the admittances for them are strictly not defined. However, since the voltage variations are in general no more than 5%, their admittances can be obtained by means of linearization. The linearization can be obtained by (1).
similarity between the mathematically in (4).

n
two
vectors,
and
is
defined
a b ai bi a1b1 an bn ,
i 1
(4)
where
a [a1 an ] and b [b1 bn ]
I V
(1)
In Euclidean space, the inner product can be intuitively viewed as what is shown in Fig. 4. As shown in the figure, the angle between vectors a and b can be obtained by (5).
where
cos
(2) (3)
a b a b
(5)
I I1 I 0 , V V1 V0 ,
and I1 is the fundamental component of the measured current when the voltage at USE is 1% of the nominal voltage value (i.e. V1) whereas I0 is the fundamental component of the measured current at the nominal voltage at USE (i.e. V0). Note that I1 and I0 can be obtained by FFT. The admittances obtained by the perturbation method are all stored in the database which will be used at stage 2. They can be used to estimate the deviations of the fundamental components of the currents when the voltage are in fluctuation. 3) Current Harmonic Spectra Since all the appliances are, strictly speaking, nonlinear in nature, current harmonics can also be used for identification purposes [3]. The current harmonics can be easily obtained by applying FFT to the steady-state current waveforms over one period. The measurements are also done at the nominal voltage conditions. Harmonics are retained for the first 15th since harmonics after that are negligibly small for most of the appliances. Moreover, to account for the utility voltage fluctuations, the fundamental components of the current spectra are estimated by means of the admittances obtained in Section III. A. 2) to improve the overall accuracy. Other harmonics are kept intact due to the assumptions that the utility voltage fluctuations are primarily related to the fundamental voltage component. In general, the fundamental voltage component will influence the fundamental current harmonics the most, despite the fact that other harmonics might also get affected. The impact on the other current harmonics are neglected in this paper to simplify the analysis. B. Stage 1 In stage 1, the measured steady-state voltage and current are used to obtain the real and reactive powers. These powers are then are compared or identified with those stored in the database. The identification tool we used is the inner product operation. The inner product essentially measures the
The stronger the similarity between the two vectors are, the closer the value will get to 1. Consequently, the the inner product between the P,Q values1 of the database and the P, Q values obtained from the measured V and I is evaluated. The ten loads among all whose values of cos closest to 1 will be passed to stage 2. Note that the great advantage of using inner product for identification tool is that it does not require time devoted for data training. This is drastically different from the NN- based NILM system, which requires a long time for training before the associated NN being used as the identification tool.
Fig. 4. The inner product in Euclidean space
C. Stage 2 As mentioned in Section I, one problem of the PQ method is that it cannot deal with the situation when the utility voltage is varied. To account for this, we proposed to use the fundamental admittance of the load described in Section III. A
P, Q values can form a 2-dimensional vector.
to estimate the change of the fundamental load current in the database due to the voltage variations. Since the current data stored in the database are all obtained under the nominal utility voltage, the proposed admittance method can be used to estimate the deviations of the fundamental components of these currents when the utility voltage is deviated from the nominal utility voltage. To estimate the deviations, (6) is employed I d Y *(Vd V0 ) , (6) where I d is the deviated fundamental current component due to the utility voltage, Vd . Therefore, the estimated fundamental component of the currents in the database can be written as I I0 Id (7) (7) will be used to update the fundamental components of the load currents for the ten appliances identified in stage 1. For illustration purpose, let us only look at the three of the ten identified loads in stage 1, and they are hairdryer (low speed), vacuum cleaner, and combined loads of TV and electric kettle under the condition that the utility voltage is 114 V ( 4% deviated from the nominal voltage). Figs. 5, 6 and 7, respectively show the updated current spectra of the hairdryer (low speed), vacuum cleaner, and combined loads of TV and electric kettle, respectively of the database under such a voltage condition, whereas Figs. 8, 9 and 10 show the actual current spectra of the three under such a voltage condition. The great resemblance between the updated database and the actual spectra are apparent from these figures.
Fig.6 The actual measured current spectrum for the hair dryer (low speed)
Fig 7. The updated current spectrum for the vacuum cleaner in database
Fig.5 The updated current spectrum for the hair dryer (low speed) in database
Fig 8. The actual measured current spectrum for the vacuum cleaner
IV.
RESULTS AND COMPARISONS
Fig 9. The updated current spectrum for the combined loads of TV and electric kettle
The system shown in Fig. 1 is used to test the proposed NILM method. In addition, the conventional PQ approach proposed in [2] is also implemented to compare against the proposed scheme. There are a total of 47 different combinations, and we tested each of them with the proposed method and the conventional PQ approach. TABLE III summarizes the detection accuracy rate. As shown in the table, the conventional approach offers great detection accuracy when the voltage at USE is not deviated from the nominal value. However, the detection accuracy is abruptly dropped when the voltage deviation is greater than 0%. The average rate is 37% On the other hand, the proposed method can maintain an average of 75% when the voltage deviation is greater than 0%. Thus, the experimental results showed that the proposed NILM algorithm can effectively enhance the overall accuracy.
TABLE III. EXPERIMENTAL RESULTS
Fig 10. The actual measured current spectrum for the combined loads of TV and electric kettle
Voltage fluctuation 0% 1% 2% 3% 4% 5% -1% -2% -3% -4% -5%
Proposed approach 97% 85% 78% 62% 60% 55% 92% 96% 80% 75% 62%
Conventional approach 97% 45% 32% 29% 28% 23% 72% 49% 40% 30% 17%
D. Stage 3 In this stage, the ten appliances identified in stage 1 will be compared with the database in terms of the current harmonic spectra. The identification tool used is also the inner product. Note that since the current harmonics are kept up to 15th harmonics, the vectors a (representing database) and b (representing the measured) are both 15-D vector. Assume that the current spectra of these loads are labeled as vector b1 to b10. The identification is performed as shown in TABLE II.
TABLE II. IDENTIFICATION BY MEANS OF INNER PRODUCT
V.
CONCLUSIONS
Loads Inner product
b1
a b1 a b1
b10
a b10 a b10
In this paper, a new NILM method is proposed. The proposed scheme consists of three cascaded stages. In the first stage, the loads are identified by means of the conventional PQ approach. In the second stage, the current spectra of the database are updated based on the linearized admittance method to account for the utility voltage fluctuations. In the final stage, these current spectra are used to compare with the measured current spectra to identify the loads which are in operation. Experimental results indicate that the proposed scheme can achieve good accuracy for all the operation modes under study. The proposed algorithm shows one way of increasing the detection accuracy and avoid using neural networks for training data for a long time. REFERENCES
Whichever has the highest score will be retained, and will be associated with the load, represented by vector a.
[1] C. Laughman, K. Lee, R. Cox, S. Shaw, S. Leeb, L. Norford, and P. Armstrong, Power signature analysis, IEEE Power and Energy Magazine, pp. 56-63, March/April 2003. G. W. Hart, Nonintrusive appliance load monitoring, Proceedings of the IEEE, vol. 80, pp. 1870-1891, 1992. M. Akbar, Z. Khan, Modified nonintrusive appliance load monitoring for nonlinear devices, IEEE International Multitopic Conference, 2007. H. H. Chang and H. T. Yang, Applying a non-intrusive energymanagement system to economic dispatch for a cogeneration system and power utility, Applied Energy, vol. 86, pp. 2335-2343, 2009. M. Zeifman and K. Roth, Nonintrusive appliance load monitoring: review and outlook, IEEE Transactions on Consumer Electronics, vol. 57, no. 1, Feb. 2011. J. Liang, S. K. K. Ng, G. Kendall, and J. W. M. Cheng, Load signature study part I: basic concept, structure, and methodology, IEEE Transactions on Power Delivery, vol. 25, no. 2, April, 2010 A. I. Albicki and A. Cole, Data extraction for effective non-intrusive identification of residential power loads, IEEE Instrumentation and Measurement Technology Conference, pp. 812-815, 1998 S. B. Leeb, S. R. Shaw, and J. L. Kirtley, Transient event detection in spectral envelope estimates for nonintrusive load monitoring, IEEE Transactions on Power Delivery, pg. 1200, 1995. S. N. Patel, et al., At the flick of a switch: detecting and classifying unique electrical events on the residential power line, Conference on Ubiquitous Computing, pp. 271-288, 2007 S. Gupta, M. S. Reynolds, S. N. Patel, ElectriSense: single-point sensing using EMI for electrical event detection and classification in the home, Conference on Ubiquitous Computing, pp. 139-148, 2010 H. Y. Lam, G. S. K. Fung, and W. K. Lee, A novel method to construct taxonomy of electrical appliances based on load signatures, IEEE Transactions on Consumer Electronics, vol. 53, no. 2, pp. 653-660, May 2007. D. Srinivasan, W. S Ng, and A C. Liew, "Neural-Network-Based Signature Recognition for Harmonic Source Identification," IEEE Transactions on Power Delivery, vol. 21, no. I, pp. 398 405, Jan 2006. H. T. Yang, H. H. Chang, and C. L. Lin, Design a neural network for features selection in non-intrusive monitoring of industrial electrical load, Proceedings of 11th International Conference on Computer Supported Cooperative Work in Design, 2007. Yu-Hsiu Lin; Men-Shen Tsai, "A novel feature extraction method for the development of nonintrusive load monitoring system based on BPANN," Computer Communication Control and Automation (3CA), 2010 International Symposium on , vol.2, no., pp.215,218, 5-7 May 2010 Reeg, C.E.; Overbye, T.J., "Algorithm development for non-intrusive load monitoring for verification and diagnostics," North American Power Symposium (NAPS), 2010 , vol., no., pp.1,5, 26-28 Sept. 2010 doi: 10.1109/NAPS.2010.5619600 Y. Du, L. Du, B. Lu, R. Harley; and T. Habetler, "A review of identification and monitoring methods for electric loads in commercial and residential buildings," Energy Conversion Congress and Exposition (ECCE), 2010 IEEE , vol., no., pp.4527,4533, 12-16 Sept. 2010 Y. H. Lin, M. S. Tsai, and C. S. Chen, "Applications of fuzzy classification with fuzzy c-means clustering and optimization strategies for load identification in NILM systems," 2011 IEEE International Conference on Fuzzy Systems (FUZZ) , pp.859,866, 27-30 June 2011 Y. S. Lai, Y. C. Chen, S. L. Tsao and T. C. Tsai, "A novel search scheme for nonintrusive load monitoring systems," 2012 IEEE International Conference on Industrial Technology (ICIT), pp.102,107, 19-21 March 2012
[2] [3] [4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
Arc quenching Equations Solving in HV Circuit Breakers by a General Diagram Bloc Model
A. Ziani and H. Moulai
USTHB, University of Science and Technology Houari Boumediene FEI/LSEI/MHVTRT, BP 32 El Alia, Bab Ezzouar Algiers Algiers, Algeria e-mail : moulaih@yahoo.fr
AbstractThis paper is devoted to the modeling and simulation of electrical arcs quenching in high voltage circuit breakers using a Matlab-Simulink software tool. The developed model enabled to succeed by a precise manner to the influence of the deionization constant on the conductance evolvement, and the plasma temperature during the arc quenching for two extinguishing mediums: air and SF6. We consider a line breaker 245kV/50kA/50Hz which is used in high voltage substations, and for which several experimental studies have been made. The results of this simulation have been compared to applied measurements. Keywords-circuit breakers; deionisation constant arc quenching; temperature;
addition to the laborious experimental measurements, numerical modeling constitutes an efficient way in order to orientate the conception of circuit breakers in the sense of their optimization while limiting the number of experimentations. Indeed, the simulation is an efficient and economic mean, commonly used in engineering for the achievement of preliminary or comparative studies, so much at the conception phase than during the normal operating of the systems. Among the useful simulation software tools, MATLAB-SIMULILNK remains one of the most used for both academic and applied research. In this work, we present the principles of the developed model for the extinction of an electric arc in a line breaker for which several experimental works have been achieved. For the conductance and temperature modeling, the Cassies model [5] will be applied for the two breaking insulating mediums: air and SF6. The values determined by numerical solving will be compared to experimental data. II. QUENCHING PRINCIPLE
I.
INTRODUCTION
Circuit breakers remain among the principal devices that are used for the control and security of electrical networks in order to ensure the interruption of the current in both normal operating or short-circuit conditions. Although appearing simple in its fundamental principle, since it consists on contacts separating, the currents cutting remains a very complicated phenomenon where several parameters intervene [1-3]. At the contacts opening, these apparatuses are simultaneously submitted to various types of constraints: dielectric, thermal and electrodynamics. Among all these constraints, the most important are those in relation with the transient phenomena associated to the electric arc occurrence at the breaking operations. However, till now, only electric arcs enable to dominate the high short circuit currents interruption process in both large interconnection networks and modest domestic installations [2-3]. The subtle raisons of their supremacy reside in their physical and thermo-electronic properties and, that confer them an incomparable ability to a fast transit from a conducting state to an insulating nature. In these conditions of extinction, the arc is violently blown when the intensity of the current passes at the vicinity of the zero current [1-4]. The short times of breaking and the strong energies liberated at the time of the arc extinction make the measurements delicate and very expensive. In this context, in
High voltage electric arcs quenching is always done at the current zero passage, and the extinction is succeed thanks to an increase of the resistance of the arc. So, after an electrical fault, the two contacts of the breaker will be separated, provoking thus the dielectric disruption of the inter-contacts insulating medium. The arc that appears consists of a column of plasma composed by ions and electrons coming from the inter-contacts, or from the electrodes metallic vapors for vacuum breakers. This column remains conducting so much that its temperature is sufficiently elevated. The electric arc is thus maintained by the energy that it dissipates by Joule effect [1-3]. For electric arc cutting in low voltage networks, the necessary voltage to its maintaining must be greater than the voltage provided by the network. This principle led to maintain a sufficiently long arc length in order to increase the arc voltage and then to succeed the arc quenching [2]. In Medium (MV) and High Voltage (HV) networks, the extinction succeeds when the thermal exchanges become greater than the thermal power given up by the arc under joule effect [1-3]. The speed of the dielectric strength recovering in
the inter contact gap depends on the value of the constant of deionization of the cutting medium [1-3]. III. ELECTRIC ARC MODELING
3], will be used in this article. This parameter will be however supposed to be stationary during all the quenching phase. IV. EQUATION OF VARIATION OF TEMPERATURE
Electric arcs breaking modeling in high voltage circuit breakers remains quite complicated especially with regard to the very important variations of the physico-chemical properties of the thermal plasma during the arc extinction phase. For that purpose, several models have been developed [1, 3]. For the HV breakers, the first basis models that permitted to obtain a macroscopic representation of the arcs are those of Cassie and Mayr [5, 6]. The Cassies model gives good results for currents greater than 100 A, and the Mayrs one is rather usable for weak currents, around the current zero. In MV and HV, the arc quenching in SF6 is practically the unique technique implemented in circuit breakers. In these voltage ranges, one finds principally the two basis models that describe the evolution of the SF6 conductance as a function of time: The model of A. M. Cassie that describes the evolution of the logarithm of the conductance g according to the arc voltage u and the parameters that define the properties of the arc, especially the constant of time and the Cassies arc voltage u c. d ln g 1 u2 = 1 (1) dt c u c
Temperature is the parameter that best describes the extinction phase of the arc. One notes that there is a difference between the electrons temperature and the weighty particles one. However, several authors suppose that all particles are at the same temperature [1-3]. Therefore, one will be able to suppose that the plasma is in thermodynamic balance during all the simulation. The temperature decrease during the extinction can be modeled by the heat equation:
.c
i2 T . KT .(T ) = h(T0 T ) + g t
(3)
Where KT is the thermal conductivity, T the temperature, the specific mass, c the specific heat at constant pressure, i the current, g the arc conductance and h the thermal exchange coefficient. The thermal transfer by conduction can be neglected with regard to the thermal exchange by forced convection [4]. In our simulation, we only take into account the cooling by forced convection. The source term in this equation is the power emitted by the arc which is i 2 g . The specific mass and the calorific capacity of SF6 have been assessed from the results of D. Koch [7] at the following conditions: pressure of 0.1MPa, =3 kg.m-3, c=1.5 J/kg/K. The value of h has been taken equal to 30, deducted from [4]. The temperature evolution has been set between 0 and 90s, consistently to experimental observations [8]. Concerning the maximal temperature level for an arc in extinction in a circuit breaker, it has been assessed by spectroscopy by Chevrier et al. [3] between 10000K and 20000K. By the use of the Matlab-simulink software, the heat equation has been solved numerically by using the ode45 code. V. PRESENTATION OF THE CONSIDERED NETWORK The equivalent circuit of the HV electrical network used in our simulations of the arc quenching is presented in the figure 1. It consists on longitudinal impedance that includes the lineic resistance and reactance of the line as well as two transversal admittances. This network is represented by an electromotive force e=245kV which debits on a line in ''pi section line'' whose characteristics are as follows: Resistance per length unit: 0.275 /km, inductance per length unit: 2.10-3 H/km, capacitance 8.6x10-9F/km. The length of the line is taken equal to 50 km. The number of sections is fixed to 1. In this survey, we consider a RLC load connected at the extremity of the line adapted such as to obtain a current of 50 kA. Figure 2 shows the Matlab-Simulink diagram bloc of the studied network.
The second model of basis is the Mayrs one, which is based on a limited number of simple and physically acceptable assumptions. It constitutes the prototype of all the family of so-called conductance models because they strain to explain the evolution of this latter according to the main parameters that define the arc.
d ln g 1 u.i = 1 dt p
Where g is the conductance of the arc u: the arc voltage i: the arc current : the constant of time p: the cooling power
(2)
The Cassies model gives good results for currents greater than 100 A, and the Mayrs one is rather usable for weak currents, around the current zero [1]. These two models take into account the constant of time effect which represents the time at the end of which the medium recovers its dielectric properties [1-3]. This parameter plays an important role in the breaking and re-breakdown phenomena [1]. Indeed, the electric arcs constant of time doesn't have a fixed value [1]. In the extinction mediums, this parameter evolves quickly at the vicinity of the current zero [1]. Experimental values of the constant of deionization time of SF6 and air, deducted from [1-
Breaker e=245kV
Line
Load
according to a 1 D geometry, while basing on the ElenbassHeller equation.
Fig. 1. The equivalent circuit of the considered electric network. VI. RESULTS AND DISCUSSIONS We chose to simulate the cutting of an electric arc in two insulating mediums: SF6 and air, by the suitable use of the models of Cassie and Mayr in a 245kV/50kA line breaker. The model is implemented in a Matlab-Simulink/ SIMPOWER SYSTEM software. This software is well adapted for modeling and simulation of non linear phenomena. It enables easily to build a diagram block and to simulate the behavior of the system in order to refine the conception of the breaking parameters. Several differential equation solvers have been tested in order to get the best convergence of the solution. We finally retain the ode45 solver with a variable time step in order to The investigation is satisfy a relative tolerance of 10-3. oriented towards the study of the influence of the deionization constant on the breaking velocity as well as the arc temperature. The used Simulink diagram block for the determination of the conductance and temperature is presented in figure 2. The simulation starts with chosen initial conditions, in accordance with the experimental measurements. The initial conductance and temperature of the arc plasma are taken equal respectively to 104 S and 10000K. The time of cutting is variable between 0 and 90s. The Cassies voltage is fixed to 30kV. The cooling by conduction is not taken into account because it is neglected with regard to the convection. Figure 3 represents the calculated evolution of the conductance of the arc in SF6 for the constants of time of 0.27s, 0.54s and 1.2s. The fastest decrease of the conductance is obtained for the constant of time 0.27s. This peculiarity has been already reported in the literature [1-3]. On the other hand, the obtained results are in good agreement with the results of J.J. Lowke et al [9] that foresee a very fast decrease of the conductance during the 100s of cutting in an arc in SF6. On figure 4, we drew the evolution of the conductance in air for a constant of time of 20s, 30s and 50s. One can note that the arc conductance become zero more rapidly for =20s. Figure 5 describes the temperature evolution in SF6 for an initial temperature of 10kK for the three considered values of . The existence of two phases (thermal expansion - cooling) for the three values of has been evidenced. The decrease of the arc temperature is obtained more quickly for = 0.27s, and the maximum temperature is obtained at 58s for =0.5s and at 38s for =0.4s, i.e. a delay of 20s. We have obtained speeds of cooling between 0.35kK/s and 0.5kK/s for = 0.27s, close to speeds obtained by J.B. Belhaouari et al. [10] that have modeled the arc temperature variations in a SF6 breaker
Fig. 2. Matlab-Simulink diagram bloc of the model
Time (s) Fig. 3. SF6 conductance variations as a function of time
Time (s) Fig. 4. Air conductance variations as a function of time
VII. CONCLUSION In this paper, the phenomenon of electric arcs extinction in high voltage circuit breakers has been simulated by Matlabsimulink for two breaking mediums: SF6 and air. The achieved model links between the high voltage line characteristics and the dynamic equation of the electric arc. Also, the model enables to foresee the conductance and temperature behavior during the extinction phase. The obtained results show well the influence of the deionization constant on the electric arc breaking behavior. Moreover, it has been evidenced the existence of two phases in the evolution of the arc temperature. A first phase during which the temperature increases under the effect of a strong thermal expansion, followed by a second phase characterized by a decrease of temperature linked to the arc extinction. REFERENCES
[1] S. Vacqui, ''L'arc lectrique'', Eyrolles collection '' Sciences et technique de l'ingnieur '', 2000. [2] S. Theoleyre, Les techniques de coupure en MT, Collection technique Schneider Electric CT N 193, 1998. [3] P. Chvrier, M .Barrault, C. Fivet, J. Maftoul, and J. Millon-Frmillon, Industrial applications of high-medium and low-voltage arc modelling, J Phys. D Appl. Phys., 30:1346-1355, 1997. [4] A. Ziani and H. Moulai, ''Extinction properties of electric arcs in high voltage circuit breakers'', Journal of Physics D: Applied Physics, Vol. 42, No 10, 2009, doi:10.1088/0022-3727/42/10/1052. [5] Cassie A M 1939 Arc rupture and circuit severity, a new theory CIGREBerlin. [6] O. Mayr Beitrage zur theorie des statischen und dynamischen lichtbogens, Arch. Elecktrotech 588-608, 1943. [7] D. Koch, Proprits et utilisation du SF6 dans les appareils MT et HT, Cahier technique n 188, Collection technique, Schneider electric. Edition fvrier 2003. [8] P. H. Schavemeker and L. Van der Sluis, An improved Mayr type arc model based on current zero measurement IEEE Trans. Power Delivery, No.15 pp. 580-84, 2000. [9] J. J. Lowke, R. E. Voshall, H. C. Ludwing, Decay of electrical conductance and temperature of arc plasmas, J. Appl. Physics, 44(8), pp. 3513 3523, 1973. [10] J.B. Belhaouari, Modlisation de lextinction dun arc de SF6 hors dquilibre thermodynamique local, Doctorat Thesis, Paul Sabatier University, Toulouse, France, N2780, 1997. [11] H. Rachard, P. Chvrier, D. Henry and D. Jeandel, Numerical study of coupled electromagnetic and aerothermodynamic phenomena in circuit breaker electric arc, Int. J. Heat and Mass Transfer, 42(9), pp. 17231734, 1999. [12] Jong Chul Lee , Youn J. Kim, SF6 arc plasma modelling for compact and environmental-friendly gas circuit breaker, Surface and Coatings Technology, Vol.201, Issues 9-11, pp.5641-5645, 2007.
Time (s) Fig. 5. SF6 temperature evolution as a function of time for different deionization constants. The initial temperature is fixed to 10kK. On figure 6 the deionization constant has been fixed to 0.27s. One first observes a fast rise of the arc temperature between 0 and 30s, passing from 17kK to 20kK for an initial temperature of 17kK, then, after 30s (i.e. for currents lower than 1300A), the arc becomes cold. Two phases are therefore visible on this figure: a thermal expansion phase owed to the injection of the Joule effect in the column of arc, and a phase of cooling linked to the blowing of SF6. This evolution in two phases has also been reported by Rachard et al. [11]. Jong Chul Lee et al. [12] have obtained also, in a numerical survey of the arc quenching in a SF6 breaker, a thermal expansion phase followed by an extinction phase.
Time (s) Fig. 6. SF6 temperature evolution as a function of time for different initial temperatures. The deionization time is fixed to =0.27s
Feature Form Extraction and Optimization of Induction Machine Faults Using PSO Technique
A.Medoued, A.Lebaroud, A.Laifa, D.Sayad
Dpartement de Gnie Electrique, Universit du 20 Aot 1955-Skikda, Alegria E-mail: amedoud@yahoo.fr
Abstract This paper presents a diagnosis method for induction machine faults investigation. The method is based on feature extraction and optimization. The feature form extraction is based on the time-frequency representation (TFR), which is designed for maximizing the separability between classes. A distinct TFR is designed for each fault class. The PSO algorithm is used for the feature form optimization. The classifier is designed with an artificial neural network. This method is validated on a 5.5-kW induction motor test bench.
Index Terms-- Induction Machine Diagnosis, TimeFrequency, ANN, PSO. I. INTRODUCTION Industrial processes require reliability and safety operation of electric motors. However, unexpected machinery failures provide loss of production, high emergency maintenance costs, damages to other related machinery, and extended process downtime. Thus, very expensive scheduled maintenance is performed in order to detect machine problems before they can result in catastrophic failure [1]-[3]. Therefore, there is a considerable demand to reduce maintenance costs and prevent unscheduled downtimes for electrical drive systems. In recent years, many research works have been carried out for the study and development of fault detection and diagnosis methods for electric machines. Recent advances of signal processing techniques, such as artificial neural networks [5], wavelets [6], etc.., have provided more powerful tools for fault diagnosis. The problem of diagnosis systems is that they use signals either in the time or frequency domain. In our approach, instead of using a time or a frequency approach, it is potentially more informative to use both time and frequency. Timefrequency analysis of the motor current makes signal properties, related to fault detection, more evident in the transform domain [7]. Traditionally, the objective of timefrequency research is to create a function that will describe the energy density of a signal simultaneously in time and frequency. For explicit classification, it is not necessarily desirable to accurately represent the energy distribution of a signal in time and frequency. In fact, such a representation may conflict with the goal of classification, generating a TFR that maximizes the separability of TFRs from different classes. It may be advantageous to design TFRs that specifically highlight differences between classes [8]-[11].
We propose to design and use the classifier directly in the ambiguity Doppler delay plane. Since all TFRs can be derived from the ambiguity plane, no a priori assumption is made about the smoothing required for accurate classification. Thus, the smoothing quadratic TFRs retain only the information that is essential for classification. This classification allows us to proceed to an optimization routine based on particle swarm technique to find the appropriate size of the feature vectors to reduce time consuming and keep signal relevant information into the vectors. In this paper, we propose a classification method based on of optimized TFR in order to extract the feature vector. Finally, a decision criterion based on a neural network classifier was used to ensure assignment to the appropriate class. The goal is the realization of an accurate diagnosis system of motor faults such as bearing faults, stator faults, and broken bars rotor faults independent from the level of load. II. CLASSIFICATION ALGORITHM. The classification algorithm is composed of the following three parts: extraction features and decision making. In the training stage, three optimal kernels are designed for separating four classes. 1) Class of healthy motor; 2) class of bearing fault; 3) class of stator fault; 4) class of broken bars. The kernel design process selects, for each class, a number of locations from the timefrequency ambiguity plane. In the decision making stage, we propose an ANN classifier with the Levenberg Marquardt algorithm. The details of each step are described in the following paragraphs. III. FEATURE EXTRACTION A. Optimal TFR For further details, we recommend the reader to review our previous work [12] and [13]. The characteristic function for each TFR is A( , ) ( , ) , represents the discrete frequency shift, represents the discrete time lag. That means the classification-optimal representation TFRi can be obtained by smoothing the ambiguity plane A( , ) with an appropriate kernel opt , which is a classification optimal
kernel. The problem of designing the TFRi becomes equivalent to designing the classification-optimal kernel opt ( , ) . This method, used to design kernels (and thus TFRs), optimizes the discrimination between predefined sets of classes. Features can be extracted directly from A( , ) opt ( , ) instead of the classification-optimal TFRi. This shortcut simplifies the computation complexity of feature extraction by reducing the calculations. IV. CLASSIFICATION USING NEURAL NETWORKS Neural networks are a relatively new artificial intelligence technique. In most cases an ANN is an adaptive system that changes its structure based on external or internal information that flows through the network during the learning phase. The learning procedure tries is to find a set of connections w that gives a mapping that fits the training set well. Furthermore, neural networks can be viewed as highly nonlinear functions with the basic the form:
7. If the sum of squared errors has not been decreased, discard the new weights, increase using v and go to step 6. 8. Else decrease
using v and stop.
Variations of the algorithm may include different values of v , one for decreasing and another for increasing it. Others may solve ( H + diag ( H )) = g instead of ( H + I ) = g , while others may select the initial according to the size of the elements on H, by setting 0 = t max(diag ( H )) , where t is a chosen value. We can see that we will have a problem if the error does not decrease after some iterations. In this case, the algorithm also stops if becomes too large.
V. PARTICLE SWARM OPTIMIZATION
F ( x, w) = y
(1)
Where x is the input vector presented to the network, w are the weights of the network, and y is the corresponding output vector approximated or predicted by the network. The weight vector w is commonly ordered first by layer, then by neurons, and finally by the weights of each neuron plus its bias. This view of network as a parameterized function will be the basis for applying standard function optimization methods to solve the problem of neural network training. Network training as a function optimization problem As mentioned previously, neural networks can be viewed as highly non-linear functions. From this perspective, the training problem can be considered as a general function optimization problem, with the adjustable parameters being the weights and biases of the network, and the Levenberg-Marquardt can be straightforward applied in this case. General Levenberg-Marquardt Algorithm The Levenberg-Marquardt consists basically in solving with different values until the sum of squared error decreases. So, each learning iteration (epoch) will consist of the following basic steps: 1. Compute the Jacobian 2. Compute the error gradient: g = J T E 3. Approximate the Hessian: H = J E
T
One goal of our approach is to minimize the size of the signal by the feature vector of very small size without losing relevant information of the signal. Hence the search for an optimum size of this vector provides a good compromise between the relevance of information and time consuming cost. In this work, the size used in the previous calculations was found by Particle Swarm optimization (PSO) method. A. PSO Introduced by Eberhart and Kennedy [4]. It is based on the analogy of swarm of birds and school of fish. In PSO, each individual called particle makes his decision using his own experience together with other individuals experience.. The main advantages of PSO algorithm are summarized as: simple concept, easy implementation, robustness to control parameters, and computational efficiency when compared with mathematical algorithms and other heuristic optimization techniques. In PSO, two different definitions are used: the individual best and the global best. As a particle moves through the search space, it compares its fitness value at the current position to the best fitness value it has ever attained previously. The best position that is associated with the best fitness encountered so far is called the individual best or pbest. The global best, or gbest, is the best position among all of the individuals best positions achieved so far (fig. 1). Using the gbest and the pbest, the ith particle velocity is updated according to the following equation:
vi k +1 = wvi k + c1rand1 ( pbesti si k ) + c2rand2 (gbest si k )
(2)
Based on the updated velocities, each particle changes its position according to the equation:
sik +1 = sik + vik +1
(3)
4. Solve
( H + I ) = g to find
5. Update the network weights
using
where, w is a weighting function, cj are acceleration factors and rand is a random number between 0 and 1, The following weighting function is usually utilized :
6. Recalculate the sum of squared errors
400 v 50Hz
w = wmax
w wmin max iter itermax
Inverter
(4)
Accelerometers ACC Load powder brake
Currents Voltages Coder
where wmax is initial weight, wmin the final weight, itermax is the maximum iteration number, and iter is the current iteration number. The parameters used are taken as follows [14]: C1=C2=2.05; wmin =0.1; wmax =0.9. Y
AS.M
ACC ACC
sk+1
vk v
k+1
Vg best
Acquisition System
Vp best
Fig.3: Signal
s
Fig.1:
Acquisition System
X Particle swarm method principle
B. Fitness Function For searching an optimized size of the feature vector based on PSO algorithm, a fitness function is needed. In this work, we consider the variance calculated for every size of the feature vector as the fitness for this size and the goal is to optimize this fitness. VI. EXPERIMENT RESULTS The experimental data are collected in Ampre Laboratory, University of Lyon. The experimental bench consists of a three-phase asynchronous-motor squirrel cage Leroy Somer LS 132S, IP 55, Class F, T C standard = 40 C. The motor is loaded by a powder brake. Its maximum torque (100 Nm) is reached at rated speed. This brake is sized to dissipate a maximum power of 5kW. Fig. 2 Power imbalance for simulating the fault of imbalance stator is obtained with a variable auto-transformer placed on a phase of the network (Fig. 2). The acquisition system used to measure these signals consists of eight differential inputs used to measure currents sampled up to 20 MHz 14-bit.
Induction machine
Powder Brake
Fig. 3 shows the device in place for the acquisition of signals: machine set up and acquisition of signals. The current signals sampling rate is 20 kHz. The number of samples per signal rises to N = 100000 samples on an acquisition period of 5s. The data acquisition set consists of 15 examples of stator current recorded on different levels of load (0%, 25%, 50%, 75%, and 100%). Different operating conditions from the machine were considered, namely, healthy, bearing fault, stator fault, and rotor fault. The training set is carried out on first ten current examples. The last five current examples are used to test the classification. Each signal is passed throw a lowpass filter and resampled with a downsamling rate of 50. Only the range of the required frequencies is preserved. By downsampling, the signal dimension has been reduced greatly. Using a lowpass filter is to avoid aliasing during downsampling. The dimension of ambiguity plane is (200 200 = 40000) points; by considering symmetry compared to the origin, we retain only the quarter of ambiguity plane, which corresponds to N = 10000. We designed three kernels: stator fault kernel, rotor fault kernel, and bearing fault kernel [3]. Fishers point locations in the Doppler-delay plane are ranged in the feature vectors {FV1,.., FVN} as training data base of the neural network . When designing a neural network, one crucial and difficult parameter to determine is the number of neurons in the hidden layers. If there are too few neurons in the hidden layer, the network may not contain sufficient degrees of freedom to form a representation. If too many neurons are defined, the network might become overtrained. Therefore, an optimum design for the number of neurons in the hidden layer is required (fig. 4) (fig. 5). In this research, we used one hidden layer with a number of different neurons to determine the suitable network.
Junction box
Coupling motorload
Fig. 2:
Motor 5.5 kW coupled with load (powder brake).
10
-2
10
-4
10
-6
10
-8
10
-10
10
-12
8 10 17 Epochs
12
14
16
Number of vectors classified
10
Performance is 1.65741e-013, Goal is 1e-012
T ra in in g -B lu eG o a l-B la c k
Dimension of the vector pattern
Fig. 4: Training diagrams for optimum case of (5) hidden neurons kernel 1
10 10
0
Fig. 7: Classification of test vectors for 35 training vectors
Introduction of the PSO In this work, the contribution of introducing the PSO is the optimization of the feature vectors size. By considering the variance as the fitness function, the size of the feature vectors is found to be 2, the first 2 with larger values of the variance (fig.8). It is important to note that the training vectors strongly correlate to the number of the classified vectors as can be seen on (Fig. 9).
Performance is 4.79085e-013, Goal is 1e-012
-2
Training-Blue Goal-Black
10 10 10 10 10
-4
-6
-8
-10
-12
10
15
20
24 Epochs
Fig. 5: Training diagrams for optimum case of (5) hidden neurons kernel 2 Table1: Misclassification results: Neuron number (3) 1/15 2/15 3/15 Actual error Neuron Neuron number number (4) (5) 0/15 0/15 1/15 1/15 3/15 2/15 Neuron number (6) 0/15 1/15 3/15
FDR1 FDR2 FDR3
Fig. 8: Feature vectors size optimization by PSO
The classification error is acceptable. It is mainly due to the limited number of the training vectors (fig. 6). Indeed, the classification error is minimized when we increase the number of training vectors to 35 vectors (10 vectors of stator currents at 0%of charge, 5 at 25%, 5 at 50%, 5 at 75% and 10 at 100% of rated charge). The results show a marked improvement in the classification process (fig. 7).
Number of training vectors
Number of vectors classified Fig. 9: Classification of test vectors based training vectors
Number of vectors classified
VII. CONCLUSION In this paper, we proposed a new fault classification scheme of induction machine based on TFR and criterion decision. We have based our classification on the ambiguity Doppler-delay plane where all the TFRs can be derived by a suitable choice of a kernel. Each type of fault was characterized by specific kernel. The classification scheme was tested with experimental data collected from the stator current measurement at different loads. The assignment of signal was made by an ANN classifier. The results verify that the new scheme with the neural
Dimension of the vector pattern
Fig. 6: Classification of test vectors for 20 training vectors
network classifier as decision criteria is able to detect and diagnose faults with high accuracy, independent of the load condition and the type of the fault. REFERENCES
[1] P. J. Tavner, B. G. Gaydon, and D. M. Ward, Monitoring Generators and Large Motors, Proc. Inst. Elect. Eng.B, vol. 133, no. 3, pp. 169180, May 1986. [2] P. Vas, Parameter Estimation, Condition Monitoring and Diagnosis of Electrical Machines. Oxford, U.K.: Clarendon, 1993. [3] A.Medoued, A.Lebaroud, A.Boukadoum, T.Boukra, G. Clerc, Back Propagation Neural Network for Classification of Induction Machine Faults, 8th SDEMPED, IEEE Symposium on Diagnostics for Electrical Machines, Power Electronics & Drives September 5-8, 2011, Bologna, Italy, pp 525-528. [4] J. Kennedy and R. Eberhart, Particle swarm optimization, in Proc.IEEE Int. Conf. Neural Netw., vol. 4, Nov. 1995, pp. 1942 1948. [5] H. Su and K. T. Chong, Induction Machine Condition Monitoring Using Neural Network Modeling, IEEE Trans. Ind. Electron., vol. 54, no. 1, pp. 241249, Feb. 2007. [6] A. Ordaz-Moreno, R. de Jesus Romero-Troncoso, J. A. Vite-Frias, J. R. Rivera-Gillen, and A. Garcia-Perez, Automatic Online Diagnosis Algorithm for Broken-Bar Detection on Induction Motors Based on Discrete Wavelet Transform for FPGA Implementation , IEEE Trans. Ind. Electron., vol. 55, no. 5, pp. 21932202, May 2008. [7] B. Yazc and G. B. Kliman, An Adaptive Statistical Time Frequency Method For Detection of Broken Bars and Bearing Faults in Motors Using Stator current, IEEE Trans. Ind. Appl., vol. 35, no. 2, pp. 442452, Mar./Apr. 1999. [8] M. Wang, G. I. Rowe, and A. V. Mamishev, Classification of Power Quality Events Using Optimal TimeFrequency RepresentationsPart 2: Application, IEEE Trans. Power Del., vol. 19, no. 3, pp. 14961503, Jul. 2004. [9] M. Davy and C. Doncarli, Optimal kernels of timefrequency representations for signal classification, in Proc. IEEE-SP Int. Symp. Time-Freq. Time-Scale Anal., 1998, pp. 581584. [10] C. Heitz, Optimum TimeFrequency Representations for the Classification and Detection of Signals, Appl. Signal Process., vol. 2, no. 3, pp. 124143, 1995. [11] B. W. Gillespie and L. Atlas, Optimizing TimeFrequency Kernels for Classification, IEEE Trans. Signal Process., vol. 49, no. 3, pp. 485496, Mar. 2001. [12] A. Lebaroud and G. Clerc, Classification of Induction Machine Faults by Optimal Time frequency Representations, IEEE Trans. on Industrial Electronics, vol. 55, no. 12, december 2008. [13] A. Lebaroud and G. Clerc, Accurate Diagnosis of Induction Machine Faults Using Optimal TimeFrequency Representations Engineering Applications of Artificial Intelligence. Vol. 22, Issues 4-5, June 2009, Pages 815-822. [14] V. Rashtchi, R. Aghmasheh A New Method for Identifying Broken Rotor Bars in Squirrel Cage Induction Motor Based on Particle Swarm Optimization Method , World Academy of Science, Engineering and Technology Vol 67, pp. 694-698, 2010.
Sustainability Assessment for an Isolated Micro-Grid using Hybird-DG System

Case Study in Marsa Alam, Egypt
Ibtihal Islam Zahran*, Eman Hassan Beshr**
AbstractEnergy is an essential element in human life, requiring a sufficient, secure and accessible supply for the sustainability of modern societies. An enormous increase has been noticed in the demand for energy since the middle of the last century as a result of industrial development and population growth. New generating stations are required continuously to cope with the increasing electrical demand. Some of those stations are environmental friendly powered by renewable energy, whereas many others are fossil fuels based, challenging the principles of sustainability. This paper focuses on two sustainability indicators; the cost of electricity generation and the greenhouse gas (GHG) emissions. Case studies are carried out on an isolated micro-grid operating by a diesel distributed generator (DG) to sustainably assess the system with and without the integration of renewable energy resources and battery storages. Keywords-Sustainability, Micro-grid, Distributed Generator, Cost of Generation, CO2 Emission, Photovoltaic model; Matlab m-file. ISC ISh K Ki LST n OACk OGC OMk q Rp Ta TASC AFCk APCk AST CFk Ck Eg Ek FCk FCRk G GCk ID Impp IOR IPh IPV Annual fixed cost of kth unit (EGP/yr) Annual Plant cost of kth unit (EGP/yr) Apparent Solar Time (minutes from midnight) Capacity factor of kth unit Capacity of kth unit (MW) Band gap energy for semiconductor (1.12eV for the silicon) Expected annual energy output of kth unit (kwh/yr) Annual fuel cost of kth unit (EGP/yr) Fixed charge rate of kth unit Solar Radiation ( kW/m2) Generation cost of kth unit (EGP/kwh) Diode current Current at point of maximum power Reverse saturation current of diode (A) Light generated photo-current Current output of a PV module TR UCk Vmpp VOC Vpv Vt Short-circuit current Leakage current Boltzmann's constant (1.380658*10-23 J/K) Temperature coefficient for short-circuit current (A/K) Local Standard Time (minutes from midnight) Number of series cells in a module Annual operating cost of kth unit (EGP/yr) Overall generation cost of the station having n generating units (EGP/kwh) Annual cost of operating labour, maintenance and supplies of kth unit (EGP/yr) Electron charge (1.6*10-19 C) Cell shunt resistance Ambient Temperature (K) Total annual station cost of a station having n generating units (EGP/yr) Reference cell temperature (K) Unit capital cost of kth unit (EGP/MW) Voltage at point of maximum power Open Circuit Voltage Output Voltage from the PV module. Thermal voltage (V) Declination of the sun (degrees) Latitude of the location (degrees) Solar hour angle in (degrees)
I.
NOMENCLATURE
II.
INTRODUCTION
Demand for electricity is escalating domestically and internationally along with regional and global environmental concerns. Driving the global energy system into a sustainable path has arisen as a major concern and policy objective. Consequently, the integration of new and renewable sources of energy became a priority in many countries all over the world. Electricity is a multipurpose energy carrier in our modern
*Electrical and Control Department, Arab Academy for Science, Technology and Maritime Transport, Cairo, Egypt. ibtihal.i.zahran@gmail.com **Electrical and Control Department, Arab Academy for Science, Technology and Maritime Transport, Cairo, Egypt. emy_beshr@yahoo.com
global economy, and it is therefore primarily linked to human and economic development. Electricity deserves precise attention with regard to its contribution to global greenhouse gas (GHG) emissions, which is reflected in the continuing development of low-carbon technologies for power generation. Greenhouse gas emissions not only cause global warming, but also other environmental concerns such as atmospheric pollution, acid precipitation, ozone depletion and forest destructions [1]. The concept of sustainable development is evolved for a livable future where human needs are met while keeping the balance with nature. Decarbonising the energy footprint of urban areas becomes therefore a critical point in the outlook of meeting challenging environmental targets set by governments in many countries and eventually fighting climate change. Many of those countries are parties in the Kyoto Protocol, an international agreement linked to the United Nations Framework Convention on Climate Change, which commits its parties by setting internationally binding emission reduction targets [2],[3]. Distributed Generators (DG) offer a quick, reliable and affordable solution to cover the gap between the shortage in supply and the demand required. There are several viable types of DG resources available in the market today, Table (I) illustrate and compare the main DG technologies as determined by the Distributed Power Coalition of America (DPCA) [4]. DGs could be used as backup if the utility power is lost, or as a source of energy for remote areas and microgrids where utility power is not available. The objective of this paper is to measure the sustainable achievement of the integration of renewable energy based DG with diesel-DG serving an isolated micro-grid. Case studies are carried out on an actual micro-grid located in Marsa Alam, in Egypt, fed by diesel-DGs. Marsa Alam is touristic place located by the Red Sea below Al-Ghurdaqah (Ghurgada), having a high potential of solar radiation as illustrated in Figure (1). A Photovoltaic (PV) DG model is proposed using Matlab m-file, to partially substitute the diesel generation. Two main factors belonging to the sustainability assessment are focused on; the cost of unit generation and the reduction in CO2 emissions. III. MICROGRIDS AND DISTRIBUTED GENERATORS MODELING
micro-grids, as it provides a power option for areas where transmission and distribution infrastructure does not exist or is too expensive to build. DGs connected to micro-grid use diverse types of technologies, some of which are low-carbon generation technologies with rising concern for global warming and environmental pollution as most countries are trying to reduce their emission levels as per the Kyoto Protocol. Global carbon commitment has been set to reduce carbon emission by 50% by 2050 [3]. Therefore, many countries are concentrating more on setting up eco-friendly plants with low-carbon generators. Micro-grids with cleaner DGs are increasingly preferred in this regard for reducing the overall environmental impact caused by power generation. Rules and regulations may be framed in future for giving credit to micro-grids that would help to substantially reduce net greenhouse gas (GHG) and particulate emissions in the environment [5]. In this paper two types of DGs are used; Diesel-based-DG and PV-based-DG. A. Diesel Generator Diesel fuel based generators are commonly used around the world. However, it is not highly recommended nowadays due to the rapid increase in fuel costs. Diesel generators are considered reliable with acceptable efficiency compared to renewable energy based generators, since it is not dependent on environmental parameters. One of the main drawbacks of diesel generation is the enormous environmental pollution it causes as it is a source of the GHG emissions. In this paper Diesel-DG is modeled as a two-state model; either on or off supplying its rated value. It is considered always on unless it passes by a scheduled or nonscheduled outage. B. PV Generator Different models were proposed previously describing the behavior of the PV module based on the identification of its parameters. This paper presents a single-diode PV module model that is completely based on datasheet values. The parameters needed from the data sheet are; ISC, VOC, Vmpp, Impp, n, TR and Ki. All other parameters affecting the power output of the PV module are calculated using the following equations; According to Figure 2, the current output of a PV module (IPV) is calculated by[6]: (1) (2) (3) Where the ambient temperature (Ta) in Kelvin and solar radiation (G) in kW/m2, are input values according to the location of implementation for the PV module. Using Shockley's Diode Equation, The diode current (ID) in ampere [6]: (4)
A micro-grid is defined as a cluster of distributed resource units and loads, served by a distribution system and can operate in grid-connected mode, islanded mode and ridethrough between the two modes [4]. Micro-grids appear to the main grid as aggregated units of loads and micro-sources, which are called distribution generators (DG). They are designed as small-scale, low voltage (LV) combined heat and power (CHP) networks for supplying electrical and heat loads to small pockets of customers. A number of micro-grids may also be integrated to form power parks for supplying larger load pockets. DGs can be used for peak shaving or as a way of helping the utility by reducing the system demand required [5]. They could be used as a stand-alone system for isolated
TABLE I. Item of Comparison/ DG Technology Dispatchable by Utility Available Capacity Efficiency Energy Density (kW/m2) Capital Cost ($/kW) O&M ($/kW)
Standard Diesel/ Gas Generator Yes 500 kW- 5MW 35% 50 200-350
SUMMARY OF DG TECHNOLOGIES Summary of Distributed Generation Technologies

Micro-turbine Generator Yes 500 kW- 25MW 29-42% 59 450-870 Photovoltaic (PV) Yes 1 1 kW-1MW 6-19% 0.02 6600 0.001-0.004 Wind Turbine No 10 kW-1MW 25% 0.01 1000 0.01 Fuel Cell Yes 200 kW-2MW 40-57% 1-3 3750 0.0017
0.01 0.005-0.0065 PV can be dispatchable with battery storage available today
(5)
(6) (7)
(8) Three parameters were usually treated in different manners in previously proposed PV models; parallel (shunt) resistance (RP), cell series resistance (RS) and ideality factor (A). In [7], the ideality factor was estimated using try and error method. RS was calculated by evaluating the slope dI/dV at VOC using the I-V curve available in the datasheet. That proposed method is considered inaccurate for the high tolerance it contained in RS calculation. In [6], the ideality factor was an assumed value, while Rs was calculated using a proposed algorithm. In this proposed model, A and Rs are both calculated, whereas RP was assigned to any large value due to its negligible effect in the PV output power. Ideality Factor (A) according to [8]:
Figure 1. Solar Radiation Distribution in Egypt
(9) Cell series resistance (RS) according to [3]:
Figure 2. One-diode PV Circuit
The PPVA (photovoltaic array power) can be calculated using the number of PV modules connected in series (NS) and parallel (NP) by [9]: (12) (10) The overall DC to AC derate factor is calculated by multiplying the component derate factors. For the PV system, the default derate factor is 0.77 [5]. Thus, (13) Principle of Newton-Raphson approximation algorithm using equation (8):
PV AC Power Newton Raphson's method is used to solve non-linear equation, in order to obtain numerical values for Ipv and Vpv to calculate the output power of the PV module (PPV) in W. (11)
(14) (15) (16) Where f'(IPV) is the derivative of the function, f (IPV)=0, IPV(i) is the present value, IPV(i+1) is the next value, Er is the relative approximate error which is calculated and compared to the pre-defined error tolerance Ea. IPV is computed iteratively and stops when Er=Ea. Solar Radiation Calculation The solar radiation (G) in kW/m2 can be calculated at the instant of any hour during the year, based on the geographic location, day and the time [10]. (17) Where, Gn is the beam radiation measured in the direction of the rays on the (n day) of the year in W/m2 and n is the day number of the year (varies between 1 and 365). (18) (19) (20)
In this paper two sustainable aspects are consider; the cost of unit generation which takes in account the minimum raw material (diesel fuel) with the contribution of free environmental recourses as the sunlight, and the other aspect is reduction in the GHG emissions due to the production of CO2 gas. A. Cost The cost of unit generation is one of the most important parameters in energy production, as it measures the cost for each kilowatt hour generated per year. It may vary depending on the fixed and running costs of the generation system. Various factors are considered; annual interest and depreciation, operation and maintenance costs, fuel costs, etc... This is done using the following equation [13]: (24) (25) (26) (27) (28) (29) B. GHG Regarding the earths environment preservation, control of GHG emissions, is one of the most important problems in the 21st century. Climate change is recognized as a global problem and can only be solved globally. Hence, it is not important at which exact location the greenhouse gas emissions reduction is achieved, but more importantly, that the reduction is done in a cost efficient way. Therefore, emission trading schemes have been developed, giving greenhouse gases a price - making it a tradable good. Carbon dioxide, CO2, is identified as one of the key drivers of global warming. The Kyoto Protocol was adopted on December 11th, 1997 in Kyoto, Japan. It entered into force on February 16th, 2005. The Kyoto Protocol can be seen as a milestone in fighting global warming, as it commits industrialized countries to reduce greenhouse gas emissions. It aims to reduce emissions of six greenhouse gases; CO2, methane, nitrous oxide, hydrofluorocarbons, perfluorocarbons, and sulphur hexafluoride [3]. The Kyoto Protocol offers three market-based mechanisms, one of which is the Emissions Trading (the carbon market). Article 17 of the Kyoto Protocol allows countries to trade their excess capacity [3], as so is the European Union (EU) Emission Trading Scheme (ETS), considered the largest multinational emission trading scheme in the world [14]. The governments of the EU Member States agree national emission caps and allocate the allowances to their industrial operators via so called national allocation
(21) (22) (23) IV. SUSTAINABILITY
According to the definition made by United Nations Environment and Development Commission, the sustainable development is a development that can meet the needs of todays world without endangering the ability of future generations to provide their own needs [11]. This could be achieved by using consumable resources of the world in the lowest possible level. The coverage of the sustainability concept in respect to electric energy production can be summarized; to increase the potential of electric energy production continuously, reducing waste materials resulting from that production in a level that can be cleaned in natural ways as well as reducing the risks related to human health in the lowest possible level and using raw materials including environmental resources in the lowest possible level [12].
plans. Plant operators have to monitor and annually report their CO2 emissions and they have to return the used emission allowances of CO2 in each year. Those installations which have allowances left over can sell them in the market or save them for future use. Those that exceed their total emissions have to pay a fine of 100 /CO2 emissions and their names are published. V. CASE STUDIES AND RESULTS
Three cases studies were carried out on an existing microgrid in Marsa Alam to investigate the sustainable impact that could be achieved by partially replacing the diesel generation by PV generation. The micro-grid is originally fed by five diesel generators each of rating 800 kW. The first case study was built on the original conditions of supplying the available load using all five diesel generators. Whereas the second case study used a PV generator rated 1600 kW and only four of the diesel generators, excluding the fifth one. A 1600 kW PV-DG replaced an 800 kW Diesel-DG considering the AC to DC derate factor according to equation (13), and the low efficiency of the PV according to Table (I). The third case is exactly the same as the second one, with an additional battery. The cases studies were simulated using Matlab m-file. The PV-DG model earlier mentioned used Sharp NU-E245 (J5) datasheet parameters; Pmax=245W ISC=37.5A, VOC=8.73V, Vmpp=30.5V, Impp=8.04A, n=60, and Ki=4.6269mA/K. whereas the hourly temperature data is collected, and the hourly solar radiation of the location is calculate along with the hourly output PV power using equations (1)-(23). The model used 6532 modules each costs 2330 EGP to produce 1600kW. Batteries are used to supply loads during generation shortages. Since the PV output power is dependent on sunlight, during night periods batteries are used to recover the PVs absence. Two batteries are used each having a capacity of 3100 Ah and costs 24800 EGP [15]. In addition, other data was collected to calculated the cost of unit generation; cost of 800kW-Diesel Generator is 1348053 EGP, cost for fuel and maintenance (including spare parts) varies according to the consumption, an average was used to be 0.3EGP/kWhr and 0.04 EGP/kWhr respectively. The load demand was recorded each our along with the outages of each generator. Table (II) shows the scheduled and non-scheduled outage of generators during the year. The case studies target to measure the sustainable impacts through two parameters; unit generation cost and CO2 emission reduction. According to [16], the diesel-DG and PVDG produce an average of 0.821 kg-CO2/kWhr and 0.105 kgCO2/kWhr respectively. Table (III) illustrates the results obtained regarding the OGC using equations (24)-(29), taking in consideration the unsupplied load and hours not supplied in each case as well as the CO2 emissions. VI. CONCLUSION The case studies results are reflected in Table (III) and Figures (4-6). Figure (4) illustrates the output PV power during
a random day with and without the battery. The battery works during the off periods of the PV and charges during the rest of the day, providing a more reliable and secure power supply. This is clearly shown in Figure (5) comparing the total energy supplied in each case. Case 1 supplied by diesel-DG is more reliable, with least unsupplied load and hours compared to case 3, whereas case 2 is the least reliable. However, the cost of unit generation for case 3 is more economical than that of case 1. Figure (6) compares the CO2 emission in each case and by each type of DG, showing a reduction in CO2 emission by 12% by replacing an 800kW Diesel DG by a 1600 kW PV-DG. Overall, the contribution of renewable energy achieved better sustainable results, however with deficiency in the continuity of providing energy. This could be overcomed with increasing the penetration of power; leading to an increase in the costs which could be waived when considering the carbon trading schemes and the taxes applied according to the Kyoto Protocol. Although Egypt is one of the Kyoto Protocol parties, yet it did not participate effectively. It is expected that all parties will soon need to contribute, and the Egyptian government should encourage PV power investments for its high solar radiation potentials through public incentives. This would lead to a decrease in the cost of the PV implementation in order to reduce GHG emissions while responding to the increasing energy demands providing reliable energy essential to maintaining competitiveness in the world market. Finally, the sustainable energy production aims to reduce to the lowest possible level the usage of natural resources, GHG emissions and usage of the electric energy production methods that cause wastes. This will contribute in the process of slowing down the global warming and climate changes, as well as new long term employment opportunities and more balanced economical development in the world will be ensured. This paper aimed to encourage the contribution of renewable energy in the electric energy production technology leading to a more sustainable world through two main sustainability parameters; the unit price of generated energy and the reduction in CO2 emissions.
Figure 3. Case Study System
TABLE II. SCHEDULED AND NON-SCHEDULED OUTAGES OF THE DIESEL GENERATORS Generator 1 Scheduled outages per month (hrs) January
10 2 8 11 6 7 9 12 9 9 3 0 6 194
Generat or 2
10 13 10 0 7 8 15 4 0 3 12 7 0 199
Generat or 3
10 0 2 9 12 0 4 3 12 0 5 10 2 179
Generat or 4
10 0 7 15 0 5 7 8 5 16 8 3 5 199
Generat or 5
10 8 0 2 3 17 0 7 5 6 2 0 4 174
Non-Scheduled outages per month (hrs)
February March April May June July August September October November December Total outage per year (hrs)
Figure 5. Comparison Between the Energy Supplied in Each Case
TABLE III.
CASE STUDIES COST CALCULATION RESULTS Case 1 Case 2

4 Diesel Power Stations Each 800kW + PV 1600kW 1.69212*1011 1.93958*1011 5.7532727*1010 1.828228*10 0.2966 213618 669 138923252.6 2598350.276
6
Case 3
4 Diesel Power Stations Each 800kW + PV 1600kW + Battery 682 kW 1.69212*1011 1.95647*1011 5.7532727*1010 2.292725*106 0.2918 20854 103 138923252.6 2598350.276
Description Energy Produced Using Diesel DG (kWhr) Total Energy Produced (kWhr) APCdiesel (EGP/yr) APCPV (EGP/yr) OGC (EGP/kWhr) Load not Supplied (kWhr) Number of hours Load was not Supplied (hr) tonneCO2/kWhr Produced by Diesel-DG tonneCO2/kWhr Produced by PV-DG
5 Diesel Power Stations each 800kW 1.95752*1011 1.95752*1011 6.655568*101

0
Figure 6. Comparison Between CO2 Emission in Each Case
REFERENCES
[1] I. Dincer, Renewable Energy and Sustainable Development: a Crucial Review, Renewable and Sustainable Energy Reviews, pp.157-175, 2000. Kyoto Protocol to the United Nations Framework Convention on Climate Change, United Nations, 1998. [Online]. Available: http://unfccc.int/resource/docs/convkp/kpeng.pdf Kyoto Protocol, MEM/03/154, Brussels, 23 July 2003. C.Smallwood,Distributed Generation in Autonomous ana NonAutomous Micro-Grids, IEEE Rural Electric Power Conference, 2000. S.Chowdhury, S.P.Chowdhury, P.Crossley, Micro-grids and Active Distribution Networks, Renewable Energy Series 6, IET, 2009. I. Houssamo, M. Sechilariu, F. Locment, G. Friedrich, "Identification of Photovoltaic Array Model Paramters. Modelling and Experimental Verification," International Conference on Renwable Energies and Power Quality, 23rd to 25th March, 2010, Granada, Spain.J. Clerk A. ElShahat, "PV Cell Module Modelling & ANN Simulation for Smart Grid Applications," Journal of Theoretical and Applied Information Technology, 2005-2010. K. Leban, E. Ritchie, Selecting the Accurate Solar Panel Simulation Model, Nordic Workshop on Power and Industrial Electronics, June 9-11, 2008. W. Zhou, H. Yang, Z. Fang, "A Novel Model for Photovoltaic Array Performance Prediction," Applied Energy 2007;84:1187-1198 J. Duffie,W. Becckman, Solar Engineering of Thermal Process,2nd Ed., N. Y. John Wiley & Sons, 1991. Bruntland GH. Our common future: the World Commission on Environment and Development. Oxford University Press; 1987. F.Iannone, S.Leva, D.Zaninelli, Hybird Photovoltaic System and Sustainability: Economic Aspects, IEEE Power Engineering Society General Meeting, 2004, Vol. 2, pp1933-1938. B.R. Gupta, Generation of Electrical Energy, Eurasia Publishing House, Ram Ngar, New Delhi-110055, 1996. A. Ellerman and B. Buchner, The European Union Emissions Trading Scheme: Origins, Allocation, and Early Results, Review of EnvironmentalEconomics and Policy, vol. 1, no. 1, pp. 6687, 2007. F.Iannone, S.Leva, D.Zaninelli, Hybird Photovoltaic System and Sustainability: Economic Aspects, IEEE Power Engineering Society General Meeting, 2004, Vol. 2, pp1933-1938. Environmental Costs of Electricity, Pace University Center for Environmental and Legal Studies, Oceana Publications, PLC inc. and Oak Ridge National Laboratories for the U.S. Department of Energy, 1994.
[2]
----0.3400 8880 80 160712344.3 -----
[3] [4]
[5] [6]
[7]
[8]
[9] [10] [11] [12]
[13] Figure 4. PV Power With and Without Battery over 24 hrs [14]
[15]
[16]
Human Motion Energy Harvesting by Design of Handheld Linear PM Synchronous Generator

M. H. Mohammadi1, M. Poshtan2
Department of Electrical and Computer Engineering American University in Dubai Dubai, U.A.E. Emails: mhmohammadi@ieee.org1, mposhtan@aud.edu2
AbstractThe authors present the experimental procedure to build a high power handheld-sized Linear Permanent Magnet Synchronous Generator (LPM SG) powered by human motion. This generator produces maximum energy for charging smallscale electronic devices such as mobile phones. Two different generator prototypes are compared through the presented theoretical and experimental results based on parallel charging of two 5V Li-ion battery supplies. The systematic choice of rare earth material PM, magnetization direction, solenoid wire diameter and number of turns have increased the generated power by more than tenfold between the discussed prototypes. Keywords-harvesting human kinetic energy; linear PM synchronous generator; power-turn density; power-mass density
In order to systematically create a power-optimized generator design, the key design parameters must first be identified. In Faradays Law of Induction, three main criteria can be used to design an LPM SG prototype. Each criterion is subdivided and elaborated in Sections II, III and IV to cover all the design parameters of the linear generator. II. PERMANENT MAGNET MATERIAL AND SIZE SELECTION New PM materials, such as NdFeB and SmCo, provide much higher magnetic field strengths than older Alnico and Ferrite materials, in the ranges of more than one Tesla [2]. As the magnetic field density B of the PM increases, the magnetic flux linked between the magnet and solenoid will further increase to induce more voltage in the solenoid. Using the hysteresis curve shown in Figure 1, the relevant points on the red-colored B-H curve define a magnets field strength. The residual field density Br is the amount of magnetic induction when there is no magnetization force H applied upon magnetic saturation. After material magnetization, the coercive force Hc defines how much of opposing magnetic field is needed to demagnetize a magnet. In order to have high Br and Hc in PMs for resisting residual magnetism loss, the maximum product of B and H in the green square area of hysteresis curves second quadrant is used to effectively compare different magnet materials. The maximum energy product BHmax is defined as the potential energy density in a magnetic material volume with the unit of kJ/m3.
I.
INTRODUCTION
Renewable energy sources such as solar, wind, hydro, and tidal require nation-wide projects to be invested by governments and multinational companies. Even though these projects require much time and effort, many people are currently not aware of the sustainability and future sight of renewable power generation and have yet to become more energy-wise in urban settings. Since the residential sector contributes to about 19% of the US energy use according to [1] and is expected to gradually increase in the following years, it is proposed to involve common people into generating their own electrical energy with the help of portable handheld generators. Despite the low-power range of basic human motion, it is both readily reliable and accessible at any time and place given a physically-fit human body. Through human motion, kinetic energy can be effectively converted into stored electrical energy for later use. This can increase the energy efficiency of our daily activities by charging mobiles phones and lighting safety LEDs during nighttime using a Linear PM Synchronous Generator. Current linear generator designs rely on the bases of Faradays Law of Induction in Equation (1), e = N d/dt.
Using Faradays Law, most linear generator manufacturers of low-power applications, such as flashlight generators, focus on creating the maximum possible number of turns using copper wires of small diameter, usually less than 0.1mm in thickness. Even though the induced voltage is higher than the required 5V peak, the generated current and subsequently, the generated power is not sufficient for widespread use.
Figure 1. Demagnetization Quadrant of the Hysteresis Curve [3]
The magnetization direction of PMs also directly affects the amount of induced solenoid voltage due to the effective flux linkage. An example of different magnetization directions such as axial and diametric fields is on Figure 2.
Figure 2. Magnetization Direction Examples: Axial (left), Through Thickness (middle), Diametrically (right) [2]
To maximize the induced voltage, the properties of four common magnetic materials are compared in Table 1. The Neodymium Iron Boron (NdFeB) material is shown to exhibit more maximum energy product BHmax, flux density, and demagnetization coercive force than other magnetic materials. For typical environments of about room temperature, the maximum working temperature of Neodymium magnets 100C will not be reached and is safe to be used in this respect. A point of consideration is that Neodymium magnets are known to be brittle and can shatter to pieces when attracted with other magnets. Therefore, Neodymium magnet material was chosen to be used in the linear generator due to its high magnetic strength and ease of manufacturing into various shapes and sizes.
TABLE I. Material Ceramic Alnico SmCo (2:17) NdFeB (N38H) COMPARISON OF MAGNETIC MATERIAL PROPOERTIES [4] Br (mT) 100 130 350 450 Hc High Low High High BHmax (kJ/m3) 26 42 208 306 Density (kg/m3) 4980 7200 8400 7470 Max Working Temp. (C) 250 550 300 120
In this research, five different NdFeB N48 grade magnet shapes were selected for testing the best magnetization direction. Their shapes and sizes, such as Ring, Rod, Sphere, Button and Cylinder, were relatively chosen to be close to the 2.1cm diameter of the 12cm long linear generator prototype tube. The prototype design structure and dimensions will be explained in Section III. The generator structure was kept constant while each magnet shape was tested in the range of 10 to 15 Hz shaking frequency by human hand motion and for different number of magnets. Since the mechanical shaking frequency is synchronous with the AC voltage frequency, the experimental shaking frequencies were measured on an oscilloscope with a voltage probe. Figure 3 illustrates the experimental data points of the induced voltage versus the number of magnets used. The best magnet configuration was found to be three Ring magnets due to its bigger size and axial magnetization direction with the best flux linkage. Figure 4 below displays the different Neodymium magnet shapes used with the highlighted triple-connected Ring magnet configuration selected for the LPM SG design.
Amongst the various magnet grades of NdFeB material, a high magnet grade of N48 was chosen. Table 2 shows how the low temperature range of Sintered NdFeB magnet grades gradually increase from the lowest N35 to the highest grade of N52. Different magnet shapes and sizes were testing upon choosing the N48 grade due to its relatively high maximum energy product of 382kJ/m3 and a magnetic field density of 1.38T. Other magnet grades with relatively higher working temperatures were available, but not chosen due to weaker magnetic strengths. The N48 working temperature of 80C is a reasonable maximum limit for the generator prototype.
TABLE II. SINTERED NDFEB MAGNET GRADES FOR LOW TEMPREATURE RANGES [2] Hcb (kA/m) 867.4 899.2 923.1 923.1 875.4 835.6 795.8 835.6 BHmax (kJ/m3) 278.5 302.4 318.3 334.2 358.1 382.0 397.9 413.8 Density (kg/m3) 7400 7400 7500 7500 7600 7700 7800 7800 Max Working Temp. (C) 80 80 80 80 80 80 80 80
Figure 3. NdFeB Magnet Shape Test for Single 12cm Long Generator with 0.50mm Solenoid Wire Thickness with 2cm of Centered Space
Grade N35 N38 N40 N42 N45 N48 N50 N52
Br (T) 1.18 1.22 1.25 1.28 1.32 1.38 1.40 1.44
Figure 4. NdFeB Magnet Shapes shown in clockwise direction from the upper-left corner: Axial Button, Diametric Cylinder, Axial Sphere, Single Axial Ring, Three Axial Rings with Stoppers, Axial Rod
III.
GENERATOR DESIGN STRUCTURE
The air gap distance between the permanent magnet and solenoid has an inverse proportional relationship with the magnetic flux linkage. As shown in Equation (2), the flux is proportional to the magnetic flux density B and the effective area A between the magnet and solenoid. Although a small air gap is required for the magnet to move freely inside the generator tube, its distance must be minimized to stop the adverse weakening flux linkage effects. The linear generator structure material can also attenuate the flux linkage of the PM. It is necessary to select thin and high magnetic susceptible insulation tubes for maximum flux linkage. In addition, the length of the generator tube leads to a higher magnet travelling velocity, directly related to the voltage induction characteristic curves. As Faradays Law includes the derivative of the magnetic flux over time, the faster change in flux linkage would generate higher induced voltage. Through experimentation, the medium shaking frequency of the linear generator was chosen to be between 7 to 13Hz for an average human adult shaking the LPM SG at a normal rate.
accounts for the substantial change of the produced current. A high conductance copper wire of 5.96x107S/m is used for its small resistive heat loss. Even though the generated emf linearly increases for higher number of solenoid turns, the total coil impedance also adversely increases with the inductive reactance of the solenoid. Since the generator tube structure and the magnet material and shape have been fixed and chosen, the correct choice of the solenoid design is crucial. A series of experiments have been conducted to find the relation between the peak power and high induced voltage with respect to both the wrapped wire diameter and the solenoid total number of turns. A. Sensitivity Analysis This analysis helped in eliminating the unneeded solution for each design parameters on way of enhancing the generator power output. The affecting variables include wire diameter, solenoid number of turns, and magnet material. The method of One-Factor-At-a-Time (OFAT) was used to test the effect of each of the above-mentioned variables by keeping the rest of the variables unchanged in the three tests explained below. B. Test 1: Changing the Wire Diameter for Magnet 1 In Test 1 experiment, the fixed dimension and shape of the 12.0cm long generator tube with 2.1cm diameter was used to find the different relations with the copper wire thickness. There were 15 different generator test cases, ranging from 0.10mm to 0.85mm thickness with approximately 0.05mm step increments in diameter size. For each wire thickness, the 2cm wire wrapping spacing was covered for an entire level with minimal overlaps to account for the solenoids geometric limitation. Test 1 used lower grade Magnet 1 to demonstrate the comparison with the stronger NdFeB magnets in Test 2. In Figure 6, the calculated and measured inverse square resistance curves are shown. From about 0.25mm to the right of the horizontal axis, the resistance per turn starts to have a smoother slope and stays within the 20m/turn range. The commercially-available flashlight generator design used wires thinner than 0.10mm resulting in high wire resistance.
Maximum transfer of magnetic field strength between the Neodymium PM and the solenoid is desired. Commerciallyavailable shaking flashlight linear generators used thick insulation tubes which attenuated the magnetic field linking with the centered solenoid. In order to fix this problem, insulation papers currently used for copper windings in electric motors and generators were chosen. It is able to withstand high temperatures and electrical currents while motors and generators are running at high load. Hence, an insulation paper with a small thickness of 0.5mm was chosen to construct and design the cylindrical generator tubes. The top and side views of the cylindrical generator tube design are shown in Figure 5. The 12cm length with 2cm of wire wrapping space ensures that the Ring magnets get head start acceleration from one end to another. This constant wire wrapping space also considers the geometric limitation of the maximum number of turns to be used.
Figure 5. NdFeB Blueprint Drawings of the Generators Cylindrical Tube Design for Side and Top Views
IV.
SOLENOID DESIGN AND EXPERIMENTATION

Figure 6. Resistance per Turn Relation with Wire Diameter for Measured and Calculated Values
As long as the magnetic flux of the PM entirely cuts through the solenoid, the induced voltage linearly increases with higher number of solenoid turns. Even though high number of turns is desired for inducing more voltage, the electromechanical energy conversion also depends on the solenoids impedance characteristic. The wire diameter has an inverse-square relation with the wire resistance, which
Figure 7 shows how the number of turns packed into the fixed space changes. From about 0.45mm to the left of the horizontal axis, the number of turns which is directly related to the induced solenoid voltage starts to sharply increase asymptotically at the risk of higher wire resistance.
wire diameter. Even though the impedance incorporates both the wire resistance and reactance, the lagging reactance may be ignored for wires smaller than the mid-range diameter due to the significantly higher resistance value. For the mediumhigh frequency range, the average saturated current density for the 15 test generators is about 5.00mA/turn. In order to evaluate the better solenoid performance, the power density, which has the unit mW/turn, was computed as a figure of merit. For the four frequency ranges and the same 15 test generators, the observed characteristic curves were obtained as shown in Figure 9. Between 0.25mm and 0.60mm wire diameter, the maximum power density data points were experimentally computed. The peak power density for the medium-high frequency range is about 0.90mW/turn.
Figure 7. Number of Turns in 2cm Space Relation with Wire Diameter for Measured and Calculated Values
In order to measure the open-circuit voltage and shortcircuit current of the linear generator, the experiment was standardized as follows. As the shaking frequency of the generator is directly related to the induced voltage, four different shaking frequency ranges were defined: low frequencies from 0 to 5Hz, medium-low frequencies from 6 to 10Hz, medium-high frequencies from 11 to 15Hz, and high frequencies from 16 to 20Hz. Upon experimentation, an average human is able to shake his or her hand to power the generator in the medium-high frequency range. In the following regressed graphs, the deviation of certain data points is attributed to the uneven shaking of the generator through human testing, spread of each frequency range, and slight nonuniformities in the hand-made generator designs. For each of the 15 wire thicknesses, the RMS open-circuit voltage was measured, and the voltage-turn density was computed to present a measure of how the voltage production varies from one wire diameter to another. The voltage-turn density is the ratio between the RMS voltage and the number of turns. This will eliminate the biasing effect of higher voltages due to higher number of turns in thinner wires. For the medium-high frequency range, the average voltage density for the 15 test generators was found to be about 4.50mV/turn. The peak AC voltage was measured to obtain the RMS ratio by dividing it with the RMS voltage. The RMS ratio shows how much area is covered by the induced voltage over one period. For sinusoidal signals, the RMS ratio is about 1.4. For the four frequency ranges and the 15 test generators, it was found that the RMS ratio was approximately 3.0. Figure 8 below shows induced periodic compressed sinusoidal voltages with a small absolute area over one period arising to a high RMS ratio of approximately 3.0 for Test 1.
Figure 9. Test 1 - Power Density versus Wire Diameter for Four Frequency Ranges
C. Test 2: Changing the Wire Diameter for Triple Ring Permanent Magnet 2 Similar to Test 1, the same generator dimensions were used this time to test the Neodymium triple Ring magnets, Magnet 2. The electrical characteristics are shown and explained below with the aid of experimental graphs. For each of the 15 wire thicknesses, the RMS open-circuit voltage was measured, and the voltage density was computed to show the variation from one wire diameter to another. Figure 10 illustrates the scatter graph data points with linear regression of each frequency range for the voltage density versus the wire diameter. For the medium-high frequency range, the average voltage density for the 15 test generators is about 8.00mV/turn. Also, the RMS ratio was calculated to be approximately 2.4 for Test 2 due to a larger covered area in the induced voltage versus time curve.
Figure 8. Sample 2 Hz Induced Voltage for 0.50mm Wire Diameter Generator
Similarly, the RMS short-circuit current for each of the 15 wire thicknesses was measured, and the current density presents the relative change in current production from each
Figure 10. Test 2 - Voltage Density versus Wire Diameter for Four Frequency Ranges
Similarly, the RMS short-circuit current was measured and the current density computed for each wire diameter. Figure 11 below shows the scatter graph data points with quadratic regression of each frequency range for the current density versus the wire diameter. The regression lines fit well to the curve, since the R2 value is above 0.97. It can be observed that the current density linearly increase up to about 0.55mm wire diameter before it starts to saturate for thicker wires. For the medium-high frequency range, the average saturated current density for the 15 test generators is about 8.00mA/turn.
wire impedance undesirably increases and reaches the solenoids geometric limitation. For the 0.28mm wire thickness, the voltage produced versus the wire turn level graph, where Level 1 has 65 turns, is in Figure 13. For Level 15 or 975 turns, the RMS voltage produced is less than 7.00V.
Figure 13. Test 3 Voltage Produced versus Wire Turn Level for Four Frequency Ranges (Each Level is 65 turns)
Figure 11. Test 2 - Current Density versus Wire Diameter for Four Frequency Ranges
The current produced versus the wire turn level with polynomial regression is graphed in Figure 14. For Level 15, the current produced is about 250mA.
Also, the apparent power density curves were obtained as shown in Figure 12. The peak power density for the mediumhigh frequency range is about 2.00mW/turn, which is more than double the power density for Magnet 1 in Test 1. Hence, Test 2 experimentally proved the greater strength of the Neodymium triple-connected Ring PM.
Figure 14. Test 3 Current Produced versus Wire Turn Level for Four Frequency Ranges (Each Level is 65 turns)
Figure 12. Test 2 - Power Density versus Wire Diameter for Four Frequency Ranges
D. Test 3: Changing the Number of Turn Levels for High Power Density Wires After experimental verification of the systematic selection of Neodymium magnets, Test 3 focused on three wire diameter thicknesses of 0.28mm, 0.35mm, and 0.40mm. Referring to the characteristic curves in Figure 12, the chosen wire diameters with maximum power densities were chosen toward the left side of the bell curves peak to increase the induced voltage to a required 5V. Let the maximum number of turns packed into the 2cm space be defined as Level 1. The next level numbers would be integer multiples of number of turns in Level 1, the same way of stacking paper sheets on top of each other. This helps analyze the relation between increasing the number of turns to an optimum peak before the
As in Figure 15, the power generated in mW shows saturation curves for the four frequency ranges. The mediumhigh frequency curve initially increases fast, but later settles down to a peak of about 1.60W for Level 15. Even though there were only six data points, the quadratic regression curves fit well to the scatter data points and the regression equation were accurately used to determine the required number of turns needed for optimum power with respect to the voltage and current curves.
Figure 15. Test 3 Power Generated versus Wire Turn Level for Four Frequency Ranges (Each Level is 65 turns)
Experiments performed on 0.35mm and 0.40mm had almost similar characteristics with differences in data points. The 0.28mm wire was also more preferable, for its solenoid is more capable of packing more number of turns within the same space for producing a higher emf. The main comparative results of Test 3 for medium-high frequency range and Level 4 are summarized in Table 3.
TABLE III. ELECTRICAL CHARACTERISTIC DATA POINTS FOR LEVEL 4, MEDIUM -H IGH FREQUENCY R ANGE OF THREE COIL D IAMETERS Coil Diameter 0.28mm 0.35mm 0.40mm RMS Voltage (V) 2.43 2.00 1.60 RMS Current (mA) 300.00 370.00 490.00 Power (mW) 729 739 782
Figure 18 shows the charging voltage waveform created by the linear generator shaking at about a 10Hz frequency. The capacitor discharges into the battery as periodic ramps with a lower designed discharge rate of around 3mV/ms, compared to the short discharge of 12mV/ms with smaller capacitance.
Figure 18. Oscilloscope Waveform of the Charging Voltage at a Single Battery Input Port with 9.4mF Smoothing Capacitance
B.
Power Generation Comparison
Figure 16 displays the designed LPM SG tube with a 2cm packed 0.28mm wire solenoid of about 975 turns. The triple Ring magnets and two sponges at either end are shown as well. Comparing the 140mW power produced in the mediumhigh frequency of Prototype 1 generator with Magnet 1 and 0.1mm solenoid wire, Prototype 2 with 0.28mm wire thickness is capable of generating more than 11 times more power. Through the conducted experiments, the desired 5V induced voltage was reached, with a high 250mA no-load current.
Both prototypes of the LPM SG have been compared in Table 4. Prototype 2 produces about 1.6W and its power mass density is about 4 times more than Prototype 1.
TABLE IV. Prototype 1 2 PROTOTYPE POWER GENERATION COMPARISON Mass (g) 56 150 Power Generated (mW) 140 1600 Power Mass Density (mW/g) 2.5 10.7
VI.
CONCLUSION
Figure 16. Designed LPM SG Prototype 2 with 0.28mm Diameter Solenoid with Triple Ring Magnets
The power generation of the LPM SG can be further improved by using stronger N52 grade Neodymium magnets with maximum magnetic flux linkage between the solenoid and PM. This can be accomplished via specifically-designed magnetization directions for the PMs and guiding the magnetic flux through a designed iron core such as a squirrel cage structure around the copper solenoid. More experimental data are needed for the three tests with a machine shaking at different frequencies as opposed to the human hand. Also, incorporating an iron core around the solenoid in the shape of a squirrel-cage, will allow the magnetic flux path to flow through low reluctance iron core compared to the high reluctance of air. This will help retain the opposing induced magnetic field and emf in the solenoid for higher power generation of the LPM SG. REFERENCES
[1] [2] [3] [4] F. Bressand et al., Wasted energy: How the US can reach its energy productivity potential, McKinsey Global Institute, pp. 6-7, June 2007. Magnet Sales. Magnets. United Kingdom. 2013 [Online]. Available: http://www.magnetsales.co.uk Shin-Etsu Chemical Co. Rare Earth Magnet Basics. 2007 [Online]. Available: http://www.shinetsu-rare-earth-magnet.jp/e/design/ Stephen P Beeby and Terence O'Donnell, "Electromagnetic Energy Harvesting," in Energy Harvesting Tecnologies, Shashank Priya and Daniel J Imman, Eds.: Springer.
V.
POWER ANALYSIS OF THE SYSTEM
A. Power Charging Circuit Based on the 0 to 20Hz frequency ranges explained in the previous experiments, the suitable shaking frequency of the human hand was found to be about 5Hz, where the mechanical shaking time constant is 200ms. The power charging circuit shown in Figure 17 consists of a bridge rectifier and a smoothing capacitor. The two 5V input ports of the 2000mAh batteries are connected in parallel to charge both at the same time. Since the generator resistance is about 23, the chosen capacitance value was chosen to be 9.4mF.
Figure 17. Power Charging Circuit of the Battery Supply
Elimination The Centrifugal Switch in Single-Phase Induction Motor(SPIM) to Increase Torque and Efficiency
Sobhan Sobhani M.Sc student Department of electrical Engineering Science and Research Branch Islamic Azad University of Tehran, Iran Sobhan.sobhani2011@yahoo.com
Mohammad Ebrahim Talebian Assistant professor Managing Director (MD) of Neka Power Plant, Mazandaran, Iran metalebian@gmail.com www.neka power plant.ir www.drtalebian.ir
Hoseyn Vafai Nejad M.Sc student Department of electrical Engineering Science and Research Branch Islamic Azad University of Tehran, Iran vafa_vaio11@yahoo.com
AbstractSingle-phase Induction Motors (SPIM) with Capacitor Start, are a sort of induction motor widely used in industry. Based on type of starting, single-phase induction motor (SPIM) is categorized as (1) Split phase, (2) Capacitor Start, (3) Shaded Pole, and (4) Repulsion Type. Today for disconnect the auxiliary winding and start capacitor of SPIM the centrifugal switch is used. Also when the motor has reached 75% of normal speed by some contacts switching is done and auxiliary winding and start capacitor is remove from the circuit. This is repeated for each startup. The centrifugal switch acts mechanically, so in over time it will be destroyed. Thus causing switch not working and the motor will be seriously affected. In this paper the centrifugal switch removed and smart devices using inductive proximity sensors (IPS) and magnetic field monitoring system (MFMS) is proposed. This system will be improved motor performance and efficiency. The simulation has been verified by Maxwell software. Result is shown at the end this document. It compared with previous approaches which show improvement. Keywords- magnetic field monitoring system (MFMS); Auxiliary winding; centrifugal switch; inductive proximity sensors (IPS);
I.
INTRODUCTION
The single-phase induction machine is the motor used in most household equipment as washing machines, refrigerators, fans. Although it has a reliable configuration, this device has two main problems: the starting current is 5 to 8 times higher than the steady state current and the centrifugal switch needs regular maintenance [1]. The centrifugal switch has some problems of the starting failure, the breaking under the worse condition. Many papers have proposed different control algorithms to improve the performance of a single-phase induction motor. Nowadays, users demand higher efficiency, reliability, miniaturization and noise immunity than conventional motor. The centrifugal switch has Poor performance in such as problems. The conventional control drives using variable frequency have not
Corresponding author Tel: +981243223850; mobile: +98 9113261732 E-mail addresses: Sobhan.sobhani2011@yahoo.com
been applied to SPIM because of starting problems, low speed operating characteristics, complexity, high cost, and controllability due to the speed estimation methods, electrical sensors used and the values of the instantaneous voltage and current Is an important variable to meet the estimated flux calculated Is. One of the methods to obtain flux sampling perspective is using Hall effects sensors that are installed in the air gap. Bug This method is that the output of the sensor varies with temperature and Compensation of these effects is difficult. Note that the flux function of the voltage and Is pending, can be used for good relations, the flux linkage The stator voltage and current can be calculated by the end of the batch sampling equations voltage, motor speed estimate. In this paper, starting torque, starting current and flux leakage are studied show that in the second set until the motor reaches it is normal speed Changes in waveform, torque, current, and flux leakage is seen. By analyzing waveforms and install a sensor, engine startup time can be obtained. In this method, the inductive proximity sensor (IPS) with magnetic field monitoring system (MFMS) has been used. Single-phase capacitor start induction motor at the run time, the sensor detects the Flux density and Leakage flux of winding .then intelligent system decides when the auxiliary winding and capacitor circuit must be disconnect. The advantages of this system are listed below: a) Better performance against mechanical centrifugal switch, so increase motor performance.
b) Engine noise at starting time, will be reduced. c) Efficiency increases because the loss will be less.
d) Centrifugal switch completely remove. e) Centrifugal switch requires maintenance the proposed system does not require maintenance. Motor reliability will increase.
f)
g) In high importance applications such as aircraft fuel
pump, this system would be more effective. h) In the high slide the system performance will be better than the centrifugal switch. i) In the variable-speed this smart system will have better performance. II. SYSTEM CONFIGURATION Sections of the proposed system include: a) b) c) The data processor unit
(a) . Front View of Sensor
Magnetic field sensor (MFMS) inductive proximity sensors (IPS)

Figure 2.magnetic field monitoring system (MFMS) [4].
Electromagnetic switch for disconnect auxiliary winding and start capacitor. Block diagram of the proposed system is shown in Fig. 1.
inductive proximity sensors (IPS)
MOROR
data processor unit
switch
Inductive sensor includes an electrical resonant circuit and energy source coupled therewith to compensate for losses in the resonant circuit. It installed on the stator. When the motor started, the stator winding leakage flux will interact with the sensor field, device is understood that the motor started. Proximity sensor effect the detection of the presence or absence of MMF in motor. Also sensor sent the signal to data processor unit. The sensitivity of the sensor is adjustable therefore it can be due to various applications and different situations. Proximity sensor circuit is shown in Fig.3.
Magnetic field sensor (MFS)
Figure 1.Block diagram of the proposed system
According to Fig.1, the sensor (MFS) scans the magnetic field of motor and then sends it to the processing unit for analysis. One way to classify the various magnetic sensors is by the field sensing range. These sensors can be arbitrarily divided into three categories low field, medium field, and high field sensing. Magnetic Sensor Technology Field Ranges is shown in Fig.2 (a), and front view of Sensor is shown in Fig.2 (b) [4].
Figure 3.Circuit of inductive sensor
When both the sensor confirmed that the magnetic flux is created then Auxiliary winding and capacitor can be removed from the circuit. This command is issued from the processor unit. In this situation, the torque-speed curve derived from a steady-state equivalent circuit is very close to the average torque-speed curve obtained from a transient state. The figure below shows a centrifugal switch. Its well known that the mechanical switch provides high limits, also could not be adjusted and calibrated.
Figure 4.centrifugal switch [6] (a) . Magnetic Sensor Technology Field Ranges
The sections (1) and (2) the probability of failure and Section (3), (5) shall be made for periodic maintenance. III. MATHEMATICAL MODEL
Equivalent circuit of single-phase induction motor is shown in fig.5.
In this section, we introduce and evaluate the objective function and for maximum efficiency and performance, the equations will be discussed. The voltage equations are as follows:
(1)
Figure 5.Equivalent circuit of induction motor
The main winding is geometrically orthogonal with Auxiliary winding. { The current equations are as follows:
(2)
Transforming the auxiliary winding and the negativesequence compensatory windings to have the same turns of the main winding, the motor would be seen as a twophase symmetrical motor and the complex currents can be described by such functions:
Figure 6.Geometrical diagrams and circuit model
The voltage induced in the main and auxiliary winding by Rotating flux is: (8) (9)
(3)
Where, (10)
Where Wm, kwm, wa, kwa represent the number of windings and the winding factor for the main, respectively auxiliary winding and is the rotating flux. The ratio of the main and auxiliary winding voltages is: By combining the above circuits, the proposed system will achieve. By reviews flux linkage waveform of single phase induction motor, specified that the first peak amplitude is higher and Varies uniformly in subsequent periods. This is shown clearly in Figure 7.
(4)
(5),(6)
Are the winding trans ratio, Wa, Wu,Wz are the turns of the windings, Kdpa,Kdpu,Kdpz are the winding coefficients. Efficiency Equation of single phase induction motor is shown below:
Figure 7.Flux linkage waveform
start
If this primary peak began to consider starting the motor, after 5 seconds it will reach its nominal speed (75% of rated speed). The simulation results demonstrate this hypothesis. Two sensors have specific roles in the project. Inductive sensor (IPS) senses motor leakage flux, which is located around the stator. In the next step sensor (MFS) to calculate the exact value of the stator magnetic field and show that the motor speed is enough And the auxiliary winding and the capacitor is not required. In Figure.8 the magnetic field of the motor, at the starting time is shown.
Switch (IES) Closed
Detection sensor IPS

YES
NO
Detection sensor MFMS

YES
NO
NO
The magnetic field is equal to the amount of regulation?

YES Switch (IES) is open after 5 seconds
END
Figure 10.Flow chart of proposed system
Figure 8.Stator magnetic field at the starting time
It shows that, at the time of motor Starting the motor to reach the nominal speed from the standstill, a strong magnetic field is created. This is the moment that after 5 seconds the motor reaches its nominal torque. Eventually, it will be demonstrated in the simulations. Equation for the magnetic field over time is as follows:
(11)
If the motor is turned on and the sensor (IPS) is realized, Then the system waits until the magnetic field of the motor reaches the specified level. After this, the system will detect that the motor is operated so by the switch (IES), the auxiliary winding and capacitor circuit will reject after 5 seconds. If the sensor (IPS) is not turn on, so the motor is switched off, therefore the control system will end. V. P ROTECTION UNIT If the rotor of motor locked, so current is extremely high. In this case, motor power supply must be disconnected quickly. Its called over current protection. By a current transformer (CT) can be measured current consumption of motor. According to this issue a module is added to flow chart, this shown in Fig.11.
start
Switch (IES) Closed
IV. SYSTEM PERFORMANCE First the circuit diagram of the proposed system will be examined. Then the proposed system flow chart is shown. The equivalent circuit of the proposed system is shown in Figure9.
The Current consumption is normal?

YES
NO
MOTOR SWITCH OFF
Intelligent electrical Switch (IES)
Detection sensor IPS

YES
NO
Detection sensor MFMS

YES
NO
The magnetic field is equal to the amount of regulation?

NO YES
Figure 9.Equivalent circuit of the proposed system
Switch (IES) is open after 5 seconds
Switch (IES) is controlled by intelligent device. The flow chart of proposed system is shown in Fig.10.
END
Figure 11.system flow chart whit protection module
VI.
S IMULATION As shown in the above figures, at the moment when the centrifugal switch acting have a sudden ripple in the waveforms. Obviously this is because of the following reasons: a) b) c) The switch is not the right time operating. The auxiliary capacitor selection is not correct. During the performance of the centrifugal switches, Motor speed was not enough.
In this is study the typical single-phase induction motor is used for simulation. First, using MAXWELL software and while there was a centrifugal switch, waveforms is specified. Then, by replacing the proposed system motor characteristics were analyzed. According to the elimination of centrifugal switch, results will show that motor efficiency is increase and system performance will be improved. Characteristics of motor before using proposed system is shown in Table 1.The motor has been named SOB-1.
TABLE I. CHARACTERISTIES OF MOTOR SOB-1
1 2 3 4 5 6 7 8
Rated Output Power (kW)

Mechanical Shaft Torque (N.m)
Efficiency (%) Power Factor Rated Slip Frequency (Hz) Centrifugal switch model capacitor value
2.27 3.74416 55.3943 0.802703 0.166667 60 lg17-154 35 F
All these problems are because the centrifugal switch works mechanically and for different operating conditions it not adjustable. In the proposed intelligent system only when sufficient flux and magnetic field in the stator is suitable, so the switch will be disconnected. This issue and choosing the right capacitor will resolve this problem (ripple in the waveforms). The following formula can used to select the suitable capacitor for starting motor [3]-[8]. Using the relation between output power, efficiency and input power, the capacitor is:
Curves of efficiency, torque and power factor of motor SOB-1, before using intelligent systems is shown below.
(12) f = 60 Hz, Vm=227V, efficiency of =77%, and an estimated KE=0.169, =2f=260=376.99, P out=2.27, the capacitor value is: C=12.42 F By using this amount of capacitor and using the proposed intelligent system, simulation will be repeated. The curves of Efficiency, torque and power factor of motor SOB-1, after using intelligent systems is shown below.
Figure 12.Efficiency curve of motor SOB-1
Figure 15.Efficiency curve of motor SOB-1
Figure 13.Power factor curve of motor SOB-1
Figure 16.Power factor curve of motor SOB-1 Figure 14.Torque curve of motor SOB-1
F. Blaabjerg, F. Lungeanu, K. Skaug, and A. Aupke, Comparison of variable speed drives for single-phase induction motors , in Proc. Power Convers. Conf., Apr. 2--5, 2002, vol. 3, pp. 13281333. [5] E. R. Collins, Jr. , Torque and slip behavior of single-phase induction motors driven from variable frequency supplies, in Conf. Rec. IEEE IAS, 1990, pp. 6166. [6] website, http://www.torq.com/pdffiles/centrifugalswitches.pdf. [7] A. S. Ba-thunya, R. Khopkar, W. Kexin, and H. A. Toliyat, Single phase induction motor drivesA literature survey , in Proc. IEMDC Conf., 2001, pp. 911 916. [8] K. Iizuka, Refrigerator control system , U.S. Patent 4 566 289, Jun28, 1986. [9] M. Chomat and T. A. Lipo, Adjustable-speed single-phase IM drive with reduced number of switches, IEEE Trans. Ind. Appl., vol. 39, no. 3, pp. 819 825, May/Jun. 2003. [10] S. Vaez-Zadeh and B. Zahedi, A steady state model including iron loss for variable speed single phase induction motors, in Proc. IEEE Power Eelecron. Spec. Conf., Orlando, FL, Jun. 2007, pp. 606611. [11] F. Blaabjerg, F. Lungeanu, K. Skaug, and M. Tonnes, Two phase induction motor drives, IEEE Mag. Ind. Appl. , vol. 10, no. 4, pp. 24 32, Jul./Aug. 2004. [4]
Figure 17.Torque curve of motor SOB-1
Characteristics of motor after using proposed system is shown in Table II.

Table II. CHARACTERISTIES OF MOTOR SOB-1
1 2 3 4 5 6 7 8
Rated Output Power (kW)

Mechanical Shaft Torque (N.m)
Efficiency (%) Power Factor Rated Slip Frequency (Hz) Centrifugal switch model capacitor value
2.27 6.71251 68.8539 0.780539 0.164667 60 12.42 F
As can be seen the efficiency increase, Also the torque of motor is improved. Actually motor performance by using the proposed system is improved. Motor power factor is slightly reduced but this is inevitable. Also the results show that by use this method will increase motor longevity. This is because that the main winding will lower current in starting time and auxiliary winding disconnected on best time.
VII. CONCLUSIONS Based on the simulation results, centrifugal switch is required to a fundamental revision. Today intelligent systems are used in the optimization problems. Determine the optimal time for switching the auxiliary winding of Motor is important in efficiency. While the proposed system appears to be expensive but it can use in such single-phase motors that are turned on and off repeatedly. Also this system can use in the various places and different weathers. The proposed system is relatively lower output ripple. REFERENCES
[1] [2] A. L. Day, Energy saver control for single phase motors, U.S. Patent4 455 521, Jun. 19 , 1984. K. Sundareswaran , An improved energy-saving scheme for capacitorrun induction motor, IEEE Trans. Ind. Electron., vol. 48, no. 1, pp. 238240, Feb. 2001. C. Mademlis, I. Kioskeridis, and T. Theodoulidis, Optimization ofsingle-phase induction motorsPart I: Maximum energy efficiency control, IEEE Trans. Energy Convers., vol. 20, no. 1, pp. 187195 , Mar. 2005.
[3]
Dynamic Fault Analysis for Type-3 and Type-4 Generic Wind Turbine Generator Models using Matlab/Simulink
Tareq Saber Abuaisha , Oliver Scheufeld , Julian Langst adtler , Albert Moser and Jutta Hanson
TU Darmstadt, Germany Email : {tareq.abuaisha, jutta.hanson}@e5.tu-darmstadt.de FGH e.V, Aachen, Germany Email : {Oliver.Scheufeld, Julian.Langstaedtler}@fgh-ma.de IAEW, RWTH-Aachen, Germany, Email : am@iaew.rwth-aachen.de
E5,
R esum eIntegration of renewable energies in general and wind energy in particular is increasing all over the globe. The awareness towards generic models or sometimes also called standard models of wind turbine generators (WTGs) is signicantly increasing. Due to their exible performance and better controllability, doubly-fed induction generator (DFIG) and fully-rated converter (FRC) based WTGs are ultimately dominating the global market. Throughout this paper, dynamic modelling and performance analysis of the generic models of these particular WTGs are achieved using Matlab/Simulink. A comparison through simulation results with manufacturerspecific models is carried out. As regards, conclusions are duly addressed. Index Termswind energy integration ; wind turbine generator ; generic WTG model ; doubly fed induction generator ; fully rated converter ; manufacture-specic model
I. I NTRODUCTION By the end of 2011, the amount of installed wind power capacity in Europe has reached approximately 97 GW. Compared to 2009, this number has increased drastically by around 27% [1]. For instance, having a shared amount of 40% among all other renewable energies, wind energy is the dominant driver of the whole renewable power generation in Germany [2]. Based on these statistics, wind power is not any more a minor player in this game, but it has to be considered as a vital and important power generation source, especially in countries with a signicant share of wind power in its generation portfolio. Moreover, this increasing integration of renewable energy has changed the system from a centralized set-up with the major power plants connected to the extrahigh voltage level to a highly decentralized one with many smaller distributed generation units connected to almost all the different voltage levels [3], [4]. For transmission and/or distribution system operators to be able to play their role in the grid, they need to carry out proper simulations and emulations on the concerned power system. These simulations do vary from the very complex and detailed types (e.g. electromagnetic transient) to the very basic and preliminary types (e.g. load ow).
In order to carry out these tasks, system operators need to model the physical behaviour of the equipment (e.g. WTGs) composing the grid. As a result, efcient models of wind turbines and wind farms which are capable of describing the physical behaviour in a proper manner is a substantial requirement. Generally manufacturer-specific models are available. However, these models usually contain much proprietary information which is being represented by a black-box. Access to these models requires a non-disclosure agreement between the dynamic model user and the turbine manufacturers [5]. In response to this challenge, the Western Electricity Coordinating Council (WECC) by its wind generator modelling group (WGMG) has initiated the development of generic wind turbine generator models for all the different types of WTGs [6], [7]. As a continuation of the WECC effort, IEEE has also established a working group to investigate and consolidate WTGs modelling issues [6]. The main purpose of generic models is to have worldwide manufacturer-independent WTG models that can be distributed, implemented and improved without any restrictions due to intellectual proprietary [5]. The remainder of this paper is arranged as follows. In Section II, an overview over fixed-speed and variable-speed WTG technologies is provided. Modelling and simulation of type-3 and type-4 generic models are described in Section III. Exemplary results for the modelling are presented in Section IV. Conclusions and outlook are provided in Section V. II. F IXED - SPEED AND VARIABLE - SPEED WTG T ECHNOLOGIES Depending on the type of power electronic converter used in the wind turbine technology on one side and on the desired speed control range on the other side, WTGs can be operated at either fixed speed or variable speed. A. Fixed-speed WTGs Fixed-speed WTGs consist of an aerodynamic rotor driving an induction (squirrel cage or wound rotor) generator or
Transformer
Transformer
IG
IG
R
(a) Speed may vary by 2-4% from no load to full load (b) Speed may vary by 2-10% from no load to full load F IGURE 1. Fixed-speed wind turbine generator technologies.
Optional
TABLE I A DVANTAGES AND D ISADVANTAGES OF VARIABLE - SPEED WTG S

Transformer
DC AC DC
Advantages Improved power quality with less icker Power factor controllability with reduced acoustic noise Improved system efficiency and Reduced fatigue Increased extracted power and reduced mechanical stress
Disadvantages Losses in power electronic converters Increased cost Voltage uctuations and harmonic penetration [9]
PMSG
AC
if WRSG
(a) FRC-based WT with a permanent magnet synchronous generator (PMSG)
Ps
WRIG
Transformer
Pr
DC AC DC AC
B. Variable-speed WTGs Variable-speed WTGs utilize power electronics to provide high controllability and efciency. They are typically equipped with forced commutated pulse width modulated (PWM) inverters and rectiers, or completely with PWM voltage source converters (VSCs) in new designs to provide a fixed voltage and fixed frequency and apply pitch control of the turbine blades [10], [11]. Table I highlights the advantages and disadvantages of variable-speed in comparison with fixedspeed WTGs. Depending on the connection of their power electronic converters, variable-speed machines can be categorized into singly-fed or doubly-fed WTGs. The singly-fed variable speed approach consists of a converter connected in series between the generator and the grid that allows a unidirectional power ow (in case a simple rectier bridge is employed in the rotorside). This converter must withstand the full power rating of the generator, and thus this conguration is known as fullyrated converter (FRC) based WT. The typical structure of this WTG is described in fig. 2a. The cost of the fully-rated converter can be decreased by using a partially rated converter. In this structure, the converted electric power is transmitted to the grid by both the stator (directly) and the rotor windings indirectly via power electronic converter. The conguration of double sided power transfer to the grid, where power conversion occurs in an induction generator driven by a wind turbine, is known as doubly fed induction generator (DFIG) based WT. The typical structure of this WTG is described in fig. 2b.
(b) DFIG-based WT with a wound rotor induction generator (WRIG) F IGURE 2. Variable-speed wind turbine generator technologies.
possibly a synchronous generator, which is directly connected through a gearbox and a shaft. They do not employ power electronics devices and are typically composed of small to medium size wind turbines. Fig. 1 shows different possible arrangements for fixed-speed wind turbines with induction generator. In the arrangement shown in fig. 1a typically a squirrel-cage induction generator is employed. Here the speed may vary by 2-4% from no load to full load. While in the arrangement shown in fig. 1b, typically a wound-rotor induction generator is employed with rotor resistances which allow speed to vary by 2-10% from no load to full load at the expense of increased losses. Capacitors are used for power factor correction [8]. In this technology the generator operates at nearly fixed speed (driven by the grid frequency). Since the speed variation from no load to full load is relatively small, i.e. almost fixed, this technology is also referred to as xed speed WTGs. Note that in some literatures the technology of wound-rotor induction generator is termed as limited variable-speed WTG, but the approach followed in this paper is to consider all WTGs which are driven by grid frequency do not employ power electronics as fixed-speed WTGs.
Regualted Bus Voltage (Vrf q ) Converter Control Subsystem ConvCtrl Power Order (Pord ) Speed Order (ref )
Terminal Voltage (Vterm ) Ipcmd Eqcmd Pgen & Qgen Generator/ Converter Subsystem GenCtrl Active Power (Pgen ) Transformer Pgen Qgen Shaft Speed (pu ) To grid
Regualted Bus Voltage (Vrf q ) Converter Control Subsystem ConvCtrl Power Order (Pord )
Terminal Voltage (Vterm ) Ipcmd Iqcmd Pgen & Qgen Generator/ Converter Subsystem GenCtrl Active Power (Pgen ) Transformer Pgen Qgen To grid
Pitch Control Subsystem PitchCtrl
Pitch Angle ( )
Wind Turbine Subsystem TurbineCtrl
Wind Turbine Subsystem TurbineCtrl
(a) Dynamic model structure of a DFIG-based WT including the corresponding subsystems
(b) Dynamic model structure of a FRC-based WT including the corresponding subsystems
F IGURE 3. Dynamic models of variable-speed generic WTG technologies [7].
As described in fig. 2a, if the employed generator is wound rotor synchronous generator (WRSG), then the DC excitation is taken from the DC link (dashed line). III. M ODELLING AND S IMULATION OF VARIABLE - SPEED WTG S As stated in table I, variable speed WTGs offer many advantages over the fixed speed technology. Moreover, with the fast developments in power electronics these advantages will be more and more pronounced in the next few years. As a result, variable-speed technology has become the state-of-theart and the natural choice for wind power penetration, especially in multi-megawatt wind turbines. For these advantages, the generic models of variable speed technology has been adopted for further investigation, modelling and development during the rest of this paper. Comparing both variable-speed technologies with each other, on one side the fully rated converter in FRC-based WT represents an increase in cost and losses. But on the other side, it enables smoother connection to the grid in which the generator transients are decoupled with more speed variation and thus higher power control capability [8]. As described in fig. 3, the generic dynamic models of type3 and type-4 WTGs are represented by a specic number of interactions among number of subsystems to achieve the functionality of a typical WTG. This functionality should include the independent control of active (torque) and reactive power and the control of generator speed and blade pitch angles, in order to achieve maximum wind power extraction [12]. A. Modelling of Type-3 (DFIG-based WT) Fig. 3a shows the dynamic model structure of a DFIG-based wind turbine including the individual subsystems together with the commands in between, as firstly described by WECC [7]. The role of every subsystem is explained as follows [13] :
Generator/converter subsystem : is represented by a regulated current source which injects proportional threephase currents into the power system based on the control commands (Ipcmd & Eqcmd ) from the converter control subsystem. Thereafter, it gives both the measured active and reactive power (Pgen & Qgen ) to the converter control subsystem as feedback and also the measured active power Pgen to the wind turbine sub-system. Converter control subsystem : it consists of active and reactive power control models. These models emulate power electronic controllers in regulating active and reactive power. Based on the measured active and reactive power and the received shaft speed pu from wind turbine subsystem, the converter control subsystem provides both ; the power order Pord and the speed order ref to the pitch control subsystem. Wind turbine subsystem : it is represented by a singlemass model (together with the other rotating masses e.g gearbox and generator shaft) and used to determine the mechanical input power Pmech and the angular speed pu of the wind turbine based on the per-unit wind speed wind,pu and the specified pitch angle . Pitch control subsystem : its primary function is to determine the desired blade pitch angle based on the desired angular speed pu and the measured active power Pgen requirements. The desired instantaneous active and reactive power (Pgen & Qgen ) are calculated using equations 1 and 2 as described in [8], [12] and then fed into the corresponding subsystems. 3 P = . (Vd Id + Vq Iq ) [MW] (1) 2 3 . (Vq Id Vd Iq ) [Mvar] 2 B. Modelling of Type-4 (FRC-based WT) Q= (2)
Fig. 3b shows the dynamic model structure of a FRC-based wind turbine including its subsystems and signal commands,
LM
25 km line
25kV/690V WTG RC Filter
1.5 1 Voltage [pu] 0.5 0
G
120 kV 50 Hz X 0/X 1 = 3 2500 MVA 120kV/25kV 47 MVA Y
Va Vb Vc
690V 50Hz 1.5MW
-0.5 -1 -1.5 0 0.1 0.2 0.3 0.4 time [sec] 0.5 0.6 0.7 0.8
F IGURE 4. Set-up system used in Matlab/Simulink to simulate type-3 and type-4 generic models.
Instantaneous angle [rad]
as firstly described by WECC [7]. Functions of the subsystems are almost identical to those of a DFIG-based wind turbine [14]. However, the wind turbine subsystem of type-4 WTG employs a simplied mechanical model. This is mainly due to the fact, that in type-4 generic model the fully-rated converter completely decouples both the generator transients and grid faults from affecting each other and thus smoother connection to the grid is enabled [12]. The above-described subsystems are implemented and simulated in Matlab/Simulink based on the parameters provided by WECC for the first generation of generic models [7], [15]. Afterwards the simulation results are compared with manufacturer-specific models in order to investigate their physical behaviour and characteristics. Results of simulation are to be discussed in the following section. IV. E XEMPLARY R ESULTS Based on the performed modelling in section III, along with the provided WECC parameters, the generic model for type-3 and type-4 has been implemented and simulated. Fig. 4 shows the set-up system used in simulating type-3 and type-4 generic models. It is similar to that used with the built-in Matlab/Simulink wind farm [16]. Fig. 5a shows the three-phase voltage measured at WTG terminal, when the input voltage is subjected to 50% voltage dip. As expected, the three-phase voltage has dropped to around 50% of its rated value due to the input voltage dip.
(a) Three-phase voltage

1.5 1 Current [pu] 0.5 0 Iaref Ibref Icref
-0.5 -1 -1.5 0 0.1 0.2 0.3 0.4 time [sec] 0.5 0.6 0.7 0.8
(b) Three-phase output current

4 2 0 -2 -4 0 0.2 0.4 time [sec] 0.6 0.8
(c) Instantaneous angular position

1.5 Active & reactive power [pu] 1 0.5 0 -0.5 -1 0 0.1 0.2 0.3 0.4 time [sec] 0.5 0.6 0.7 0.8 Pgen Qgen
A. Simulation Results of DFIG-based WT DFIG-based WT has been taken as example, whilst same principle can be applied on FRC-based WT. Fig. 5b shows the three-phase injected current into the power system from the dependent current sources. The irregularities which occur at the points 0.3 sec and 0.5 are due to voltage dropping and recovering respectively, which will cause phase shifting due to reactive current increase. In order to perform the required alignment between the synchronous rotating frame and the stator magnetic ux, the instantaneous angular position of the stator magnetic ux has to be estimated. The estimated angle is shown in fig. 5c. The instantaneous ux position varies linearly from to . The measured instantaneous active and reactive power (Pgen & Qgen ) will be calculated using equations 1 and 2 respectively. Later they will be fed into the corresponding sub-systems of type-3 generic model. Fig. 5d shows the instantaneous measured active and reactive power at the WTG terminal.
(d) Instantaneous measured active and reactive power F IGURE 5. Simulation results of type-3 generic model during a 50% voltage dip measured at WTG terminal.
B. Comparing the Simulation Results of the Generic Models with Manufacturer-specific Models The comparison is done by calculating the absolute difference between the simulation results of the generic model and that of a manufacturer-specific model as shown in equation 3. (3)
P (t) = PGeneric Model (t) PManufacturer Model (t) , Q(t) = QGeneric Model (t) QManufacturer Model (t) , where 0 t 7 sec and t = 103 sec
Active power [pu]
1 0.5 0 0 1 0.5 0 0 0.5 0 1 2 3 4 5 6
Pgen
Active power [pu]
1 0.5 0 0 1 0.5 0 0 1 0 -1 1 0 -1 1 2 3 4 5 6
Pgen
6 P(t)
7 Comparison result
6 P(t)
Comparison result
7 Qgen
7 Qgen
Reactive power [pu]
-0.5 0 0.5 0 -0.5 0 1 2 3 time [sec] 4 5 6 7 1 2 3 4 5 6 Q(t) 7
Reactive power [pu]
6 Q(t)
Comparison result
Comparison result
3 4 time [sec]
F IGURE 6. Comparing the active and reactive power of type-3 generic model with a manufacturer-specific model during a 25% voltage dip.
F IGURE 8. Comparing the active and reactive power of type-4 generic model with a manufacturer-specific model during a 25% voltage dip.
S1 Zg
G
25 kV 50 Hz = 74.89 X/R = 5 Z1 S2 25kV/690V 1.5 MVA
WTG Vnom = 690V fnom = 50Hz Pnom = 1.5MW
Z2
the reference per-unit reactive power Qgen is zero. This is valid for both Matlab/Simlulink type-3 and type-4 generic models and also for the manufacturer-specific models simulated in DigSILENT PowerFactory. The simulation time is around 7 sec. This presents an appropriate time-frame for short-term voltage stability studies (typically from 5 30 sec) [18]. Additionally, the step-size is 103 sec in both simulation platforms. Fig. 6 and fig. 8 show the simulation results of the active and reactive power of type-3 and type-4 generic models compared to a manufacturer specic model. The results are matched in the pre-fault region whereas the mismatch in the fault and post-fault regions is due to the strategies employed in the manufacturer-specific models to force a specic behaviour for active power recovering. This is not the case for the generic models. It is also noticeable that the reactive power Qgen would increase during the fault in order to support the voltage. The post-fault region (3.25 - 7 sec) of active power difference curve P (t) implies that manufacturer-specific model requires larger post-fault time to recover active power. Comparing fig. 6 with fig. 8, the oscillations in the reactive power curve Qgen are a typical characteristic of type-3 generic WTG. They perish in case of FRC-based WT because the full size converter decouples the generator transients to generate a smooth response. The small peak value of the difference curve of reactive power would result from a closer match to the manufacturer-specific model. In future research protection apparatus can be employed along with the generic models in order to imitate the behaviour of the manufacturer-specific model by forcing a comparable behaviour on active and reactive power after fault clearance.
F IGURE 7. Set-up system for the comparison of type-3 and type-4 generic models with manufacturer-specific models.
In order to allow a comparison between type-3 and type-4 generic models on one side and manufacturer-specific models on the other side, the following precautions are to be considered [17] : The set-up system shown in fig. 7 is used for the simulation and comparison. The ratio between Z1 and Z2 will generate the desired voltage dip. When switch S2 is closed, the voltage on WTG terminal will be calculated using the voltage divider principle. The vendor-specific models are simulated in DigSILENT PowerFactory, whereas in Matlab/Simulink a programmable voltage source was employed to generate a comparable effect. Moreover, because the induction generator of type-3 generic model is emulated by dependent current sources, another speciality of Matlab/Simulink is that it cannot connect two blocks in series when the first block is modelled as a current source. To overcome this problem an RC-filter was connected as shown in fig. 4. The maximum per-unit active power Pgen is 1 p.u. and
V. C ONCLUSION DFIG and FRC based WTs offer many advantages over the fixed speed technology including exible performance and better controllability. In this paper, the first generation of type-3 (DFIG) and type-4 (FRC) generic models has been implemented and simulated in Matlab/Simulink. The simulation results were compared with manufacturerspecific models in order to investigate the physical behaviour and characteristics. Afterwards, the implemented models were added to the Matlab/Simulink library to serve as basis for future research and development. In order to analyse the dynamic behaviour of the generic models, a three-phase fault in which the voltage drops to 25% of its initial value was studied as an example. The obtained results show that the generic models can be employed when an estimated behaviour is sufcient. Their intended use is for load ow simulations and dynamic fault analysis, where the highest level of accuracy is redundant. Additionally due to their public availability, a good feature of the generic models is that they can be developed and improved to describe the concerned systems in the best possible manner. ACKNOWLEDGMENT The authors would like to thank the anonymous wind turbine manufacturers for providing the manufacturer-specic models which served as benchmarks for the validation of type-3 and type-4 generic models as well as Professor Istv an Erlich (Institute of Electrical Power Systems, University DuisburgEssen) for the allocation of the aggregated generic model of the variable-speed machine. R EFERENCES
[1] EEG, German renewable energy sources act, [Online], November 2012. [2] BWE, German wind energy association, [Online], December 2011. [3] C.-H. Lo and N. Ansari, Decentralized controls and communications for autonomous distribution networks in smart grid, IEEE Transactions on Smart Grid, vol. 4, no. 1, pp. 6677, March 2013. [4] H. Weiss and K. Ince, Decentralized supply system with renewable energy based on quasi-eight-level inverter, in IEEE Region 8 International Conference SIBIRCON, Irkutsk Listvyanka, Russia, July 2010, pp. 524529. [5] NERC, Standard models for variable generation, NERC, 116-390 Village Blvd., Princeton, NJ 08540-5721, USA, Tech. Rep., August 2010, North American Electric Reliability Corporation. [6] M. Asmine, J. Brochu, J. Fortmann, R. Gagnon, Y. Kazachkov, C.E. Langlois, C. Larose, E. Muljadi, J. MacDowell, P. Pourbeik, S. A. Seman, and K. Wiens, Model validation forwind turbine generator models, IEEE Transactions on Power Systems, vol. 26, no. 3, pp. 1769 1782, August 2011. [7] A. Ellis, Y. Kazachkov, E. Muljadi, P. Pourbeik, and J. Sanchez-Gasca, Description and technical specications for generic wtg models a status report, in IEEE/PES Power Systems Conference and Exposition (PSCE), May 2011, pp. 524529, WECC Working Group on Dynamic Performance of Wind Power Generation & IEEE Working Group on Dynamic Performance of Wind Power Generation. [8] J. Machowski, J. W. Bialek, and J. R. Bumby, Power System Dynamics: Stability & Control, 2nd ed. UK, Poland: John Wiley & Sons Ltd, 2008, pp. 265297. [9] F. D. Kanellos and N. D. Hatziargyriou, The effect of variablespeed wind turbines on the operation of weak distribution networks, IEEE Transactions on Energy Conversion, vol. 17, no. 4, pp. 543548, December 2002.
[10] A. Dadhania, Modeling of doubly fed induction generators for distribution system power ow analysis, Masters thesis, Ryerson University, Toronto, Ontario, Canada, 2010. [11] I. S. Wander, Modeling of synchronous generator and full-scale converter for distribution system load ow analysis, Masters thesis, Ryerson University, Toronto, Ontario, Canada, 2011. [12] F. Blaabjerg, F. Iov, Z. Chen, and K. Ma, Power electronics and controls for wind turbine systems, in IEEE International Energy Conference, Department of Energy Technology, Aalborg University, October 2010. [13] K. J. Faria, Doubly-fed induction generator based wind power plant models, Masters thesis, The University of Texas at Austin, Austin, Texas, USA, 2009. [14] J. Fortmann, S. Engelhardt, J. Kretschmann, C. Felters, M. Janssen, T. Neumann, and I. Erlich, Generic simulation model for dg and full size converter based wind turbines, in 9th International Workshop on Large-Scale Integration of Wind Power, Qubec/Canada, November 2010. [15] WECC, Wecc wind power plant dynamic modeling guide, Approved by: M&VWG/TSS, November 2010, guideline. [16] Hydro-Qubec and I. The MathWorks, Simpowersystems, August 2012, matlab Simulink demo of Wind Farm- DFIG Average Model, power wind dg avg.mdl, Richard Gagnon, Hydro-Quebec. [17] I. Erlich, F. Shewarega, and O. Scheufeld, Modeling wind turbines in the simulation of power system dynamics, in Scientic Journal of Riga Technical University, Power and Electrical Engineering, vol. 27, August 2010, pp. 4147. [18] A. Perdana, Dynamic models of wind turbines, Ph.D. dissertation, Chalmers University of Technology, Goeteborg, Sweden, 2008, a Contribution towards the Establishment of Standardized Models of Wind Turbines for Power System Stability Studies.
Adaptive Control with Tuning Function Control Design Applied to Class-E/F Inverter
S. Zerouali A. Hadri Hamida S.M. Mimoune A. Allag A. Bensalem
Dep. of LMD. S.T dep. of Electrical engineering University of Batna University of Constantine Batna, Algeria Constantine, Algeria e-mail: sakina_z@yahoo.fr am_hadri@yahoo.fr dep. of Elec. Eng. dep. of Elect. Eng. dep. of Elec. Eng. MSE Laboratory University of Biskra University of Batna Biskra, Algeria Biskra, Algeria Batna, Algeria s_m_mimoune@yahoo.fr
AbstractAt the radio frequency the induction plasma generator is very attractive for several industrial applications especially in material processing, so an inductively coupled plasma generator is developed in this paper. A high-frequency power supply classE/F inverter is required to generate the magnetic field at frequency ranging from hundreds of kilohertz to tens of megahertz offer high efficiencies at high power densities. A resonant inverter based on zero-voltage switching and an adaptive control via a backstepping based on Lyapunov theory is used as a power supply. This paper presents the circuit modelling, simulation and control considerations. Keywords Plasma Generators, Buckstepping control, Tuning functions. class-E/F Inverter,
I. INTRODUCTION With the development of power semiconductor devices, many new circuit techniques and control schemes, research about high frequency circuits using advanced power devices such as MOSFETs, IGBTs and so we have been performed for high power applications, it has made it possible to implement high frequency inverters for induction heating, dielectric heating, and plasma generation. These applications generally require power levels from watts to megawatts at a single frequency ranging from hundreds of kilohertz to tens of megahertz [1]. Soft switching techniques have been used in power converters to reduce switching losses and alleviate electromagnetic interference (EMI); it has been studied based on its topology [2]. The various resonant inverters which are class-D, class-E, class-F and class E/F etc. inverter using power devices such as MOSFETs and IGBTs offer reduced switching loss by effective means of soft-switching technique [4]. The class-E inverter is an example of a resonant inverter which allows obtaining, almost the sinusoidal currentvoltage at the frequency above (several of KHz to several of MHz) [3] and is a well-known resonant converter that can operate at these frequencies with very high efficiency and produce up to several kilowatts of power [1][4]. It is a single-ended or pushpull topology where a transistor is soft switched, and therefore, its switching losses are significantly reduced.
A. Comparison and Motivation for class-E/F inverter Comparing class E to the two class-F tunings, several advantages and disadvantages are apparent. Class E has the advantage of being capable of strong switching operation even with a very simple circuit, whereas class F allows this only as a limiting case using a circuit with great complexity [4]. Whereas the class-E amplifier is limited only by the intrinsic switching speed of the active device, class-F amplifier tunings may find their switching speed dominated by the limited number of harmonics, which have been utilized in the waveforms. Additionally, class E has the advantage of incorporating the output capacitance of the transistor into the circuit topology. The simple class-F implementation will not work in the presence of large output capacitance since the harmonics that were intended to be open circuited at the transistor will instead be capacitive. Classes F and 1/F also have advantages. First, they present more desirable waveforms. It is desirable to have waveforms with low peak voltage and rms current, (it is clear that class-F and F1amplifiers can perform better in these respects) [4]. B. Adaptive control This paper presents an approach to adaptive control of class E/F resonant inverter via a backstepping tuning function control design. This design removes several obstacles from adaptive linear control. Since the design is based on a single Lyapunove function incorporating both the state of the error system and the update law, the proof of global uniform stability is direct and simple. Moreover, all the error states except for the parameter error converge to zero [9]. These issues motivate the search for more desirable strongswitching amplifier tunings, which would, ideally, have some of the best features of both the class-E and class-F tunings. If such a harmonic tuning strategy is to be found, it would ideally have the following features: Incorporation of the transistor output capacitance into the tuned circuit. ZVS to eliminate discharge loss from this capacitor. Simple circuit implementation.
Use of harmonic tuning to achieve improved waveforms for better performance. II. SYSTEM CONFIGURATION Fig.1 shows the system configuration of class-E seriesparallel (LCL) resonant inverter for inductively coupled plasma generator. It consists of one or more switches (MOSFET IRFP) connected with output parasitic capacitance and a freewheeling drain-source diode.
Argon, gaz
800
is
E G
Ls
Cs
ilo
Ro Lo
vcp
Cp
Fig.2. Class E inverter for inductively coupled plasma generator
1000
Lf D1 D3 Cf G
Ls
Cs
t e n s io n V T r ( V )
600 t e n s io n V c 3 ( V ) 3.05 3.1 3.15 3.2 3.25 temps(s) 3.3 3.35 3.4 x 10
-5
500
400 200 0
Power D4 supply
Cp
D2
-200
-500
3.05
3.1
3.15
3.2 3.25 temps (s)
3.3
3.35
3.4 x 10
-5
c o u r e n t s i1 e t i L (A ) 3.05 3.1 3.15 3.2 3.25 temps(s) 3.3 3.35 3.4 x 10

-5
Driver
Optimal control
Plasma reactor
10 5 0 -5
Fig.1. Class E inverter for inductively coupled plasma generator
The output resonant equivalent circuits were constructed by the output capacitor, matching transformer and plasma reactor. The load is modelled by the equivalent impedance which is varied during the heating process [7]. A. Operation of a class-E resonant inverter The class-E resonant inverter is power topology specially suited for high frequency operation due to its low switching losses [8]. The key point of a class-E resonant inverter is the capacitance C voltage evolution after switch S is turned off. To minimize the switching losses, transistor S must be turned on while diode D is in conduction, thus providing to minimize the switching losses, transistor S must be turned on while diode D is in conduction, thus providing zero-voltageswitching. Capacitor C also operates as a turn-off snubber, further reducing the switching losses. The switch must operate with ZVS commutations. Failure to do so will result on capacitor discharging through the main switch thus increasing turn-on losses strongly. The voltage waveform applied to the resonant tank has strong harmonic content. B. Steady-State analysis topology The applied analysis method is based on a state-space description of the circuit and calculations of its properties using a dedicated program written in MATLAB fig.3. The power switch of the inverter are turned on and off during each a constant interval T, the circuit variables namely, voltages vc , vcs , vcp and inductor currents iLs , iLr are chosen as the state variables, such as:
C o u re n t i1 (A )
-10
3.05
3.1
3.15
3.2 3.25 temps (s)
3.3
3.35
3.4 x 10
-5
Fig.3. The waveforms of the switching voltage (a) switching current (b) and is and iL currents of class-E inverter.
III. CLASS-E/F2 RESONANT INVERTER A. Analysis method The applied analysis method is based on a state-space description of the circuit and calculations of its properties using a dedicated program written in MATLAB. The main points of this method for the Class E/F2 inverter are presented below [4]. The following assumptions are used throughout the analysis 1) The transistor acts as a switch with a resistance of RTon for the ON interval, infinite resistance for the OFF interval, and zero switching times. 2) The shunt capacitance C1 is linear including the transistor output capacitance. 3) All passive elements are linear, ideal, and time invariant. The transistor is driven at frequency f and at any duty cycle D, where the duty cycle is defined as the switch ON time divided by the switching period T. The circuits in Fig. 3 can be described by a normalized set of state equations of the form:
dx / dt = Ax + Bu where x = [ x1 , x 2 , x3 , x 4 , x5 , x6 ] is the state vector. x1 = vc1 , x 2 = v c 2 , x3 = v c3 , x 4 = il1 , x5 = i LR ,
B = 0 1 L1 0
T
(3)
= Ax + Bu , y = Cx , x
where x = [ x1 , x 2 , x3 , x 4 ] is the state vector.
(1)
x1 = vcp , x2 = vcp x2 = vcs , x3 = ils , x4 = iLo .
(4)
il1(A), iLo(A)
6 4 2 0 -2 -4
Fig.4. Circuit diagram of class E/F2 inverter.
-6
a11 0 0 A= 1 L 1 1 L 2 1 L3
0 0 0 0 1 L2 0
0 0 0 0 0 1 L3
1 C1 0 0 0 0 0
1 C1 1 C2 0 0
R L2 0
1 C1 0 1 C3 0 0 0
1.6
1.65
1.7
1.75
1.8 T(s)
1.85
1.9
1.95 x 10
-5
Fig.5. Input and output current il1 and ilo.

vc1 400 300 500 vc2
(5)
200 0 100 0 1.6 1.65 1.7 vc3 800 600 400 200 0 -200 1.65 1.7 1.75 1.8 x 10
-5
1.75
1.8 x 10
-5
-500 1.6
1.65
1.7 il1
1.75
1.8 x 10
-5
2.8 2.7 2.6 2.5 1.65 1.7 1.75 x 10

-5
Denoting the matrix A: for the ON interval a11 = 1 / C1 RTon and a11 = 0 for the OFF interval. The steady-state solution is completed by adding the softswitching conditions. The ZVS condition demands that the transistor voltage at the instant the transistor is switched on should be zero. B. Design procedure The design procedure is explained with an example circuit of the Class E/F inverter. The specifications of the example circuit are described in the table I.
TABBLE 1: DESIGN PARAMETERS FOR THE CLASS-E AND CLASS-E/F2 INVERTER Description Operating frequency The peak transistor voltage the rms transistor current Parameters L1, L2 and L3 Parameter C1, C2 and C3 Resistance of the ON interval The load resistance Input voltage Value f = 1 MHz VTm = 450V ITrrms = 5A L1 =270H, L2 =16.8 H, L3 =8 .3H C1=1.64 nF, C2=1.99 nF, C3=0.71nF RTon =0.182 R =20 .33 including RL2 VI =200V
Fig.6. Capacitors voltage vc1, vc2, vc3 and il1 current:

il0 5 0 -5 1.62 1.64 1.66 1.68 1.7 ic3 4 2 0 -2 -4 1.62 1.64 1.66 1.68 1.7 1.72 1.74 1.76 1.78 1.8
-5
1.72
1.74
1.76
1.78 x 10
-5
x 10
Fig.7. Load current ilo and parallel capacitor current.
The analysis shows that the Class E/F inverter has the best performance, and therefore, the remainder of this paper will focus on this one. V. ADAPTIVE BACKSTEPPING DESIGN The control objective is to generate a feedback control input u(t) for the plant with unknown parameters , such that all closed loop signals are bounded, and the plant output y (t ) = x1 (t ) tracks a given bounded reference output yr(t) with r (t ) [10], [11], [12]. r (t ) , pounded derivatives y y Consider the plant of class E series resonant inverter: 1 = a1 x1 a2 x2 + b1u x 2 = a3 x1 x (6) y = cx1 Where the specific parameters are: a3 = 1 / C and b = 1 / l . a1 = R / l , a2 = 1 / l ,
VI. SIMUTATION RESULTS The theoretical results have been verified experimentally using the Class E/F2 inverter circuit in Fig. 6. The basic Class E circuit is extended by an input filter (L3 and C3) and parasitic series resistances (RL1, RL2 and RL3) of the inductors. They are used to calculate the output power and total efficiency of the inverter.
Actuator
System
Kresslimier Filters
Where the vector K T = [K1 K 2 ] is chosen so that the matrix: K1 1 A0 = A KC T = is Hurwitz, and hence P exists K 2 0 such that
T PA0 A0 P = I , P = PT
(13)
Adaptive Buckstepping procedure Step 1
With the help of the those filters our state estimate is = + T x (14) (15)
Step 2
and the state estimation error = xx
Step
By exploiting the structure of f ( y, u )T we denote the filter by: T = [ ] (16) and show that to the special dependence of f ( y, u )T of u, the equation for and are governed by:
Fig.8 Adaptive Buck-stepping Control Scheme.
Where the specific parameters are:
a1 = R / l , a2 = 1 / l ,
a3 = 1 / C and b = 1 / l . 1. State estimation filters: We start by representing the plant (6) in the observer canonical form:
= A +eu 0 1 = A0 + e2 y
(17)
the vector in (11) can be obtained from the filter (17) through the algebraic expression 3 = A0 (18) What has been achieved thus far is a static relationship between the state x and the unknown parameter :
x = + T +
a b 0 1 = x x y 1 + u 0 0 a2 a3 0
y = Cx
(8) (9)
(19)
Where
0 1 a1 b A= , a = a a B = and C = [1 0] . 0 0 0 2 3 In this situation, we are able to express (7), (8) as where
In conclusion, from (19) and the expressions of KREISSELMEIR filters an equivalent expression for the is virtual estimate x
3 = A0 x
a A + b
i i 0 0 i =1
= Ax + f ( y, u ) x
u y 0 f ( x, u )T = 0 0 y
(10)
For the plant (6) with relative degree 2, an adaptive backstepping control design procedure consists of two steps. It starts with its output y ,
= x2 yC T a y
From the algebraic expressions (19) we have
(21)
and the parameter vector T = [b a ] For state estimation we employ the filters
= A + Ky 0 T = A T + f ( y , u )T 0
(11)
x2 = 2 + ( 2 ) T + 2
(22)
T
= b2 + 2 + [0 2 ] + 2
(23)
(12)
Substituting both (22) and (23) into (21), we obtain the : following two important expressions for y
= 2 + T + 2 y
= b 2 + 2 + + 2
and
T
(24) (25) (26)
Where 1 is the tuning function
, and Were (37) is used as the actual update law for = , where = ( (y r + 1 )C )z1 (38) 1 1
Substituting (38) and (38) into (36) we obtain ~T z z + c z 2 + b V 1 1 1 1 1 1 2
= K 2 2 1
(39)
Where the regressor and the truncated regressor are defined as
Step 2: from (29) with the help of (26) we obtain
= 2 , 2 yC T
, = 0, 2 yC T
(27)
All of these states are available for feedback. Our design task is to force the output y to asymptotically track the reference output yr while keeping all the closed loop signals bounded. We employ the change of coordinates z1 = y yr (28) y r 1 z 2 = 2 (29) is the estimate of = 1 / b , our goal is to regulate Where z [z1 z 2 ]T to zero.
1 T ~ (40) + 2 1 y Since our system is augmented by the new state z2, we augment the Lyapunov function (35) as 2 = 2 2 z
V2 = V1 +
1 2 1 T z2 + P 2 4d1
2
(41)
The derivative of V2 satisfies

1 1 2 2 c2 z 2 d2 V2 c1 z1 z 2 y + 2d 2 2
2 2 c1 z1 c2 z 2
(42)
Step 1: Let the tracking error z1 = y yr and y r 1 , where 1 a function to be introduce z 2 = 2 designed is. Then, from (28) and (29) we get: T ~y r + bz2 1 = b1 + 2 + + 2 b z (31) Scaling the first stabilizing function 1 as 1 1 = Choosing the design function 1 as:
T
And the control law
r u = 2 + y
(43)
IV. SIMUTATION RESULTS OF THE SYSTEM WITH BACK-STEPPING CONTROL DESIGN In this section, we depict some simulation results by using adaptive Back-stepping controller in different disturbance conditions.
TABLE 2: DESIGN PARAMETERS FOR THE OBSERVER BASED CONTROLLER Description Liapunov gains Observer gains Damping gain Adaptation gains
x 10
-5
(32)
1 = c1 z1 d1 z1 2 Results in the system

~ + bz r + 1 ) 1 = c1 z1 d1 z1 + 2 + T + 2 b( y z 2
(33)
(34)
3.5 3
Value c1=2e-5, c2=0,02 k1=0,01 , k2=2 d1= 1e-5, d2=0,01 g1=10, g2=10
and considering the first partial positive definite function: 1 2 1 ~2 ~ b ~2 1 T V1 = z1 + P + + (35) 2 2 2 4d1
We examine the derivative of V1 as:
~ ~ +b z ( y r + 1)C) b( y r + 1) V1 = z1 c1z1 d1z1 + 2 + ( 2 1 ~ b T T 1 4d1
2.5
R o e s tim (H )
]
(36)
1.5
0.5
~ ~ terms in (36) we To eliminate the unknown indefinite and choose = sgn(b)( y r + 1 ) (37)
0 0
0.5
1.5
2
temps (s)
2.5
3.5 x 10
4
-5
Fig.9 Estimation of parameters theta1
x 10 8
6
t ta 3 s ti m
10% and 50% of uncertainties, and from these results, it is observed that the proposed observer-based backstepping controller is good in adapting the uncertainties and load disturbances (refer the estimation of the load current in fig.13, the adaptation is not exceeding 0.2 s). VI. CONCLUSION
-2 0
0.5
1.5
2
temps (s)
2.5
3.5 x 10
4
-5
Fig.10 Estimation of parameters theta3

7 x 10
4
E s tim a tio n d e t ta 1
The design class-E and E/F series-parallel resonant inverter power supply for an inductively plasma generator system has been presented. The variable load is highly inductive and requires a several kW active power at a frequency of several MHz Based on a detailed topology investigation, a resonant circuit supplied by a voltage source class-E and class-E/F03 inverter are chosen. An analysis of the circuit and basic design rules are given. These E/F tunings allow strong-switching operation as in class E, but show a greater tolerance for transistor output capacitance and present waveforms approaching those of the more desirable class-F inverse, the family exhibits a traedoff between circuit complexity and performance. The tracking performance of the observer-based back-stepping controlled allowing operation of the inverter with the MOSFET switching-losses is investigated and simulation results verifying the operation of the control.
4
-5
0.5
1.5
2.5
3.5 x 10
Temps (s)
Fig.11. Estimation of parameters teta2 (RO)

3
REFERENCES
[1] H. Fujita, H. Akagi " A 2MHz, 2 KW Voltage-Source Inverter for LowTemperature Plasma Generators: Implementation of Fast witching with a Third Order Resonant Circuit", IEEE Trans. on Industrial Electronics, Vol. 43, No. 1, pp. February 1996. [2] N.J. Park, D. Y. Lee and D.S. Hyun "Study on the new control scheme of class-E inverter for IH-Jar application with clamped voltage characteristics using pulse frequency modulation" IET Electr. Appl., 2007, 1, (3), pp. 433-438. [3] Zbigniew Kaczmarczyk, "High-Efficiency Class E, EF2, and E/F3 Inverters", IEEE Trans. on Industrial Electronics, Vol. 53, No. 5, pp. 1584-1593 OCTOBER 2006 [4] D. Y. Lee, D. S. Hyum and I. Choy, " A New Hybrid Control Scheme Using Active-Clamped Class-E Inverter with Induction Heating Jar for High Power Application", IEEE 2002. [5] B. Grzesik, Z Kaczymarczyk, J, Junak, "A Class E Inverter the influence of Inverter Parameters on its Characteristics", IEEE 1996, pp. 1832-1837. [6] Xiying Li,"Linear quadratic tracking optimum controller model design to optimize high frequency power supply performance", Thesis for the Degree of Mater of Science in Engeering, Youngstown State University, March 1999. [7] Sibylle Dieckerhoff, Michael J. Ryan and Rik W. De Doncker " Design of an IGBT-based LCL-Resonant Inverter for High-Frequency Induction Heating" IEEE. 1999. pp. 2039-2045. [8] J. Ribas, J. Cardein, J. Garcia, M. A. Dalla-Costa, J. M. Alonso, "Electronic Ballast for Metal Halid Lamps based on a Class E Resonant Inverter Operating at 1 MHZ. IEEE 2005, pp. 600-605. [9] John Wiley and Sons Adaptive Control Design and Analysis Gang Tawo, University of Verginia, 2003. [10] Francesc Pozo Montero, "Nonlinear control of uncertain systems", Doctoral Theses, Polytechnics Ctalunya University, Barcelona, Nov.2004. [11] Lars Walln "Dynamic Tyre Models in Adaptive Slip Control" Department of Automatic Control, Lund Institute of Technology, March 2001. [l2] S. Mohamed, J. C. Ahmed, "Development of Low-Cost Multifunction Sensors for Lightweight Fire and Forget Antitank Weapon System", IEEE Trans. Industrial Electronics, Vol. 30, No. 1, Feb. 2010, pp. 27-33.
1 e rre u r Z (A )
-1
-2
-3 0
0.5
1.5
2 temps(s)
2.5
3.5 x 10
4
-5
Fig.12. Steady state tracking error z1 and z2.
L e c o u re n t d e c h a rg e e t la r f r e n c e (A )
6 4 2 0
-2 -4 -6 -8 0 0.5 1 1.5 2 2.5 3 3.5 x 10 4

-5
temps (s)
Fig.13. Estimation of the load current
The Figs. 9, 10 and 11, represent the estimation of the load current and the estimation of load parameters theta1, theta2 and theta 3 for Disturbance of the load parameters of 50% L. In order to demonstrate the adaptability and robustness of the proposed controller, an uncertain condition is considered for the simulations, where the load parameters are assumed to be
Energy Storage for Microgrids Islanding Operation

A. S. A. Awad, and M. M. A. Salama
Electrical and Computer Engineering Department University of Waterloo Waterloo, ON, Canada asamir@uwaterloo.ca; msalama@uwaterloo.ca
AbstractThis paper investigates the application of energy storage systems to enable microgrids (Grids) islanding operation. Inverter-based Distributed Generation (DG) is the most common type found in modern distribution systems. Basically, many of these DGs rely on renewable energy sources as prime movers, and hence their output power is limited to a certain maximum value at every instant. Therefore, satisfying demand power is almost impossible when loads are only supplied by such DGs during autonomous mode of operation. Traditional solutions for this problem may include load controlling, i.e. load shedding or load dumping, so as to reduce power mismatch between generation and demand. Alternatively, energy storage devices can be adopted and controlled to act as a slack source in order to supply or consume any power mismatch once islands are formed. The aforementioned two solutions are compared in this paper so as to conclude operational guidelines to successfully enable Grids islanding operation. Simulations are conducted using PSCAD/EMTDC package. Keywordsmicrogrids; distributed generation; energy storage systems; islanding; load shedding
T. H. M. EL-Fouly
CanmetENERGY, Natural Resources Canada Varennes, Qubec, Canada telfouly@nrcan.gc.ca Grids can operate in two modes: grid connected mode (during normal conditions), or islanding mode (during any grid disturbance). Islanding is defined in IEEE standards [5] as a condition in which a portion of the utility system that contains both load and distributed resources remains energized while isolated from the remainder of the system. Island formation mainly intends to improve service reliability and power quality for served customers. On the other hand, some issues related to safety of personnel and equipment imped islanding operation, and hence all DGs are usually forced to be shut down upon islanding detection and remain disconnected until the utility grid is safely restored. However, intentional islanding has been recently allowed in the IEEE standards [6]. Further, some Grid islanding projects have been implemented by British Columbia Hydro and Hydro Quebec to improve system reliability [3]. DERs are usually modular units with small capacity; they can be either DGs, Distributed Storage (DS) units, reactive power sources, or hybrid mix between two or more devices. Two main types of DGs exist according to their interface with the distribution network: conventional rotating machines, i.e. synchronous or induction, and inverter (electronically coupled) based DGs.
I.
INTRODUCTION
Power systems are now evolving from the conventional regulated one, with centralized generation connected to the transmission networks, to the de-regulated structure that allow for small generators to be connected directly to the distribution networks. Consequently, distribution networks become active, and hence they are usually referred to as active distribution networks [1]. Microgrids (Grids) are defined in [2] as the connection of modular generation units ( tens of kW or less) to low voltage (LV) distribution networks. Grids have also been defined as entities that coordinate the operation of Distributed Energy Resources (DERs) consistently and in a decentralized way [3]. Moreover, Grids can be defined as LV networks with island operation capability, and comprising of DERs and some controllable loads which all of them can be managed through Grid management system [4]. In general, Grids are small-size low or medium voltage networks that are used to supply power and/or heat to a small community, as shown in Fig. 1.
Figure 1. Typical Grid structure The authors would like to thank the Government of Canada for financially supporting this research through the Program on Energy Research and Development.
Distribution feeder
Distribution feeder
Inverter based DG units are very common nowadays; they are characterized with fast dynamic response through the interface converter and capability to limit fault current to almost 200% of normal current. However, they provide inertia-less generation compared to conventional rotating machines [7]. Therefore, they may lead to voltage and/or frequency instability due to any power mismatch between the supply and the demand after islanding detection. Further, such DGs are usually driven by renewable energy sources as prime movers, and hence their output power is limited to a certain maximum value at any instant. Consequently, their ability to satisfy demand requirements, and thus to enable Grids islanding operation is limited. Therefore, relying on intermittent renewable sources, such as wind turbines or photovoltaic arrays, may not improve service reliability. An alternative solution is to adopt dispatchable energy sources, e.g. diesel generators or fuel cells, in order to successfully enable Grids islanding [8]. The authors in [9] proposed a control technique for inverter based DGs which allows for intentional islanding operation. The proposed approach consists of two operation schemes, one for grid connected mode and the other for islanding operation. The inverter control switches from P-Q control (during grid connected mode) to voltage control (during islanding mode). This control strategy is based on the assumption that the inverter is supplied from an ideal dc battery, and hence the DG is able to satisfy any amount of demand power during islanding operation. From the aforementioned discussion, it can be concluded that if the Grid has neither conventional rotating DGs nor dispatchable inverter based DGs, then energy storage are essential for Grids islanding operation to balance supply and demand. Energy storage systems further provide seamless transition from grid connected to islanding mode of operation. Some of the available energy storage options for Grids are flywheels, batteries and super-capacitors. Flywheels can be used as central storage for the whole Grid, while batteries can be installed as central storage or at the dc link of each individual DG. Super-capacitor is an expensive option compared to batteries and flywheels [10]. According to the best of authors knowledge, no previous work has addressed the feasibility of different schemes to enable successful islanding operation. Therefore, the ultimate goal of this work is to investigate through simulations two schemes which might be considered to enable Grids islanding operation, as follows: 1- Load shedding/dumping scheme with inverter based DG, operating under Maximum Power Point Tracking (MPPT) and controlled as proposed in [9]. 2- Energy storage scheme with the same DG as above. Simulations are conducted using PSCAD/EMTDC package, and the results are discussed so as to provide operational guidelines to enable Grids islanding operation in the future. In section II, the system under study and the
modeling approach are introduced. Section III presents the simulation scenarios and the results. Finally, the conclusions of this research work are summarized in section IV. II. SYSTEM UNDER STUDY
The system under study, shown in Fig. 2, represents a typical Grid which comprises of inverter based DG, inverter interfaced energy storage module, and a controllable load. The controllable load is assumed to be consisting of clusters of loads; each cluster has a separate circuit breaker so that load shedding or dumping can be easily achieved. For sake of simplicity, the amount of load to be shed or to be dumped is assumed to be pre-known and equal to the load-generation mismatch just before the disturbance. System parameters as well as controllers coefficients are given in the Appendix. The control strategies for the inverter based DG and the storage module are discussed in the following subsections. A. Inverter Based DG Control The DG control circuit is shown in Fig. 3 in which the dc buck converter is used to control the dc battery terminal voltage (Vb) to half of the battery internal emf in order to achieve maximum output power from the battery, and thus simulating the performance of MPPT based DG. This is achieved by comparing the measured and the reference battery terminal voltages to produce an error signal. This error signal is then used as an input to a Proportional Integral (PI) controller which generates the duty cycle control signal of the dc buck converter in order to control Vb. Further, the DG inverter control is based on Sinusoidal Pulse Width Modulation (SPWM). It is designed to set the dc link voltage (Vdc) at 600 Volts which is required for output ac voltage of 380 Volts (rms, line to line). By controlling Vdc, the inverter output active power (P) is maintained constant at the maximum power available from the dc battery, which is assumed to be equal to 0.1 MW in the whole analysis. For this purpose, PI controller is also deployed to set the phase angle of the modulating signal in order to control the dc link voltage, and hence the output active power. The output reactive power (Q) is further controlled by another PI controller which feeds the magnitude of the modulating signal. Finally, the modulating signal is compared to a carrier signal so that switches firing signals are generated.
Figure 2. System under study
Figure 4. The control circuit for inverter interfaced energy storage module
III.
Figure 3. The control circuit for inverter based DG
SIMULATION SCENARIOS AND RESULTS
The previous control strategy is adopted in normal grid connected mode of operation; however after switching to islanding mode, the control circuit structure is almost the same except that Point of Common Coupling (PCC) voltage (V) is controlled by adjusting the modulating signal magnitude. This is achieved by a selector switch that operates based on islanding detection algorithm. It should be emphasized that islanding detection is not the main objective of this paper; therefore, it is assumed that islanding is detected in a good timely manner and a signal is sent to the selector switch to change the control mode after islanding. Moreover, the inverter modulating frequency is set to 60 Hz in order to operate properly during both grid and islanding modes of operation. B. Inverter Interfaced Energy Storage Control The energy storage module is also interfaced through SPWM inverter. It is designed such that its terminal voltage is maintained constant at 1 p.u, as shown in Fig. 4, and hence it operates as a slack (reference) bus during islanding mode of operation. The main characteristic feature of the energy storage module is that it is always connected to the system with the same control structure. Therefore and upon islanding detection, it instantly supplies or absorbs any power mismatch between the DG source and the load, assuming that its State of Charge (SOC) permits this feature. At the first sight, it may be thought that this assumption is not totally true since SOC might be at its minimum or maximum level at a given instant, and hence no further discharging or charging is possible. However, this assumption is acceptable when there are several DS units that might have different SOCs allowing for different power import/export scenarios. This concept of DS is very promising to enable islanding operation as it will be proved later on.
The system under study is investigated under two different operational scenarios. In each scenario, the system is running in normal grid connected mode before a grid disturbance is detected at t=5 sec. and the feeder breaker is opened so that an island is formed between the DG, the controllable load, and with/without the energy storage module. Moreover, three load conditions are considered in each scenario as follows: Case 1: matching load, i.e. equal to the DG output power before islanding instant (0.1 MW + J 0.03 MVAR), Case 2: 80% of matching load (0.08 MW + J 0.024 MVAR), Case 3: 120% of matching load (0.12 MW + J 0.036 MVAR) Scenario I: Inverter Based DG and Load Shedding/Dumping Scheme In this scenario, the inverter based DG operates alone to satisfy the load active and reactive powers. It is designed to operate at its maximum output power of 0.1 MW and 0.03 MVAR before islanding as discussed earlier. Further, the inverter control switches to voltage control mode after islanding detection, and load dumping/shedding schemes are applied in cases 2 and 3, respectively, in order to match the DG output power. Load dumping/shedding scheme is simulated through one loads breaker which operates after four cycles from the islanding detection instant. Figures 5-8 present the DG output active and reactive powers, PCC voltage, and frequency, respectively, before and after islanding detection for the three loading conditions. Figures 5 and 6 reveal how the DG output active and reactive powers remain constant at their pre-defined values before and after islanding detection due to sake of load control scheme. It is also clear that the DG inverter maintains acceptable voltage and frequency values as long as the demand magnitude is equal or less than the maximum output power (cases 1 and 2). On the other hand, if the demand power exceeds the maximum power value (case 3), the PCC voltage drops significantly as shown in Fig. 7. This can be justified as inverter based DGs are characterized with very fast response, and hence they are reluctant to go out of step very quickly. Therefore, other means should be adopted to pick up the initial power mismatch and prevent voltage collapse during islanding mode of operation.
Figure 5. DG output active power for scenario I
Scenario II: Inverter Based DG and Inverter Interfaced Energy Storage Module This scenario integrates the energy storage module to the Grid with the same DG adopted in the previous scenario. However, the load shedding/dumping schemes are assumed to be not active in this scenario. Figures 9 and 10 show the DG and the storage module active powers before and after islanding detection. The DG output active power remains constant after switching to islanding control mode, and the energy storage module consumes/supplies the power mismatch between the DG and the load, as shown in cases 2 and 3, respectively. The results reveal the unique feature of energy storage as discussed earlier. Therefore, adopting energy storage eliminates the need to load shedding/dumping as in the previous scenario. Similarly, the DG output reactive power does not change after islanding detection, as shown in Fig. 11. Moreover, the PCC voltage and the system frequency prove the superior performance of applying energy storage in distribution systems, as indicated in Figures 12 and 13, respectively, since they are maintained within their permissible values, and hence the island is allowed to operate successfully.
Figure 6. DG output reactive power for scenario I
Figure 9. DG output active power for scenario II
Figure 7. PCC voltage for scenario I
Figure 10. Energy storage output active power for scenario II
Figure 8. System frequency for scenario I
stability in terms of voltage and frequency fluctuations after islanding detection. Therefore, energy storage systems are recommended in future smart grids to enable islanding operation, and hence making use of available DERs to improve system reliability. APPENDIX: SYSTEM PARAMETERS
Grid rated voltage Grid resistance & inductance (RG & LG)
Figure 11. DG output reactive power for scenario II
22 kV 0.85 , 0.005H 1.7 , 0.01 H 1 MVA 0.380 / 22 kV 0.01 p.u 150 kVA 2 kV, 10 ohms 0.6 kV 0.001 H, 732 F 0.5, 0.5 0.1, 0.1 0.1, 0.01
Line resistance & inductance (Rline & Lline) Coupling transformer rated power Coupling transformer rated voltage Coupling transformer leakage reactance DG rating Battery voltage and internal resistance (for inverter based DG) Battery voltage (for energy storage module) Output LC filter Controller 1 coefficients (Kp & KI) Controller 2 coefficients (Kp & KI)
Figure 12. PCC voltage for scenario II
Controllers 3-5coefficients (Kp & KI)
ACKNOWLEDGMENT The authors would like to thank the Government of Canada for financially supporting this research through the Program on Energy Research and Development. REFERENCES
[1] "Enabling Tomorrows Electricity System"; Report of the Ontario Smart Grid Forum, Available: http://www.ieso.ca/imoweb/pubs/smart_grid/ Smart_Grid_Forum-Report.pdf [2] J. A. P. Lopes, C. L. Moreira, and A. G. Madureira, "Defining control strategies for MicroGrids islanded operation", IEEE Transactions on Power Systems, vol. 21, 2006, pp. 916-924. [3] N. Hatziargyriou, H. Asano, R. Iravani, and C. Marnay, "Microgrids", IEEE Power and Energy Magazine, vol. 5, 2007, pp. 78-94. [4] H. J. Laaksonen, "Protection Principles for Future Microgrids", IEEE Transactions on Power Electronics, vol. 25, 2010, pp. 2910-2918. [5] "IEEE Recommended Practice for Utility Interface of Photovoltaic (PV) Systems", IEEE Std 929-2000, 2000, p.i. [6] "IEEE Guide for Design, Operation, and Integration of Distributed Resource Island Systems with Electric Power Systems", IEEE Std 1547.4-2011, 2011, pp. 1-54. [7] F. Katiraei, R. Iravani, N. Hatziargyriou, and A. Dimeas, "Microgrids management", IEEE Power and Energy Magazine, vol. 6, 2008, pp. 5465. [8] A. Llaria, O. Curea, J. Jimnez, and H. Camblong, "Survey on microgrids: Unplanned islanding and related inverter control techniques", Renewable Energy, vol. 36, 2011, pp. 2052-2061. [9] H. Zeineldin, E. F. El-Saadany, and M. M. A. Salama, "Intentional islanding of distributed generation", IEEE Power Engineering Society General Meeting, 2005, pp. 1496-1502 Vol. 2. [10] N. W. A. Lidula and A. D. Rajapakse, "Microgrids research: A review of experimental microgrids and test systems", Renewable and Sustainable Energy Reviews, vol. 15, 2011, pp. 186-202.
Figure 13. System frequency for scenario II
IV.
CONCLUSIONS
This paper has discussed state of the art in enabling Grids islanding operation. Dispatchable energy sources are the key for ensuring successful islanding operation. However, the common application of intermittent renewable sources which are usually interfaced through inverter units leads to the emergence of energy storage as a vital solution to enable Grids islanding operation. The latter conclusion is supported through simulation analysis in which load controlling scheme with inverter based DG is investigated and compared to the application of energy storage module. The results conclude that load shedding/dumping schemes along with inverter based DG control is not sufficient to maintain voltage stability once islands are formed. This is mainly due to the fast response and the inertia-less production from such DGs. On the other hand, energy storage application ensures system
Sliding Mode Control Applied To UPS Inverter Using Norm of the State Error
Hamza Makhmreh
Electrical and Electronic Engineering Department Eastern Mediterranean University Gazimagusa, Mersin 10, Turkey Hamza.ahmed@alumni.emu.edu.tr
Abstract In this paper we introduce a new approach to sliding mode control of UPS inverters. The slope of the sliding line is defined as a nonlinear function of the error vector norm, with the effect that the sliding time of the state trajectory is reduced norm of both voltage error x1 , and its derivative x 2 ; which multiplied by a factor ( ). The norm value is saturated for small and large value of ; which represents the slope of the sliding line. Another parameter ( ) is also introduced, which is a gain multiplied by the result of the saturated function. Key words: UPS inverter, sliding mode control.
Osman Kukrer
Electrical and Electronic Engineering Department Eastern Mediterranean University Gazimagusa, Mersin 10, Turkey osman.kukrer@emu.edu.tr In this paper the norm of the error function which is a combination of both the voltage error x1 , and its rate of change
x2 factorized by factor called ( ). This function is called the

norm function which is used to determine the value of the sliding line. The norm function is designed such a way so that a large value of the sliding line slope is applied to the system when the states are far away from the origin; and a small value of the slope is used when the states are close to the origin. The mathematical derivation shows that the best performance of the system when is chosen to be zero. The simulation is compared to three level hysteres is and rotating sliding line based approaches, and the new approach shows fast response, low total harmonic distortion and small RMS voltage error II. SLIDING MODE CONTROL USING NORM APPROACH
I.
INTRODUCTION
A UPS inverter is required to deliver a pure sinusoidal output waveform, with low total harmonic distortion at fast response when a disturbance occurs at the load side due to linear or nonlinear loads. Many approaches to control UPS inverters have been introduced, such as repetitive controller; which based on Internal Model (IM) theory. Perfect tracking performance could be maintained if the model of the reference is included in the closed loop feedback controller [1]. Deadbeat control method is using the inverse discrete model of the inverter a zero steady state error could be maintained within very small settling time [2]. Adaptive control scheme is used without any prior knowledge about the plant parameters. The controller adapts itself such a way that high performance is achieved [3]. There are many continuous-time sliding mode controllers applied to UPS inverters. In [4] basic equation for calculating the switching frequency is derived. The switching frequency is reduced to the half by using a three level hysteresis sliding function to control the inverter; this property has been proved using the simulation and experiment. The system has the property of simple implementation with good disturbance rejection. In [5] the sliding line function is derived from a fuzzy logic controller. Two values of the sliding line are applied to the system input; large value when the system is affected by disturbances and small one when the system is in the normal operation. The sliding function behavior makes the sliding line rotates at the state plane. The derived function improves the system response.
A. Mathematical Formulation
+
Q1
Q3
i
ic
C
io
+
Vs
vo
Load
vi
Q2
Q4
__
Figure 1. Single phase UPS inverter.
The mathematical equations which represent the operation of the inverter are given as [4]
di L dt = vi vo
(1) (2)
i = ic + io
Where vi = uVs represents the inverter output voltage, and u is the control input function, which is defined to be of three level nature. By defining the output voltage error as and (2) can be written as
During the sliding mode when s = 0 , we obtain the following equation which describes the dynamics of the output voltage error
x1 = vo vo , and its rate of change as, x2 = vo v , then (1)

i x1 = x 2
i ii ii ii* x2 = x1 = vo vo
x1 = x2 = x1 =
x12 sgn( x1 )
(1 2 x12 )
1 2
(9)
(3)
notice that if x1 < 0, then (9) becomes
(4)
x2 =
x12
(1 2 x12 )
1 2
(10)
Here the sliding surface which is defined as
s = x1 + x2
(5)
is designed such that is a nonlinear function of the error norm, as shown in Fig. 2. In the interval between e1 and e2 on the linear part of the x characteristics
The term under the square-root in the denominator of (10) should always be positive, so that the system always has a solution. The solution (10) is given by solving the following equation
(1 ax1 ) x1
2
dx1 = dt ,
(11)
= xn
where
(6)
is the slope of the linear part of the function. In (6)
such that a = 2 . Equation (11) can be written as
xn is the norm of the error, represented by x1 and x2 as

xn = x = x1 + ( x2 )
2 2
(7)
1 ax12 12 1 + a sin ( a x ) = ( t t1 ), (12) x1 x11

and the solution is given by
t = sgn( x11 ) sgn( x12 ) + 1 sin 1 ( ) (13)
where is a parameter suitably chosen to ensure that the two terms under the square-root have the same physical unit and are comparable in magnitude.
2
The slope
1
e1
xn = x
(2 1 ) x where = and 11 represents the 1 [(1 + 2 )(1 + 2 )] 2 2 1 initial value of the error voltage when the state trajectory starts to slide on the line s = 0 , and x12 represents its final value, while t represents the time needed to slide from
x11
to
e2
reach x12 as shown in Fig. 3. Notice that for the error norm approach when the states come close to the equilibrium point, the slope of the sliding line decreases rapidly till it reaches the point x12 . This is because of the nature of the algorithm used here which aims to have a large slope when the trajectory hits the sliding line and starts to slide on it. The value of the slope at that moment is given by 2 = e2 .
Figure 2. The proposed nonlinear function in terms of error norm.
Squaring both sides of (6), substituting in (7) and solving for xn yields
xn =
x1 (1 x1 )
2 2
(8)
x2 x11
=0 Reaching Mode
200
100
Output Voltage and Current
t
x1 < 0 x1
x12
Sliding Mode
-100
-200 0.06
0.065
Time
0.075
0.08
Figure 3. State trajectory of the phase plane.
Figure 4. . Output voltage and current waveform of the norm approach.

40
After that, when the state reaches the point x12 the slope of the sliding line is decreased to 1 = e1 , which is a lower value of the slope that leads to a reduced steady state error.
30
20
B. Design of The SM Controller The aim is to reduce the sliding time as much as possible, such that the controller returns the system to the origin when a disturbance is applied to the system. The plot of (13) for various values of and , with respect to sliding time shows that should be chosen as small as possible such that the reaching time is reduced. For = 0 the norm function will reduce to the absolute function of the voltage error. Using the absolute value of the voltage error the number of design parameters is reduced to be only one instead of two, which is much better and easier to work with.
III. SIMULATION RESULTS Using Matlab-Simulink the new approach has been tested for a triac-controlled resistive load ( R = 3 ) with firing angle
O u tput V oltag e E rror
10
-10
-20
-30
-40 0.06
0.065
Time
0.075
0.08
Figure 5. Output voltage error of the norm approach.
Fig. 6 shows the state trajectory of the norm approach. Its obvious that the chattering problem is a bit large for the threelevel hysteresis approach [4], and its reduced to a lower level (the zigzag width is reduced) by making use of the rotating sliding line approach. The chattering problem is almost eliminated using the norm approach which is established analytically from the system parameters and which is achieving the system requirements. The new approach has been compared with the three level hysteresis [4], and the rotating sliding line [5] approaches. The design parameters are the same as those given in [4] and [5], except the value of the hysteresis width which is chosen to be 30000 for all the approaches. The comparison results are summarized in Table 1. The RMS error of the output voltage is defined as
90 simulating a sudden disturbance occurring on the load side. Depending on the previous argument in section (2.B) the value of is chosen to be zero for best performance. The
system parameters are Vm = 200 V, Vs = 300 V, L= 250 H , C=100 100 F , e1 and e2 are 5 and 35 resp., 1 and 2 are 8 and 16 resp. The output voltage and current waveforms are shown in Fig. 4. The amplitude of the output voltage for the norm approach is larger which means that the steady state error is lower as shown in Fig. 5. From the same figure its easy to notice that the system convergence to the reference very fast for the norm approach.
1 T RMS error = x12 dt T 0
(14)
The sliding mode control based on the error norm approach is capable of reducing the RMS error compared with the other approaches. It also yields the smallest error in the amplitude of the fundamental output voltage (0.75% compared with 2.36% for RSMC and 1.99% for 3LH). The THD produced by the proposed approach is slightly better than that of 3LH and slightly larger than that of RSMC.
5
x 10
IV.
CONCLUSION
-2
-4
A new approach of sliding mode control is introduced which is derived from the error norm. This results in a time varying slope of the sliding mode control. The behavior of the norm function is useful for reducing the sliding time, which means that the response time is decreased. Simulation results show the benefit of using such a function which noticeably improves the response time, increases the amplitude of the fundamental voltage and reduces the RMS value of the voltage error. REFERENCES
-30 -20 -10 0 10 20 30 40
X2
-6
-8 -40
X1
[1]
Figure 6. Phase portraits for the norm of the state error approach. TABLE I. COMPARISON OF THE THREE SLIDING MODE CONTROL
METHODS FOR OUTPUT VOLTAGE
[2]
Performance parameter THD (%) RMS Error Fun. Amp. (Volt) (volt) No. of Design parameters
3LH 2.89 4.90 196.02 Two
RSMC 2.65 4.96 195.28 Two
Error norm ( = 0) 3.03 4.41 198.38 One
[3]
[4]
[5]
Li, Min, et al. "Comparison of two repetitive control strategies of UPS inverter on saber." IEEE International Forum on.Strategic Technology (IFOST), 2011 O. Kukrer, and H. Komurcugil, Deadbeat control method for single phase UPS inverters with compensation of computation delay, IEEE Proc. Inst. Elect. Eng. Elect. Power Appl., vol. 146, pp. 123-128, January1999 H. Deng, R. Oruganti, and D. Srinivasan. "Adaptive digital control for UPS inverter applications with compensation of time delay." IEEE Nineteenth Annual Con. in Applied Power Electronics, vol. 1, pp. 450455, 2004. O. Kukrer, H. Komurcugil, and A. Doganalp, "A three-level hysteresis function approach to the sliding-mode control of single-phase UPS inverters, IEEE Transactions on Industrial Electronics, vol. 56, no. 9, pp. 3477-3486, 2009. H. Komurcugil, "Rotating-Sliding-Line-Based Sliding-Mode Control for Single-Phase UPS Inverters. IEEE Transactions on. Industrial Electronics, vol. 59, no. 10, pp. 3719-3726, 2012.
Load Uncertainty Based Transmission Network Expansion Planning

Chandrakant Rathore
Electrical Engg. Department S V National Institute of Technology Surat, India Chandrakant.r@rediffmail.com
Ranjit Roy
Electrical Engg. Department S V National Institute of Technology Surat, India rr@eed.svnit.ac.in
Abstract Transmission expansion planning (TEP) is more robust and reliable when it is able to tackle the uncertainty generated by load. In this, paper in order to handle the demand uncertainty a mathematical model is presented. In particular, the uncertainty of each bus load analyzes individually and is expressed in interval number. To solve the DC-Static transmission network expansion planning problem a metaheuristic population based algorithm named Gbest Artificial Bee Colony (GABC) is proposed. Garvers 6-bus test system is used to simulate the proposed method and results shows that minimum investment cost can be found considering the uncertainty in demand. Keywords- Artificial bee colony algorithm, direct current (DC) model, interval, resizing, uncertainty
expressed in probabilistic model or interval model [14]-[17]. Moreover Monte Carlo simulation method is used for the load uncertain but this method is time consuming. There are few literatures reported in TEP problems related to load uncertainty as examples, [17] presented two mathematical models considering load uncertainty, [18] presented concept of the minimum load cutting problem, multiyear planning considering load uncertainties [19], demand uncertain and risk aversion [20] and wind farm and load uncertainties in [21] respectively. In this work, load uncertainty is expressed in interval number and Gbest Artificial Bee Colony (GABC) algorithm [22]-[23] which is a population-based search procedure is adopted to solve the DC-Static transmission network expansion planning (STNEP) problem. The main contributions of the paper are:a) Load uncertainty model is considered in STNEP problem. b) Gbest ABC algorithm is used to obtain the optimal solution. c) It is observed that with same minimum investment cost with same line expansion the proposed STNEP problem able to cater future load without load shedding phenomena. The rest of the paper is organized as follow: Section II describes the mathematical model for transmission expansion planning. Section III presents overview of Artificial Bee Colony Algorithm on TNEP problem. Section IV presents the results obtained and discussion and finally conclusion in Section V. II. MATHEMATICAL MODEL FOR TEP Generally, for the TNEP problem DC power flow model is widely used. Many other models are reported in [24] as an alternative to DC model. A. Traditional STNEP model The objective function for the TNEP problem is to minimize the transmission line investment cost of the system. In this paper, the problem can be formulated based on lossless DC power flow model as follows [7]-[8], [24],
I.
INTRODUCTION
The aim of the transmission network expansion planning (TNEP) problem is to determine what, where and when new transmission lines to be installed in order to meet the future increase in load with minimum investment cost. How to satisfy the network constraints and reliability of the systems in uncertain nature of demand is a critical issue. So in order to adopt this problem in this paper uncertainty of individual load bus is formulated via DC power flow model. Transmission expansion planning (TEP) is divided into two categories static and dynamic. Static TEP is simulated for a specified planning horizon and dynamic TEP for different stages in a specified planning horizon. There are various mathematical, heuristic and meta-heuristic techniques used to solve the TEP problem starting from linear programming approach [1], dynamic programming [2], interactive method [3], hierarchical decomposition [4], simulated annealing for long term TNEP [5], constructive heuristic algorithm (CHA) applied for solving AC TNEP problem in [6]. Application of CHA in a branchand- bound to solve DC model static TEP problem is presented in [7], Branch-and-Bound algorithm [8], Artificial techniques such as, ANN [9], Fuzzy systems [10], Discrete particle swarm optimization (DPSO) algorithm [11], and Projection-Adapted Cross Entropy method [12], novel differential evolution algorithm (DEA) [13] are reported in literatures respectively. The load is uncertain in the transmission line planning problem is due to many reasons. TEP will be different from that under definite load. Generally the load uncertainty is
Minimize C Subjected to f f n |f | P 0 n P n n P n P n P f 0 (2) (3) (4) (5) (6) ,

CL
(1)
shedding of system and is (US $/MW) component that maximizes the supplied demand at each bus i in the system. The values of and are chosen in such way that there will be no need for additional artificial generation and the optimized solution of P is as high as possible. The optimal solution for TEP problem is when there is no load shedding, i.e. r = 0.
III.
OVERVIEW OF ARTIFICIAL BEE COLONY ALGORITHM
where C , CL , n and represents, the total transmission investment cost in US $, construction cost of each line in branch i-k in US $ and the number of circuits added in each right-way-of i-k. represents set of all right-of-way paths for candidates network expansion. f is power flow in branch iis the thermal limit of the circuit P and P are the real k, f power generation and the corresponding demand at each node are i, and are the voltage angle at i and k bus, n , n the number of existing circuits, maximum number of added circuits in branch, n the total integer number of circuits added to the branch i-k. is the susceptance between buses i and k, P ,P represents the maximum and minimum real power generation, the maximum, lower limit of generation capacity at bus i respectively. Equation (1) represents the total investment cost of new lines added, (2) and (3) represents power balance constraint at each node and voltage balance constraint in each loop both are based on Kirchhoffs law, (4) represents the power flow limit for each branch, (5) represents power generation limit at each bus, and (6) link expansion limit for each branch. B. Modeling considering uncertainty in load The DC-STNEP model considering individual load demand and the load at each bus i between interval [P ,P is represented as [17]: Minimize C , CL n
Nr
Karaboga and Bastruk [25] have introduced the Artificial Bee Colony (ABC) algorithm which is meta-heuristic algorithm based on intelligent behaviour of honey bee swarm. The position of a food source signifies a possible solution of the optimization problem and the fitness solution of the problem a corresponding nectar amount of a food source. It consists of three groups of bees namely employed bees, onlooker bees, and scout bee. Bees going to food source randomly, they carry information and share it with other bees waiting in the hive regarding location and the profitability of that particular food source are called employee bees. The bees waiting in the dance area for making decision to choose a food source based upon information given by the employed bees known as onlooker bees and a bee which carrying out random search around the swarm to find food source is called the scout bee. ABC generates arbitrarily distributed initial population P of Ns vectors of candidate solutions as in (15),
P X , X , . . , XN
T
(15)
Each candidate solution X is a D-dimensional vector containing as many integer-valued parameters in (16) as D is the number of optimized parameters.
X x , , , x , , . . , xD, , i = 1, .........., Ns
(16)
In order to update the new candidate food position, employee bee uses (17) to search for new position from the old in memory
v x x x
NP
(7) (8) (9) (10) (11) (12) (13) (14)
(17)
Subjected to f P r P f n n |f | n n f P P P P 0 r 0 n P P n P 0
where is a random number between [-1, 1], k {1, 2..., Ns} and j {1, 2..., D} are randomly chosen indexes. Onlooker bee selects a food source according to the probability calculated by using (18) associated with that food source pp .
pp
f N f
(18)
where fitness is the fitness value of the solution i, Ns is the number of food source. In ABC, if a position cannot be enhanced during set number of cycles, then it is assumed to be abandoned. This control parameter is named limit. Assume that the abandoned
where r represents the load shedding at each bus i, P represents the maximum and minimum level of and P load at bus i. (US $/MW) is the penalty to penalize the load
source is x and j {1, 2..., D}, then the new food source found out by the scout replaced the abandoned one by using (19)
x x rand 0,1 x x
(19)
The control parameters for GABC algorithm to obtain optimal solution for Garvers 6-bus test system are as follows: 50 employee bees, 500 onlooker bees, limit is 4 and C is 1.5 and maximum number of cycles is taken as 500. IV. RESULTS AND DISCUSSION Static transmission network expansion planning problem is solved for Garvers test systems by applying proposed algorithm and is implemented in Matlab 7.9. To validate the performance of GABC algorithm the results obtained are compared with other optimization methods. The static TEP problem was analyzed for three different cases. The best results for minimum investment cost, Cinv in US $ is obtained by proposed algorithm after 20 trial runs.
A. Garvers 6 bus system Garvers system [1] is used as a first test system in this paper, which comprises of 6 buses, 15 new candidate links, and 760 MW of demand all the necessary system data can be found in [24] as shown in Fig 1. The upper bound of new line in each right-of ways is 4. In this system, bus 6 is a new generation bus that needs to be connected with existing network.
For each candidate source position v is produced and estimated by the artificial bee, its quality is compared with its old one. If its found better than old one, it is replaced by old one and if not old one is retained in memory. In ABC algorithm the solution search equation described as in (17) has good at exploration but poor at exploitation. In order to achieve good optimization performance the exploration and exploitation abilities should be equally balanced. Therefore to achieve this, Eqs (17) is modified to improve the exploitation as follows [23]
vij = x ij + fij * (x ij - x ki ) + ij (yi - x ij )
(20)
Where the term added in (17) is gbest term is the jth element of the global best solution, and is a uniform random number in [0, C] where C is a non-negative constant. The modified ABC was named as Gbest-guided ABC (GABC) algorithm.
A) Implementation of GABC to TEP problem
This section provides application of GABC algorithm to solve STNEP problem as follows:
1) Input parameter and initialization step: After receiving required input parameters such as population size, number of food source, number of employed bees, onlooker bees and limit value. Lower and upper bounds of candidate solution are specified. Algorithm generates randomly distributed initial population of Ns size by (15). In this TEP problem formulation, each candidate solution Xi is
Figure 1. Garvers 6-bus test system [1]
an integer value. X n , , , n , , . . , nD, , i {1, 2..., Ns}, D is the number of possible and nji represents the number of possible lines between each branch j and i. 2) Employee bees serach phase: In this algorithm, to optimize food source employed bees select the position by using (20). Penalty factor method is used to handle the constraints. After calculating the position, the value of objective functions will be calculated and it keeps the best solution obtained in memory. 3) Onlooker bees serach phase: Based on the probabilities, find out from (18), onlooker bees search for food source in swarm, and if the newly obtained food source dominates the old one by using (20) that will update its position. 4) Scout bee serach phase: If a solution found in onlooker search phase cannot be improved further for certain numbers of cycles, then scout bee finds a new source randomly and replace the abandoned source using (19). If the maximum number of cycles is not reached, jump to step 3. The algorithm is terminated when it reaches to its number of cycles.
1) With Traditional Model a) Considering fixed generation and load demand. The optimal results found by GABC algorithm are as follows: Transmission line investment cost Cinv = 200,000 US $ which is same found in [8], [9], [13] and [26] having additional following topologies n2-6 = 4, n3-5 = 1& n4-6 = 2 and with line losses is 55.848 MW respectively. Fig.2 shows the optimal solution which indicates that no lines are overloaded.
Figure 2. Optimal solution for Case 1(a)
b) Considering generation resizing ,i.e. generation values are scheduled between . The optimal solution found in this case is Cinv= 110,000 US $, same as in [8], [26] and [17] with following network configuration added additional n3-5 = 1& n4-6 = 3 and line loss is 52.495 MW. Fig 3 shows the optimal solution which indicates that no lines are overloaded and resizing of generation results in less number of new line addition.The convergence curve characteristic for both the cases is shown in Fig 4 indicates that the proposed algorithm is able get optimal solution within 50 iterations.
TABLE I.
RESULTS OF GARVER 6 BUS SYSTEM FOR CASES 1(a) AND (b)
Results of static TEP Best, US $ Average, US $ Worst, US $ Standard deviation Loss MW Line additions for the best result Number of lines added
Case 1 (a) 200,000 228,800 272,000 36189 55.848 n2-6 = 4, n3-5 = 1 & n4-6 =2 7
Case 1 (b) 110,000 136,500 140,000 7451.598 52.495 n3-5 = 1& n4-6 = 3 4
2) With 5% load uncertainty model with generation resizing In this case 5% uncertainty in demand is considered as shown in Table II. Using this model two good solutions are found by the proposed algorithm after multiple runs whereas [17] found three. The results found are as: a) Investment cost Cinv = 110,000 US $ with the following lines added to the initial network: n2-6 = 3 & n3-5 = 1 and line loss is 49.096 MW. b) Investment cost Cinv = 110,000 US $ with the following lines added to the initial network: n3-5 = 1 & n4-6= 3 and line loss is 52.086 MW.
Figure 3. Optimal solution for Case 1(b)
Table III shows the total supplied demand and individual demand in each solution. Solution 2(b) is the best solution among two because it supplies more demand than the traditional planning. The optimal solution obtained with case 1(b) is same as with case 2(b) which indicates with same investment cost and same line configuration load shedding can be avoided.
TABLE II. ACTUAL AND 5% UNCERTAIN DEMANDS AT EACH LOAD BUS Actual Demand (MW) 80 240 40 160 240 Interv. Demand (MW) [76 ; 84] [228 ; 252] [38 ; 42] [152 ; 168] [228 ; 252]
Bus No Figure 4. Cost convergence curve for Garver 6-bus system 1 2 3 4 5
Table I presents detail results obtained by GABC for both the cases. Results obtained for case 1(a) and 1(b) indicates that generation resizing leads to less investment cost and line losses as compared to fixed generation.
TABLE III. Load Bus 1 2 3 4 5 Total (MW) SOLUTIONS OBTAINED BY GABC Solution 1 (MW) 76 234.214 41.765 153.549 228 733.559 Solution 2 (MW) 84 230.794 42 167.891 251.899 776.585
V.
CONCLUSIONS
The total demands supplied by GABC in two solutions are 733.559 & 776.585 MW whereas by CBGA in [17] found three solutions and with supplied total demands of 764.15, 742.86 and 777.17 MW. Hence the results obtained by the proposed algorithm are competitive as it supplies the total demand needed in traditional planning and also supplies approximately the maximum demand as found by CBGA.
3) With 10% load uncertainty model with generation resizing In this case 10% uncertainty in demand is considered as shown in Table IV. Using this model two good solutions are found by the proposed algorithm. The results found are as: a) Investment cost Cinv = 110,000 US $ with the following lines added to the initial network: n3-5 = 1 & n4-6= 3 and line loss is 51.656 MW. b) Investment cost Cinv = 110,000 US $ with the following lines added to the initial network: n2-6 = 3 & n3-5 = 1 and line loss is 49.625 MW.
TABLE IV. ACTUAL AND 10% UNCERTAIN DEMANDS AT EACH LOAD BUS Actual Demand (MW) 80 240 40 160 240 Interv. Demand (MW) [72 ; 88] [216 ; 264] [36 ; 44] [144 ; 176] [216 ; 264]
This paper presents a mathematical model which considers load uncertainty at each load buses. In this study load uncertainty is expressed with 5% and 10% of base load.The advantage of this model is that it optimizes the demand in each bus individually. GABC is used to solve the problem of DCSTNEP. The results obtained illustrates that considering the uncertainties in the planning period it is possible to know that for the new demand at a specified bus transmission expansion plan is capable to supply all demand with same minimum investment cost without load shedding. REFERENCES
[1] L.L. Garver, Transmission network estimation using linear programming, IEEE Trans Power Appl. Syst. vol. PAS-89, no. 7, pp. 16881697, Sept./ Oct. 1970 Y.P .Dusonchet, A.H. El-Abiad, Transmission planning using discrete dynamic optimisation, IEEE Trans. Power Appl. Syst., vol. PAS-92, no. 4, pp.1358-1371, April 1973 A.O. Ekwue, B.J. Cory, Transmission system expansion planning by interactive methods, IEEE Trans. Power Appl. Syst., vol. PAS-103, no. 7, pp. 1583-1591, July 1984 R. Romero, A. Monticelli, A hierarchical decomposition approach for transmission network expansion planning, IEEE Trans. Power Appl. Syst. vol. 9, no. 1, pp. 373-380, Feb. 1994 R. Romero, R.A. Gallego and A. Monticelli, Transmission expansion planning by simulated annealing, IEEE Trans. Power Appl. Syst. vol. 11, no. 1, pp. 364-369, Feb. 1996 M.J. Rider, A.V. Gracia, R. Romero, Power system transmission network expansion planning using AC model, IET Gener. Transm. Distrib., vol.1, no. 5, pp.731-742, Sept. 2007 R. Romero, E.N. Asada, E. Carreno and C. Rocha, Constructive heuristic algorithm in a branch-and-bound applied to transmission network expansion planning, IET Gener. Transm. Distrib., vol. 1,no. 2, pp. 318-323, March 2007 M.J. Rider, A.V. Garica and R. Romero, Transmission system expansion planning by a branch-and-bound algorithm, IET Gener. Transm. Distrib., vol. 2,no. 1, pp. 90-99, Jan. 2008 Al-Saba Tawfiq, El-Amin Ibrahim, The application of artificial intelligent tools to the transmission expansion problem, Elect. Power Syst. Res., vol 62, no. 2, pp 117-126, June 2002 A.S. Sousa and E.N. Asada, Combined heuristic with fuzzy systems to transmission system expansion planning, Elect. Power Syst. Res, vol. 81, no. 1, pp 123-128, Jan. 2011 Y. X. Jin, H Z. Cheng, J Y. Yan and L. Zhang, New discrete method for particle swarm optimization and its application in transmission network expansion planning, Elect. Power Syst. Res., vol. 77, no. 34, pp. 227-233, March 2007 A. Eshragh, J. Feliar, A. Nazar, A projection-adapted cross entropy (PACE) method for transmission network planning, Int. J. Elect. Power Energy Syst., vol. 2, no.2 pp. 189-208, June 2011 T.S. lm, G.A. Taylor, M.R. Lrving and Y.H. Song, Differential evolution algorithm for static and multistage transmission expansion planning, IET Gener. Transm. Distrib., vol.3 no. 4, pp. 365-384, 2009 A. M. L. da Silva, S. M. P. Riberiro, V. L. Arienti, R. N. Alan, M. B. Do Coutto Filho, Probabilistic load flow techniques applied to power system expansion planning, IEEE Trans. Power Syst., vol. 5,no. 4, pp. 10471053, Nov. 1990. B. G. Gorenstin, N. M. Campodonico, J. P. Costa, M. V. F. Pereira,Power system expansion planning under uncertainty, IEEE Trans.Power Syst., vol. 8, no. 1, pp. 129136, Feb. 1993 J. Choi, T. Tran, A. El-Keib, R. Thomas, H. Oh, R. Billington, A method for transmission system expansion planning considering
[2]
[3]
[4]
[5]
[6]
[7]
[8]
Bus No 1 2 3 4 5
[9]
[10]
[11]
Table V shows the total supplied demand and individual demand in each solution. Solution 3(a) is the best solution among two with zero load shedding it supplies more demand than the traditional planning.
TABLE V. Load Bus 1 2 3 4 5 Total (MW) SOLUTIONS OBTAINED BY GABC Solution 1 (MW) 88 220.234 43.959 175.308 263.983 791.486 Solution 2 (MW) 87.050 262.628 44 148.926 216 758.605
[12]
[13]
[14]
[15]
[16]
probabilistic reliability criteria, IEEE Trans. Power Syst., vol. 20, no.3, pp. 16061615, Aug. 2005. I. de J. Silva, M. J. Rider, R. Romero, C. A. F. Murari, Transmission network expansion planning considering uncertainty in demand, IEEE Trans. Power Syst., vol. 21, no. 4, pp. 15651573, Nov. 2006. P. Wu, H. Cheng, J. Xing, "The interval minimum load cutting problem in the process of transmission network expansion planning considering uncertainty in demand." IEEE Trans. Power Syst., vol. 23, no. 3, pp.1497-1506, Aug. 2008 M. C. da Rocha, J. T. Saraiva, "A discrete evolutionary PSO based approach to the multiyear transmission expansion planning problem considering demand uncertainties", Int. J. Elect. Power Energy Syst., vol. 45, no. 1, pp. 427-442, Feb. 2013 D. Delgado, J. Claro, "Transmission network expansion planning under demand uncertainty and risk aversion", Int. J. Elect. Power Energy Syst., vol. 44, no. 1, pp. 696-702, Jan. 2013 H. Yu, C. Y Chung, K.P. Wong, J.H. Zhang, "A chance constrained transmission network expansion planning method with consideration of load and wind farm uncertainties". IEEE Trans. Power Syst., vol. 24, no. 3, pp. 568-1576, Aug. 2009 W.Gao, S. Liu, L.Huang, "A global best artificial bee colony algorithm for global optimization". J. Comput. Appl. Math. vol. 236, no. 11, pp. 2741-2753, 2012 G.Zhu, S.Kwong, "Gbest-guided artificial bee colony algorithm for numerical function optimization". Appl. Math. Comput. vol. 217, no. 7, pp. 3166-3173, 2010 R. Romero, A. Monticelli, A. Garcia and S. Haffner, Test systems and mathematical models for transmission network expansion planning, IET Proc-Gener. Transm. Distrib, vol. 149, no. 1, pp. 27-36, Jan. 2002 B. Basturk, D. Karaboga, An artificial bee colony (ABC) algorithm for numeric function optimization, In IEEE swarm intelligence symposium, pp. 12-14, May 2006 I.J.Silva, M.J.Rider, R.Romero, A.V.Garcia,C.A. Murari,Transmission network expansion planning with security constraints, IEE Proc. Gener. Transm. Distrib., vol. 152, no. 6, pp. 828-836, 2005
[17]
[22]
[18]
[23]
[24]
[19]
[25]
[20]
[26]
[21]
Wind Turbine Effects on Power System Voltage Fluctuations

A. Erduman, IEEE Student Member, H.C. Klkran*, B. Kekezolu, A. Durusu, IEEE Student Member, M. Tanrven
Department of Electrical Engineering Yildiz Technical University Istanbul / TURKEY
Abstract Every passing day wind energy becomes more indispensable part of power systems. So this increasing require the analysis of wind turbine impact on the grid. In this study, possible effect of wind turbine which is connected to Turkish distribution grid are analyzed. Distribution grid system is modeled with NEPLAN power system analysis simulator program. Finally, results are presented. Keywords-wind turbines; grid connection; voltage fluctation
I.
INTRODUCTION
Global energy demand increase too fast and that push the world into a search of new and alternative energy sources. Wind power is a preferable choice because of its several advantages like easy to build and technological availability. But the increase in wind power rate on grid. For researchers, this problem has become a significant research area. There are many studies in literature which are related to impacts of wind turbines on grid. Soma Biswas et. al. presented a new placement method for distributed generation systems to reduce voltage sag problem [1]. Min Min Kyaw and V.K. Ramachandaramurthy proposed a control system for on-grid wind turbines to control voltage regulation [2]. Mansour Mohsenia and Syed M. Islam compared international grid codes for wind turbine integration [3]. In a power system, providing constant voltage continuously is one of the important criteria to evaluate power systems
TABLE I.
performance. Possible changes in power systems alter bus voltages. Because of this reason, wind turbines effects on voltage must be analyzed before integration of wind turbines to grid. In this study, possible effects of a wind turbines on voltage is examined during wind turbines are connected to the grid at middle voltage level. A part of Turkish Power System is chosen and it is modeled by using NEPLAN power system analysis simulator program [4]. 7 buses of this system have been examined. Results which are obtained from this model have been interpreted under the light of standards. In Section 2, grid codes of wind turbines are defined and limits of voltage variations are presented. In Section 3, power system model is analyzed and the results are interpreted in Section 4. Section 5 concludes this paper. II. STANDARDS
At the moment of a wind turbine connection, several parameters should be considered. Correspondingly, governments have made rules which cover parameters and limits that should be taken into account while a wind turbine is connecting to the grid. In Turkey grid code requirements are stated in TS EN 61400-21 Wind Turbines- Section 21: Measurement and Evaluation of Power Quality Characteristics of Wind Turbines which are connected to grid [5], Electricity Market Grid Regulation Appendix 18 [6] and Pertaining to the Implementation of Unlicensed Electricity Production on Electricity Market Regulation" [7].
VOLTAGE CRITERIA FOR UNLICENSED RENEWABLE ENERGY PRODUCTION IN TURKEY Current Level 16A< Nominal Voltage Level 230 Volt Un Un Maximum Voltage Level 264,5 Volt 1,1xUn 1,1xUn The Longest Cleanup Time (Over voltage) 0,2 sn. Minimum Voltage Level 195,5 Volt 0,9x 0,9x The Longest Cleanup Time (Low voltage) 1,5 sn. -
Low Voltage
Unlicensed
High Voltage
16A> 16A>
In unlicensed energy generating systems, voltage levels are defined according to nominal current. As it can be seen in Table I, voltage range is 195,5 V 264,5 V for systems whose currents are lower than 16 amperes. Maximum voltage level is defined as 1,1*Un (1,1*Nominal voltage) and minimum
voltage level is defined as 1,1*Un (1,1*Nominal voltage) for systems, even if they are low or high voltage systems. Fault clean up time range is defined between 0,2-1,5second.
TABLE II.
VOLTAGE CRITERIA FOR LICENSED RENEWABLE ENERGY GENERATION IN TURKEY
Nominal Operating Voltage
Nominal Voltage Range 0,45xUn < U < 0,70xUn
Longest Fault Cleanup Time 150 ms t(ms)=4000xUn-2100 150 ms t(ms)=1500xUn-150 Minimum Voltage Level 0,9x 0,9x
Licensed
Area 1
Un 0,70xUn < U < 0,9xUn U=0
Area 2
Un 0 < U < 0,9xUn Nominal Voltage Level Un Un Maximum Voltage Level 1,1xUn 1,1xUn Longest Fault Cleanup Time ( Over-Voltage) -
In licensed energy generating systems which are similar to unlicensed energy producing systems, maximum voltage level is defined as 1,1*Un (1,1*Nominal voltage) and minimum voltage level is 1,1*Un (1,1*Nominal voltage). In Table II, the variations of operating voltage and fault clean up time during the fault for licensed energy generating system are showed. III. CASE STUDY
5 buses are load bus which are in Fig 1. B1 and B3 are generator bus. A 60 MW wind turbine is connected to B1 bus.3 generators whose rated power is 200 MVA, are connected to B3. In Table III, loads of the system is given. As it can be seen in Table III, total active power demand is 386 MW and reactive power demand is 44MVAr [9].
TABLE III. Active Power LOAD1 LOAD2 LOAD3 80 35 124 27 120 LOAD PARAMETERS Reactive Power 6 7 18 0 13 Power Factor 0,997 0,98 0,99 1 0,99
In this study it is observed that possible voltage variations while grid connected wind turbines are running. A power system which is in Marmara Region, Turkey, has been modeled in NEPLAN power system analysis simulator and system has examined by using this model. A. Description of Considered Power System Power system which is used in this study, has 7 buses and its nominal voltage rating is 154 kV. This system is in Marmara Region, Turkey. One-line diagram can be seen in Fig 1. B7 bus is the point of connection to grid. [8].
B1 TR1 WTs B3 LOAD5 L3 TR3 B4 GEN TR2 L4 LOAD1 L5 B5 L6 TR4 LOAD2 TR6 LOAD4 LOAD3 GRID TR5 L7 L1 L2 TR7 B6 B7 B2
LOAD4 LOAD5
There are 7 power transformers in the system. Wind turbines are connected to the grid over a transformer whose turn ratio is 30/154 kV. All transformers technical data are given in Table IV. Number B7 bus power is the highest one among the all buses.[8]. In this system, type of conductors are 795MCM and 954MCM which are used in transmission lines. Number L6 and L7 consist of two lines. Systems transmission line data are given in Table V. The longest line of the system is the L5 line whose length is 67,273 km. Beside, number L4 line whose form is 954 MCM, the shortest line and its length is 0,5 km.
Figure 1. Power System which is observed
TABLE IV. Quantity TR1 TR2 TR3 TR4 TR5 TR6 TR7 1 3 2 3 2 1 3 Primer Voltage (kV) 154 154 154 154 154 154 154
TECHNICAL PARAMETERS OF POWER TRANSFORMERS Seconder Voltage (kV) 30 15,75 33,6 31,5 34,5 33,6 34,5 %Ukr(1) 11,6 10 11,6 9,5 12 11,61 12,12 Nominal Power (MVA) 100 220 100 50 100 50 1000 Connection Type YNynO YNynO YNynO YNynO YNynO YNynO YNynO
There are 20 wind turbines which are 3 MW, are connected to B1. VESTAS V90 wind turbine model is used in the system [10]. Wind turbine has a 90m rotor diameter and generates electricity via an induction generator. Fig. 2 shows the relation between wind speed and output power of wind turbine.
TABLE V. Length (km) L1 L2 L3 L4 L5 L6 L7 16 25 33,334 0,5 68,273 2 45,53
TRANSMISSION LINE PARAMETERS

X (ohm/km) 0,379987 0,379987 0,379987 0,403 0,379987 0,379987 0,379987 R (ohm/km) 0,0759974 0,0759974 0,0759974 0,069 0,0759974 0,0759974 0,0759974 C (f/km) 0,009384 0,009384 0,009384 0,09 0,009384 0,009384 0,009384 Figure 2. Power Curve of Wind Turbine
Simulation studies carried out by using real wind data for considered area. Weibull distribution function of used wind speed is shown in Fig. 4[11]. B. Results and Discussions In this paper, the examined power system is modelled and analyzed in NEPLAN power system analyzer program. NEPLAN model of the system has shown in Fig 3.
x >=
1,94
x >=
112,03
x >=
222,13
Lejant
Ar ykl Ekipmanlar Hat Ykler 30,000 kV 154,000 kV 34,500 kV 6,760 kV 1,000 kV 15,750 kV
Figure 3. Model of the System in NEPLAN Power System Analysis Simulator Program
Two conditions, whether wind turbines are connected to the system or not, are observed separately. In addition to this, in case of a wind turbine is connected to B2 and B6 buses, has analyzed. By using this analysis, it is clearly observed that how bus voltages change after wind turbines are connected to the system. Bus voltage variations are obtained for each bus of the system.
Figure 4. Considered wind profile
TABLE VI.
VARIATIONS AT BUSES ACCORDING TO WIND TURBINE CONNECTION

WTs are Connected to B1 (kV) 159,45 157,805 158,527 158,501 155,766 155,783 154 WTs are Connected to B2 (kV) 158,97 157,796 158,482 158,454 155,476 155,493 154 WTs are Connected to B6 (kV) 159,324 157,702 158,534 158,51 155,96 155,977 154 WTs are Connected to B1, B2 and B6 (kV) 159,342 158,07 158,535 158,507 155,585 155,601 154
BUS B1 B2 B3 B4 B5 B6 B7
Without Wind Power (kV) 152,412 153,189 157,259 157,224 152,026 152,043 154
In this study, 60 MW wind turbine was connected to B1, B2 and B6 respectively and in that cases voltage variations in the system are observed. Results of every situation are presented in Table VI. After the connection of wind turbines to the buses, voltage of all buses increased. In Table VI, it can be seen that highest voltage increase occurred at B1 when 60 MW wind turbine was connected to same bus (B1). In this case, voltage increase is %4,62 at B1 bus. It is detected that B1 was the most affected bus after connection of wind turbines. On the other hand, after all process, B7 did not react and stayed stable. The reason of this, B7 is connected to the grid and in other words it is a bus with high short-circuit power. Also, the second least affected bus is B3. The whole buses of the system did not exceed the voltage limits which are stated in standards as +%10*Un (Un : nominal voltage) during the steady-state. In Section 2, it is mentioned detailed that bus voltage variations have to be tested during both steady state and fault conditions. For this purpose, single-phase fault and 3-phase fault conditions are implemented separately to buses which wind turbines are connected.
Figure 6. Characteristic of B2 bus After Fault Occurs
ACKNOWLEDGMENT Authors would like to thank the Yildiz Technical University for full financial support of the project no: 2012-04-02-DOP04. REFERENCES
[1] Soma Biswas, Swapan Kumar Goswami and Amitava Chatterjee, Optimum distributed generation placement with voltage sag effect minimization, Energy Conversion and Management, vol. 53, pp 163 174, 2012. [2] Min Min Kyaw and V.K. Ramachandaramurthy, "Fault ride through and voltage regulation for grid connected wind turbine", Renewable Energy, vol 36, pp 206-215, 2011. [3] Mansour Mohsenia and Syed M. Islam, " Review of international grid codes for wind power integration: Diversity, technology and a case for global standard ", Renewable and Sustainable Energy Reviews, vol.16 , pp. 3876 3890, 2012. [4] http://www.neplan.ch [5] TS EN 61400-21. Rzgr trbinleri - Blm 21: ebekeye bal rzgr trbinlerinde g kalitesi karakteristiklerinin llmesi ve deerlendirilmesi; 2011. [6] Elektrik Piyasas ebeke Ynetmelii, EK-18, EPDK,2013. [7] Say : 28229,Elektrik Piyasasinda Lisanssiz Elektrik retimine likin Ynetmeliin Uygulanmasina Dair Tebli,Mart 2012. [8] TEA.YTDB ,Etd Ve Programlama Daire Bakanl,Trkiye Ulusal Elektrik Andaki Havai Hatlarn Trafolarn ve Genratrlerin Elektriki Karekteristikleri,SA-2004/1 [9] T E A ,YT D B.,Sistem letme Ettleri Mdrl, 2012 PUANT (YAZ) Yk artlarinda Yk Akii, Faz Ve Faz Toprak Kisa Devre Etd ,Se-2012/3 [10] http://www.gov.pe.ca/photos/sites/envengfor/file/950010R1_V90GeneralSpecification.pdf [11] T.Girolu, A.Durusu, A. Erduman, R.Yumurtac, M.Tanrven,Evaluation Of Wind Power Potential And Economic Analysis For atalca In Marmara Region, , NURER, III. International Conference on Nuclear and Renewable Energy Resources, 2012.
Fig. 5, Fig. 6 and Fig. 7 show voltage variations after a single-phase fault and 3-phase fault occurred at B1, B2 and B6 respectively (fault period: 150 ms). Results of this analysis showed that the most affected bus was the B1 when a fault occurred in the system. Bus voltage variations of the system stayed in the limits which are stated in standards, after fault conditions are implemented. IV. CONCLUSION
In this study, voltage variations are investigated while a wind turbine is connecting to the grid. To analyze this variations, an existing system in the Marmara Region of Turkey is chosen and modelled in NEPLAN. Using this model, two conditions are studied, with or without a 60 MW wind turbine is connected to the system. In the light of this analysis, it has been seen that bus voltage increase with the wind turbine. However, bus voltage remained in the limits stated by regulation standards. On the other hand, it is obviously seen that voltage limits are exceeded when a wind turbine with a rated power over than 60MW is connected to system or same wind turbine is connected to a bus with a lower short-circuit power. Therefore, bus voltage variations must be examined whenever a wind turbine is connecting to the grid.
Multi-stage PHEV smart coordination in distribution systems

M. F. Shaaban
Department of ECE University of Waterloo Waterloo, Canada mshaabanuw@gamil.com
E. Akhavan Rezai
Department of ECE University of Waterloo Waterloo, Canada seakhava@uwaterloo.ca
E. F. El-Saadany
Department of ECE University of Waterloo Waterloo, Canada ehab@uwaterloo.ca
AbstractThis paper proposes a new coordination method for plug-in electric vehicles (PHEVs) charging in smart grid. Unlike the previously published work, this method guarantees feasibility of the optimal charge decision. That is because; the proposed method is composed of two stages that aim to maximize the PHEV owners satisfaction and minimize the local distribution company (LDC) operating costs without violating system constraints. The proposed method is simple and easy to be implemented for online applications. Simulation results on a typical distribution network are provided to demonstrate the effectiveness of the proposed method. Index Terms-- Distribution networks, electric vehicle, energy management, smart parking lot.
I(i,k,t), IMAX(i,k) The actual current and maximum allowable current in pu between buses i and k. NT The total number of vehicles that are served during the 24 hours span PDG(i,t), QDG(i,t) The generated active and reactive powers in pu from DG at bus i and time t PG(i,t), QG(i,t) The active and reactive generated powers at bus i and hour t respectively in pu PL(i,t), QL(i,t) The active and reactive power demands respectively at bus i and hour t in pu Ploss(t) The total system power loss in pu at time t PNL(i,t), PPHEV(i,t) The active power demands in pu of the normal load and the PHEV load respectively PPHEV(C(i),t) The active power consumed or delivered in kW by charger C at bus i and time t Sbase The base power for the pu system in kVA SC0(n), SCRCH(n), The initial, reached and desired SOC in % for SCDES(n) vehicle n SCF(C(i),t) The final reached SOC at time t+t due to the decision at time t for the vehicle connected to charger C at bus i SOCI(C(i)) The initial SOC in % of the vehicle connected at charger C and bus i upon plugged-in SOCREQ(C(i)) The required SOC in % for the connected vehicle to charger C at bus i T The duration in min of the sliding window V(i,t), (i,t) VMAX, VMIN X, PCH Y(i,k), (i,k) t The magnitude in pu and angle of the voltage at bus i The maximum and minimum voltage limits respectively in pu The charging decision as a percentage of the allowable charging power and the charger allowable charging power in kW The magnitude in pu and angle of the admittance of the line between buses i and k The success factor for the coordination process measured over a 24 hours span The time step in min used for coordination, which also represents the controller cycle
I. A. Acronyms AER DG LDC NLP PHEV PVDG SOC UNC V2G VCC B.
NOMENCLATURE All-electric range Distributed generation Local distribution company Non-linear programming Plug-in hybrid electric vehicle Photo-voltaic-based DG State of charge Uncoordinated charging Vehicle-to-grid Vehicles control center
Symbols BT(C(i)) The battery capacity in kWh of the vehicle connected to charger C at bus i C(i) Chargers indices at bus i CT(i) fBAT(C(i),t) G(i,k) i, k Total number of chargers at bus i Function that represents the battery characteristics of the vehicle connected to charger C at bus i and time t The conductance in pu of the line between buses i and k. Bus indices
The 3rd International Conference on Electric Power and Energy Conversion Systems, Yildiz Technical University, Istanbul, Turkey, October 2-4, 2013 2
II.
INTRODUCTION
Environmental concerns due to emission levels have arisen in the last few years. Records [1] show that the transportation sector has the second highest emissions share. low or zero emissions vehicles have gained a lot of interest in the last few years due to oil prices and environmental concerns. Due to the recent development of renewable sources of energy and the availability of electric power systems almost everywhere, the electrification of vehicles is considered the best option. Therefore, a good solution is to transfer to low emission vehicles in transportation, where the PHEVs are considered best option. However, studies show [2] that charging can impose a potential risk to the system, if the charging process is not managed properly. This is because, it is anticipated that PHEVs will be crowded in certain geographical areas, which can have significant stress on local distribution equipment. The additional load imposed by high penetration of PHEVs is expected to cause severe consequences, such as [2]: Increased losses. Phase imbalance. Voltage limits violation. Power quality problems. Transformer degradation. Fuse blowouts.
However, the work presented in [6-11] rely on the current status of the system and do not consider all the system technical constraints, such as voltage limits, feeders capacities, and power flow balance. Therefore, the feasibility of the implementing the outcomes is not guaranteed. Moreover, considering a time window in future would improve the efficiency of the proposed methods. On the other hand, some work proposes coordination of PHEV charging in a time window in future, such as [12]. However, the work in [12] does not consider power system technical constraints as mentioned before. In [13], a coordinated charging for PHEV to minimize system losses is proposed, where the system technical constraints are considered. However, the PHEV owner satisfaction is not guaranteed, as the method aims only to maximize the benefits of the LDC. The presented work in this paper aims to fill in some gaps that exist in previous work. These gaps can be summarized as follows: Dealing with each individual vehicle in the system rather than group of vehicles. Consider system technical constraints. Guarantee feasibility of the charging decision through two stage optimization method.
Different solutions are being explored by grid operators to deal with the expected growing penetration levels of PHEVs. These solutions can be summarized into two main categories, as follows: Uncoordinated PEVs charging, this requires upgrading the system infrastructure to meet the excess demand. Some researchers [3] proposed utilizing DG units, which could be economically better solution compared to upgrades. Coordinated PEVs charging, which utilizes the two way communication infrastructure in the smart grid, such as [4]. This will lead to improved performance of the smart grid under high penetration of PEVs compared to uncoordinated charging [5]
The rest of this paper is organized as follows. The problem description and the proposed system structure are presented in Section III. The proposed VCC is presented in Section IV. Section V and VI introduce a sample case study, and results and discussion, respectively. Finally, conclusions are presented in Section VII. III. PHEV COORDINATION SYSTEM The structure of the proposed coordination system is shown in Fig. 1, where the smart grid is divided into territories. Each territory serve one or more parking lot and managed through an aggregator. The aggregator is responsible for collecting vehicles information and normal load point information and sends it to the VCC. The VCC receives the information from each aggregator in the system and run the optimization algorithm, which is explained in next section.
The presented work focuses the second category of solutions, where the PHEV charging/discharging is coordinated to improve the performance of the smart grid. Several works have been proposed for PHEVs coordination. A real-time smart coordinated method is proposed in [6], where PHEV owners preferred charging zones are considered. In [7], a demand side management of PHEVs as dynamically configurable dispersed energy storage is proposed. A V2G control scheme is proposed in [8] to coordinate PHEVs charging/discharging. A method to manage the V2G ancillary services is proposed in [9]. An intelligent method to control PHEV charging loads in response to time-ofuse price is proposed in [10]. In [11], smart metering and demand side management in distribution system are proposed for PHEV charging coordination; however, real-time implementation and power system constraints are not considered.
Figure 1. The structure of the proposed SRTCS
It is assumed that the vehicle driver will enter his vehicle charging code and the parking duration. Then, based on this information and the battery SOC, the maximum final SOC is calculated and the vehicle owner is required to enter the
required final SOC, which should be less or equal to the maximum final SOC. Further, the VCC reaches to optimal charging/discharging decisions, which are sent to aggregators and the process starts again after a time delay to insure equal time step. The outcomes of the VCC are the decisions for the duration of a fixed sliding window [14], as shown in Fig. 2. However, only the decisions for the next time step (i.e. t+t) are implemented and the rest of outcomes are used as initial condition for solving the optimization problem in the next time step; thus, these outcomes are tuned with the new information available in the next time step.
QG (i ,t ) QL (i ,t ) =
V
k =1
(i ,t )V( k ,t )Y(i ,t ) sin( (i ,t )
+ ( k ,t ) (i ,t ) ) i , t (3)
P G(i ,t ) = P DG(i ,t ) , Q G(i ,t ) = QDG(i ,t )
i > 1, t
(4) (5) (6) (7)
PL ( i , t ) = PNL ( i , t ) + PPEV ( i , t )
CT ( i )
i, t
i, t
PPEV ( i , t ) =
C( i )
PCH ( C ( i ) ,i ,t ) X ( C ( i ) ,t )
PCH ( C( i ) ,i ,t ) = f BAT CH ( C( i ) ,t ) (SCF( C ( i ),t ) )
CH ( i ) S base
i, C ( i ) , t
2) Voltage limits: The voltage at each bus should be within the technical limits for all time steps.
Vmin V(i,t) Vmax
i,t
(8)
3) Feeders capacity limit: The feeders currents should be maintained below their capacity limits.
Figure 2. Sliding window
I (i ,k ,t ) I MAX ( i ,k )
i, t
(9)
IV.
PROBLEM FORMULATION
In this section, the proposed two-stage optimization problem formulation is explained. In the first stage, the SOC of the vehicles connected to the system is maximized up to the desired value set by the vehicle owners. Further, the reached SOC levels in this stage are set as constraints for the second stage, which aims to minimize the system losses. This proposed structure guarantee feasibility, as the constraint in the second stage is already known to be feasible solution from the first stage. The proposed optimization algorithm addresses the grid operators concerns, where the customer satisfaction comes first, and then as a second objective, the system costs are minimized as possible. The details of the two stages are as follows: A. Stage 1 In this stage, the objective is to maximize the PHEVs batteries SOC, while maintaining system technical constraints. This problem is defined as NLP, and can be described as:
CT ( i )
4) Charging decision: Currently, the available chargers in the market have only the ability to be enabled or disabled. When enabled, it charges the vehicle battery with maximum allowable power depending on the battery characteristics and the charger ratings. This implies a binary decision variable for the charging decision, which makes the optimization problem mixed integer non-linear programming. However, in this work, it is assumed that the charger can be controlled up to the maximum allowable power. Therefore, the charging decision is assumed to be continuous as in (10). X ( C ( i ) , t ) [0,1] i, C(i ) , t (10) 5) SOC constraints: The vehicles SOC should be limited to the desired SOC entered by the user, as in (11). The vehicles SOC is updated each time step as in (12) and (13). SCF( C( i ) , t ) SC REQ (C ( i ) ) i, C ( i ) , t (11)
SCF ( C ( i ) , t =1) = SCI ( C ( i ) ) + 100 % X ( C ( i ) , t =1) PCH ( C(i) ,i , t =1) ( t / 60 ) / BT ( C ( i ) )
SCF( C ( i ) , t +1) = SCF( C ( i ) , t ) + 100% X ( C ( i ) , t ) PCH ( C(i) , i , t ) ( t / 60) / BT( C ( i ) ) i , C ( i ) ,1 < t
MAX
X
SCF
C( i )
i, C (i )
( C ( i ) ,T )
(1)
(12) (13)
Subject to: 1) Power flow constraints: Power balance constraints at each bus can be implemented as in (2) and (3). The generated power at each bus is the generated power from the DG connected at this bus, as in (4). The total active power consumed by load is the sum of the power consumed by normal load and PHEV load, as in (5). Consumed power at each bus due to PHEV load depends on the charging decision, charger ratings, and the charger efficiency, as in (6). The charger allowable power is assumed to be function in the SOC of the battery. This function depends on the battery characteristics, as in (7).
B. Stage 2 The objective in this stage is to minimize the system losses, while maintaining system technical constraints, where the final reached SOC in the first stage is utilized as constraints. This problem is defined as NLP and can be described as:
MIN
X
P
t =1
loss ( t )
(14) (15)
2 Ploss( t ) = 0.5 G( i ,k ) (V(2 i ,t ) + V( k ,t ) 2 V( i ,t )V( k ,t ) cos( ( k ,t ) ( i ,t ) ) ) i k
PG(i,t ) PL(i,t ) = V(i,t )V(k ,t )Y(i,t ) cos((i,t ) + ( k ,t ) (i ,t ) ) i, t

k =1
(2)
Subject to the same constraints in stage one, except (11) which is replaced by (16). SCF( C( i ) ,T ) = SC FINAL ( C( i ) ) i, C ( i ) (16)
V.
CASE STUDY
To evaluate the proposed PHEV coordination structure, a sample case study was examined on 38-bus system [15], which contains a mix of residential, commercial and industrial customers being supplied from main substation as shown in Fig. 3. The total system peak load is 4.37 MVA. The system under study is modeled in a MATLAB software environment, while the VCC is modeled in GAMS software. The results are sent back to the MATLAB software for the updating of the batteries SOC. The time step of the process is assumed to be 10 min (i.e. t =10 min), and the simulation duration is 24 hours. However, this time step can be reduced significantly depending on the system and the hardware used for implementation. The maximum run time for the proposed algorithm in the system under study is less than 10 seconds, where the hardware utilized for simulation is quad core 2.8 GHz processor and 6 GB RAM. Four parking lots are considered in this system on buses 8, 12, 23, and 32, as shown in Fig. 3. The number of charging locations available in each parking lot varies based on the case study. Three levels of chargers are commonly available for PHEV charging, as shown in Table I [16]. For simplicity, all chargers are assumed to be 2nd level with 7.2 kW rating, and all vehicles are assumed to have all electric range of 50 miles. Moreover, it is assumed that all vehicles batteries characteristics are the same as given in Appendix 1.
Two case studies are introduced. The first case study represents the low penetration of PHEV, and the second case study represents high PHEV penetration. The difference between the two cases is the number of charging spots available at each parking lot.
TABLE I. PHEV CHARGERS RATINGS [16]
TYPE Level 1R Level 2R Level 3R
SPECIFICATIONS 110/120V, AC, 15 -20 amps Does not require installation and can use standard 120v electrical outlet Typical charge times: 8-12 hours 208-240V, AC, 15-30 amps Requires special installation Typical charge times: 3-8 hours Known as DC fast charging 440V, DC, 125 amp, 55kW or higher Requires special installations in the station and the vehicle DC Fast Charging typically returns 50% of a PHEV batterys capacity in under 30 minutes.
Figure 3. 38-bus distribution test feeder
5
A. PHEV charging coordination with low penetration In this case study, the total PHEV penetration is 18% of the total system load. To evaluate the performance of the proposed approach, it is compared to the uncoordinated charging. The uncoordinated approach charges any connected vehicle to the system without considering system technical limits. The normal load of the system is shown in Fig. 4 [17]. The results of the simulation are shown in Fig. 5 and Table II. As shown, both the uncoordinated and the coordinated method achieve 100% successful charging, where the success factor is measured as the mean success of charging for all vehicles in the system over the 24 hours under study, as in (17).
PHEV load (MW)
1.00 0.80 0.60 0.40 0.20 0.00
Time of day Uncoordinated Smart
1 = N T
n = NT SOCRCH ( n ) SOC0( n ) SOC DES ( n ) SOC0 ( n ) n =1
Figure 5. PHEV consumption for low penetration case
(17)
It is shown that the proposed two-stage coordination method can significantly reshape the charging requirements of the PHEVs connected to the system so that during peak the charging is limited and either performed before or after the peak period based on the forecasted data. The proposed coordination strategy charging curve peaks coincide with the valleys in the normal system load curve. The smart charging results in 5.22% less system losses over the day compared to uncoordinated charging. On the other hand, the performance of the uncoordinated charging depends on the arrivals and departures of the vehicles.
TABLE III. PHEV CHARGING COORDINATION OUTCOMES WITH
LOW PENETRATION
B. PHEV charging coordination with high penetration A penetration of 42% of PHEV is assumed in this case. This penetration is beyond the system technical limits if the vehicle charging is uncoordinated. The results for the two approaches are shown in Fig. 6, and Table II.
TABLE IIIIVI. PHEV CHARGING COORDINATION OUTCOMES WITH HIGH
PENETRATION
UNC (%) Feasibility 100% Infeasible
SRTCS 91.11% Feasible
UNC (%) Feasibility 100% Feasible
Proposed 100% Feasible
PHEV load (MW) MW
5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0
Normal load (MVA)
As shown, the proposed approach significantly shifted the PHEV demand to off-peak period as possible using the two stage optimization. Although, the uncoordinated approach achieves a success of 100%, it is not feasible as it violates one or more of the system constraints described in Section IV. On the other hand, the proposed method achieves satisfactory success without violating system constraints. Moreover, the proposed charging method results in much lower system losses, where the system losses increase by 75% in the uncoordinated scenario compared to base case without PHEVs. On the other hand, the losses increase only 29% in the coordinated scenario.
2 1.5 1.00 0.90 0.80 0.70 1 0.60 0.50 0.40 0.30 0.20 0.5 0.10 0.00 0 6 AM 10 AM 2 PM 6 PM 10 PM 2 AM 6 AM 10 AM Time of day Time of day
Time of day Figure 4. Normal load consumption [17]
Uncoordinated
Smart
Figure 6. PHEV consumption for high penetration case
6
VI.
CONCLUSION
This paper proposed a structure to manage the dynamics associated with the coordinated charging of PHEVs in smart parking lots. The two stage optimization problem formulation guarantees feasibility of outcomes, where the customer satisfaction is given the first priority then the system losses come in second priority. However, the customer required level of charge is not guaranteed depending on the grid conditions. Moreover, a sliding window optimization is applied to make use of the forecasted normal load data and coordinate the vehicles charging based on it. Two case studies were simulated for a typical distribution network with different sizes of parking lots. The simulation results prove the effectiveness of the proposed methodology. The proposed structure is simple, fast, and easy to be implemented in practical. VII. APPENDIX The Li-ion battery characteristics in [18] are adjusted to the utilized chargers ratings in the presented case studies in Section V.
8 7 6 5 4 3 2 1 0 0 20 100 90 80 70 60 50 40 30 20 10 0 60 Time (min) Figure 7. Charging power and the SOC of a typical Li-ion battery [ 18] 80 100 120
P SOC 40
REFERENCES
[1] U.S. Enviromental Protection Agency. [Online]. Available: http://www.epa.gov/climatechange/Downloads/ghgemissions/ [2] R. Liu, L. Dow, and E. Liu, A Survey of PHEV Impacts on Electric Utilities, in Proc. ISGT11, pp. 1-8, Oct. 2011 [3] M.F. Shaaban, Y.M. Atwa, E.F. El-Saadany, "PEVs Modeling and Impacts Mitigation in Distribution Networks," IEEE Trans. Power Syst., vol.28, no.2, pp.1122,1131, May [4] H. Khayyam, H. Ranjbarzadeh, V. Marano, Intelligent Control of Vehicle to Grid Power, Journal of Power Source, 2012, pp.1-9 [5] R.A. Verzijlbergh, M.O.W. Grond, Z. Lukszo, J.G. Slootweg, M.D. Ilic, "Network Impacts and Cost Savings of Controlled EV Charging," IEEE Trans. on Smart Grid, vol.3, no.3, pp.1203,1212, Sept. 2012 [6] S. Deilami, A. S. Masoum, P. S. Moses, and M. A. S. Masoum, Realtime coordination of plug-in electric vehicle charging in smart grids to minimize power losses and improve voltage profile, IEEE Trans. Smart Grid, vol. 2, no. 3, pp. 456-467, Sept. 2011. [7] C. Pang, P. Dutta, and M. Kezunovic, BEVs/PHEVs as dispersed energy storage for V2B uses in the smart grid, IEEE Trans. Smart Grid, vol. 3, no. 1, pp. 473-482, March 2012. [8] Y. ota, H. Taniguchi, T. Nakajima, K. M. Liyanage, and J. Baba, Autonomous distributed V2G (vehicle-to-grid) satisfying scheduled charging, IEEE Trans. Smart Grid, vol. 3, no. 1, pp. 559-564, March 2012.
SOC (%)
P (kW)
[9] C. Quinn, D. Zimmerle, and T. H. Bradley, An evaluation of state-ofcharge limitations and actuation signal energy content on plug-in hybrid electric vehicle, vehicle-to-grid reliability, and economics, IEEE Trans. Smart Grid, vol. 3, no. 1, pp. 483-491, March 2012. [10] Y. Cao, S. Tang, C. Li, P. Zhang, Y. Tan, Z. Zhang, and J. Li, An optimized EV charging model considering TOU price and SOC curve, IEEE Trans. Smart Grid, vol. 3, no. 1, pp. 388-393, March 2012. [11] G. Strbac, C.K. Gan, M. Aunedi, V. Stanojevic, P. Djapic, J. Dejvises, P. Mancarella, A. Hawkes, D. Pudjianto, S. Le Vine, J. Polak, "Benefits of Advanced Smart Metering for Demand Response based Control of Distribution Networks," Imperial College, London, 2010. [12] Y. He, B. Venkatesh, and L. Guan, Optimal scheduling for charging and discharging of electric vehicles, IEEE Trans. Smart Grid, vol. 2, no. 3, pp. 1095-1105, Sept. 2012. [13] E. Sortomme, M.M. Hindi, S.D.J. MacPherson, S.S. Venkata, "Coordinated Charging of Plug-In Hybrid Electric Vehicles to Minimize Distribution System Losses," IEEE Trans. on Smart Grid, vol.2, no.1, pp.198-205, March 2011 [14] H. Yifeng B. Venkatesh, L. Guan, "Optimal Scheduling for Charging and Discharging of Electric Vehicles," IEEE Trans. Smart Grid, vol.3, no.3, pp.1095,1105, Sept. 2012 [15] D. Singh, R. Misra, D. Singh, "Effect of Load Models in Distributed Generation Planning," IEEE Tran. Power Sys., vol. 22, no. 4, pp. 22042212, Nov. 2007. [16] Commonwealth Edison Company, "Initial Assessment of the Impact of the Introduction of Plug-in Electric Vehicles on the Distribution System," ILLINOIS COMMERCE COMMISSION, Dec. 2010. [17] E. Lopez, H. Opazo, L. Garcia, L.; P. Bastard, "Online reconfiguration considering variability demand: applications to real networks," IEEE Transactions on Power Systems, vol.19, no.1, pp.549,553, Feb. 2004 [18] F. Marra, G. Y. Yang, Ch. Trholt, E. Larsen, C. N. Rasmussen, S. You, Demand Profile Study of Battery ElectricVehicle under Different Charging Options, in Proc.2012 IEEE Power PES General Meeting (GM).
Lifecycle-Based Design and Optimization of Electrical Motor-Drives Challenges and Possibilities

Toomas Vaimann, Ants Kallaste, Aleksander Kilk
Department of Electrical Engineering Tallinn University of Technology Tallinn, Estonia
Anouar Belahcen
Department of Electrical Engineering Aalto University Espoo, Finland
AbstractThis paper describes the procedure of lifecyclebased design and optimization of electrical motor-drives as well as related methodologies to be developed. As the electric motordrives are the most electrical energy consuming apparatuses, thorough and in depth optimization of the drives during their design process is essential to support energy efficient and environmental friendly development in electrical engineering and industry processes. Lifecycle-based optimization would grant a possibility to evaluate a wider range of variables that affect the environment and energy consummation of the drives. An overview of existing optimization techniques and related literature is presented. Necessity of new methodologies is pointed out. Keywords Design methodology; electric machines; lifecycle; motor drives; optimization.
that a design and optimization procedure for such machines accounting for all the above mentioned factors is possible to implement and it will result in a lifecycle energy and resource efficient electric motor-drives to be used in several sectors of the industry. However, the development of this design and optimization methodology requires both technical design tools and market data. II. OPTIMIZATION CHALLENGES
I.
INTRODUCTION
Nowadays there are increasing global concerns and needs for energy and earth natural resources saving. In what concerns the electrical energy, the most energy consuming apparatuses are the electrical motor-drives with 1067 TWh in 2005 in European countries only, which represents 30% of the total electricity consumption and 427 Mt of CO2 emissions. This use-phase energy is estimated to increase to 1252 TWh by the year 2020 if no measures are taken [1]. Besides the use-phase energy, motor-drives require a large amount of natural resources such as iron, copper, permanent magnet, and other minerals as well as the energy needed to extract, process, manufacture, and distribute these materials. The European legislation on the ecodesign emphasizes the role of the lifecycle of goods from raw material to their final disposal [2]. Induction machines are the most used types in the industry processes, whereas permanent magnet machines are used in more demanding processes, although they are spreading very quickly due to electric and hybrid vehicle technologies. An alternative machine to the induction and the permanent magnet machines is the permanent magnet assisted synchronous reluctance machine or in some case the synchronous reluctance machine. Authors of the paper believe
This paper has been supported by Tallinn University of Technology base financing support fund (project B19 Design and Optimization Methodology for Electrical Machine-Drives).
Several research works dealing with the optimization of electrical machines have been published [3]-[7]. The main results from these works is the appropriateness of an optimization algorithm or another for the electrical machines and the ways to reduce the computation time related to these optimization tasks. In some cases the work is very narrow and concentrated on a given property of the machine such as torque ripple [7]. If the lifecycle energy consumption of electrical machines is to be cut, more general optimization methodologies need to be developed. Such methodologies need to account for the load cycles of the machine and frequency converter as well as the energy used for producing the machine and other related components. From the electric-drive simulation point of view, the coupling of the field and circuit equations in the electrical machine as well as the inclusion of the frequency converter supply in the solution of the finite element simulations seems to be the state-of-art [8]-[16]. These simulation methods are nowadays well documented and possible to implement for the design of electrical motor-drives. Efficient implementation and reduction of these methods and models as well as their application in the optimization of electrical-drives is an important and challenging task that has been planned to be carried out by the authors of the paper in view of new knowledge on the design of lifecycle-optimal electric drives. Electrical machines are dissipating energy and thus heating up during their operation. The losses of these machines can be estimated correctly if the finite element method is used to simulate them. The temperature rise of the machine is one of the most important limiting conditions and thus needs to be accurately estimated. The thermal analysis of electrical
machines and frequency converters can be carried out in two ways, either with finite elements [17], [18] or with thermal networks [19]-[23]. The finite element method is very accurate but rather time consuming, whereas the thermal network does not require large amount of computations and is still accurate enough. One of the drawbacks of the thermal network method resides in the fact that it is not able to identify hotspots. Thus a compromise between accuracy and computing time is required. When dealing with the optimization of electrical machines, one of the most crucial aspects is the computation time. Solving an optimization problem on a single PC is not acceptable, for it will take days or even weeks of computation. The trends in this area are towards distributed computation either by using supercomputers or clusters of PCs known as cluster or cloud computation [24], [25]. The cloud computation can be implemented on exiting PCs within a given organization and taking advantage of the exiting idle computer resources with very little investment, mainly consisting of software cost and its installation. III. OPTIMIZATION POSSIBILITIES
The labor input can also be taken into account as a specific cost:
Pl [ EUR / kWh ]. Wmec
gl =
(2)
This cost does not include the material manufacturing cost as this latter has already been included in the material cost. Usually, the labor cost is very small for series manufacturing but for large-size machines the labor cost can represent a large part of the total cost. Instead of merging the above costs and penalties in a single objective function for the optimization, the methodology we propose is to keep these different factors in their original units and define separate objective functions for the optimization. Such a multi-objective optimization procedure will result in the so-called Pareto-Front of optimal solutions and let the designer the freedom to choose the desired solution among the optimized ones according to the specifications and needs. It is clear that such a methodology requires time-varying market data and geopolitical information about the environment friendliness of certain mining processes. In the EuP directive [1], such task is not included and the materials environmental aspects are left of from the responsibility of the machine designer and manufacturer. IV. ENVIRONMENTAL IMPACT
Electrical motors and drives are dissipative and they waste part of the electrical energy supply in the form of heat. Such wastes of energy are referred to as losses, which are usually expressed in terms of power (kilowatts kW), but can be integrated over the lifecycle of the motor to give the total lost energy in kilowatt-hours (kWh) if the duty cycle of the drive is known. The energy consumed by the motor-drive, including the energy spent on production and disposal of the motor-drive, as well as the losses might be presented in a specific form with respect to the mechanical energy produced:
Wtotal , Wmec
we =
(1)
One of the key aspects of the proposed design and optimization methodology is the assessment of environmental impact of electrical machines throughout their life-cycle. As the electric motor-drives very widely used and are the most electrical energy consuming apparatuses, environmental issues must be taken into account during the design and optimization, although these factors are usually left out of the process. The environmental pollution regarding electrical machine industry starts already in the phase of exploitation of minerals from which metals are extracted to manufacture components needed for the construction of electrical machines (Al, Cu, Fe, etc.) [26]. As permanent magnet electrical machines are used widely and their usage is growing, environmental impact of extraction of rear earth metals, used for permanent magnets must also be noted. The strongest permanent magnets available and usually used in electrical machines are NdFeB (neodymium-ironboron) magnets. The main material used to create these magnets is neodymium (Nd), which is a rare earth metal. In high power wind turbines, the mass of needed NdFeB alloy can be up to 2000 kg [27]. Over 90% of the extraction of neodymium is done in China [28]. The extraction process of this material requires the addition of several strong chemicals. The main problem with this process is that all the toxic waste ends up in nature, as the reservoirs used to collect the waste are very poorly maintained [29]. In addition to raw material production, the negative impact on environment goes on during the exploitation of the machine itself. The polluting during the life-cycle of the machine is continued mainly through development of heat, generation of external magnetic fields, noise, vibration, emissions of volatile substances in electrical materials, loss of lubricants in service etc [26]. It should also be noted that when the unity power of the electric machine is larger, the values of these polluting parameters are higher as well.
where Wtotal is the total energy dissipated and Wmec the produced (or converted) mechanical energy. The specific consumption of materials has to be converted to a specific cost:
gm =
mi
Gi
Wmec
[ EUR / kWh ] ,
(2)
where pmi are specific costs of materials and Gi are material weights. In the literature, the optimization is usually focused on maximizing the power efficiency, and minimizing the material cost without regards to the impact of different materials on the environment. Such a methodology does not give insight into the impact of the used energy and materials on the environment. In the approach proposed by the authors, some environmental factors need to be introduced in the optimization too. They could be penalty coefficients kmi (1/kg) depending on the material and its environmental impact, which leads to the definition of the specific environmental impact
ke =
mi
Gi
Wmec
(3)
Disposal and scrapping of the used machines on the end of their life-cycle also poses and environmental risk. Scrapping of the machines causes problems because of their breakup components, as sorting of materials and their reintroduction into the circuit will result in polluting operations [26]. At the disposal of the electrical machines, the negative environmental impact appears in gas emanations, smoke, dust and other substances resulting from dismantling, in order to return the materials to the economic cycle [26]. As described, there is a remarkable impact on the environment throughout the whole life-cycle of the electrical machines and electric motor-drives, all of which should be taken into account during the design and optimization process. Life-cycle based optimization has been a subject of several researches [30]-[33] and the need for development of such methodologies is rising. V. PROPOSED DESIGN AND OPTIMIZATION METHODOLOGY
The hypothesis of the authors is that a power-efficient motor-drive is not necessarily lifecycle-energy efficient neither it is sustainable. The other hypothesis is that and optimization procedure for sustainable design of such motor-drive is possible to implement and generalize based on the existing methods and algorithms described in the previous section.
Fig. 2. Optimization process of the proposed methodology.
The proposed methodology and the research dealing with it will synthesize new knowledge on how the design of electrical motor-drives should be carried out in view of contributing to sustainability and energy-efficiency. Such contribution is expected through the minimal use of environment-spoiling materials such as rear earth permanent magnets, and better conformity with the operating cycles of the motor-drive. Also it will contribute to the improvement of optimization methodologies especially when applied to complex and coupled problems. Such a methodology could be the use of different algorithm at different levels of the optimization, e.g., Genetic Algorithm at component level and Particle Swarm Algorithm at system level.
Fig. 1. Models to be developed, research areas to be investigated and the interactions between them.
The proposed methodology consists of building an optimization procedure based on different simulation tools of the motor-drive and on knowledge of the energetic contents of the materials used in the drive as well as the knowledge of the impact of different materials on the environment and use these procedures for a motor-drive of approximately 10-40 kW power. The motor drive consists of a permanent magnet assisted synchronous reluctance motor supplied from a fourquadrant frequency converter. The different models to be built as well as the interaction between these models and the final results are shown on Fig. 1. The optimization process of the proposed methodology is shown on Fig. 2. Main challenges of the proposed methodology and their scattering at different level of the research modes are shown on Fig. 3. From the figure one can easily see that the research consists of evenly distributed contributions at the fundamental and applied research levels as well as the sector of marketing and economy.
Fig. 3. Repartition of research challenges according to their contribution in the fundamental, applied and market level.
VI.
CONCLUSION AND FUTURE WORK
Taking into account different legislative regulations concerning energy efficient sustainable solutions and lowering the CO2 emissions, design and optimization of different energy consuming apparatuses is of high importance. As electrical motor-drives are the most electrical energy consuming devices in the electrical energy field, also their optimization procedure must be developed to meet the standards of the future world. For that, lifecycle-based design and optimization methodologies must be researched more deeply than before.
The authors of the paper are engaged with developing an optimization procedure for electrical motor-drives, considering the different above mentioned factors. The proposed procedure will be used for the design of an efficient and sustainable motor-drive consisting of an integrated permanent magnet assisted synchronous motor and frequency converter. The development of the frequency converter itself will not be the part of the study but its appropriate control will be investigated. Further steps of the ongoing research are building of a finite element based electromechanical model of permanent magnet synchronous reluctance machine accounting for the losses; building of an accurate and fast thermal model of permanent magnet synchronous reluctance machine; formation of an upto-date database of materials used in the electrical machines and drives; further development of an optimization procedure enabling distributed computation; construction of a prototype of the sustainable and optimized motor-drive and related measurements to verify the feasibility of the proposed design and optimization methodology. If successful, the research will generate invaluable knowledge that will later serve as background for the optimization of different energy systems such as electric vehicles and smart grid topologies with distributed generation. REFERENCES
Commission regulation (EC) no 640/2009, Implementing directive 2005/32/EC of the European Parliament and of the Council with regard to Ecodesign requirements for electric motors, Official Journal of the European Union. [2] Directive 2009/125/EC of the European Parliament and of the Council, Establishing a framework for the setting of Ecodesign requirements for energy-related products, Official Journal of the European Union. [3] K. Idir, L. Chang, and H. Dai, Error-based global optimization approach for electric motor design, IEEE Transactions on Magnetics, vol. 34, no. 5, pp. 2861-2864, 1998. [4] X. D. Xue, K. W. E. Cheng, T. W. Ng, and N. C. Cheung, Multiobjective optimization design of in-wheel switched reluctance motors in electric vehicles, IEEE Transactions on Industrial Electronics, vol. 57, no. 9, pp. 2980-2987, 2010. [5] P. Ragot, M. Markovic, and Y. Perriard, Optimization of electric motor for a solar airplane application, IEEE Transactions on Industry Applications, vol. 42, no. 4, pp. 1053-1061, 2006. [6] B. S. Rahman, and D. K. Lieu, Optimization of ring magnet geometry for field harmonic control in electric motors, IEEE Transactions on Magnetics, vol. 28, no. 6, pp. 3380-3386, 1992. [7] N. Chen, S.L. Ho, and W. N. Fu, Optimization of permanent magnet surface shapes of electric motors for minimization of cogging torque using FEM, IEEE Transactions on Magnetics, vol. 46, no. 6, pp. 24782481, 2010. [8] O. A. Mohammed, and S. Ganu, FE-circuit coupled model of electric machines for simulation and evaluation of EMI issues in motor drives, IEEE Transactions on Magnetics, vol. 46, no. 8, pp. 3389-3392, 2010. [9] M. D. Otto, and D. V. Otto, Computer simulation of electric motor drive systems including the power electronic network, IEEE Transactions on Industry Applications, vol. 28, no. 5, pp. 1072-1080, 1992. [10] M. C. Costa, S. I. Nabeta, and J. R. Cardoso, Modified nodal analysis applied to electric circuits coupled with FEM in the simulation of a universal motor, IEEE Transactions on Magnetics, vol. 36, no. 4, pp. 1431-1434, 2000. [11] P. Zhou, D. Lin, W. N. Fu, B. Ionescu, and Z. J. Cendes, A general cosimulation approach for coupled field-circuit problems, IEEE Transactions on Magnetics, vol. 42, no. 4, pp. 1051-1054, 2006. [12] S. L. Ho, S. Niu, and W. N. Fu, An equivalent parameter extraction method of transient electric circuit and magnetic field coupled problems based on sensitivity computation of system equations, IEEE Transactions on Magnetics, vol. 47, no. 8, pp. 2068-2075, 2011. [1]
[13] G. Benderskaya, H. De Gersem, and T. Weiland, Integration over discontinuities in field-circuit coupled simulations with switching elements, IEEE Transactions on Magnetics, vol. 42, no. 4, pp. 10311034, 2006. [14] H. De Gersem, R. Mertens, D. Lahaye, S. Vandewalle, and K. Hameyer, Solution strategies for transient, field-circuit coupled systems, IEEE Transactions on Magnetics, vol. 36, no. 4, pp. 1531-1534, 2000. [15] Y. Kano, K. Watanabe, T. Kosaka, and N. Matsui, A novel approach for circuit-field-coupled time-stepping electromagnetic analysis of saturated interior PM motors, IEEE Transactions on Industry Applications, vol. 45, no. 4, pp. 1325-1333, 2009. [16] S. L. Ho, H. L. Li, W. N. Fu, and H. C. Wong, A novel approach to circuit-field-torque coupled time stepping finite element modeling of electric machines, IEEE Transactions on Magnetics, vol. 36, no. 4, pp. 1886-1889, 2000. [17] Y. Huang, J. Zhu, and Y. Guo, Thermal analysis of high-speed SMC motor based on thermal network and 3-D FEA with rotational core loss included, IEEE Transactions on Magnetics, vol. 45, no. 10, pp. 46804683, 2009. [18] J. P. Bastos, M. F. Cabreira, N. Sadowski, S. R. Arruda, and S. L. Nau, A thermal analysis of induction motors using a weak coupled modeling, IEEE Transactions on Magnetics, vol. 33, no. 2, pp. 17141717, 1997. [19] R. Yabiku, R. Fialho, L. Teran, M. E. Ramos, and N. Kawasaki, Use of thermal network on determining the temperature distribution inside electric motors in steady-state and dynamic conditions, IEEE Transactions on Industry Applications, vol. 46, no. 5, pp. 1787-1795, 2010. [20] A. Boglietti, A. Cavagnino, and D. Staton, TEFC induction motors thermal models: a parameter sensitivity analysis, IEEE Transactions on Industry Applications, vol. 41, no. 3, pp. 756-763, 2005. [21] S. Mezani, N. Takorabet, and B. Laporte, A combined electromagnetic and thermal analysis of induction motors, IEEE Transactions on Magnetics, vol. 41, no. 5, pp. 1572-1575, 2005. [22] D. Staton, A. Boglietti, and A. Cavagnino, Solving the more difficult aspects of electric motor thermal analysis in small and medium size industrial induction motors, IEEE Transactions on Energy Conversion, vol. 20, no. 3, pp. 620-628, 2005. [23] A. Boglietti, A. Cavagnino, M. Pastorelli, D. Staton, and A. Vagati, Thermal analysis of induction and synchronous reluctance motors, IEEE Transactions on Industry Applications, vol. 42, no. 3, pp. 675-680, 2006. [24] S. Androutsellis-Theotokis, and D. Spinellis, A survey of peer-to-peer content distribution technologies, ACM Computing Surveys, vol. 36, no. 4, pp. 335371, 2004. [25] L. Youseff, M. Butrico, and D. Da Silva, Toward a unified ontology of cloud computing, Grid Computing Environments Workshop GCE '08, pp.1-10, 2008. [26] D. Necula, N. Vasile, and M. F. Stan, Maintenance a solution to reduce the environmental impact of electrical machines, Electrical and Electronic Engineering, vol. 1, no. 2, pp. 55-59, 2011. [27] P. Fanning, and T. Shelley, Turbine technology gears up, Eureka Magazine, 9 June 2010, www.eurekamagazine.co.uk. [28] J. Pyrhnen, J. Nerg, P. Kurronen, J. Puranen, and M. Haavisto, Permanent magnet technology in wind power generators, in Proc. 19th International Conference on Electrical Machines ICEM 2010, pp. 1-6, 2010. [29] S. Parry, and E. Douglas, In China, the true cost of Britains clean, green wind power experiment: Pollution on a disastrous scale, Daily Mail, 26 January 2011. [30] F. Klocke, D. Lung, R. Schlosser, B. Dobbeler, and S. Buchkremer, Ecological lifecycle assessment of an electric drive for the automotive industry, in Proc. 2nd International Electric Drives Production Conference (EDPC), pp. 1-6, 2012. [31] V. Debusschere, B. Multon, H. Ben Ahmed, and P. E. Cavarec, Life cycle design of a single-phase induction motor, Electric Power Applications, vol. 4, no. 5, pp. 348-356, 2010. [32] M. Torrent, E. Martinez, and P. Antrada, Life cycle analysis on the design of induction motors, International Journal of Life Cycle Assessment, vol. 17, pp. 1-8, 2012. [33] S. Petrushin, and R. Gubajdulina, New principles of mechanical engineering organization, in Proc. 7th International Forum on Strategic Technology (IFOST), pp. 1-4, 2012.
Sequence Hybrid Calculation Technique An Approach for Clustering Power Systems Analysis
P. Wirasanti ! ". #rt$ohann ! %. Hoppe ! H. Saffour ! &. %orton'
South Westphalia (ni)ersity of Applied Sciences*Campus Soest! +,bec-er .ing '! /0101 Soest! (Germany) ' (ni)ersity of 2olton! &eane .oad! 2olton! (U.K)
AbstractClustering power systems approach has been recently discussed as a sustainable strategy for distributed generation. Aim of this concept is to turn a passive distribution network to be an active network by clustering power systems into small areas and establishing downsized conventional control strategy. Consequently, each cluster can be operated based on each area. Obviously, the cluster network structure can be realized as many interconnected grids, which can be resulted in the comple ity of systems analysis. !n order to handle that issue, analysis based on sequence components and hybrid calculation technique is proposed. "his method is able to include the behavior of interconnected cluster into the analysis. "herefore, the cluster analysis results in decoupling power systems analysis. #urthermore, the proposed method is able to solve the unbalanced conditions of distribution e.g. asymmetrical feed in power and untransposed power cable. "he proposed method is verified by comparing the results with three phase $ewton %aphson load flow calculation and &!g'!()$" *ower#actory program. !n addition, the proposed method can be utilized not only for the clustering power systems analysis, but also for the general purpose system analysis. KeywordsClustering Power Systems, Distribution Network, Asymmetrical Three Phase Load low Calculation, !ybrid calculation, Se"uence Com#onents
-ept the main idea of "4TS#=" interconnected grid ; =><. Consequently! the clustering power systems concept intends to cluster the power systems into se)eral areas! called cluster area. This means that each cluster networ- structure! which lin-s to another cluster area! can be described li-e interconnected grids. To clarify the cluster networ-! o)er)iew of clustering power systems concept is displayed in ?ig. . 3t shows only the part of distribution system operator le)el and down to the local le)el. .egarding the cluster concept figure! the networ- structure can be represented as interconnected networ-s. ?urthermore! it can be described by the hierarchical system@ superordinate! ordinate and subordinate le)els. This hierarchical networ- system structure pro)ides the possibility to repeat the networ- buildup process. 3n order to turn the ordinary passi)e distribution system to be the acti)e systems! the clustering approach announces the distribution management system 6&%S7 for the cluster automation application. The application of &%S is the cluster controller! which ta-es place for each cluster area management. To accomplish the &%S functionality! the management and control functions based on cluster concept ha)e been continuously de)eloped. ?or instance! the multi=le)el secondary control approach ; '<! this method adopts the traditional hori8ontal "4TS#=" secondary control function and applies to each cluster area in hori8ontal and )ertical way ; A<. This means that the clustering power strategy downsi8es the traditional control architecture to the distribution le)el. Therefore! the clustering power systems concept can achie)e the tas- of ad$ustment of all generating units within its areas of responsibility. Then it results in the energy balancing within each cluster area. According to another application! which &%S has to be capable of is a cluster management and optimi8ation process. #ne of the most important functions for this process is the cluster load flow analysis. To achie)e the cluster analysis! it can be seen that the analysis method has to be handled with interconnected clusters in order to describe the beha)iors of interconnected cluster e.g. superordinate and subordinate cluster. Therefore! the interconnected )oltages are selected for that purpose because the )oltage is a systems state )ariable. Consequently! this paper presents an asymmetrical three phase power systems analysis using hybrid calculation approach based on sequence components technique. The approach of hybrid calculation is ta-en into account9 because the hybrid
3.
34T.#&(CT3#4
.ecently! &istributed 5eneration 6&57 technologies become more potential contributors of electricity supplied to electric utilities. 3t leads to increase the grid integration ratio of &5. Thus! the trend of decentrali8ed power systems has been considered as a future of energy supply systems. According to the decentrali8ed systems! it can be predicted that the energy flow process can be changed from unidirectional to bidirectional! which means that the power flow can be re)ersed from a distribution le)el to a transmission le)el. %oreo)er! the distribution systems will be changed from a passi)e control area to be an acti)e control area. To o)ercome and reali8e the mentioned issues! many research pro$ects ha)e been established. The clustering power systems approach is also included9 this approach has been introduced and successfully promoted during last two years. The main idea of this concept is to coe:ist the &5 with the con)entional power systems. Therefore! the system structure and the control approach are introduced and de)eloped based on the con)entional system ; 1<. The cluster networ- structure
This research wor- is supported by the 5erman B?ederal %inistry of "ducation and .esearchC and administrated by the BPro$e-ttrDger E,lichF 6PtE7C.
traditional bus admittance matri: has to be firstly modified by sorting groups of -nown parameters and un-nown parameters. To clarify the hybrid concept! ;UL< is defined as a )ector group of -nown )oltage bus and ;UN< is )ector group of un-nown )oltage bus. #n the other hand! ;IL< is )ector group of un-nown current bus and ;IN< is )ector group of -nown current bus. .egarding this assumption! the bus admittance matri: is rearranged and portrayed in 6 7.
[I N ] [Y NN ] [I ] = [Y ] L LN
[Y NL ] [U N ] [Y LL ] [U L ]
(1)
According to rearranged matri:! the admittance group of ;YNN< is a square admittance matri:! which is related to )oltage )ector ;UN<. Similarly! the square matri: ;YLL< is corresponded with the )oltage )ector ;UL<. The rest parts of the sorted admittance matri: are defined with matri: ;YNL< and ;YLN<! respecti)ely! and those two are the transposed matri: of each other. 3n order to obtain the character of hybrid calculation! the current )ectors and )oltage )ectors are twisted! as shown in 6'7
?ig. . #)er)iew of clustering power systems on distribution networ-
calculation method is able to sol)e the mi:ing system9 consists of )oltage sources and current sources model. Since! the cluster approach is intended to deal with the distribution le)el! the sequence components technique is utili8ed for the asymmetrical analysis purpose. The de)eloped algorithm can deal with the unbalanced conditions of load and generation as well as the asymmetrical condition of distribution networtopology e.g. untransposed power cable. ?urthermore! this method is intended to include the beha)ior of interconnected cluster into the analysis. Therefore! the cluster analysis results in decoupling power systems analysis as well. 3n order to ensure and )erify the de)eloped analysis method! the traditional three phases 4ewton .aphson calculation are used as reference for the results comparison. 3n the following sections of this paper! the bac-ground of hybrid calculation! the proposed analysis method including the algorithm flow chart and the )alidation through case study are presented. 33. #G".G3"W #? HH2.3& CA+C(+AT3#4 APP.#ACH
[I N ] [U N ] [U ] = [H ] [I ] L L
(2)
To find out the hybrid matri: ;H<! the equation relationship in 6 7 has to be ta-en into account. The rearrangement of 6 7 based on 6'7 are calculated and written in 6>7 and 617. 3t is able to do that because the equation relationship is a linear equation.
[I N ] = { [Y NN ] [Y NL ][Y LL ] [Y LN ] } [U N ] + [Y NL ][Y LL ] [I L ] (3)
[U L ] = [Y LL ] [Y LN ][U N ] + [Y LL ] [I L ]
(4)
Combining 6>7 and 617 in matri: form! consequently! hybrid matri: can be obtained as portrayed in 6/7. 3t can be noticed that the position 6 ! 7 of hybrid matri: is described by admittance matri: term! position 6'!'7 is an impedance matri: form and the rest of matri: is a dimensionless term. This calls a character of hybrid matri:.
[I N ] [Y NN ] [Y NL ][Y LL ] [Y LN ] [U ] = [Y LL ] [Y LN ] L
As mentioned that the clustering power systems analysis is required! the interconnected )oltages ha)e to be included in the analysis method in order to represent the beha)iors of interconnected clusters ; <! ; ><. Consequently! the hybrid method is selected for that purpose! because the hybrid calculation method is able to analy8e the system with the mi:ing of )oltage sources and current sources ;1<. To clarify hybrid calculation method! it is worthy to illustrate its deri)ation. 5enerally! the classical power systems analysis uses a bus admittance matri: in order to define a system structure. This fact is also a base of hybrid method as well. 2ut the concept of hybrid method is to analy8e the system! which is consisted of mi:ing of )oltage sources and current sources. Therefore! the
[Y NL ][Y LL ] [U N ] (5) [Y LL ] [I L ]
According to character of hybrid calculation! this supports the requirement of cluster systems analysis! which the interconnected )oltages ha)e to be built=in to describe the beha)ior of interconnected clusters. .egards this approach! it has been already e:amined and it succeeds in decoupling system analysis. This decoupling opportunity has been pointed out as an ad)antage of this approach because it can reduce the analysis comple:ity of future power systems. ?urther information can be found in ; < and ; ><. (nfortunately! the pre)ious studies are based on single phase system or
symmetrical systems. This is not the case for future system! since the trend of decentrali8ed power systems has been considered as sustainable energy supply systems. The distributed generation units can cause the asymmetrical problem. ?or e:ample! the feed in power of photo)oltaic systems under 1./-W is allowed to feed in as single phase system ; I<! ; 0<. &ue to that reason! this paper is aimed to handle asymmetrical problem based on hybrid calculation. 3n summary! the approach of clustering power systems and asymmetrical hybrid calculation are a powerful tool to analy8e the entire power systems. 333. P.#P#S"& C+(ST". SHST"%S A4A+HS3S A+5#.3TH%
where d stands for diagonal element! c is compensation or non=diagonal element and 120 represents each sequence components! respecti)ely. As defined that the asymmetrical effect can be e:tracted by compensation term. The compensated part can be written as a )ector! shown in 6J7.
[I N 'A!c ] [U = [H L 'A!c ]
'A !c
[U ]
N 'A ] [ I L 'A ]
(7)
Since &5 units become more potential and are dramatically penetrating in distribution networ-! they can cause the asymmetrical problem due to single phase feed in power in high integration area ; I<. To o)ercome that problem! this paper introduces sequence hybrid method based on clustering power systems approach. 2efore discussion on the proposed analysis method! it is noteworthy to ha)e a short o)er)iew of general asymmetrical power systems analysis. Commonly! the asymmetrical power analysis method! which is focused on distribution networ-! can be summari8ed into three main approaches@ the lateral )oltage drop analysis ;/<! the complete three phase 4ewton .aphson calculation ; J<! and the calculation based on sequence components ;J<! ;I<! ; /<! ; K<. Comparing these three methods! it is ob)ious that the best powerful method is the approach of sequence components because it pro)ides the possibility to transform the system into three decoupling systems@ positi)e! negati)e! and 8ero sequence. Those three systems are able to calculate in parallel process9 it means that it consumes less computational en)ironment and results in fast calculation. According to that ad)antages of sequence components! the combination between sequence components technique and hybrid load flow calculation method is selected for the cluster analysis. Howe)er! the sequence components approach is originally de)eloped for the symmetrical power systems. Therefore! the implementation of this approach for the distribution networ- or asymmetrical networ- needs to be clarified ;K<! especially how the symmetrical components technique can be used for the asymmetrical networ- and non=three phase section. To o)ercome this issue! a compensation technique for asymmetrical component has been already presented and discussed in ;J< and ; /<. 4e)ertheless! this paper points out the compensation technique in different way. 5enerally! the sequence components matri: elements are located only at the diagonal position! when the networ- is symmetrical. This can be implied that the asymmetrical component can be represented by the non=diagonal elements of the sequence components matri:. .egarding this approach! the compensation technique is targeted to e:tract the effect of non=diagonal elements. Thus! 6'7 can be written in sequence components framewor- and e:tracted into two parts! as shown 6K7.
[I N 'A ] [U = [H L 'A ]
Since the compensated part can be written as a )ector and it is utili8ed as subtractions! therefore! equation 6K7 can be rewritten and obtained as symmetrical system as following equation.
[I N 'A ] [I N 'A!c ] = [H [U L 'A ] [U L 'A !c ]
'A !d
[U ]
N 'A ] [ I L 'A ]
(8)
Currently! the asymmetrical components are remo)ed from diagonal part and described as compensation )ector. 3t is ob)ious that 6I7 is described as symmetrical system. Consequently! the sequence hybrid equation can be decoupled based on each sequence framewor-! as illustrates in 607.
[I N ] [I N !c ] [U ] [U ] = [H L L !c
!d
[U ] ]
[I L ]
N
[I N ' ] [I N ' !c ] [U N ' ] [U ] [U ] = [H '!d ] [I ] L' L ' L ' !c [I N A ] [I N A!c ] [U N A ] [U ] [U ] = [H A !d ] [I ] LA L A L A!c
(9)
'A !d
[U ]
] [U N 'A ] + [H 'A !c ] [I L 'A ] [I L 'A ]

N 'A
(6)
According to decoupled sequence hybrid equation and compensation process! the asymmetrical load flow calculation based on sequence hybrid technique is de)eloped and utili8ed for cluster system analysis. 3n order to figure out the proposed algorithm! the flow chart is illustrated in ?ig. '! where lower suffi:es 120 stand for each sequence respecti)ely! abc is for three phase systems and i is a bus number. The upper suffi: k stands for iteration step and 0 is an initial )alue. The transformation matri: from sequence components frameworto three phase system is defined by ;T,i<. 3t is noteworthy to mention that the calculation algorithm is di)ided the calculation into two main parts@ the compensation part and the decoupling part. #nly in the decoupling part the algorithm is separated and calculated based on positi)e sequence! negati)e sequence and 8ero sequence component. The rest of algorithm is calculated based on complete system description. .egarding proposed algorithm! the target )alues or input )alues are defined by two types! -nown apparent power buses ;SNabc,i< and -nown )oltage buses ;ULabc,i<. #b)iously! the load flow solutions are the )oltage of -nown power buses ;UNabc,i<.
?ig. '. Proposed sequence hybrid load flow calculation flow chart
To clarify the proposed algorithm! it is noteworthy to e:plain in steps as following@ ?irstly! the target apparent power ;SNabc,i< is di)ided by initial bus )oltage )ector ;UNabc,i0<! in order to get the initial in$ection current )ector ;INabc,i0<. Afterwards! the initial in$ection current )ector has to be transformed to sequence components framewor- and combined with the -nown )oltage buses )ector ;UL120,i<. This combination will be forwarded to be an initial input of hybrid calculation. The first loop of iteration! the compensation process is not started yet. When the process is entered in the sequence hybrid calculation! the compensation process will be acti)e after the first iteration loop. After the hybrid calculation! the results of all sequence components are combined and forwarded to the compensation bloc-. 3n the same time! they are transformed into three phase system in order to update a new current in$ection )ector +astly! when the load flow solution or un-nown )oltage )ector ;UNabc,i< is con)ergent! then the calculation is stopped. 3n summary! the proposed algorithm of sequence hybrid load flow calculation pro)ides the simplest way to sol)e clustering power systems concept as well as the general asymmetrical load flow calculation. 2y the asymmetrical element compensation process! the system matri: can be
decoupled. This leads to parallel computation process! which results in fast computational time. 3n order to figure out the application of cluster system calculation! the case study is gi)en in the ne:t section. 3G. CAS" ST(&H
To )erify the proposed algorithm and cluster analysis method! the 3""" >J nodes test feeder is ta-en into account as a base networ-. 3n order to represent the character of cluster networ-! the >J nodes feeder is modified! as depicted in ?ig. >. The modified networ- is described with three cluster le)els9 node J00 is defined as superordinate le)el! node J'/ and JJ/ are subordinate le)el. The other modification point is substation model. 3t is replaced with '>A*1.I-G './%GA &yn/ distribution transformer. All cable and load configurations are -ept as pro)ided data ;'A<. As introduced that the cluster analysis is aimed to process based on each cluster! thus it requires the interconnected clusters )oltages to be included into the analysis in order to describe their beha)iors. 3n order to assume or get a good interconnected cluster )oltage! the test system is firstly e:amined by three phases 4ewton .aphson method. The superordinate cluster is defined as slac- bus. 2oth subordinate clusters are defined as in$ection power bus! where subordinate clusters in$ects S2 L 1'M$' -GA and subordinate clusters' in$ects SA2C L KAM $ /-GA. After the calculation! the interconnected clusters )oltages are obtained! respecti)ely. According to those results! the interconnected clusters )oltages are ta-en and applied for proposed cluster analysis method. The interconnected clusters )oltages are stated in table .
TA2+" 333.
+#A& ?+#W ."S(+TS C#%PA.3S#4 #? S"+"CT"& 4#&"S
> Phases 4. 4ode JA' JA1 J'A J'1 J'0 J>A J>' J>1 J>I J1 J1' NGAN ;p.u.< A.0I A.0J0' A.0IA/ A.0I'' A.0KJ/ A.0K11 A.0/IA A.01I1 A.0>0J A.01A A.0I 1 angle GA ;deg< = / .'1KI = / .'11 = / .''A = / .AK/ = / .>0/1 = / ./'1I = / .K>'> = / .JJI = / .I0 / = / .011K = / .'10/
Sequence Hybrid NG2N ;p.u.< A.0I A.0J0' A.0IA/ A.0I'' A.0KJ/ A.0K11 A.0/IA A.01I1 A.0>0J A.01A A.0I 1 angle G2 ;deg< = / .'1KI = / .'11 = / .''A = / .AK/ = / .>0/1 = / ./'1I = / .K>'> = / .JJI = / .I0 / = / .011K = / .'10/
?ig. >. %odified 3""" >J 4odes Test ?eeder TA2+" 3. APP+3"& G#+TA5"S #? C+(ST". 4#&"S
4ode J00 J'/ JJ/
NGAN ;p.u.< .AAA
angle GA ;deg< A.AAA
NG2N ;p.u.< .AAA
angle G2 ;deg< = 'A.AA I0.A>/ I0.'>I0
NGCN ;p.u.< .AAA A.0KA A.0K0
angle GC ;deg< 'A.AA => ./1K => .KJ'
To obser)e the )oltage profile of e:amined interconnected clusters system and to )alidate the proposed sequence hybrid analysis method! the comparati)e load flow results of selected nodes of phase A are shown in table >. 3t is ob)ious that the comparati)e results are identical. %oreo)er! it can be obser)ed that the )oltage profile of e:amined system is )ery unbalance. This can be referred to the description in 3""" test feeder document! since the >J nodes test feeder is defined with )ery unbalanced loading character. Additionally! the proposed algorithm is also aimed to be e:amined with &3gS3+"4T program. (nfortunately! the program cannot e:amine the 3""" test feeder because it is not able to describe the asymmetrical cable parameter into &3gS3+"4TPs cable model. Howe)er! the proposed method is e:amined with &3gS3+"4T! but in different test scenario. 3t is noteworthy to mention that the comparati)e results are identical. 3n summari8ation! this case study e:presses the application of sequence hybrid method for the clustering power systems analysis. Since! the cluster analysis is aimed to operate based on each cluster area! therefore! the )oltage of interconnected clusters needs to be included in order to describe the beha)ior of the interconnected clusters. After in)estigation! the proposed sequence hybrid method can fulfill those aspects. ?urthermore! the comparati)e results pro)ide the reliability of proposed method. This means that the proposed algorithm can function properly. Currently! the proposed sequence hybrid method is already )erified through case study. Thus! the ne:t issue! which is necessary to discuss is algorithm performance.
A.0K' = / .//> A.00' A.0IA = / .'1I A.0I/
After performing the proposed algorithm! the interesting issue which is worthy to discuss is the in$ection power of superordinate cluster node and subordinate cluster nodes. Since! the assumed interconnected clusters )oltages are obtained from the results of 4ewton .aphson analysis! thus! the calculation of slac- in$ected power from proposed sequence hybrid analysis must result in same )alue. "specially! the in$ection power of subordinate clusters has to produce the same )alue because these in$ection powers are gi)en in 4ewton .aphson analysis. The in$ection power of cluster nodes is portrayed in table'. Clearly! the results are identical. The same results of in$ection power are one of )alidations of proposed analysis method.
TA2+" 33. CA+C(+AT"& 34E"CT3#4 P#W". #? C+(ST".S 4#&"S
4ode J00 J'/ JJ/
PA ;-W< I/K.A' A.AA 'A.AA
OA ;-)ar< /I>.0J A.AA /.AA
P2 ;-W< KA .I' 1'.AA 'A.AA
O2 ;-)ar< >I/.0' ' .AA /.AA
PC ;-W< 0AA.11 A.AA 'A.AA
OC ;-)ar< 'K1.I A.AA /.AA

?ig. 1. Performance comparison between > phases 4. and sequence hybrid
The algorithm of three phases 4ewton .aphson method and sequence hybrid method are de)eloped based on %atlab .'A 'a! running on 3ntel Core ' &uo CP( >.AA 5H8 .A% 1.AA52 >'2it=System. The comparison based on iteration step and computational time is illustrated in ?ig. 1. According to performance comparison! it can be noticed that the proposed algorithm need more iteration steps but less computational time than three phases 4ewton .aphson method. This is ob)ious! because the proposed algorithm is basically based on current iteration process! which is generally faster than 4ewton .aphosn process. Howe)er! it is worthy to mention that the load flow algorithms can be speeded up. This is a matter of accuracy or fast calculation. 3t can be obtained by changing the ma:imum allowable error. ?urthermore! the computational time can be more speeded up! when the algorithm is con)erted to another programing language e.g. C code. C#4C+(S3#4 As the integration of &5 based &".s into the distribution networ- is significantly increasing! the challenge of combining large numbers of &5 in the power systems has to be carefully clarified and managed in order to a)oid the system stability problem. To handle that issue! it requires analysis tool to obser)e the problem. Consequently! the proposed algorithm is de)eloped to deal with the problem in the distribution le)el. The sequence components technique is selected and utili8ed for the asymmetrical analysis purpose. The de)eloped algorithm can deal with the unbalanced conditions of load and generation as well as the asymmetrical networ- condition e.g. multi=phase distribution systems. ?urthermore! this analysis method is de)eloped to support the clustering power systems approach. As regards clustering approach! it is aimed to operate based on each cluster area. Therefore! the )oltage of interconnected clusters needs to be included in the analysis to represent the beha)ior of the interconnected clusters. According to this fact! hybrid calculation method is selected to deal with this issue. Since the analysis is intended to include the beha)ior of interconnected cluster into the analysis! therefore! the cluster analysis results in decoupling systems analysis as well. The case study e:amination pro)ides the )erification of the proposed method@ the comparati)e results are identical. This means that the algorithm is reliable and can function properly. 3n conclusion! this proposed sequence hybrid analysis method is not only used for cluster analysis purpose! but it is also able to be utili8ed in the general power systems analysis. .egarding clustering power systems concept! it has been recently promoted as a sustainable strategy for the future grid. This approach gi)es the opportunity to establish the con)entional control and management strategy down to distribution networ-. To -eep this ad)antage! the cluster analysis based on sequence hybrid load flow calculation is de)eloped to be one of cluster applications. The proposed method pro)ides the simplest way to sol)e load flow problem of cluster systems in order to figure out optimum )alue for management process. Therefore! it pro)ides the chance to reali8e the automated cluster management and optimi8ation in order to empower the distribution networ- and entire system. ?inally! this clustering power systems philosophy is a
candidate to be one of the best solutions for the future smart gird applications. ACQ4#W+"&5%"4T This research wor- is supported by the 5erman B?ederal %inistry of "ducation and .esearchC and administrated by the BPro$e-ttrDger E,lichF 6PtE7C. ."?"."4C"S
; < ;'< ;>< "4TS#="! (CT" B#peration Handboo- 3ntroductionC! ?inal )'./ "! '1.AK.'AA1! Euly 'AA1. "4TS#="! (CT" #peration Handboo- = P +oad=?requency Control and Performance! .elease 0.A>.'AA0. Gerband der 4et8betreiber G&4 e.G. beim G&"W! BTransmission Code 'AAJ= 4et8= und Systemregeln der deutschen Rbertragungsnet8betreiberC Gersion . ! 6'AAJ7. Q. Heuc-! Q. &. &ettmann! &. Schul8! "lec-trische "nergie)ersorgung! Gieweg! 6'AAJ7! pp. >K/=>I0. William H. Qersting! &istribution System %odeling and Analysis! C.C Press! 6'AA'7 .. ?. Arritt! .. C. &ugan! &istribution System Analysis and the ?uture Smart 5rid! 3""" Transactions on 3ndustry Applications! Golume@ 1J ! 3ssue@ K pp.@ '>1>='>/A %. A. A-her! Q. %. 4or! A. H. A. .ashid B3mpro)ed Three=Phase Power=?low %ethods (sing Sequence ComponentsC 3""" Transactions on Power Systems! Golume@ 'A ! 3ssue@ > pp.@ >I0= >0J S. P. Thang B?ast Three Phase +oad ?low %ethodsC 3""" Transactions on Power Systems! Golume@ 3ssue@ > pp.@ /1J= //1 P. Wirasanti! ". #rt$ohann! A. Schmelter! &. %orton! BClustering Power Systems Strategy for ?uture &istributed 5enerationC! 3""" SP""&A%! Sorrento! 3taly! &ecember 6'A '7 P. Wirasanti! ". #rt$ohann! %. Hoppe! A. Schmelter! &. %orton! B"mpowering the &ecentrali8ed Power Systems with %ulti +e)el Clustering Power Systems StrategyC! 3AST"& P"S! Phu-et! Thailand! April 6'A '7 P. Wirasanti! ". #rt$ohann! S. Ealoudi! &. %orton! B&ecoupling Power Systems Analysis (sing Hybrid +oad ?low CalculationC! 3""" P"S 3S5T! %anchester! (Q! &ecember 6'A 7 ". #rt$ohann! P. Wirasanti! %. +ingemann! W. Sinsu-tha)orn! S. Ealoudi! &. %orton! B%ulti +e)el Hierarchical Control Strategy for Smart 5rid (sing Clustering Concept.C! 3CC"P= 3nternational Conference on. Clean "lectrical Power .enewable "nergy .esources 3mpact! 3taly! Eune 6'A 7 ". #rt$ohann! P. Wirasanti! W. Sinsu-tha)orn! S. Ealoudi! 4. Alamin! &. %orton! BAnalysis of 3nterconnected Power Systems by Hybrid calculationC! 3C."PO! +as Palmas! Spain! April 6'A 7 W. Sinsu-tha)orn! ". #rt$ohann! %. +ingemann! S. Ealoudi! 4. Hamsic! A. Schmelter! &. %orton! B.ecycling Con)entional Control Strategy and Hierarchy for ?uture &5 ControlC 3nternational Symposium on 3ndustrial "lectronics! 2ari! 3taly! Euly 6'A A7 Q. %. 4or! %. A. A-her! S. 5. B&ecoupled Symmetrical Component ?rame Power System %odels for Smart 5ridC! 3""" P"S 3S5T! %anchester! (Q! &ecember 6'A 7 %. T. Qamh! .. 3ra)ani BThree=Phase %odel and Power=?low Analysis of %icrogrids and Girtual Power PlantsC C35." conference on Power Systems Canada! 6'AA07 .. 5. Wasley! %. A. Shlash B4ewton=.aphson algorithm for >=phase load flowC Euly 6 0J17 2%(! BTahlen 8ur "ntwic-lung der erneuerbaren "nergien in &eutschland ab dem Eahr 00AC! a)ailable at@ http@**www.erneuerbare= energien.de S%A! BG&" Anwendungsregel 1 A/ September 'A 'C! a)ailable at@ http@**www.sma.de 3""" Test ?eeder! a)ailable at@ http@**www.ewh.ieee.org*soc*pes*dsacom *testfeeders*inde:.html
;1< ;/< ;K<
;J<
;I< ;0<
; A<
; <
; '<
; ><
; 1<
; /<
; K<
; J< ; I<
; 0< ;'A<
Transmission Swtiching to Relieve Voltage Violations in Low Load Period

W. M. Guo, Q. Wei
Henan Electric Power Research Institute Zhengzhou 450052, Henan Province, China. 719277394@qq.com
G. J. Liu , Y. Wang
School of Electrical Engineering, Shandong University Jinan 250061, Shandong Province, China
X. K. Zhang
Shandong Electric Power Corporation Jinan 250001, Shandong Province, China
AbstractWith the development of modern power system, extra-high voltage (EHV) AC transmission lines are widely used. Because of large capacitive charging, long EHV AC lines may cause voltage violations in low load period. The leading-phase operation of synchronous generators and switching on shunt reactors are two common measures to relieve voltage violations in low load period. In this paper, we propose that opening some EHV lines in low load period, namely transmission switching is a more economic and efficient measure. The problem of finding an optimal topology to relieve voltage violations in low load period is formulated as a mixed integer non linear programming problem. After some simplifications, the model is converted to an easily solved mixed integer linear program. The modified IEEE 30-bus test system is used to evaluate the applicability and effectiveness of the proposed method. Index Terms-- Transmission switching, voltage violations, EHV power system
constraints. As for switching on shunt reactors, it needs a large amount of investment. Whats more, a considerable percentage of shunt reactors is switched on only during the low load period and most of time they have to remain idle. Compared with above two approaches, opening some EHV lines, namely transmission switching, can relieve voltage violations in a more economical way. When an EHV line is open through control of transmission line circuit breakers, the amount of charging power injection is reduced. In addition, as the power carried by this line is shifted to other remaining lines, other remaining lines will consume more reactive power. To implement transmission switching, only some communication and switching equipment are required, most of which have already been installed in nowadays power system. In fact, optimizing power system operation through transmission switching is not a wholly new problem. In the early 1980s many studies have been conducted dealing with line and bus-bar switching to improve system security, reduce the network loss and/or generation cost, or to relieve voltage violations or line overloads caused by line or transformer outages[2]-[5]. In [6], the problem of finding an optimal generation dispatch and transmission topology to meet a specified load is formulated as a mixed integer linear program, with the binary variables representing the operating state of each transmission line. A significantly saving in generation cost can be achieved if transmission line can be switched offline temporarily compared with the dispatch over a fixed network. More importantly, transmission switching does not intrinsically cause security and reliability problems. In [7], it is proved that transmission switching can reduce generation cost while satisfying strict N-1 standards. In [8], a co-optimization formulation of the unit commitment and transmission switching while ensuring N-1 standards is presented. Our investigation extends this literature by proving the degree to which transmission switching could relieve EHV network voltage violations in low load period. The difficulties of this problem mainly lie in the discrete performance of switching actions and AC constraints, which are hard to be modeled. Using a binary variable indicating the line is in
I.
INTRODUCTION
In order to meet the need of long-distance and largecapability electricity transmission, EHV transmission lines are widely used. While transmitting the load power, EHV transmission lines both produce and consume reactive power. In low load period, the amount of reactive power consumed by EHV transmission lines significantly reduces and the problem of reactive power surplus can be very serious, which may result in network voltage exceeding specified upper limit. For example, it has been reported that the risk of voltage violations is pretty high in some regional power grids of China during the Chinese New Year holiday when the load can decrease as much as 40%. To absorb surplus reactive power, power system operators usually take the following measures, namely the leading-phase operation of synchronous generators and switching on shunt reactors. However, the leading-phase operation of synchronous generators can influence real power generation, resulting in that power system cant operate in the most economical way [1]. Besides, the amount of reactive power absorbed by synchronous generators through the leadingphase operation is very limited due to generator static stability
Project Supported by National Natural Foundation of China (NSFC) (51077087, 51007047,51177091)
service or out of service, in this paper the problem of relieving voltage violations in low load period through transmission switching is formulated as a mixed integer nonlinear program (MINLP). Due to its complexity, this MINLP model is very difficult to solve. In order to apply transmission switching to large scale EHV network, this model is converted to a mixed integer linear program (MIP) based on some simplifications. Indeed, transmission switching itself represents a step change in power systems, which can be considered to be similar to a contingency. Some may argue that transmission switching can cause potential threats to power system security and reliability, e.g., some unstable transients may get triggered. In this paper, opening some EHV lines is suggested only in low load period, when the possibility of the reduction of system security by such switching actions is relatively low. We are not suggesting relieving voltage violations through transmission switching at the expense of secure and reliable power system operations. EHV transmission lines that are open in low load period may be switched back into the system in other heavy load periods. This paper is structured as follows: Section II provides a detailed analysis on relieving voltage violations through transmission switching in low load period using a simple 3bus system. Section III gives a description on the basic MINLP formulation and the MIP formulation for this problem. Section IV presents the results and analysis for the modified IEEE 30-bus system. Section VII summarizes the main findings of this work. II. RELIEVE VOLTAGE VIOLTATIONS THROUGH TRANSMISSION SWITCHING
G 1 2
Fig.1. Three-bus system
TABLE I . LINE DATA From bus 1 1 2 To bus 2 3 3 R(pu) 0.0022 0.0022 0.0111
a
X(pu) 0.0222 0.0222 0.0111
BCAP a(pu) 1.0050 1.0050 0.5025
BCAP =half total line charging susceptance
TABLE II. BUS DATA Bus number 1 2 3 Voltage schedule(pu) 1.03 Pload (pu MW) 0 0.5 0.3 Qload (pu MVAR) 0 0.5 0.3
The reactive power source in EHV power system mainly consists of synchronous generators, transmission lines, capacitors and reactors. Among them, synchronous generators can generate reactive power in normal operation states and absorb reactive power through leading phase operation. Transmission lines both produce and consume reactive power. Capacitors can only generate reactive power and reactors can only absorb reactive power. Here we mainly focus on transmission lines. Generally, the reactive power produced by transmission lines is proportional to the square of the voltage and it can be assumed relatively constant considering that the voltage must be kept within about 5% of nominal voltage. The reactive power consumed by transmission lines is proportional to the square of the current. Since the current in low load period is much smaller than that in heavy load periods, a large amount of reactive power will be generated and the network voltage may exceed the specified upper limit. Besides the leading-phase operation of generators and switching on shunt reactors, opening some EHV lines in low load period can also relieve voltage violations. To give a detailed analysis, a 3-bus system shown in Fig.1 is used as an example. The nominal voltage of this 3-bus system is 500kV and the network voltage upper limit is 525kV, namely 1.05pu. The line parameters are obtained from a real 500kV transmission line. Table I, II give the concrete data of the lines and buses respectively.
Here we focus on the reactive power flow across transmission line and the Newton power flow results are shown in TABLE III.
TABLE III. TRANSMISSION LINES FLOWS Bus i 1 1 2 Bus j 2 3 3 Reactive power flow from bus i to bus j(pu MVar) -2.292 -2.331 -0.641 Reactive power flow from bus j to bus i (pu MVar) 0.141 0.180 -0.48
As can be seen from TABLE IIIthe net reactive power value generated by transmission lines reaches as much as 5.423pu. The sum of reactive power demand is 0.8pu and the generator at bus 1 has to absorb as much as 4.623pu reactive power, which may exceed its reactive power absorption capability. Besides, in the given condition, the voltage of bus 2 is 1.0556pu and the voltage of bus 3 is 1.0565pu, both of which have already been above the specified 1.05pu. To relieve voltage violations, some uneconomic generators have to be started or additional shunt reactors have to be switched on. Considering transmission lines are lightly loaded in low load period, opening some transmission lines wont cause many threats to the security of the system. For example, the line from bus 2 to bus 3 can be open if the remaining two lines could work reliably. After opening the line from bus 2 to bus 3, the reactive power flow is shown in TABLE IV,.
TABLE IV. TRANSMISSION LINES FLOWS Bus i 1 1 Bus j 2 3 Reactive power flow from bus i to bus j(pu MVar) -1.644 -1.852 Reactive power flow from bus j to bus i (pu MVar) -0.5 -0.3
Qk , ji = zk (V j2bk V j2 Bk VV i j g k sin ji + VV i j bk cos ji )
(7) (8) (9) (10) (11) (12)
P +P
g gn k , n
k , n
Pk ,n = Pnd
k , n
From the results, it is known that the net reactive power generated by transmission lines drops to 4.296pu. Now the voltage of bus 2 is 1.0415pu and the voltage of bus 3 is 1.0463pu. Both of them are within the specified range. Under this condition, generator at bus 1 absorbs 3.496pu reactive power, which is only 76% of that when all the three lines are closed. In fact, the network voltage can be very different depending on which line is switched off. The comparison result for the 3-bus system is summarized in TABLE V.
TABLE V. VOLTAGE PROFILE AT EACH BUS Bus number 1 2 3 Voltage when line 2-3 is open(pu) 1.03 1.0415 1.0463 Voltage when line 1-2 is open(pu) 1.03 1.0587 1.0586 Voltage when line 1-3 is open(pu) 1.03 1.0586 1.0611
Q + Q
g gn k , n
k , n
Qk , n = Qnd
k , n
V min Vi V max
min i max
min max Qg Qg Qg
As can be seen from TABLE V, relieving voltage violations in low load period through transmission switching can work only if the right line is open. For example, opening the line from bus 1 to bus 2 or the line from bus 1 to bus 3 will make voltage violations more serious. So it is very necessary to give a systemic search method to find out the sets of lines that should be switched off in low load period. III. FORMULATION AND MODELING
Where, Vi denotes the voltage of bus i; Vnominal denotes system nominal voltage and a typical value is 1.0pu; zk denotes the status of transmission line k(open, zk =0;closed, zk =1); Sk denotes the MVA rating of line k; Pk,ij denotes the real power across line k from bus i to bus j; Qk,ij denotes the reactive power across line k from bus i to bus j; gk denotes the conductance of line k; bk denotes the susceptance of line k; Pg denotes the real power generated by generator g; Pnd denotes the real power demand at bus n; Qg denotes the reactive power generated by generator g; Qnd denotes the reactive power demand at bus n; Vmin and Vmax denote minimum and maximum system voltage; min and max denote minimum and maximum voltage angle; Qgmin and Qgmax denote the reactive power generation limits of generator g. The objective is to minimize the sum of bus voltage deviations. Here, it is assumed that Pg is constant and the control actions are generator reactive outputs and the status of each transmission line. For the sake of simplicity, N-1 security constraints are not included in this model, but there is no reason that they could not be included other than the usual problems with convergence and computation time.
Most of the existing ACOPF methods optimize generator outputs, transformer taps and reactive power compensation over a fixed network to minimize system operation cost or transmission losses. Transmission switching to relieve voltage violations can also be incorporated into ACOPF with a binary variable zk representing the line is open or closed. Considering AC constraints and binary variables, a basic formulation can be defined as follows:
Minimize:
iPQ
V V
i
no min al
s.t.: zk Sk2 ( Pk2,ij + Qk2,ij ) zk Sk2

zk Sk2 ( Pk2, ji + Qk2, ji ) zk Sk2
Pk ,ij = zk (Vi 2 g k VV i j g k cos ij VV i j bk sin ij )
Pk , ji = zk (V g k VV i j g k cos ji
2 j
Clearly, the above model is a mixed integer nonlinear program, which is very complex to solve. As far as we know, existing solvers are mainly to solve mixed integer linear programming problems. As for such a complex MINLP with non-convex constraints, there are no solvers to give a satisfactory optimal solution, especially for large scale EHV (1) power systems. One common approach is to use a linear approximation of this problem. In order to convert the objective into a linear (2) form, two variables + and - are introduced to replace the i i absolute value form in (1). These two variables satisfy the (3) following constraints: (4)
i+ max ( 0, Vi V no min al ) no min al Vi ) i max ( 0, V
(13)
VV i j bk sin ji )
(5)
Thus the objective changes into the flowing expressions,

Minimize:
iPQ
Qk ,ij = zk (Vi 2bk Vi 2 Bk VV i j g k sin ij + VV i j bk cos ij )
+ i
+ i )
(14)
(6)
After a simple analysis, it can be found that the objective (14) is equivalent to (1).
Because transmission lines are lightly loaded in low load period, it is assumed that the possibility of line flow violations is pretty low when a limited number of lines are open. Therefore, constraints (2)-(5), (8) and (11) are ignored. To ensure system security, power flow constraints are checked in the end of problem solving procedure. Thus, (6)-(10) and (12) are incorporated into the last model. Among them, (6) and (7) are nonlinear expressions. For convenience, firstly all transmission lines are assumed to be closed. In (6) and (7), gk can be assumed to be zero as line resistance is much smaller than line reactance for EHV transmission lines. (6) and (7) can be written as:
After a simple analysis, it can be found that zkVi is equivalent to si. A further detailed description can be found in [9].To limit the number of lines allowed to open, we have,
(1 z ) j
k k
(25)
Where, j is the number of lines allowed to be open. To ensure system security, j will be increased gradually so that every time there is only one new line to open, then a transient stability simulation and a Newton power flow program will be run. If there are any security violations, the line will be kept closed and the MIP model will be run again. IV. IEEE 30-BUS TEST CASE-RESULTS AND ANALYSIS To evaluate the applicability and effectiveness of the proposed method, the IEEE 30-bus test system is used. The susceptance of transmission line in the standard IEEE 30-bus test system is not as large as that of real EHV transmission lines. In order to simulate an EHV power system, the susceptance of transmission line is intentionally increased. The modified line data is given in TABLE VI.
TABLE VI . MODIFIED LINE DATA From bus 1 1 2 3 2 2 4 5 6 6 9 6 9 9 12 12 12 12 12 14 16 To bus 2 3 4 4 5 6 6 7 7 8 6 10 11 10 4 13 14 15 16 15 17 BCAP (pu) 0.005750 0.001852 0.001737 0.003790 0.001983 0.001763 0.004140 0.001160 0.008200 0.004200 0.002080 0.005560 0.002080 0.001100 0.002560 0.001400 0.002559 0.001304 0.001987 0.001997 0.001932 From bus 15 18 19 10 10 10 10 21 15 22 23 24 25 25 28 27 27 29 8 6 To bus 18 19 20 20 17 21 22 22 23 24 24 25 26 27 27 29 30 30 28 28 BCAP (pu) 0.002185 0.001292 0.006800 0.002090 0.008450 0.007490 0.001499 0.002360 0.002020 0.001790 0.002700 0.003292 0.003800 0.002087 0.003960 0.004153 0.006027 0.004533 0.002000 0.005990
Qk ,ij = Vi 2bk Vi 2 Bk + VV i j bk cos ij Qk , ji = V j2bk V j2 Bk + VV i j bk cos ji
(15) (16)
As can be seen from (15) and (16), the relation between reactive power flow and bus voltage is nonlinear due to the square of variable Vi and the product of the variables Vi and Vj. Using the method of Taylor series expansion, Vi 2 and Vi Vj can be written as:
Vi 2 = 2Vi 1.0
VV i j = Vi + V j 1.0
(17) (18)
In (15) and (16), cosij and cosji can be assumed to be 1.0 as ij and ji are usually small. Substituting (17) and (18) into (15) and (16),
Qk ,ij = (Vi V j )bk 2Vi Bk + Bk
(19) (20)
Qk , ji = (V j Vi )bk 2V j Bk + Bk
(19) and (20) give the linear approximation of the relation between reactive power flow and bus voltage. From numerical experiments, it is found that they are accurate enough to give a satisfactory solution. Whats more, we will run a Newton power flow program to obtain the exact power flow results after the optimization procedure. Taking the binary variable zk into account, (19) and (20) can be written as,
Qk ,ij = zk ( (Vi V j )bk 2Vi Bk + Bk )
(21) (22)
Qk , ji = zk ((V j Vi )bk 2V j Bk + Bk )
They are nonlinear relations for containing the product of continuous and binary variables. In [9], a method to convert this kind of nonlinear relations into linear forms is proposed. Taking zkVi as an example, it can be found that zkVi can be replaced by si , which is a new continuous variable. si satisfies the following constraints:
V min zk si V max zk
0 Vi si (1 zk )V max
(23) (24)
Firstly, a comparison is made between the bus voltages obtained using (17) and (18) and those obtained using the Newton power flow method, which can be considered as actual value. Taking the original network that all lines are closed as an example, the results are shown in Fig.2.
1.1 1.08 1.06 V o ltag e 1.04 1.02 1 0.98 0.96 1 3
Newton method
using Eq(19)and Eq(20)
As the increase of the number of the lines allowed to open, the sum of bus voltage deviation will decrease, but at a decreasing rate. (See Fig.4).
0.8 0.7
the sum of voltage deviation
0.6 0.5 0.4 0.3 0.2 0.1 0 0 2 4 6 8 the number of lines allowed to open
Fig.4. Sequences of line openings
11 13 15 17 19 21 Bus number
23 25 27 29
Fig.2. Voltage using two methods
As can be seen from Fig.2, the bus voltages obtained using (19) and (20) are quite close to those obtained using Newton method. The average error is 0.00371pu. As shown in Fig.2, we will also run a Newton power flow program after the optimization. Hence, the method based on (19) and (20) are accurate enough to give the voltage variation trend as transmission lines are open. To show the viability and effectiveness of transmission switching in reliving voltage violations, here it's assumed that generators reactive power outputs are constant. Thus the only control action during the optimization procedure is changing the status of transmission line. For further testing the correctness of this solution, these 41 lines are opened one at a time, and a Newton power flow program is run to calculate the sum of bus voltage profile deviation after a line is open. After removing the nonconvergence solutions, Fig.3 gives the top ten minimum deviation solutions.
0.7
V. CONCLUSIONS In this paper it is proposed that transmission switching can be used to relive voltage violations in low load period for EHV power system. As only some communication and switching equipment are required, transmission switching can relieve voltage violations in a more economical way. A basic MINLP formulation is given. To avoid the computational intractability, this model is converted to a mixed integer linear program (MIP) based on some simplifications. This paper is a first step in analyzing the potential benefits of transmission switching to relieve voltage violations in low load period. More researches, particularly in the areas of stability need to be done in the future. REFERENCES
[1] [2] [3] [4] P. Kundur, Power System Stability and Control. New York: McGrawHill, 1994, pp. 417-424. Rolim J G, Machado L G B, "A study of the use of corrective switching in transmission systems," IEEE Trans. Power Systems, vol.14, pp.336341, 1999. Wrubel J N, Rapcienski P S, "Practical experience with corrective switching algorithm for on-line applications," IEEE Trans. Power Systems, vol.11, pp.415-421, 1996. Lee CY, Chen N, "Distribution factors of reactive power flow in transmission line and transformer outage studies, " IEEE Trans. Power Systems, vol.7, pp.194-200, 1992. Granelli G, Montagna M, Zanellini F, Bresesti P, "Optimal network
reconfiguration for congestion management by deterministic and genetic algorithms," Electric Power Systems Research, vol.76, pp. 549-556,
the sum of voltage deviation
0.6 0.5 0.4 0.3 0.2 0.1 0 9-10 27-28 12-15 6-10 10-21 10-20 15-18 27-30 12-16 the line to open 1-2
[5]
Fig.3. the top ten minimum deviation solutions
As shown in Fig.3, the minimum deviation corresponds to opening the line 9-10. Then the number of lines allowed to be open in (25) is set to be 1, the results of the proposed model shows that line 27-28 should be open. The two solutions are different. In fact, opening the line 9-10 is infeasible for the voltage of bus 9 is 1.064pu after the line 9-10 is open, which exceed the upper limit. Consequently, the optimal feasible solution is to open line 27-28, which is identical to the proposed model.
[6] [7] [8] [9]
2006. Fisher E B, ONeill R P, Ferris M C, "Optimal transmission switching," IEEE Trans. Power Systems, vol.23, pp. 1346-1355, 2008. Hedman K W, ONeill R P, Fisher E B, "Optimal transmission switching with contingency analysis," IEEE Trans. Power Systems, vol.24, pp. 1577-1586, 2009. Hedman K W, Ferris M C, O'Neill R P, "Co-optimization of generation unit commitment and transmission switching with N-1 reliability," IEEE Trans. Power Systems, vol.25, pp. 1052-1063, 2010. Jose M. Arroyo, Francisco D. Galiana, "On the solution of the bilevel programming formulation of the terrorist threat problem," IEEE Trans. Power Systems, vol.22, pp. 789-797, 2005.
Experimental Validation on Stator Fault Detection via Fuzzy Logic

Hamid Fekri Azgomi1, Javad Poshtan2
Department of Electrical Engineering Iran University of Science and Technology (IUST) Tehran, Iran E-mail1: hfekri@elec.iust.ac.ir E-mail2: jposhtan@iust.ac.ir
Abstractthe detection of faults in induction motors is becoming increasingly important. The main difficulty in this task is the lack of an accurate analytical model to describe a faulty motor. A fuzzy logic approach may help to diagnose induction motor faults. This work presents a reliable method for the detection of stator winding faults based on monitoring the line current amplitudes. In this method, fuzzy logic is used to make decisions about the stator motor condition. In fact, fuzzy logic is reminiscent of human thinking processes and natural language enabling decisions to be made based on vague information. Therefore, this paper applies fuzzy logic to induction motors fault detection and diagnosis. The motor condition is described using linguistic variables. Fuzzy subsets and the corresponding membership functions describe stator current amplitudes. A knowledge base, comprising rule and data bases, is built to support the fuzzy inference. The induction motor condition is diagnosed using a compositional rule of fuzzy inference. Experimental results are presented in terms of accuracy in the detection motor faults and knowledge extraction feasibility. The preliminary results show that the proposed fuzzy approach can be used for accurate stator fault diagnosis. Keywords; Fault Detection; Induction Motor; Fuzzy Logic; Stator; Experimental Data.
Majid Poshtan
Department of Electrical Engineering American University in Dubai Dubai, UAE E-mail: mposhtan@ aud.edu
stator winding faults [2]. Furthermore, it is worth noting that faults in stator winding are remarkably significant due to the effects they exert on the functional features of induction motors. It is the main reason making it the important subject that should be considered. In the context of fault in induction motor, various methods can be divided into a few general categories of which referred to model based, signal based and smart methods [1]. As for smart methods, it categorized to fuzzy logics, neural networks and genetic algorithm. The fuzzy logic approach can be incorporated to the context of fault detection and in effect to the linguistic concepts of the motor performance in different circumstances and in presence of fault condition; that is to say, it turns the combination of expert knowledge and linguistic concepts into rules that delineate the functional status of motor and deal with the problem of fault detection [3], [4]. The purpose of the present study is to design e detector system to determine the presence of fault in stator winding of induction motor by means of fuzzy logic. To this end, analytic model of induction motor dynamic has been considered in the next section and the model along with the faults in the stator winding come under analysis. In the part III fault detection via fuzzy logic has been reviewed, after that the main stages to design proper fault detector system have been reported. Afterwards it has been shown that with this method, different kinds of fault in stator winding can be detected. The online process of fault detection is one of the most important advantages of this method. Simulation results are presented to verify the accuracy of motor fault detection and knowledge extraction feasibility. The preliminary results show that the proposed fuzzy approach can be used for accurate stator fault diagnosis. Finally, since one of the most important issues in induction motor fault detection is the evaluation of the method using practical test data, in section IV this method is implemented on experimental data and the accuracy of this method is verified.
I.
INTRODUCTION
An induction motor is viewed as the beating heart of the modern industry due to its extensive application, low maintenance cost, easy control as well as high reliability. In modern industrial communities, induction motor has been growingly receiving attention in various industries including power station, automotive, petrochemical, and aerospace. Given the fact that motors are prone to failure, the problem of monitoring and prevention of unexpected failure are incorporated to the monumental challenges we face in industry in spite of high reliability of these machines. Among pressures with which an induction motor may encounter, it can be referred to electrical, mechanical and thermal ones [1]. Researchers have always dealt with the faults seen in induction motors, namely the fault related to stator winding, asymmetry, rotor and eccentricity ones. Their extensive investigations have shown that 30 to 40 percent of faults in induction motors are linked to stator winding. A report, which was offered by the research institution, Electric Power, addressed the issue, saying that almost %38 of the faults in induction motor are
II.
OFFLINE INDUCTION MOTOR DYNAMIC MODELING
A. Healthy model To carry out a dynamic model, it needs to be concentrated on the basic equations of induction motor. As has been mentioned in different references, dynamic equations in a three-phase induction motor with a healthy status without any faults in stator winding stand as follows[5]; The dynamic model of Induction motor with stator winding fault in dq stationary reference frame can be described as follows: Stator equations Rotor equations And speed equation
the currents, voltages and electromagnetic flows by i, v and , the resistance, leakage and mutual inductance by r, and , the phases a, b and c by indexes a, b and c. With regards to these equations, the output of three-phase currents of stator is shown in Fig. 1. As it has been shown in Fig. 1, we see that three-phase stator currents are symmetric ones in healthy condition. B. Faulty model To simulate the fault in induction motor, assuming that some parts of a stator phase become a short circuit which is showed in Fig .2, dynamic equation turns out to be as follows: For stator equations
Where if indicates short-circuit current and represents the degree to which short circuit is going on in winding [6]. Moreover, it should be noted that rotor equations remain unchanged. (3) & (4)
With regards to these equations, current equation in dq reference can be written as follows Where constant coefficients are equal to Also for torque equation
Figure 2. Stator windings with interturn fault on a single phase.

Figure 1. Three-phase stator currents without fault (healthy condition)
The r and s subscripts indicate the values that are associated rotor and stator circuits respectively. Representing
As for the equation of short circuit winding flux And torque equation changes ( )
Finally the currents of stator, rotor, and the winding having fallen short in circuit are calculated as follows: ( )
Figure 4. Three-phase stator currents with 20% fault
As it has been shown in Fig. 3 and Fig. 4, we see that with %10 fault the asymmetry in three-phase currents is less than with %20 fault one. III. FAULT DETECTION OF STATOR USING FUZZY LOGIC
The Constant coefficients in the above equations are as follows With respect to these equations, the results of performed simulations for disabled stator with %10 and % 20 faults have been shown in Fig. 3 and Fig. 4.
The general trend in the problem of fault detection by means of a fuzzy logic is that after receiving the three-phase stator's current signals, first the root mean square (RMS) of each phase is calculated and then entered the system of fuzzy inference [3]. In this inference system, three major sections must be taken into consideration; the first one is related to the input of the system that has to enter the system in a fuzzy mode, the second one is related to fuzzy inference base and, in the end, the final section in which fuzzy output needs to be an actual one by an appropriate de fuzzier. Now, each phase is explained in what follows. In the primary phase as linguistic concepts, the membership functions are determined; that is to say, as we call the rms values of three-phase currents Ia, Ib, Ic, we can create a fuzzy input by the use of linguistic concepts - very small (VS), small (S), medium (M) and big (B) [7]. In order to obtain optimal fuzzy membership functions, the simulation data using by analytical model which had been mentioned in the previous section, are employed. A sample of fuzzy membership functions for each phase of stator current is presented in Fig. 5. As for the external part of the detection system, it has been considered the same amount as in internal part. We consider the "Stator Condition" (SC) as suitable output and constitute an appropriate membership function of output using linguistic concepts; good (G), damaged (D) and seriously damaged (SD). A sample of these functions is showed in Fig. 6. The important part of fuzzy fault detection system is a rule base as well as inference engine, where rules are derived from expert knowledge and come under analysis with respect to input and output membership functions presented on a fuzzy structure. Therefore, regarding how motor will operate, we need to write rules in the form of If-Then one with respect to data basis such as:
Figure 3. Three-phase stator currents with 10% fault
Figure 5. The sample of membership functions for input current
Figure 7. Stator condition in healthy mode
Figure 6. The sample of membership functions for stator condition (SC)
If (Ia is VS) Then (SC is Seriously Damaged)
In the same way, and with respect to motor alignment in a good state, and with the deletion of common and unnecessary rules from all states, we can arrive at 14 optimal rules that have been shown in Table. I. In what follows, the results obtained from performed simulations in the process of fault detection can be observed. (Fig. 7 - Fig. 9)
TABLE I. RULE BASE OF FUZZY FAULT DETECTOR SYSTEM
Figure 8. Stator condition in faulty mode (% 10)
If part Rule Number 1 2 3 4 5 6 7 8 9 10 11 12 13 14 Ia VS VS VS B B M S M S S S M S M S S M S M S M B M S S S M M S S Ib Ic
Then part SC SD SD SD SD SD SD G G D D D D D D
Figure 9. Stator condition in faulty mode (% 20)
As observed, in the first state - Fig. 7- the result obtained from a state without any failure is shown, which is obviously agreed upon with respect to the output membership function In Fig. 8, the status of stator is declared to be damaged as had been figured In the end, the status of failure is detected as much of %10 correctly in Fig. 9. It is worth noting that the current fluctuations at the onset of outputs because of the time between transient state to the steady state. IV. EXPERIMENTAL VALIDATION OF PROPOSED METHOD
In this section, the evaluation of the proposed method is presented using experimental data. These data have been taken from a practical induction motor that is shown in Fig .10. Three phase current signals are presented in Fig. 11. As it can be observed, the noisy signals are because of the natural noise in the laboratory condition.
Also these signals in the presence of the Interturn faults have been shown in Fig .12 and Fig .13.
As it can be observed in these figures, with %10 fault ,the asymmetry in three-phase currents is less than with %20 fault one. The output signal of fault detector system (Stator Condition) has been shown in Fig .14 Fig .16. As it can be seen in Fig. 14, the SC index is approximately between 0.26 and 0.28 which belongs to Good (G) membership function.
Figure 10. Experimental Setup
Figure 14. Experimental output result (healthy mode)
Figure 11. Experimental result (three phase currents in healthy mode)
Figure 15. Experimental output result (%10 fault mode)
Figure 12. Experimental result (three phase currents in faulty mode-%10)
Figure 16. Experimental output result (%20 fault mode)
In Fig .15, the SC index has reached to estimated value of [0.40, 0.45] which referred to Damaged (D) state. This alteration is due to %10 fault occurrence in stator winding at 5 sec.
Figure 13. Experimental result (three phase currents in faulty mode-%20)
Finally, the SC index is 0.85 in Fig .16 which refers to Seriously Damaged (SD). This alteration is due to % 20 fault occurrences in stator winding at 5 sec.
These figures show that motor is in a healthy condition and after 5 seconds, %10 and %20 of one phase winding goes to short circuit. (Fig. 12 and Fig. 13)
V.
CONCLUSION
[2]
In this paper, the realization of the fault in stator winding in a three-phase induction motor has been considered. In the first state, the induction motor was simulated by dynamic equations to that effect. Afterwards, the equations were revisited by accounting faults in one of the phases. As for the issue of fault realization, the fuzzy logic and its application in clustering have been used. The advantages of this method are the high accuracy, easy implementation and independence to motor model during the fault detection process. It is Noteworthy that the analytical modeling of induction motor that is explained in this paper, is an offline process which leads to knowledge extraction of different situations of the motor conditions. This knowledge is used to generate proper fuzzy fault detector system. however, it should be noted that its capability of being adjusted to various conditions of a motor is among the extremely practical advantages of such method in modern industry; which means that we are able to meet the requirements of the objective through making a slight change in the mechanism of fault realization in case motor parameters undergo changes at any respect. Finally, evaluation of the proposed method has been approved using experimental data and its accuracy was verified.
[3]
[4]
[5] [6]
[7]
[8] [9]
[10]
[11]
REFERENCES
[1] S. Nandi, H.A. Toliyat, and L. Xiaodong, "Condition Monitoring and Fault Diagnosis of Electrical Motors-A Review, " IEEE Transactions on Energy Conversion, Vol. 20, no. 4, pp. 719-729, 2005.
[12]
[13]
A. Siddique, G.S. Yadava, B. Singh, "A Review of Stator Fault Monitoring Techniques of Induction Motors, " IEEE Transactions on Energy Conversion, Vol. 20, no. 1, pp. 106-114, 2005. P.V. Jover Rodriguez and A. Arkkio, "Detection of Stator Winding Fault in Induction Motor Using Fuzzy Logic, " Applied Soft Computing, Vol. 8, no. 2, pp. 1112-1120, 2008. S.M.A. Cruz and A.J.M. Cardoso "Diagnosis of Stator Inter Turn Short Circuits in DTC Induction Motor Drives, "IEEE Transaction on Industry Applications, Vol. 40, no. 5, pp. 13491360, 2004. P.C. Krause, O. Wasynczuk, S.D. Sudhoff, "Analysis of Electric Machinery, "IEEE Press, New York, 1996. R.M. Tallam, T.G. Habetler, and R.G. Harley, "Transient Model for Induction Machines with Stator Winding Turn Faults, "IEEE Transactions on Industry Applications, Vol. 38, no. 3, pp. 632-637, 2002. F.S.V. DAngeloa Marcos, M. Palharesb Reinaldo, H.C. Ricardo Takahashic, H. Loschid Rosangela, M.R. Lane Baccarinie, M. Caminhasf Walmir "Incipient Fault Detection in Induction Machine Stator Winding Using Fuzzy-Bayesian Change Point Detection Approach, " Applied Soft Computing, Vol. 11, no. 1, pp. 179-192, 2011. L.A.Zadeh, "Fuzzy sets "Information and Control, vol.8, pp. 338-353, 1965. M.S. Ballal, Z.J. Khan, H.M. Suryawanshi, and R.L. Sonolikar, "Adaptive Neural Fuzzy Inference System for the Detection of Inter Turn Insulation and Bearing Wear Faults in Induction Motor, "IEEE Transactions on Industrial Electronics, Vol. 54, no. 1, pp. 250-258, 2007. R. Isermann, "Model-Based Fault Detection and Diagnosis: Status and Applications, "Annual Reviews in Control, Vol. 29, no. 1, pp. 71-85, 2005. F. Bagheri, H. Khaloozadeh and K. Abbaszadeh, "Stator Fault Detection in Induction Machines by Parameter Estimation Using Adaptive Kalman Filter, "Iranian Journal of Electrical and Electronic Engineering , Vol. 3, no. 3, pp. 72-82, 2007. J. Penman, H. G. Sedding, B. A. Lloyd, and W. T. Fink "Detection and Location of Inter turn Short Circuits in The Stator Windings of Operating Motors, "IEEE Transactions on Energy Conversion, Vol. 9, no. 4, pp. 652658, 1994. R. Tallam, et al. "Neural network based on-line stator winding turn fault detection for induction motors, "IEEE Industry Applications Conference, Vol. 1, pp. 375-380, 2000.
A combined approach to multi-objective optimization of capacitor placement in radial distribution networks

Igor Trach
Department of modeling power systems Institute of Electrodynamics NAS Ukraine Kyiv, Ukraine E-mail: igor.trach@ied.org.ua
Abstract A problem of capacitor placement in nodes of radial distribution networks is multi-objective because it has been formulated by using several objectives. The main criteria for capacitor placement and options selection are usually considered as the following functions: minimization of active losses and minimization of investment cost. The problem solution does not always give a positive result in terms of engineering logic, the need to consider imprecision of technical information, specific features of the networks requirements, inadequate pricing in conditions of market economics in transition periods etc. Therefore, an additional third criterion, which is based on the Technical-Coenose approach, is proposed in order to obtain a specific solution for radial distribution network. This criterion requires ensuring maximum compliance of capacitor parameters (such as their conductivities, capacities or power) to the specifically normalized hyperbolic rank-parameter dependence. The proposed combined algorithm uses three mentioned criteria in order to find an acceptable solution in the space solutions formed by method of concessions by making the permissible deviations from the optimal value of each criterion. The proposed method allows to find the vector of ranked capacitor conductivities with specific nonlinear dependence, such as hyperbole. The comparison is made between two approaches: usage of two main criteria and proposed combined algorithm. Calculation procedures are based on the modified version of Particle Swarm Optimization approach. The medium voltage distribution networks with balanced loading and radial configuration with possibility of installation of parallel compensation are considered for analyses. A radial distribution feeder with 27 nodes is used as an example to show the proposed method of capacitor placement. The results show that obtained solutions of shunt capacitor banks placement satisfy the formulated problem and show effectiveness of the prospect of practical purposes applying. Keywords Capacitor placement, Radial network, Particle Swarm Optimization, Technical coenosis, Method of concessions, Multi-objective optimization.
Yuri Zubiuk
Electrosphere.ltd. Kyiv, Ukraine E-mail: yurizua@gmail.com
additional limitations when choosing an objective function. In this paper, the method of CB choice and placement is developed based on the combined approach, which includes the combination of traditional optimization criteria, additional criterion (based on the Technical-Coenose approach presented below), method of concessions and modified PSO-method. II. PROBLEM METHOD DESCRIPTION AND SOLUTION
The first main criterion for capacitor placement and selection options usually considers the following function used to minimize active losses [1]:
P =
P min
i i =1
(1)
where - P the total active power losses and
P
i =1
losses in the i node of distribution system, n number of nodes. The function (1) may have the following constraints: The voltage magnitude at each node must be maintained within the limits:
U min U i U max
(2)
where U min , U max the minimum and maximum allowable voltage. CB reactive power in the i node must have the following limits:
Qc min Qci Qc max
(3)
where
Qci
CB reactive power in the i node.
current I i in the i branch after CB installing must satisfy the branch nominal magnitude:
I.
INTRODUCTION
A reduction of active power losses is one of the major problems of radial distribution networks. A significant part of active power losses occurs due to the flow of reactive currents. An installation of capacitor banks (CB) at certain points of radial distribution feeders improves the voltage profile at the nodes, and reduces active power losses. That is why the problem to find an optimal sizing and placement of capacitors is very important. And the network (feeders) regimes are
I i I nomi
[1]:
(4)
The second main criterion is the cost function defined as

Bmin = K p P +
K
i =1
nc
ciQci +
K
j =1
nl
Lj L j min
(5)
where K p the annual cost per unit of active losses P ($/(kW*year)), Kci the annual cost of reactive power Qci corresponding i CB ($/(kVAr*year)), nc total number of CB, K Lj - the annual cost of conductor in j section ($/km), Lj the length of conductor in j branch (km), nl total number of branchs. Analysis of problem solutions using criteria (1) and (5) shows that they do not always give a positive result in terms of engineering logic, the need to consider imprecision of technical information, specific features of the networks requirements, inadequate pricing in conditions of market economics in transition periods etc. Therefore, additional criteria and methods taking into account these aspects are needed. For this purpose it is proposed to use the concession method approach for multi-objective optimization. The application of concession method allows to theoretically find the quasi-optimal solution, but it can be interpreted as one which can be used as optimal for practical purposes. The concessions method is used to solve multicriteria problems (1) and (5) with constraints (2)-(4). Concessions method provides a set of solutions of (1) and (5), taking into account the permissible concessions in the following form: Criterion of losses (L-criterion).
H (r ) = H1 / r , r = 1,2,..., d
(8)
where r - parametric rank; H1 parameter value of individual with rank r = 1; rank factor characterizing the degree of curve distribution steepness, d - total number of ranks. Analysis of numerous electrotechnical systems showed that the are most effective when 0 , 5 1 , 5 [3]. According to (8), the vector of ranked CB conductivities has to be selected in correspondence with the following condition: yTC( r )= yTC(r=1)/r, ( 9)
The set of solutions YLost is formed from n-dimension vectors yLost (y1, y2, ..., yn), which are within the range of permissible concessions 1 of minimal active losses (1).
Pmin P (1 + 1)Pmin
Application of TC-approach (9) allows to formulate an additional criterion for vector of CB conductivities y(y1,y2,,yn) choice: 1) similarity of searching vector y of ranked CB conductivities to hyperbolic dependence (9), 2) minimization of the magnitude of CB conductivity of the first rank y(r=1). In this case, the Technical-Coenose approach takes the form of the following criterion: Criterion of Technical Coenose approach (TCcriterion): YTC YCost,
nor yTC
nor ( r )= yTC (r =1) /r ,
(6)
In our case, the variables are the conduction capacitors y1, y2, ..., yn. Criterion of cost (C-criterion). The set of solutions YCost is formed within the set of solutions YLost YCost YLost. The set of solutions YCost consists of vectors yCost (y1, y2, ..., yn). Vectors yCost are in the range of permissible concessions 2 of minimum cost min :
min B (1 + 2 ) min
|ydiff|=
(10) |ymax |,
nor ( yTC y nor )2
0 |ydiff| 3
nor yTC ( r =1)
min
(7)
where |ydiff| - the distance between normalized vectors nor | yTC y nor |, 3 - permissible concessions of the distance between normalized vectors. The unit length of normalized nor |=1, | nor |=1. The concession allows to vectors is | yTC y 3 estimate the similarity of searching vector y of ranked CB conductivities to hyperbolic dependence in accordance with TC- riterion. The effective technique for solving the optimal CB placement problem is Particle Swarm Optimization (PSO) because of its advantages such as simple implementation, small computational load, and fast convergence [4]. In the PSO-method each particle moves in the multi-dimensional set of solutions. Moreover, the particle position xk is determined by the interaction with all other particles of swarm. All particles move to the global extremum; the process occurs at the end of the relevant performance of breakpoint criteria. One of the classical optimization algorithms of continuous nonlinear functions, namely A modified particle swarm
Also the set of solutions YCost can be determined separately from YLost, as an intersection of the set of solutions YLost=YLost YCost . Technical-Coenose approach [2, 3] is used as an additional technical criterion, according to which the vector of CB conductivities yTC (y1, y2, ..., yn) has the form of a ranked parametric distribution that corresponds to hyperbolic Hdistribution of certain quantitative characteristics. The system of installed CB is considered as the system of TechnicalCoenose type. At the same time, in the selected set of CB, the type is considered to be their certain numerical parameter: rated power, conductivity or capacitance. TC-approach is based on bringing the ranked parametric distributions to the specific hyperbolic dependence [2, 3]. By analogy to biological coenoses the electrotechnical systems can be considered as TC [2], where a certain family of electrical products (for example CB in our case) can be arranged by a particular species-forming parameter:
optimizer [5], is used in this paper. To calculate the position and velocity for n-dimension of m particles, the following expressions are used at each k iteration: vk+1 = wvk + a1rnd(pbestk - xk) + a2 rnd (gbestk xk), xk+1 = xk + vk+1 where xk - the current position of particle, vk - particle velocity, a1, a2 - constant acceleration, pbest best position found by particle, qbest - best position found by all particles, w - inertia factor, rnd - random number between 0 and 1, m number of particles. The feature of this PSO-algorithm is the ability to control the speed of convergence process using the number of particles in the swarm and the inertia factor. After PSOmethod modification, intermediate results of each iteration computation xk are not discarded, but are grouped in the appropriate set of solutions. In addition, at the beginning of the process a suitable total number of particles and the ratio of inertia are chosen. As a result of the modified PSO-method, a set of solutions is formed, where each m particle corresponds to the ndimensional vector of CB conductivities y(y1, y2, ..., yn) consisting of n CB set out in the relevant feeder nodes. A combination of criteria (6), (7) and (10) gives the following multi-objective algorithm for CB placement in radial distribution networks:
Technical and economical data of conductors are taken as R=0,1576 [ /km], X=0,2277[ /km], Cost=151[$/km]. In the first two branches of the network the conductors technical data is: z01=0,1233 + j0,4127[], z02=0,014+ j0,6051[]. Base values for calculations are: Sb = 24 MVA, Ub = 23 kV. The types of wires were chosen previously and this procedure is not described in the example. The conductor selection problem, as it was done in some studies [6,8], is not considered. Capacitor size and cost [7] where prices indicated in $/(kVAr*year) are presented in Table I.
TABLE I j kVAr Price j kVAr Price j kVAr Price j 1 150 0.500 7 1050 0.2280 13 1950 0.211 19 2850 0.183 25 3750 0.183 2 300 0.350 8 1200 0.1700 14 2100 0.176 20 3000 0.180 26 3900 0.182 3 450 0.253 9 1350 0.2070 15 2250 0.197 21 3150 0.195 27 4050 0.179 4 600 0.220 10 1500 0.2010 16 2400 0.170 22 3300 0.174 28 >4050 0.179 5 750 0.276 11 1650 0.193 17 2550 0.189 23 3450 0.188 6 900 0.183 12 1800 0.187 18 2700 0.187 24 3600 0.170
Using the modified PSO-method, the set of solutions YLost is formed by the L-criterion (6). Each m particle corresponds to the n-dimensional vector of CB conductivities yLost(y1, y2, ..., yn).
kVAr Price j kVAr
Within the set of solutions YLost, the set of solution YCost is formed by applying the C-criterion (7), and consists of a set of vectors yCost(y1, y2, ..., yn).
Price
Within the set of solutions YCost, the set of solutions YTC is formed by applying the TC-criterion (10), which consists of the set of solutions yTC(y1, y2, ..., yn). Within the set of solutions YTC, the vector y(y1, y2, ..., yn), which corresponds to the normalized approximating hyperbole nor (r), where r=1,2,,n, is chosen. yTC The proposed algorithm allows to find the vector of CB conductivities with special nonlinear hiperbolic dependence in accordance with considered TC-approach. III TEST REZULTS With an application of the proposed algorithm, the radial network [6] of 23 kV voltage, which consists of n = 27 nodes and 21 loads, was studied. The calculation was carried out using the constraints (2)(4) and in the absence of sources of higher harmonic components of currents, for a period of 8760 hours per year. The cost of energy losses is taken as 0,168$/KWh [7].
In Table II, electrical parameters of network (feeder) [6] are given, where Z line resistance []; S [kVA], PF total power and power factor of the load node respectively. At the beginning of the first iteration of the calculation the B with fixed capacities (rated power) were installed in each node. Thus, the vector of CB conductivities y(y1, y2, ..., yn) had n=27 variables. Subsequently, as the result of solution determination set, some CB took zero values of the rated power. In Fig. 1, the normalized vector of ranked CB conductivities ynor (marked by squares) is shown. It is similar nor (marked by to the normalized hyperbolic curve yTC
nor asterisks), which has a low value of first rank yTC ( r =1) =0,6013
and rank factor =0,62. Comparison of CB installed power QC for two variants respectively based on: - Variant I. L- and C- criteria are applied, - Variant II. L-, C- and TC- criteria are applied.
TABLE II
nl 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Z () 0,123+j0,4127 0,014 + j0,6051 0,7463+j 1,0783 0,4052+ j0,5855 1,152 + j1,665 0,5261+ j0,7601 1,2263+ j1,1142 2,861+ j2,600 3,188+ j2,897 1,284+ j1,167 0,697+ j0,6335 1,999+ j1,833 0,9123+ j0,8365 1,236+ j1,133
S 0 0 0 1350 0
PF nl 15 16 17
Z () 2.883 + j2,644 3,213 + j2,946 1,284 + j1,166
S 1125 1300 300
PF 0,93 0,75 0,9 0,8 0,85 0,8 0,8 0,93 0,89 0,88 0,9 0,75
maximum size of the CB according to criteria of variants II (2550 kVAr , node 11). The results of the CB placement are shown in Table IV according to criteria of variants I and II. When Variant II (10) was applied, in comparison with Variant I (7), the active losses P increased by 2.03% and the cost increased by 1.87%. When using TC-criterion in Variant II, this slight increase of active losses and cost allows to use this criterion for searching CB sizing and placement.
y, yCostmin, [pu]
0,8 18 0 19
0,6973+ j0,6335 1150 1,999 + j1,8325 0,9123+ j0,8365 1,236 + j1,1332 2,884 + j2,644 3,213 + j2,946 1,294 + j1,187 350 400 1100 125 965 982
1300 0,95 20 0 1250 0 1313 21
0,9 22 23
0.25
0,8 24
0.2
1100 0,95 25 400 0,75 26
0,7027+ j0,6443 1300 3,3234 + j1,948 1,517 + j0,8893 0 875 200 0
0.15
1350 0,85 27 1225 0,82 0
0,9 0
0.1
nor y nor , yTC

0.05
0.7
0.6
10
15
20
25
30
r
0.5
Figure 2
0.4
TABLE III
0.3
0.2
QC, (kVAr) nl
riteria Variant I II
QC, (kVAr) nl
I riteria Variant II
QC, (kVAr) nl
I riteria Variant II
0.1
1
0 0 5 10 15 20
0 6450 300 1350 150 900 0 750 0
0 0 0 1800 0 600 0 1200 0
10 1350 11 600 12 300 13 600 14 900 15 450 16 900 17 0 18 900
900 2550 300 1200 450 150 1350 300 900
19 20 21 22 23 24 25 26 27
0 450 600 0 300 600 600 450 300
450 150 750 600 900 750 450 600 300
25
2 3
Figure 1
4 5 6 7 8 9
Fig. 2 shows the searching vector y of ranked CB conductivities (marked by squares), which is similar to the hyperbolic curve yTC and vector yCostmin (marked by asterisks), that corresponds to the minimum cost Bmin (7) after ranking vectors, r- rank. These variants CB placement and sizing are shown in Table III. The maximum size of the CB according to criteria of variants I (6450 kVAr, node 2) is much higher than the
TABLE IV
Criteria of Variant I Criteria of Variant II
Applicable criteria
[7]
Active losses (p.u.) Cost ($/year) Magnitude of conductivity of CB of the first rank y(r=1) (p.u.) .
0.03096 1.121*106 0.26875
0.03159 1.142*106 0.10625 [8]
distribution systems with nonlinear loads, Universities Power Engineering Conference, UPEC 2008. 43rd International, pp. 1 5, 2008. Y. Baghzouz, and S. Ertem, "Shunt Capacitor Sizing for Radial Distribution Feeders with Distorted Substation Voltages," IEEE Trans. Power Delivery, vol. 5, pp. 650-657, Apr. 1990. T. Halil, and A. Gorpinich, Selection of Optimal Conductors Sections and Capacitors Placement in Radial Distribution Systems by Selective Particle Swarm Optimization, Scientific Papers of Donetsk National Technical University, vol. 11, pp.406-413, 2011.(Rus)
The result of the T-criterion application is a significant reduction of total overcompensation. The decrease of maximum overcompensation is the following: - 2223 kVAr at node 11 for Variant II (with the Triterion), - 6450 kVAr at node 02 for the Variant I. IV CONCLUSION In addition to the traditional an additional third criterion, which is based on the TC-approach, was proposed in order to obtain the specific solution for CB sizing and placement in radial distribution network. The concessions method was used to solve the multi-criteria problems in conditions of information uncertainty. A combination of three criteria, concessions method and PSO-method gives a combined approach to multi-objective optimization of CB placement and sizing in radial distribution networks. The proposed method allows to find the vector of ranked CB conductivities with specific nonlinear dependence, for example hyperbolic dependence according to considered TC-approach. An application of the proposed combined approach allows to obtain the solutions with the CB placement in larger number of nodes with small difference between the maximum and minimum CB parameters (rated power, conductivity or capacitance) with slight increase in active losses and cost. The application of the proposed combined method, in comparison with traditional one, allows to reduce the capacitive overcompensation in the network nodes with CB at relatively low active power losses and seems to be useful for practical purposes.
References
[1] A.A. Eajal, and M.E. El-Havary, Optimal Capacitor Placement and Sizing in Distorted Radial Distribution Systems. Part II: Problem Formulation and Solution Method, 14th IEEE Internat. Conference on Harmonics and Quality of Power (ICHQP), Bergamo, Italy, pp.1-6, 2010. B.I. Kudrin,Electrical supply of industrial enterprises, Eds. Moscow: Inermet engineering, 2005, p. 672. V.I. Gnatyuk. The Law of Technocoenosis Optimum Construction, Eds. oscow: Publishing ouse of Tomsk State University - Center for Systems Research, 2005, p. 384. M. AlHajri, M. AlRashidi and M. El-Hawary. A novel discrete particle swarm optimization algorithm for optimal capacitor placement and sizing, 2007 Proceedings of Canadian Conference on Electrical and Computer Engineering (CCECE), pp.1286-1289, 2007. Y. Shi, and R. Eberhart, A modified particle swarm optimizer, The 1998 IEEE International Conference on Evolutionary Computation Proceedings, p. 6973, 1998. M. Vahid, A.A. Hossein, M. Kazem ,Maximum loss reduction applying combination of optimal conductor selection and capacitor placement in
[2] [3]
[4]
[5]
[6]
Investigating DC Link Current Ripple and PWM Modulation Methods in Electric Vehicles
Onur Satilmis
Yildiz Technical University Department of Electrical Engineering Istanbul-Turkey emese@yildiz.edu.tr
Abstract Batteries are one of the key components for Electric Vehicles (EVs). While batteries supply energy to output at normal working mode by converting internal chemical energy to electric energy (discharging), they restore energy at regenerative working mode as opposed to discharging (charging). During each working mode, battery current is not smooth DC and it has ripple. The ripple current influence battery lifetime negatively. Simulations which include ideal battery model and ideal DC bus capacitor model do not give realistic results about battery ripple current. This paper presents comparative investigations and evaluations about battery current waveform, battery current harmonic spectrum and the ratio of battery current harmonic components to battery average current for different modulation methods and operating states for motor (i.e. speed and torque) by using high frequency battery model and high frequency DC bus capacitor model. Keywords-electric vehicles; electric vehicle batteries; DC bus capacitors; power electronics.
Erkan Mese
bus current rms value and harmonic content depend on the load current magnitude, the modulation index (Mi), PWM method and the load power factor (cos) [6]-[7]. This paper provides comparative investigations and evaluations about battery current for different modulation methods and operating states. In Section II, PWM methods will be summarized. In Section III, DC bus capacitors and high frequency capacitor model will be discussed, EV batteries and high frequency battery model will be mentioned in Section IV. Then simulation results which are obtained in MATLAB/Simulink will be investigated and evaluated in Section V and finally conclusions about this study will be given in Section VI.
I.
INTRODUCTION
EVs have a history more than a hundred years have vanished for a long time due to they have not competed with internal combustion engines (ICEs). However EV researches have began to rise in recent decades again due to global warming, legal obligations, rising petroleum prices and concerns about petroleum reserves [1]-[3]. Batteries are key components which enable electric vehicle development. Batteries performance and lifetime are affected by working temperature. While low working temperatures reduce battery performance, high working temperatures shorten battery lifetime. For instance; lead acid batterys lifetime and nickel cadmium batterys lifetime decrease 50% and 16-18% at 95oF operating temperature, respectively [4]. Battery temperature rise due to additional losses caused by ripple current. Battery manufacturers recommend that battery ripple current should not be above 5% of the battery Ahr capacity (e.g. 5A rms for every 100Ah of battery capacity) [5]. As a result, battery ripple current is a really important issue for its lifetime. The fundamental power electronic interface between battery and AC traction motor is a three phase two-level voltage source inverter (VSI) as shown in Fig. 1. DC bus current (ii) consists of battery current (ibatt) and capacitor current (icap). DC bus current harmonic content occurs at carrier frequency and its sidebands, carrier frequency multiples and their sidebands with PWM operation [6]. DC
Figure 1. Three phase two-level VSI and traction motor from left to right.
II.
REVIEW OF PWM METHODS
Basic PWM structures can be classified as open loop schemes and closed loop schemes [8]. Closed loop schemes are preferred in cases variable load conditions to fit new conditions, and in order to compensate dc bus variations [9]. In the most applications of the three phase voltage source pulsewitdh modulated (VS-PWM) converters, current feedback loop is used to control ac load currents due to the better performance and faster response [10], [11]. In AC motors, when current control is used, dependence on stator parameters reduce and instant torque control is provided [11]. Basically, in the current control, AC phase currents are forced to follow reference current signals. Current control techniques for three phase VS-PWM converters are divided into two main groups, linear and nonlinear controllers. While linear controllers include stationary controller PI, synchronous vector controller (PI), state feedback controller, predictive and deadbeat controllers, nonlinear group includes hysteresis, delta modulation and on-line optimized controllers. In contrast to
nonlinear controllers, linear controllers, due to their structures, can be used together with carrier-based PWM (CB-PWM) methods and thus CB-PWM methods advantages are exploited such as better waveform quality, fixed switching frequency and implementation simplicity [10]. CB-PWM methods have two main implementation techniques as the triangle intersection technique and the direct digital technique. In the triangle intersection technique, the switching instants are created by comparing modulation wave with a triangular carrier wave [7]. In the direct digital technique, a reference vector is obtained by using space vector theory from three reference waves and this reference vector is generated from space vector diagram by selecting available space vectors. The direct digital technique is more complex than the triangle intersection technique for simulation and implementation because of processing steps [7], so triangle intersection technique has been used in this study to perform CB-PWM methods. If the neutral point on the ac side is isolated such as most three phase ac motor drive and utility interface applications, phase currents depend on only the voltage difference between phases. Thanks to this speciality, at the triangle intersection technique, a zero sequence signal (ZSS) can be added to reference modulation waves. The added ZSS extends voltage linearity range, decreases switching losses, and improves the waveform quality without produce phase voltage distortion. The added different ZSSs create different modulation methods. A generalized signal block diagram for a triangle intersection technique with ZSS is illustrated in Fig. 2. These modulation methods can be divided in two groups as continuous PWM (CPWM) and discontinuous PWM (DPWM) [7], [12]. In the CPWM methods, triangular wave and modulation wave always intersect and so switchings occur. However in the DPWM methods, modulation wave is linked to triangular wave peak point for a while, so switching and related losses does not occur. In this manner the switching loss characteristics of DPWM is better than CPWM methods [7]. Moreover, while CPWM methods have better waveform quality than DPWM in low modulation index ranges, in higher modulation index ranges is vice versa under equal inverter average switching frequency [7], [12].
Va* + + Vc* + + Vc** + + + Vb* Vb** + S5 Va** + S3 S1
and iq* references currents and errors are obtained. PI compensators convert error signals so that, vd and vq voltages are obtained. After that vd and vq voltages are converted to va*, vb* and vc* reference voltages by using inverse Park transformation. Lastly switching instants are formed in PWM modulator with the help of reference voltages [10], [11].
- Vdc + d* q* + d
d-q
+ -
Ed PI
Vd d-q Vq abc
Va* Vb* Vc*
Eq
PWM Modulator
Inverter
PI
A B
abc
Load
Figure 3. Synchronous vector controller.
Modulation index (Mi) which is input voltage utilization level is important for performance characteristics of a modulator [7]. Mi
Where Mi is modulation index, V1m is fundamental component magnitude of the line to neutral inverter output voltage and V1msix-step is fundamental component magnitude of the line to neutral inverter output voltage for sixstep operation. B. SPWM SPWM is the best known CB-PWM method. In the SPWM, three reference sinusoidal waves and a triangular carrier wave are compared and switching instants are formed. SPWM has narrow range of linearity, Mi max0.785 [12].
Calculation of ZSS
ZSS
Carrier Wave
Figure 2. A generalized signal block diagram for a triangle intersection technique with ZSS.
Figure 4. Carrier waves (left), ZSSs, reference and obtained modulation waves.
A. Synchronous Vector Controller (PI) Basically, in this method which is illustrated in Fig. 3., AC load currents iA, iB and iC are transformed id and iq currents by using Park transformation. These currents are compared id*
C. Space Vector PWM (SVPWM) In this method, ZSS is generated by scaling 0.5 with minimum magnitude reference wave which is determined by comparing Va*,Vb* and Vc*. For instance; if
Va*Vb* and Va*Vc*, ZSS is identified as 0.5Va* and Mimax0.907 [7]. D. DPWM1 In this method, ZSS is generated with the help of maximum magnitude reference wave which is determined by comparing Va*,Vb* and Vc*. For instance; if Va*Vb* and Va*Vc*, ZSS is identified as sign(Va*)Vdc/2-Va* and Mimax0.907. DPWM1 has minimum switching losses at unity power factor operating condition [7]. Carrier waves, ZSSs, reference modulation waves and obtained modulation waves of SPWM, SVPWM and DPWM1 are shown in Fig. 3, respectively. III. DC BUS CAPACITOR AND HIGH FREQUENCY CAPACITOR MODEL (HFCM)
and optimization for power electronics systems from 5 kHz to 20 kHz frequency range [20].
Figure 6. The equivalent circuit of high frequency battery model.
V.
SIMULATION RESULTS
DC bus capacitor is one of the most important passive components in EV inverter [13]. Electrolytic capacitors are used in conventional designs. Generally, film capacitor technology is used as a DC bus capacitor for automotive applications [14] due to various benefits. Hence this study is based on the film capacitor. The equivalent circuit of a capacitor is shown in Fig. 5 [15]. This circuit can also be used to simulate a film capacitor via MATLAB/Simulink. There are four components which are ESR, C, Rp, ESL. Power losses in capacitor is proportional with ESR. ESL is the total series inductance of the capacitor winding including any internal connections. C is nominal capacitance and which is given at 25oC and 1kHz [16]. Rp is parallel resistance and it is about leakage current [17]. ESR, C and ESL are affected from temperature and frequency [13], [17]-[18]. In addition, C is also affected applied voltage [18]. ESL can be ignored at low frequency [19] but due to high frequency operation in this study, ESL will be taken into account.
Battery current has been analyzed for different modulation methods and operating states (speed (n) and torque (T)) at 8kHz the carrier frequency. Fig. 7 shows the diagram to simulate the system. DC bus has been created with a 410V NiMH battery and a 350uF DC bus film capacitor. A permanent magnet synchronous motor (PMSM) has been used as a load. PMSM parameters are: Rs=0.3 Ld=5.9745 mH Lq=5.9745 mH Pole pairs=16 Trated=358.9 N.m As seen from data, a simplification is made by selecting surface mounted PM machine. The simplification allows us to set D axis current to be zero and to observe the effect of Q axis current on the modulation schemes. Furthermore, surface mounted PM machines are sometimes preferred as generator in the series hybrid vehicle architectures.
- Vdc +
HFBM HFCM
d*=0 q* + -
+ -
Ed PI
Vd d-q Vq abc
Va* Vb* Vc*
Eq
PWM Modulator
Inverter
PI
thetam (rad)
d
d-q
A B
abc
C
thetam (rad) K1
Figure 5. The equivalent circuit of a capacitor.
IV.
BATTERIES AND HIGH-FREQUENCY BATTERY MODEL (HFBM)
PMSM
PI
+ nref (rpm)
nm (rpm) K2
wm (rad/s)
An EV battery is characterized some performance metric such as specific energy (Wh/kg), specific power (W/kg), efficiency, maintenance, cost, lifetime, safety, environmental adaptation [2]. Lead-acid battery, nickel-based batteries and lithium-based batteries can be considered for EV application [2]. Due to availability of the related data, Nickel Metal Hydride type batteries are considered in this investigation. A high frequency battery model shown in Fig. 6 has been used to realize realistic observation of battery and capacitor ripple currents for an EV application. It consists of two resistors and two inductors. The model can be used to ripple current analysis, passive component design, control strategy design
Tm (N.m)
Figure 7. The equivalent circuit of simulation model.
In Fig. 8, Fig. 9 and Fig. 10, ibatt, icap and ii waveforms are shown for two different torque and two different speed values at SPWM, SVPWM and DPWM1 method, respectively. As shown in Fig. 8, Fig. 9 and Fig. 10, while SPWM and SVPWM methods have small and similar fluctuations, DPWM1 method has big fluctuations according to mean current value, especially at low torque and high speed.
(a)
(b)
(c)
Figure 8. Current waveforms at SPWM (a) T=358.9 N.m, n=220 rpm; (b) T=358.9 N.m, n=80 rpm; (c) T=70 N.m, n=220 rpm.
(a)
(b)
(c)
Figure 9. Current waveforms at SVPWM (a) T=358.9 N.m, n=220 rpm; (b) T=358.9 N.m, n=80 rpm; (c) T=70 N.m, n=220 rpm.
(a)
(b)
(c)
Figure 10. Current waveforms at DPWM1 (a) T=358.9 N.m, n=220 rpm; (b) T=358.9 N.m, n=80 rpm; (c) T=70 N.m, n=220 rpm.
In Fig.11, Fig.12 and Fig.13, battery current harmonic spectrums are shown for two different torque and two different speed at three different modulation methods. Harmonic content is obtained as a base value battery current DC component (Idc).
(a)
(b)
(c)
Figure 11. Battery current harmonic spectrums at SPWM (a) T=358.9 N.m, n=220 rpm; (b) T=358.9 N.m, n=80 rpm; (c) T=70 N.m, n=220 rpm.
(a)
(b)
(c)
Figure 12. Battery current harmonic spectrums at SVPWM (a) T=358.9 N.m, n=220 rpm; (b) T=358.9 N.m, n=80 rpm; (c) T=70 N.m, n=220 rpm.
(a)
(b)
(c)
Figure 13. Battery current harmonic spectrums at DPWM1 (a) T=358.9 N.m, n=220 rpm; (b) T=358.9 N.m, n=80 rpm; (c) T=70 N.m, n=220 rpm.
As shown in Fig. 11, at rated torque and n=220 rpm, the dominant harmonics are at the carrier frequency. However twice the carrier frequency harmonics become more dominant with decreasing speed at the same torque and with decreasing torque at the same speed. While the carrier frequency harmonic magnitudes (Idc as a base value) at rated torque and n=220 rpm are decreasing with decreasing speed at the same torque and decreasing with torque at the same speed, twice the carrier frequency harmonics magnitudes increase. As shown in Fig. 12, at rated torque and n=220 rpm, the dominant harmonics are at the carrier frequency. However twice the carrier frequency harmonics become more dominant with decreasing speed at the same torque and with decreasing torque at the same speed. While the carrier frequency harmonic magnitudes at rated torque and n=220 rpm are decreasing with decreasing speed at the same torque and decreasing with torque at the same speed, twice the carrier frequency harmonics magnitudes increase. As shown in Fig. 13, in all cases, the dominant harmonics are at the carrier frequency and battery current frequency. Their magnitudes at the rated torque and n=220 rpm increase with decreasing speed at the same torque and with decreasing torque at the same speed. As shown in Fig. 11, Fig.12 and Fig.13, SPWM and SVPWM have similar harmonic spectrums and generally SVPWM dominant harmonic magnitudes are lower than SPWM. However, DPWM1 has rather different harmonic spectrum from SPWM and SVPWM and generally its
dominant harmonic magnitudes are bigger than SPWM and SVPWM. Battery current rms value (Ibatt) consists of a dc component (Idc) and harmonic components (Ih) as described in (2). Harmonic components create battery ripple currents and additional losses. Based on equations (2) and (3), battery current distortion ratio (IBD) can be described as in (4). I2batt = I2dc + I2hI2h = I2batt I2dc IBD = Ih / Idc
In Fig. 14.a, Fig. 14.b and Fig. 14.c, IBD variations with speed at different torques are given for SPWM, SVPWM and DPWM1 methods. Also, in Fig. 15.a, Fig. 15.b and Fig. 15.c IBD variations with speed are given for different modulation methods where T=358.9 N.m, T=210 N.m and T=70 N.m. IBD was multiplied by 10000. As shown in Fig. 14, IBD is lower at low torque values than high torque values for high speed - battery current is low at these intervals because of low torque values - at the SPWM and SVPWM methods, but for DPWM1 the situation is vice versa. However, IBD is higher at low torque values than high torque values for low speed battery current is high at these intervals because of high torque values at the SPWM and SVPWM but DPWM1 has different behavior at these intervals. As shown in Fig. 15, IBD is similar at SPWM and SVPWM for all speed values especially at low torque and their IBD values is lower than DPWM1. While DPWM1 has
(a)
(b) Figure 14. IBD variations with speed for different torque (a) SPWM; (b) SVPWM; (c) DPWM1.
(c)
(a)
(b)
(c)
Figure 15. IBD variations with speed for different modulation methods (a) T=358.9 N.m; (b) T=210 N.m; (c) T=70 N.m.
big IBD at low speed and low torque values, its IBD values is near to other PWM methods at high speed and high torque values. Because at high torque and speed ranges, Mi value is high and zero vector times are low, and so DPWM1 method has similar pulse pattern to SPWM and SVPWM. Furthermore, DPWM1 method has lower switching losses than other methods especially at unity power factor. I BD values, battery current harmonic spectrums and waveforms also change with capacitor constants and battery constants in Fig. 5 and in Fig. 6, respectively. High IBD causes high losses and so increasing battery temperature and as a r esult batterys lifetime reduces. Selection available modulation method, operating state and as well as using a capacitor which has low ESR, low ESL and high capacity decreases fluctuations, harmonic contents and IBD values. VI. CONCLUSION In this paper, battery current has been analyzed for EVs by using high frequency battery model and high frequency capacitor model via Simulink. Widely known modulation methods were considered. Battery current analyses and harmonic spectrums are provided for different speed, torque and modulation methods. IBD has been described and IBD variations with speed and torque have been obtained for each PWM method. As shown that, IBD values for SPWM and SVPWM are better than DPWM1. DPWM1 method has very high IBD values except high speed and high torque. IBD values, battery current harmonic spectrums and waveforms are also influenced with battery constants and capacitor constants. IBD value should be minimized to avoid reducing battery lifetime because of heating. So SPWM or SVPWM method should be preferred for longer battery lifetime, also if possible SVPWM. Of course, battery current is not a single criterion about selection PWM method. It should be evaluated together with other criterions such as range of linearity, waveform quality, switching losses, implementation simplicity, etc. This study can be helpful in sizing battery and DC bus capacitor as well as in selection of PWM method and operating conditions. The results are valuable in shedding light for electric vehicles and other similar battery powered inverter applications. VII. ACKNOWLEDGEMENTS This work was partially supported by Marie Curie International Reintegration Grant within the 7th European Community Framework Programme under contract number PIRG05-GA-2009- 248493.
REFERENCES
[1] [2] C. C. Chan, The state of the art of electric, hybrid , and fuel cell vehicles, Proc. of the IEEE, vol. 95, no.4, pp. 704718, April 2007. M. Ehsani, Y. Gao, S. E.Gay and A. Emadi, Modern Electric, Hybrid Electric, and Fuel Cell Vehicles: Fundamentals Theory, and Design, CRC, 2005. C. C. Chan and Y. S. Wong, The state of the art of electric vehicles technology, in Power Electronics and Motion Control Conference, 2004. IPEMC 2004. The 4th International, vol. 1, 14-16 August 2004, pp. 46-57. Megger, Battery Testing Guide. N. Kutkut, Output AC Ripple Effect Technical Note. U. Ayhan and A. M. Hava, Analysis and characterization of DC bus ripple current of two-level inverters using the equivalent centered harmonic approach,in Energy Conversion Congress and Exposition (ECCE), 17-22 September 2011, pp. 3830-3837. A. M. Hava, R. J. Kerkman, and T. A. Lipo, Simple analytical and graphical methods for carrier-based PWM-VSI drives, IEEE Trans. on Power Electronics, vol. 14, no. 1, pp. 49-61, January 1999. J. Holtz, Pulsewidth modulation for electronic power conversion, Proc. IEEE, vol. 82, pp. 11941214, Aug. 1994. Muhammad H. Rashid, Power Electronics Handbook, Academic Press, 2001. Marian P. Kazmierkowski and Luigi Malesani, Current control techniques for three-phase voltage-source PWM converters: A survey, IEEE Trans. on Ind. Electronics, vol. 45, no. 5, pp. 691-703, Oct. 1998. Luigi Malesani and Paolo Tomasin, PWM current control techniques of voltage source converters A survey, Industrial Electronics, Control, and Instrumentation, 1993. Proceeding of the IECON 93., International Conference on, vol. 2, pp. 670-675, November 1993. M. P. Kazmierkowski, R. Krishnan, and F. Blaabjerg, Control in Power Electronics, Elsevier Inc., 2002. H. Wen, X. Wen, J. Liu, X. Guo, and F. Zhao, A low-inductance highfrequency film capacitor for electric vehicles, in Proceeding of International Conference on Electric Machines and Systems, 8-11 October 2007, Seoul, Korea, pp. 2046-2050. W. Wondrak, A. Nisch, S. Pieger, A. Rodewald, M. Wagner, A. Willikens, and P. Wurster, Requirements on passive components for electric and hybrid vehicles, CARTS Europe 2010, 10 -11 November 2010, Munich, Germany. G. L. Johnson, Solid State Tesla Coil, 2001. Power Film Capacitor Application Guide, Cornell Dubilier. Film Capacitors: General technical information. EPCOS. J. Pelcak, Equivalent circuit model for tantalum and niobium oxide capacitors for use in simulation software AVX. General technical information of film capacitors for use in electronics, CAPAKOR. J. Wang, K. Zou, C. Chen, and L. Chen, A high frequency battery model for current ripple analysis, in Applied Power Electronics Conf. and Expo. (APEC), 2010 25th Annual IEEE, 2010, pp. 676-680.
[3]
[4] [5] [6]
[7]
[8] [9] [10]
[11]
[12] [13]
[14]
[15] [16] [17] [18] [19] [20]
Battery Characterization for Vehicular Applications Using Hardware-in-Loop Real-Time Simulation

Damon Bazargan
ECE, University of Manitoba Winnipeg, Canada umbazarg@cc.umanitoba.ca
Shaahin Filizadeh
ECE, University of Manitoba Winnipeg, Canada shaahin.filizadeh@ad.umanitoba.ca
Garry Bistyak
ECE, University of Manitoba Winnipeg, Canada garry_bistyak@umanitoba.ca
Abstract This paper develops a hardware-in-loop simulation test bench for testing battery storage systems under realistic operating conditions. It consists of an actual battery interfaced with a real-time simulation model of the system in which it is embedded. The setup can also be used to assess the accuracy of mathematical battery models. The developed test-bench is used for simulation of a hybrid electric vehicle. Moreover three mathematical battery models are also developed and examined using comparative assessment of their results with those produced by the real battery-in-loop simulator. Keywords: Battery modeling, Hardware in the loop simulation, Vehicle backward model.
I.
INTRODUCTION
This paper describes a hardware-in-loop (HIL) real-time simulation setup for emulation of actual stress conditions, and characterization and comparison of battery models. In contrast to conventional computer models, a HIL simulation scheme uses actual batteries in concert with a real-time simulator. Measurements taken on such a HIL setup can be readily used to compare the actual battery performance with what is predicted by its models. Additionally such a setup can also be used to expose an actual battery to different operating conditions, stress levels and usage profiles to assess its performance under conditions that may be prohibitively difficult or costly to replicate in practice. In the following sections, a HIL real-time simulation setup is described in which Li-ion batteries are interfaced to a model of a hybrid electric vehicle. The setup is then used to assess the batterys dynamic behaviour in a given driving cycle, in which the battery is exposed to several cycles of charge and discharge of different intensity and length. In parallel with the HIL setup, three mathematical models of the battery are also run concurrently. Voltage and current profiles of the HIL setup and those of the three models are then compared to assess the accuracy of the models. II. BATTERY MODELING
Plug-in hybrid electric vehicles (PHEV) and hybrid electric vehicles (HEV) are well-known solutions for increasing vehicle energy efficiency and reducing green-house gas emissions [1,2]. A hybrid vehicle uses two or more energy sources in a synergic manner. For example an internal combustion engine (ICE) and a battery energy storage system (BESS) are common components of a typical hybrid electric drive-train [2]. Analysis and design of a battery energy storage system often relies on a model that represents its steady state or dynamic behaviour. Much effort has been spent to develop battery models for such purposes [3-8]. Since battery characteristics change over time, both in the short-term due to operating conditions and in the long-term due to aging, all battery models experience deviations from actual battery behaviour. For example in many battery models the internal resistance of a battery is assumed to be constant, although research shows that it is indeed a function of such parameters as the temperature and the state of charge (SoC) [9]. Other important parameters such as efficiency and capacity are often assumed to be constant despite their variation with operating conditions and the usage history of the battery. In particular, as the battery ages both the efficiency and the capacity decrease [10-13] adversely affecting simulation results from models that do not include these variations. Batteries are used in a wide range of applications, which expose them to greatly varying usage profiles and hence different stress levels [14]. For example in an uninterruptable power supply, the battery is rarely used. On the other hand in a vehicular application the battery is continuously charged and discharged. The nonlinear nature of a battery combined with the specific demands of its environment results in inaccuracies in the battery simulation model.
This section describes three commonly used mathematical battery models. These models all use parameters that can be empirically obtained using tests on the battery. As Li-ion battery technology is gaining popularity in vehicular systems, this study will focus on this chemistry. In particular the models are developed for the LPF26650P lithium ion battery. Specifications of this battery are listed in Table I. A. Battery Models
The three battery models used here, i.e. the Thevenin, the Randle, and the Shepherd, are all based on equivalent electric circuits and are commonly used in the simulation of PHEVs and HEVs. The Thevenin model is depicted in Fig. 1. It is composed of a constant dc voltage source and a resistance. The dc voltage source represents the open-circuit voltage (OCV) and the resistor represents the internal resistance of the battery. The Randle model is depicted in Fig. 2. It is composed of two resistors, a capacitor and a constant dc voltage source. The values of the passive elements can be determined as described in [9]. The constant dc source is defined by the OCV of the battery. Due to the RC circuit combination the Randle model provides an approximation of the batterys first order
dynamic behavior. During charge and discharge events the Randle model reacts smoother in comparison to the Thevenin model as a result of the equivalent RC time constant.
TABLE I. LPF26650P LITHIUM ION BATTERY SPECIFICATIONS Parameter Nominal Capacity @ C/5 (Ah) Average Operating Voltage @ C/5 (V) Weight (g) Continuous Discharge (A) Pulse Discharge (A) 30 Seconds Charge Current (A) Charge Voltage Cut-off (V)
Rint
A VFull VExp
(2) (3) (4)
3 QExp
Value 2.6 3.2 80.52 10 26 2.6 3.65
(VFull VNom A(exp( BQNom ) 1)).(Q QNom ) QNom
The values of VExp, QExp, VNom, QNom, VFull are determined by extracting information from the discharge curve of the battery obtained at a constant discharge rate of 0.2C [6]. Table II shows the parameters of the three models for the considered lithium ion battery.
TABLE II. MODEL PARAMETERS Parameter A B K R1 R2 C R V0 Value 0.55 83 0.12 0.25 0.05 750 0.3 13.4 Model Shepherd Shepherd Shepherd Randle Randle Randle Thevenin/Shepherd Shepherd
Fig. 1: Thevenin model.

R2 R1
Fig. 2: Randle model.
As an example, Fig. 4 shows a verification of the developed Shepherd model. It depicts the discharge traces of the battery (from a fully charged state) at a constant rate of 0.2 C. The figure shows the output of the Shepherd model as well as the recorded terminal voltage of the actual battery. It is clearly seen that as the SoC of the battery drops its terminal voltage declines due to the change of its internal voltage. The Thevenin and Randle models are not able to represent such behaviour due to their SoC-invariant parameters (as shown later in section II.B). B. Battery Model Comparison To compare the three battery models, a depleted lithium ion battery is charged using a pulsed charging method. It consists of a series of constant current pulses interleaved with periods of rest [15]. The actual battery and the three models are subject to the same pulsed charging regime. The measured battery terminal voltage and the terminal voltage predicted by the models are shown in Figs. 5-7. As shown, both the Randle and the Thevenin model produce estimations of the terminal voltage that progressively become closer to the actual terminal voltage as the battery SoC increases. The Randle model additionally shows exponential rise and fall of the terminal voltage. Both models, however, have considerable error particularly at low SoC levels. This is caused by the fixed-value internal voltage sources found in both models.
The Shepherd model is depicted in Fig. 3. It is composed of a variable dc voltage source and a constant resistance which represents the internal resistance of the battery. The value of the internal resistance must be measured. The variable dc voltage source is a function of the state-of-charge (SoC) of the battery and is described in (1).
Rint
Fig. 3: Shepherd model.
V V0 K
Q Q i(t )dt
where V0 is the nominal voltage and Q is the battery capacity in Ah. The constants A, B and K in (1) are defined as follows.
A exp( B i(t )dt )
(1)
14 13.5
Model
13 12.5 12 11.5 11 10.5 10
Voltage (v)
Actual
2000
4000
6000
8000
10000
12000
14000
16000
18000
Fig. 7: Comparison of the Shepherd model and the actual battery.
Time (s)
Fig. 4: Verification of the Shepherd model.
III. HARDWARE-IN-LOOP (HIL) SIMULATION SETUP In this section a HIL real-time simulation setup is described. In this setup an actual battery is interfaced with a real-time simulation model of a larger system in which the battery is embedded. In particular the system developed in this paper considers battery-in-loop simulation of a hybrid-electric vehicular powertrain. A. Description of the HIL Setup HIL simulation is an alternative to complete model-based electromagnetic transient (EMT) simulation. It is the closest alternative to field experimentation, provides exhaustive testing opportunity, and prevents costly and damaging failures of the actual system. HIL simulation has been used for controller assessment for many years especially in vehicle, aerospace and power industries [16]. In the particular case of battery storage systems, HIL simulation eliminates the need for precise modeling of the battery and greatly enhances the accuracy of simulation results. The HIL simulation setup in this study consists of a device under test (i.e. a battery), a real-time simulator and an interface, which is a high bandwidth amplifier with adequate voltage, current and power ratings, as shown in Fig. 8.
Fig. 5: Comparison of Thevenin model and the actual battery.
The Shepherd model, on the other hand, is capable of tracking the actual battery terminal voltage as shown in Fig. 7. In this model the voltage source value depends on the SoC of the battery. In the beginning of the experiment, when the SoC is low, the Shepherd model is different from the actual measured results. It however quickly adjusts its internal voltage and as the SoC level increases its output matches the actual battery profile.
Fig. 6: Comparison of the Randle model and the actual battery.
Fig. 8: HIL configuration.
The amplifier mimics the effect of the simulation on the battery, by sinking or sourcing a certain mount of electric current as specified by the real-time simulator. In this paper, a reduced-scale battery-in-loop simulation is used, i.e., the actual battery interfaced is a small-scale version of the actual battery storage system. This allows easier implementation as it involves smaller voltage, current and power values. It is also less hazardous and less expensive than a full-scale test setup [9,17]. In particular, interfacing a full-scale battery storage system requires an amplifier with large ratings, which may be uneconomical. A scaling factor of 300 (applied to the power demand from the battery) is used in this case. B. Description of Real-Time Vehicular Simulation Model
Froll Fr (1
v )mg sin( ) 3600
(8)
where specifies the road grade, a is the air density (kg/m3), Af is the vehicle frontal area (m2), CD is the aerodynamic drag coefficient, and Fr is the rolling resistance coefficient. The required power is calculated as in (9). Table III shows the vehicle parameters that are used in the simulations in this paper.
P Fprop.v
Software implementation
(9)
In this paper, the real-time simulator of Fig. 8 is used to simulate a parallel hybrid-electric vehicular powertrain over a given driving cycle. Fig. 9 shows a diagram of the developed HIL simulation, in which the real-time model and physical components are clearly marked. As it is shown a driving cycle specifies the instantaneous acceleration and velocity of the vehicle. Using a backward vehicular model, the required power of the vehicle is calculated. The required power is provided by the battery and/or the internal combustion engine, as determined by a vehicle controller. Although the HIL system alleviates the need for mathematical modeling of the battery, it can be used to assess the accuracy of such models through comparison with the results obtained from the actual interfaced battery. Therefore, in the developed HIL simulation setup, the three mathematical models described in section II are run in parallel with the rest of the system. Throughout the simulation, the batterys terminal voltage profile is recorded and compared with those predicted by the three models. The results are then used to assess how closely the three models can predict the behavior of the battery in a realistic scenario. The vehicle dynamics block calculates the required power according to the instantaneous velocity and acceleration of the vehicle. The instantaneous velocity of the vehicle is defined by the driving cycle, and is used for calculation of the acceleration. The Fprop is the propulsion force provided by the vehicle power train, and has to overcome the opposing forces of aerodynamic drag Fa, rolling resistance Froll and road grade force Fg . Newtons second law of motion is then applied as follows.
Fprop Fa Froll Fg m dv dt
Driving cycle (Velocity vs time)

Instantaneous velocity
Vehicle dynamics
Demand power
Vehicle controller
Demand power from battery
Internal combustion engine

Hardware implementation
Actuator controller
Battery terminal voltage
Battery current
Actuator
Battery
Fig. 9: Flow diagram of the used backward model. TABLE III. VEHICLE SPECIFICATIONS Parameters Vehicle mass (m) Frontal area (Af) Aero dynamic drag coefficient (CD) Rolling resistance coefficient (Fr) Motor efficiency Regenerative braking efficiency Value 1330 kg 2.23 m2 0.26 0.01 0.8 0.7
(5)
where m is the vehicle mass (kg), and v is the vehicle speed (m/s). The aerodynamic drag, the rolling resistance and the grade forces are given as follows.
Fg mg sin( )
(6) (7)
Fa
1 a CD A f v 2 2
For this study a parallel-hybrid control strategy is used. In a parallel hybrid the two sources of power, i.e. the engine the electric motor, are mechanically coupled. The vehicle controller shares the required power between the two. The electric motor and the internal combustion engine (ICE) are independently controlled. Their contributions to the propulsion power are added up by a power-split gear device. Fig. 10 shows how the demand power is shared. The battery is the primary source of power, as long as the required power does not exceed the maximum electric power capability of the electrical path. Once the demand power exceeds the capability of the battery, the ICE is engaged and provides the rest. The engagement of the battery is also affected by its SoC. If the SoC of the battery is less than a given threshold, it cannot be used for propulsion and the ICE is engaged earlier. Apart from
Battery SoC
charging using the ICE, regenerative braking harvests part of the kinetic energy of the vehicle to charge the battery during braking.
Maximum Engine Power
Voltage(v)
16
15
14
Maximum Motor Power
13
12
Motor power Engine power
11
Demand power
10
500
1000
Fig. 10: Vehicle controllers power-sharing strategy.
1500 2000 Time(s)
2500
3000
Fig. 12: Battery voltage profile.

10 8 6 4
Current(A)
IV. RESULTS AND DISCUSSION A weekday driving cycle for the city of Winnipeg, shown in Fig. 11, is used for simulations in this study. Figs 12 and 13 show the battery voltage and current when the real-time simulation setup of the vehicle is run over the given driving cycle. As seen, the battery voltage and current change dynamically to provide the required power for the vehicle in conjunction with the ICE, as decided by the vehicle controller.
100 90 80 70
Velocity (Km/h)
2 0 -2 -4 -6 -8 -10 0 500 1000 1500 2000 Time(s) 2500 3000
60 50 40 30 20 10 0 0 500 1000 1500 2000 Time(s) 2500 3000
Fig. 13: Battery current profile.
Note that the SoC of the battery in this experiment varies between 100% to 60%. The SoC trace is obtained using a direct coulomb counting method. In this relatively high SoC range the Shepherd model is able to track the battery voltage successfully, as shown in Fig. 7.
T final Vbattery
Fig. 11: Weekday driving cycle of the city of Winnipeg.
RMS
Vmodel dt Vnominal
TABLE IV. RMS ERROR VALUES
(10)
Figure 14 shows voltage profiles predicted by the three battery models as well as that of the actual battery. The Thevenin and the Randle model profiles match exactly (except for the small exponential rises and falls of the Randle model that are indistinguishable in the figure due to the time scale). To judge the quality of the models it is required to devise a quantitative measure of how closely they predict the voltage profile of the battery during HIL simulation. A normalized root mean square (RMS) error indicator is given in (10). This indicator is calculated (see Table IV) for the three models using samples of their output voltage and the actual voltage of the battery. It is seen that the Shepherd model produces the closest output among the three models.
RMS1 (Thevenin model) RMS2 (Randle model) RMS3 (Shepherd model)
2.66 2.81 1.50
V. CONCLUSIONS The paper described a HIL simulation setup for testing actual battery storage systems and demonstrated its use in a vehicular system. The developed setup is not only a superior simulation platform for nonlinear and time-variant battery storage systems, but also provides a unique opportunity to assess the accuracy of battery models under different operating
points, and stress and usage profiles. It was shown that all three battery models considered had appreciable errors in predicting the terminal voltage of the battery, although the Shepherd model produced the closest results of the three. It is important to note that Figs. 5-7 are depicted for the tested battery at low SoC ranges. Since in a HEV the SoC of the battery remains higher than a minimum threshold, the Shepherd model can predict the voltage profile of the actual battery better than the two other battery models.
15 14.5 Thevenin model 14 13.5
Voltage(v)
13 12.5 12 11.5 11 2040 Shepherd model
Actual bettery
2041
2042
2043
2044
2045 2046 Time(s)
2047
2048
2049
2050
Fig. 14: Battery voltage profile comparison.
Battery storage systems are nonlinear and highly operatingpoint dependent systems whose performance is significantly affected by the stress level imposed on them. The developed HIL setup is able to accurately mimic the actual stress conditions on a real battery in a specific application and therefore provides a realistic test environment. The same functionality can be used for testing other operating-condition dependent systems as well. VI. REFERENCES [1] J. Tollefson, Charging up the future, Nature, vol. 456, pp. 436440, Nov. 2008. [2] C. C. Chan, The state of the art of electric and hybrid vehicles, Proceedings of the IEEE, vol. 90, no. 2, pp. 247 - 275, Feb. 2002. [3] O. Tremblay1 and L. Dessaint, Experimental validatio n of a battery dynamic model for EV applications, Journal of World Electric Vehicle, vol. 3, May 2009. [4] A. Ramamurthy, Advanced lithium ion battery modeling and power stage integration technique, Energy Conversion Congress and Exposition (ECCE), pp. 14851492, 2010. [5] L. Gao, Dynamic lithium-ion battery model for system simulation, IEEE Transactions on Components and Packaging Technologies, vol. 25, no. 3, Sept. 2002. [6] O. Tremblay, A generic battery model for the dynamic simulation of hybrid electric vehicles, Proc. Vehicle Power and Propulsion Conference, 2007. [7] S. Chen, Modeling of lithium-ion battery for energy storage system simulation, Power and Energy Engineering Conference, Mar. 2009.
[8] M. Ceraolo, New dynamical models of lead-acid batteries, IEEE Transactions on Power Systems, vol. 15, no. 4, pp. 1184 - 1190, Nov. 2000. [9] D. Bazargan, Hardware-in-Loop Simulation of Battery Storage Systems for Power System Applications, M.Sc. dissertation, University of Manitoba, Nov. 2012. [10] B. Liaw, R. Jungst, G. Nagasubramanian, H. Case, and D. Doughty, Modeling capacity fade in lithium-ion cells, Journal of Power Sources, no. 140, pp. 157-161, Aug. 2005. [11] M. Broussley, S. Herreyre, P. Biensan, and P. Kasztejna, Aging mechanism in li-ion cells and calendar life prediction, Journal of Power Sources, vol. 97, pp. 13-21, Aug. 2001. [12] J. Dogger, B. Roossien, and F. Nieuwenhout, Characterization of li-ion batteries for intelligent management of distributed grid-connected storage, IEEE Transactions on Energy Conversion, vol. 26, no. 1, p. 256-263, Mar. 2011. [13] J. W. Stevens, A study of lead-acid battery efficiency near top-of-charge and the impact on PV system design , Photovoltaic Specialists Conference, Conference Record of the Twenty Fifth IEEE, no. 26, pp. 1485 - 1488, May 1996. [14] E. I. Baring-Gould et al., Detailed Evaluation of Renewable Energy Power System Operation: A Summary of the European Union Hybrid Power System Component Benchmarking Project, National Renewable Energy Laboratory, May 2005. [15] L. Chen , A Design of an Optimal Battery Pulse Charge System by Frequency-Varied Technique, IEEE Transactions on Industrial Electronics, vol. 54, no. 1, Feb. 2007. [16] A. Bouscayrol and J. Irwin, Control and Mechatronics , CRC Press, 2011. [17] A. Allegre, Reduced-scale-power hardware-in-the-loop simulation of an innovative subway, IEEE Transactions on Industrial Electronics, vol. 57, no. 4, pp. 1175-1185, Apr. 2010.
High Frequency Power Assessment on Overhead Transmission Lines

Abdelkrim Boudjemaa, Azzeddine Nacer, Tarik B. Berbar and Bachir Lehouidj
University of Science and Technology Houari Boumediene, FEI/LSEI, BP 32, Bab Ezzouar, Algiers, Algeria.
nacer3dz@gmail.com
AbstractThe transmitted power along an incurved overhead two conductors transmission line is numerically evaluated in the lossy ground case at high frequency. The technique uses the powerful and versatile chain ABCD matrix of a transmission line segment to frequency characterization in various and complex configurations. This technique is used to calculate power attenuation on any distance from the high frequency generator along an incurved overhead two conductors transmission line over the 1-30MHz range. Keywords- power line, multiconductor, high frequency, propagation constant
The overhead transmission line is not strictly TEM because ground and wires are not perfect conductors (Fig.2). However, under some conditions, its behavior is quasi-TEM mode and the radiated field is supposed negligible. C.R. Paul [5] clarifies the widely used decoupling technique for the determination of propagation constants and characteristic impedances. In real situation, the overhead transmission line is always incurved. The sag is the distance between the lowest level of the incurved wire and the ideal straight line (Fig.3). It is the consequence of mechanical properties of the structure [6].
I.
INTRODUCTION
The power line is working for power delivery at low frequencies (50 or 60Hz), however it can be used as a communication channel at high frequencies (1-30MHz). An important research effort is made in many countries to enhance the performance of Broadband Power Line (BPL). An important limitation of this channel is the high attenuation that reduces the effective available span along line. This work is essentially focused on power efficiency of the incurved overhead two wires transmission line at several distances from the source.
II.
MODAL METHOD FOR OVERHEAD LINE
The modal method was first introduced in 1926 by J.R. Carson [1] in the case of industrial frequency applications. It is extended by M. D'Amore & al [2] and R.G. Olsen & al [3] to the high frequency range (1-30MHz). This method is used for the study of coupling between parallel cylindrical conductors (Fig.1). Diameter, conductivity and magnetic permeability of each conductor wire are explicitly taken into account, and hence the influence of electrical and magnetic ground characteristics can be evaluated. Using the extended method, P. Amirshahi & al [4] show that the transmission capacity of the Multiconductor Transmission Line (MTL) is very high in ideal and theoretical situation. The assumption in the modeling of a MTL by means of transmission line equations, is that the electric and magnetic fields in the plane transverse to the direction of propagation of the line, are quasi-TEM.
Figure 1. Parallel two-conductor transmission line with loss
III.
UNIFORM TWO-CONDUCTOR TRANSMISSION LINE
Voltages and currents waves are propagating along the two conductor transmission lines. Figure 2.b represents the voltages and currents of a set of two coupled parallel conductors, the ground is also a conductor. Where:
( z ) = [V , V ] T : Vector of phasor line voltage (with V 1 2 respect to the reference ground) at the distance z from the origin, and ( z ) = [I , I ] T : Vector of phasor line current at the I 1 2 distance z from the origin.
=V (0) and I =I (0) : Voltage and current phasors at V 0 0 the origin.
For computing the frequency response of a multiconductor transmission line at a distance z from the origin, we must determine the voltage and the current phasors. The ABCD matrix of the uniform two-conductor transmission line gives relation between input and output vectors as follows:
A V 0 = I0 YC B
Z V ( z ) B C Z A I Y C C ( z)
(2.a)
= cosh (z ) A
(a)
(2.b)
I01 I02 V01 V02
I1(z)
= sinh (z ) B
V1(z)
(2.c)
^ ^ Zc
I2(z) V2(z)
^ Z L
=Z 1 is the characteristic admittance matrix of Where Y C C the MTL. Using equation (2), we obtain the voltages and currents at an arbitrary position along the line.
(b) Figure 2. Uniform overhead two-conductor transmission line. (a) Description. (b) Electrical model
The MTL is characterized by two matrices [2]: the per (2 2) and the per-unitunit-length series impedance matrix Z (2 2) . The elements of length shunt admittance matrix Y these two matrices are calculated using the formulas presented in [2] and used in [4] under the usual quasi-TEM restrictions:
d << , hi << w , ai << hi , ai << d
(a)
I01 I02 V01 V02
I1(L)
Where and w denote respectively wavelengths in vacuum and conductors. The propagation matrix of the MTL and the are related to matrices Z characteristic impedance matrix Z C by the following expressions [5]: and Y
Y 2 = Z =Y 1 Z C (1.a)
G v Y vi
I2(L) V1(L) Z vi G i V2(L)
^ Z L
(b) Figure 3. Overhead two-conductor line with sag. (a) Description. (b) Electrical model
(1.b)
IV.
INCURVED TWO-CONDUCTOR TRANSMISSION LINE
Here, it is assumed that all the eigenvalues of the matrix 2 are distinct and therefore this matrix is diagonalizable. One known matrix requires exists in the performing this eigenvalues. method for calculating the square root of a that it should be diagonalizable, however, it literature [7] many numerical algorithms operation even in case of possible multiple
A. Description We can identify the curve of the overhead line (Fig.3.a) with the following function: h( z ) = hmin cosh (acosh (hmax / hmin ) (2 z / L 1) ) (3)
The incurved line (Fig.4) is subdivided into N sections of length l that must be a small fraction of the wavelength of the signal. Then the number N and the length l verify the following conditions:
= sinh (l (h ) ) B m m The current delivered by the source is:
(5.e)
N 16 L
f max c
(4.a)
G Z = Y Z +G +Z I 0 vi L i v L vi
)(
V 0
(6)
The load voltage and the current over the line are then calculated with relation (5).
l c 16 f max (4.b)
y
h(z 0)=hmax h(z 1)=h(z 5) h(z 2)=h(z 4) hmin
C. Power along a nonuniform transmission line Our aim is to compute the power delivered by the source and the power dissipated by the load. Hence it is necessary to know currents and voltage of the source and the load. For communication applications, the load is matched to the line. The load matrix impedance is equal to characteristic impedance matrix of the last section of the line.
l z0 z1
l z2
l z3 L z4
l z5 z6
We consider the common mode excitation of the MTL (Fig.6): V01 = V02.
The attenuation of the two conductors transmission line can be deduced by comparing the total active power Pin entering into the system with the power PL flowing into the load.
Figure 4. Subdivision of overhead two-conductor line with sag ( N=6)
G v Yvi
Figure 5. Electrical model of subdivided two-conductor transmission line (N = 4)
Z vi Gi
Figure 6. Common mode excitation of the MTL.
B. ABCD-Matrix for a nonuniform transmission line The electrical model (Fig.5) can be formulated as follows:
( L) G V v = I ( L) Yvi V Z vi 0 I G i 0
1
The total active power over the transmission line is [8]: Pin = 1 T I * Re V 0 0 2
(5.a)
(7)
While the dissipated active power in the load is [8]: PL = 1 T ( L) I * ( L) Re V 2
G v Y vi
Z vi G i
= M
m= N
1 m
(5.b)
(8)
Where (T) denote the transposed not conjugated and (*) the conjugated not transposed.
A 1 m M m = (h ) B Y C m m Z (h ) B m C m Z (h ) A (h ) Y C m m C m (5.c)
The power gain or efficiency of the transmission is given by the following expression [8]: GP = PL Pin
= cosh (l (h ) ) A m m
(9)
(5.d)
D. Enchainment of regularly spaced posts In the case of an enchainment of the k identical cells (Fig. 7), the relationship between the input and the output becomes: ( k L ) G V v = Y ( ) I k L vi
y
hmax hmin Lt L L
g
0.95
Z vi G i
0.9
V 0 I 0
(10)
0.85 Gp
h=10m sag=1m sag=2m sag=3m sag=4m
0.8
0.75
0.7
0.65
10
12
14 16 f (MHz)
18
20
22
24
26
28
30
x
g,
Figure 8. Power efficiency of the two-conductor line for some values of sag. One cell. Lt=100m, d=1m, h=10m. Average soil.
(a)
G v Yvi Z vi G i G v Yvi Z vi G i
0.95
0.9
(b) Figure 7. Chain of two identical cells of overhead line with sag. (a) Description. (b) Electrical model.
0.85
h=10m sag=1m sag=2m sag=3m sag=4m
In relation (10), we compute the power k of The ABCD matrix of the line between two posts. These expressions are used to study lines that have a curvature. It becomes easy to quantify the effect of the sag of an overhead line on the frequency response.
Gp 0.8 0.75 0.7 0
10
12
14 16 f (MHz)
18
20
22
24
26
28
30
Figure 9. Power efficiency of the two-conductor line for some values of sag. One cell. Lt=100m, d=1m, h=10m. Wet soil.
V.
NUMERICAL APPLICATION
0.9 0.85 0.8 0.75 0.7 0.65 0.6 0.55 Gp 0.5 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0 2 4 6 8 10 12 14 16 f (MHz) 18 20 22 24 26 28 30 0.45 h=10m sag=1m sag=2m sag=3m sag=4m
We make a numerical application to evaluate the effect of the sag on the power efficiency at high frequency. During simulation, it is assumed in all cases that spacing between wires is d = 1m, the maximal height of the wires is equal to hmax = 10m. The transmission line has a span of L = 100m and is working in common mode excitation (V01 = V02). We consider average (g = 0.005S/m, g = 13) and wet (g = 0.01S/m, g = 30) soils. Then we show (Fig.8, 9, 10 and 11) the power gain for the uniform straight line (hmax = 10m) and for the incurved line with many values of the sag (1, 2, 3 and 4 m). We carry simulations for one cell (Lt = 100m) and five cells (Lt = 500m). We notice that the power gain is minimal between 4 and 6MHz, it decreases until 0.15 for an overhead line above ordinary soil with sag of 4m. It is also underlined that the wet soil always shows better power gain. For the case Lt = 500m (Fig.10 and 11), if the sag is equal to 4m, the power efficiency is near the half of the value calculated for an ideal straight line.
Figure 10. Power efficiency of the two-conductor line for some values of sag. Five cells. Lt=500m, d=1m, h=10m. Average soil.
1 0.95 0.9 0.85 0.8 0.75 0.7 0.65 Gp 0.6 0.5 0.45 0.4 0.35 0.3 0.25 0.2 0 2 4 6 8 10 12 14 16 f (MHz) 18 20 22 24 26 28 30 h=10m sag=1m sag=2m sag=3m sag=4m 0.55
Figure 11. Power efficiency of the two-conductor line for some values of sag. Five cells. Lt=500m, d=1m, h=10m. Wet soil.
VI.
CONCLUSION
This work shows clearly that the sag of an overhead line has a significant influence on the power efficiency and therefore on the useful range of a signal transmitted over this kind of line when the common mode is excited.
REFERENCES
[1] [2] J.R. Carson, Wave propagation in overhead wires with ground return, Bell Syst. Techn. Journal, Vol.5, 1926, pp.539-554. M. D'Amore, M.S. Sarto, A New Formulation of Lossy Ground Return Parameters for Transient Analysis of Multi-Conductor Dissipative Lines, IEEE-T-Power Del., Vol.12, n1, January 1997, pp.303-314. R.G. Olsen, M.D. Wu, High Frequency Propagation Losses on an Open Wire Transmission Line Above Dissipative Earth, IEEE-T-Broadcast., Vol.34, n2, June 1988, pp.292-300. P. Amirshahi, M. Kavehrad, High-Frequency Characteristics of Overhead Multiconductor Power Lines for Broadband Communications, IEEE Journal on Selected Areas in Communications, Vol.24, n7, July 2006, pp.1292-1303. C.R. Paul, Decoupling the Multiconductor Transmission Line Equations, IEEE-T-MTT, Vol.44, n8 , August 1996, pp.1429-1440. M. Muhr, S. Pack, R. Schwarz, S. Jaufer, Calculation of Overhead Line Sags, 51st Internationales Wissenschaftliches Kolloquium, Technische Universitt Ilmenau, 11-15 September 2006. N.J. Higham, Functions of Matrices: Theory and Computation, SIAM, Philadelphia, PA, USA, 2008. S.J. Orfanidis, Electromagnetic Waves and Antennas, Rutger University, USA, 2008.
[3]
[4]
[5] [6]
[7] [8]
Power Converters Variable Structure Control surface design

Bashar Khasawneh, Maha Sabra, M. Zohdy
Dept. Electrical and Computer Science Oakland University Rochester, MI USA Bykhasaw/Msabra/Zohdyma@oakland.edu
AbstractVariable structure control (VSC) is an effective method for robust control of nonlinear systems. In this paper, VSS design is applied to Buck Boost DCDC converter. The algorithm developed in [1] extended to introduce a new concept for DCDC converters. The main contribution of this paper is the introduction of the duty cycle as a main part in the design of the sliding surface. We will show how to design the sliding surface and maintain control on it with load changes that requires a duty cycle change. Generalized results based on switching duty cycle are presented. Simulation of different duty cycles also presented. Keywords-DCDC converters; Variable structure; surface design; sliding surface; hyper plain
The states are the inductor current and capacitor voltage forming the following
Figure 1. Buck-boost converter
x= i v I. INTRODUCTION The use of variable structure control is increasing due to the wide range of applications that requires high switching control. The improvement in switching devices and their efficiencies, improved MOSFETs and IGBTs made faster switching on the order of hundreds of KHz possible. In a variable structure system (VSS) the structure is changed to achieve the desired output; the structure change is typically accomplished through defined switching based on the states of the system. A variable Structure System (VSS) structure is intentionally changed to achieve a desired performance. This intentional structure change is typically accomplished through fast switching between different control actions [1], [2], [5], and [7]. This makes VSS and Variable Structure Control (VSC) an excellent and reliable control method for the power electronics systems such as the DCDC discussed in this paper. II. CONVERTERS MODELING In this paper we will extend the VSC control to apply to DCDC converters as a function of the changing duty cycle. First we will go through the modeling of Buck-Boost converter and develop an average state space model as done in many text books and published papers. The buck-boost converter is shown in fig. 1. In the operation of the BuckBoost converter, the output voltage can be less or greater than the input voltage hence the name buck boost. The switch denoted by S in Fig. 1 can take the values 0 for open and 1 for closed [9], [10], [11].
(1)
First position, the switch is closed and the corresponding equations can be found to be A = B = 0 0 0 1/RC (2)
1/L (3) 0 Second position, the switch is open the corresponding equations can be found to be A = B = 0 1/C 1/L 1/RC (4)
0 (5) 0 Using (1) through (5) and duty cycle d, then the system can be represented by the well-known average model as follows where : is the input voltage. x= A d B d III. A (1 B (1 d) x d) u (6)
DESIGN OF A SLIDING SURFACE FOR THE BUCKBOOST
The design of a hyperplane determines the desired behavior of the VSC. Conceder the linear time invariant system given by the state space representation of the BuckBoost converter
x = Ax s = Cx Where x R ,u R ,s R
Bu
(7) (8)
(n-m)=1 and we chose the Eigen value to be -2 as an example and the associate Jordan matrix J=-2 Solve equation (15) AG GJ = BL (15)
L is an arbitrary m x (n-m) matrix; chosen to be as an example A (1 B (1 d) d) L= [2] Solving for the variables G we get G= [G1; G2] = [0; -0.5] Solve for the generalized B using the following B|G = B (16) G Solving the above equation we get the generalized vectors below. B = [0.005; 0] G = [0, -2] Let C = NB where N is m x m matrix selected as the identity matrix I
A= A d and B= B d s is the sliding surface
The control low will be of the following form in term of s(x) as described in [1], [3] u (x) = u (x), u (x), s (x) s (x) 0 0 (9)
It is assumed that the matrix CB is non-singular and both C and B are full rank. In this paper we will demonstrate the process of selecting proper sliding surface for a DCDC converter and the effect of system parameters on it .The condition to guarantee the motion on the surface s(x) = 0 is given by ss 0 (10) In the neighborhood of s for all i=1, 2m. and the system states will satisfy the following equations: s(x) = Cx = 0 s(x) = Cx = 0 (11) (12)
The sliding surface s(x) = Cx for the buck boost converter is given by the following equation: s(x) = 0.005x 1.25 10 x (17) The above sliding surface corresponds to a duty cycle d=0.2. The coefficient of x is small with respect to the coefficient of x . Changing the duty cycle changes sliding surface and a new one is generated. If we change d=0.5 then the new surface is found following the same steps as above to be: s(x) = 0.025x 0.8 10 x (18) Finding the sliding surfaces for incremented duty cycles is shown in TABLE. 1 where Q and W are defined in the next section
TABLE I. Duty 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Q 1.0e-2 0.5e-2 0.33e-2 0.25e-2 2.0e-2 0.17e-2 0.14e-2 0.13e-2 0.11e-2
SLIDING SURFACE COEFFICIENTS VS. D
Using (12) and (7) we get the equivalent control discussed by Utkin [3], [4]. It is developed to derive the sliding mode equations into the manifold s(x) = 0 and then the solution to s(x) = 0 is called the equivalent control u . u = (CB) CAx (13)
Then using (13) and (7) we obtain the equivalent system in sliding mode [5] x= I B(CB) A x = A x (14) The generalized inverse approach selects a switching that assigns the (n-m) free Eigen values and the associated Eigen vectors [1], [4]. Given the Buck-Boost converter with the average model as in (6), we will design a switching surface based on the generalized inverse approach. The converter is basically a two state converter where the states are defined by the inductor current and the capacitor voltage. Typical application is to control the output voltage to a desired value. First we define n the number of states, therefore n=2, then m =1. The generalized inverse approach procedure [1],[2],[6]: Select (n-m) eigenvalues for the system in 8 and form associated Jordan matrix J
C W 2.22e-6 1.25e-6 0.95e-6 0.83e-6 0.80e-6 0.83e-6 0.95e-6 1.25e-6 2.22e-6
Going through the mathematical analysis to get (16) interim of the duty cycle and for the given converter
parameters, we will present the relation between the surface s(x) and the duty cycle d. This will reduce the computational analysis for each duty cycle and will produce a direct expression that depends on the duty cycle. Using the converter parameters, L, C and R and using the average model containing the duty cycle d we solve (15) Given the following using the circuit average model: A= G= J=J 0 Where l is an arbitrary positive number. BL = l
L
IV.
EXAMPLE
Let L = 10mH, C=100uF and R=1 with l=2, N=1 and the pole placed at -10. Using Matlab we find the following results: S(x) = 0.05x S(x) = 0.0167x 6.25 10 x 4.16 10 x (32) (33) For a 0.6 duty ratio the surface changed to the new surface To generalize the results for the buck-Boost converter and from the derivation above we can find the surface based on the given parameters directly. S(x) = Qx Where Q = W=
L L
C ( LR C R ) C ( LR ( CL ) ) C( R ) ( CJL ) R )
0
( C R )
(19)
R
Wx
(34)
G1 G2
(20) (21) (22)
J L
( ) CJL
CJ(
(35)
For simplicity using the previously defined, a, b, n, m and p we get W=

L(J (
L J
Using equation (15) and substituting (19) to (22) we get the following:
L ( R
)( )(
J J
J)
(36)
G2
)
JG1 = JG2
L C
(23) G1 =
C R
G2
(24)
Based on the circuit parameters and the desired Jacobean matrix we can vary the sliding surface of the power converter instantly as the duty cycle changes to accommodate the desired output. V. SURFACE DESIGN SIMULATION To simulate the behavior of the surface with the change in duty ratio the states coefficients Q and W in (34) were computed and plotted below:
For simplicity let a=

( R )
,b =
,n =
C R
,m =
,p =
Rewrite (23) and (24) we get the following equations (a mG2 J)G2 bG1 = n (25)
W 0 x 10
-7
Variable Structure Surface S(x)
JG1 = p
J J J
J
(26)
-0.5
Solving (25) and (26) for G1 and G2 we get G1 = G2 = (27) (28)

J
-1
-1.5 0.1
0.2
0.3
0.4
0.5 0.6 Duty Ratio (d)
0.7
0.8
0.9
0.1
0.05
Now from equation (16) B = B G G Where g denotes generalized (30)
0 0.1
0.2
0.3
0.4
0.5 0.6 Duty Ratio (d)
0.7
0.8
0.9
Figure 2. The coefficients of the proposed hyper plane
Using (27) to (30) and solving for the generalized B matrix we get
G T
B =
B G B G
B G G B G
(31)
Vin Duty Vout d
[d]
[Vin]
1/L Step Product2 Gain1 1 s
Duty cycle Calculation [IL
0.5 d u t y c y c le
IL
Integrator1
IL
1/L Product1 [U] Gain3

Vc
Vc S(x)
[S]
[d]
Subsystem
1 Constant3 Product 1/C Gain2 1 s

Vc Vc
-0.5 0.02
-1 Gain4
vc
[Vout]
Integrator 1/(R*C) Gain

Vc
0.01 S(x) 0 0 0.02
0.04 Time
0.06
0.08
0.1
Figure 3. Open loop buck boost converter
Figure 6. The S-d plot with time
As a result of the simulation, for the given parameters x coefficient is the dominant factor in forming the surface. We conclude that the inductor is the main player in the Buck Boost converter. In the design of a DCDC converter the proper sizing of the inductor is necessary to reach the desired output, and as seen from the hyper plane design it has the major effect on the reaching conditions. A Buck Boost converter is simulated with a constant duty cycle using a pulse generator to show the effect of the different duty cycles on the surface design.
50 0 -50 V
Fig. 6 shows the behavior of the surface with a constant duty ratio over time. In a closed loop VSC this surface is what the local control is designed to maintain to achieve the desired performance. The initial control to reach the desired surface is achieved by the equivalent control. A closed loop controlled buck boost converter is simulated as shown in Fig. 7
Vin Duty Vout d
[d]
[Vin]
1/L Step Product2 Gain1 1 s
Duty cycle Calculation [IL]
IL
Integrator1
Vout
1/L
IL Vc Out1
[S]
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
Product1 [U]
Gain3
Vc
[d]
Subsystem
51 Vin 50 49 V
1 Constant3 Product 1/C Gain2 1 s
Vc Vc
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
-1 Gain4
vc
[Vout]
Integrator 1/(R*C)
10 Amp IL 5
[Vout]
Gain
Vc
65
V2DC*
[Vin] Constant
0.01
0.02
0.03
0.04 0.05 0.06 Time [sec]
0.07
0.08
0.09
0.1
[IL] [d] [S]
out_VSC.mat To File [Vout]

V2DC
Gate Driv e
[U]
Figure 4. Output voltage and current of the Buck Boost converter open loop control d=30%
0.02
Controller
Figure 7. Closed loop Buck Boost converter model
The simulation results for a desired 60v output are shown in Fig. 8, 9 and 10 below.
S(x) 0.01
0.1
0.2 0.3 duty cycle
0.4
0.5
Figure 5. X-Y plot of the d vs. S for different duty cycles
100 Vout 50 0 V V
100 Vout 50 0
0.02
0.04
0.06
0.08
0.1
0.02
0.04
0.06
0.08
0.1
100 Vin 50 0 V V
100 Vin 50 0
0.02
0.04
0.06
0.08
0.1
0.02
0.04
0.06
0.08
0.1
40 Amp 20 0 Amp IL
30 20 10 0 0 0.02 0.04 0.06 Time [sec] 0.08 0.1 0 0.02 0.04 0.06 Time [sec] 0.08 0.1 IL
Figure 8. Output voltage and current of the Buck Boost converter

0.1
Fig. 11 shows the result for a step command in the voltage from 30v to 60v and the increase in the duty cycle.
0.1
0.08
0.06
0.08
S(x)
0.04
0.06
S(x)
0.02
0.04
0.02
-0.02
0.1
0.2
0.3
0.4
0.5 0.6 duty cycle
0.7
0.8
0.9
-0.02
0.1
0.2
0.3
0.4
0.5 0.6 duty cycle
0.7
0.8
0.9
0.8 0.6 duty cycle 0.4 0.2 0 -0.2 4 2 1 S(x) 0 0 0.04 0.02 Time 0.08 0.06
Fig. 12 shows the system reaching the equilibrium point for the first command and then for the second voltage command.
0.6
0.4 duty cycle

3 0.1
0.2
Figure 10. The S-d plot with time
Fig. 9 shows the d- S plot and Fig. 10shows the d-S plot over time. The results show how an appropriate control will force the system to the desired surface in the shorts path. Now we will explore the effect of a change in voltage demand on the behavior of hyper plain. A closed loop system is simulated with a step voltage demand.
-0.2 4 2 0 0.02 S(x) -2 0 Time 0.06 0.04 0.1 0.08
Fig. 13 showing reaching the surface for the step change in the output voltage .
Next we will simulate the effect of load change on the hyper plain. The load was increased to double the original value, the results are shown next. Fig. 14 shows the output voltage being maintained and the change in the current. Again the duty has minimum impact and the surface stays in the same plain.
100 Vout 50 0 V
Fig. 15 and fig. 16 show the system reaching the equilibrium points for a change in the load. VI. CONCLUSION
0.02
0.04
0.06
0.08
0.1
In this paper we have developed a relation between the hyper plane and the switching duty cycle for a Buck-Boost converter (34) to (36). The relations are simulated and the results are shown. By relating the hyper plain to the duty cycle, computation time of the method discussed in [1], [2] are reduced. REFERENCES
[1] M. Zohdy, M.S. Fadali and J. Liu, "Variable structure control using system decomposition," Automatic Control, IEEE Transactions on, vol. 37, no. 10, pp. 1514-1517 1992. [2] M.A. Zohdy, M.S. Fadali and J. Liu, "Variable structure dynamic output feedback," in American Control Conference, Proceedings of the 1995, 1995, pp. 1518-1522 vol.2. [3] V. I Utkin, and K.D. Yang, Methods for constructing discontinuity planes in multidimensional variable structure systems, Automat. Remote Contr. Vol. 39, no. 10, part 1 pp. 1466-1470, Oct. 1987. [4] V. Utkin, "Application of equivalent control method to the systems with large feedback gain," Automatic Control, IEEE Transactions on, vol. 23, no. 3, pp. 484-486 1978. [5] Liu, Wen-Jeng. "Decentralized control design for a class of non-linear large-scale systems with matched and unmatched uncertainties." Transactions of the Institute of Measurement and Control 33.5 (2011): 631-644. [6] O.M.E. El-Ghezawi, S.A. Billings and A.S.I. Zinober, "Variablestructure systems and system zeros," Control Theory and Applications, IEE Proceedings D, vol. 130, no. 1, pp. 1-5 1983 [7] R.A. DeCarlo, S.H. Zak and G.P. Matthews, "Variable structure control of nonlinear multivariable systems: A tutorial," Proceedings of the IEEE, vol. 76, no. 3, pp. 212-232 1988. [8] Lee, Chan-Soo, et al. "Integrated BiCMOS Control Circuits for HighPerformance DC-DC Boost Converter." IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 28, NO. 5, MAY 2013 . [9] Jun Liu, Ping Yang, Xucheng Lin and Shaoxiong Zhou, "Modeling and simulation of DC/DC converters based on double-loop control," in Power Electronics Systems and Applications, 2009. PESA 2009. 3rd International Conference on, 2009, pp. 1-4. [10] Tze-Yee Ho, Mu-Song Chen, Chih-Hsien Lin and Chen-Wen Chang, "The design of a flyback converter based on simulation," in Electronics, Communications and Control (ICECC), 2011 International Conference on, 2011, pp. 3996-3999
100 Vin 50 0 V
0.02
0.04
0.06
0.08
0.1
40 Amp IL 20 0
0.02
0.04 0.06 Time [sec]
0.08
0.1
0.1
0.08
0.06
S(x)
0.04
0.02
-0.02
0.1
0.2
0.3
0.4
0.5 0.6 duty cycle
0.7
0.8
0.9
0.8 0.6 duty cycle 0.4 0.2 0 -0.2 4 3 2 1 S(x) 0 0 0.04 0.02 Time 0.08 0.06 0.1
Excitation Current Control of A Claw Pole Automotive Alternator

M. Gkay Unutulmaz
Calibration and Controls Ford Otomotiv Sanayi A.. Kocaeli, Turkiye
Lale T. Ergene
Faculty of Electrical and Electronic Engineering Istanbul Technical University Istanbul, Turkiye The Lundell type of alternator with claw pole rotor is the first choice of the vehicle manufacturers as an electrical power generator. This type of alternator can satisfy the required electrical power of the vehicle with low cost. However, as increasing the required power, the alternator manufacturers prefer to apply some different design methods to the machine while they keep main construction as the same. The benefits of these methods are performance increase, cost reduction, size and weight reduction of the alternator. Also voltage regulators and alternator output rectifiers are other subjects of total efficiency increase methods. The Lundell alternator can be classified in AC type electrical machines. Lundell machine basically is a synchronous machine with a DC excitation however it has a special rotor construction. This construction is more robust to the centrifugal forces than the ordinary rotor design of other type of synchronous machines. II. FIELD WINDING MODELING
AbstractThis paper overviews the Lundell type of automotive alternator with claw poles. The machine construction will be summarized and the field winding will be the focus point. Mathematical equations of the field winding, field current and the excitation circuit losses will be determined. A simulation model that contains the detailed excitation circuit parameters will be established. Results will be compared with different PI controller coefficients. Keywords; alternator, claw pole, field winding, field winding losses, excitation current control
INTRODUCTION I. Consumer preferences and competitive pressure force the vehicle manufacturers to add new specifications to their products. The rising trend in the comfort and the safety results with more electrical power requirement. Not only new technologies and specifications but also changing mechanical driven components to electrical driven, the consumption of the electrical power has been rising dramatically. Claw pole Lundell type of alternator has been using as a component of the vehicle to charge the battery and supply the electrical loads for years. In fact, the alternator was capable of supplying required electrical power of the vehicle but as the technological growth of the vehicles and electrification of the vehicle systems, the efficiency of the traditional charging system is called into question.
A. Rotor and Claw Pole Construction The excitation current and the rotor magnetic flux of the machine are DC, as a result of this neither hysteresis nor the Foucault losses are assumed to be generated in the rotor core. Consequently, there is no need to build the rotor core with laminated sheet metal. The rotor magnetic field is created with a rotor coil which is coiled around a magnetic core. The current flows through the slip rings. When the current starts to flow, the rotor becomes an electromagnet and magnetic field is created. This strong magnetic field flows through the two main pieces of rotor core and each of them becomes a pole N or S.
Figure 1. Lundell type of alternator with claw poles. [1]
Figure 2. Claw pole rotor construction. [2]
The shape of the rotor core, in other words, claws of the rotor results the magnetic field flow alternately with respect to the stator reference. The two ends of the rotor winding are attached to the slip rings. The excitation current can be controlled from outside of the machine via the brushes. B. Equivalent Circuit and Field Winding Mathematical Modelling The claw pole Lundell alternator is classified in synchronous AC machines. The excitation circuit consists of slip-ring voltage drop and losses, excitation winding resistance and inductance. On the other hand, the 3 phase stator contains armature winding resistance and inductances. The detailed equivalent circuit of this machine that contains excitation circuit and armature circuit is given below.
The switching element will be assumed as a MOSFET and the duty cycle refers to the on position duration of the period. The excitation circuit can be modeled in more detailed as in the figure below.
Figure 5. Detailed view of the excitation circuit
The excitation circuit commonly works in three different stages with respect to MOSFET working regions which are conduction mode, switching mode and cutoff mode. A. Conduction Mode The MOSFET is driven by a constant PWM frequency with a duty cycle D. During the status of conducted MOSFET, the power is supplied from the battery and current flows over the MOSFET drain to source resistance, slip ring and brush contact resistances, excitation current resistance and the excitation winding inductance.
Figure 3. One phase equivalent circuit of the alternator
Here the rf is the field winding resistance and Lf is the field winding inductance. On the stator side, Ef is the induced voltage La and L are the self and leakage inductances and rs refers to the stator winding resistance. Field winding general mathematical expression can be determined by applying the Kirchhoffs rules to the equivalent circuit above. Vf = Ifrf + Lf dIf/dt The excitation circuit self inductance can be written as, Lf=f/If (2) (1)
Figure 6. Equivalent circuit while the switching element is in conduction mode
III. EXCITATION CIRCUIT MODELING Step-down SMPS is controlling the excitation current by switching the battery voltage with a modulation frequency. Thus results with a reduced output voltage from the SMPS which is applied to the excitation winding. Due to the high pulse width modulation (PWM) frequency and inductance effect, the current ripples could be negligible and the current can be approximated as a constant average value.
The conducted MOSFET loss can be approximated like [5] PCondMode=(If)2.RDSon.D (3) Here, RDSon is the drain to source resistance of the switching component in on position and D is the duty cycle. B. Switching Mode Switching position is the transient state between the on and off positions. The voltage and current waveforms for all working regions are plotted in the figure 7. Here the switching losses can be calculated by applying a triangular area geometry. Calculating the switching power dissipation during the off-on and on-off transitions results with the equation below [5], PSwMode=1/2.Vcutoff.Icon.(tswon+tswoff).fsw (4)
Figure 4. Duty cycle refers to the high side ratio of the period
Pbl=2.Rb.(If)2
(7)
where Rb is one brush contact resistance and If is the rotor average current. IV. EXCITATION CURRENT CONTROLLER DESIGN The claw pole Lundell alternator machine is a variable speed generator and excitation current is the component that used to regulate the output voltage directly. However, this voltage regulation can not be very fast in case of transient situations because of the high excitation circuit inductance. The induced voltage equation can be written for the claw pole alternators as [3,4],
Figure 7. Working regions of a switching element
E = If.km.wg
(8)
The point here is switching losses are directly proportional to the PWM frequency. If frequency is high to make a smooth excitation current, also switching losses will be high. The total duration of the switching element is so small that the whole period can be divided into two main phases of conduction mode and cutoff mode. C. Cutoff Mode In a period, the duration except the conduction state of the MOSFET is the off position and showed in a period as 1-D. At that period, the energy collected in the inductance flows over the fly wheeling diode.
Here, If is the excitation current and the wg is the alternator angular mechanical speed. km refers to the machine constant and contains machine construction parameters. The Lundell alternator voltage output of a conventional 12V charging system is around 14-15V due to the battery state of charge and actual electrical loads. If the battery is discharged or additional electrical loads are active then the output voltage is much higher than 12V and the alternator can supply more current both to charge the battery to supply the loads. Because of DC voltage is needed to charge the battery, an AC/DC rectifier is has to be used. The manufacturers prefer to use 3phase full bridge diode rectifier due to its low cost. In figure 9, the block diagram of the charging system model is given. In this paper, the focus point is the excitation current control.
Figure 8. Equivalent circuit while the switching element is in cutoff mode.
Figure 9. Conventional vehicle electrical charging system diagram
Here the wasted energy over the diode can be calculated as PCOffMode=IDiode.VDiode.(1-D) (5)
D. Rotor Copper Loss Rotor copper loss is a result of the winding resistance. The loss is proportional to the winding resistance and the square of phase current. The loss is formulated as Prcl=3.Rs.(If)2 (6)
In the MATLAB model, the excitation circuit will be modeled as a RL load. Because the PI controllers are easy to calibrate with acceptable success, the other vehicle systems are generally controlled by the PI controllers and PI control method is used commonly in automotive industry. In this model, excitation current will be controlled by a PI controller instead of other type of control theories to prevent the complexity. General overview of the excitation current controller model is in figure 10. The first component of the model is a voltage set point generator. This block supplies the required DC voltage for battery charging and the loads. The second component contains the switching element mathematical model which has a 1 kHz switching frequency. Also a PI controller is plugged in to the model to control the excitation current. The third parameter of the model includes general solution for the excitation circuit which is figured out in the figure 5.
where; Rs is the rotor resistance and If is the one phase stator current. E. Slip Ring Brush Loss The excitation current flows from the controller to the winding over a slip-ring and brush mechanism. The contact resistance of the brushes results with a loss and can be formulated as
This part of the model contains field winding resistance and inductance, brush resistance and switching element forward resistance. Fourth block of the model solves the voltage inducement and other stator equations. The final part of the model calculates the losses of the excitation current control circuit and the outputs are the losses of the circuit elements.
In figure 12, the integral controller also included the controller to eliminate steady state error. Different integral gains are applied to the controller. The observation here is, integral gain can eliminate the steady state error however, if the gain is calibrated too much, this results with over shooting of system output.
Figure 12. DC voltage setpoint and actual DC voltage comparison with PI controller. Figure 10. MATLAB Simulink model of the excitation current controller
V. RESULTS The model ran with different PI controller coefficients to compare the controller performance and the settling time. The output graphics and controller characteristics of different PI data sets are presented below. In the figure below, the model ran with only a P controller. Different P gains are applied to the model and it is observed that using a P controller results with a steady state error. The amount of the error is depends on the P gain and as the P gain increases, the steady state error decreases in direct contradiction.
Figure 13 is the result of eliminating the steady state error by applying an integral gain. But, the system still has slow response to input.
Figure 13. current with P gain equals to 5 and I gain equals to 10
Figure 14 shows the system behavior with increased P gain and steady state error elimination with increasing the integral gain.
Figure 11. DC voltage setpoint and actual DC voltage comparison with P controller
smaller because the voltage set point is around 14V and nearly constant during a high alternator speed.
Figure 14. Decreased system response time with PI controller
In the next figure current response of the excitation circuit to 14V DC input is plotted. In addition, current characteristics with constant P and variable I gains are plotted.
Figure 16. Duty cycle and actual voltage comparison
Figure 17 compares the duty cycle and the actual voltage response of the controller. System start behavior is similar with the observations from previous figure. In addition, step responses of the system against different voltage set points are observed.
Figure 15. Excitation current characteristics.
The losses are also calculated with selected PI controller gains P=10 and I=50. With a 1.8A field current and 14.8V output voltage, the main part of the electrical losses is field winding resistive loss with a 75% of the total electrical losses. Slip-ring brush loss is 17% of the electrical losses. MOSFET conduction loss is directly depends on the switching element. In this model, the conduction loss is around 1.2% of total electrical losses during a 40% duty cycle. Fly wheeling diode holds a portion of 7% of the loss because of the voltage drop over the diode. Switching loss of the MOSFET is so small that can be omitted in low frequencies. In the model, controller effort directly regulate the switching element duty cycle. As a result of this, controller effort and duty cycle are directly proportional to each other. In figure 16, system start characteristics with different PI controller coefficients are investigated. Duty cycle is high at the start time to fill the RL circuit with required energy. Overshoots of the controller are observed because of the high integral gains. After the first system start, the duty cycle gets
Figure 17. Controller effort and duty cycle with P gain equals to 4 and I gain equals to 50
In figure 18, 19 and 20; loaded conditions of the alternator are showed. Head-rear lamps and full illumination lamps are used as loads. Load values are determined with the measurements from the vehicle. Load states are showed in figure 20. DC bus voltage set point and the actual voltage are plotted of a part of the simulation in figure 18. Voltage-drops can be seen during 100ms before the excitation controller satisfy the voltage set point. The change in the excitation current at the transition of loads is plotted in figure 19.
Voltage Setpoint vs Actual DC Bus Voltage Under Load Conditions Voltage Setpoint Actual Voltage
14.4 Voltage (V) 14.2 14 13.8 13.6 13.4 570 571 572 573 574 575 576 Time (Second) 577 578
579
580
VI. CONCLUSION The excitation current controller output simulated with the measured excitation circuit parameters as resistances and inductances of an alternator. A PI controller is used to control the field current. The P gain is chosen to reduce response time and steady state error and I gain is determined to satisfy minimum over and undershoots as well as steady state error. The final response time of the model calculated at around 100ms. The response time can be re-calibrated by changing PI coefficients to reduce the voltage spike concerns.
Figure 18. DC bus voltage set point and the actual voltage in load condition
Excitation Current Under Load Conditions 2.9 2.85 2.8 Current (A) 2.75 2.7 2.65 2.6 2.55 2.5 570 571 572 573 574 575 576 Time (Seconds) 577 578 579 580
REFERENCES
[1] D.M. Whaley, W.L. Soong, and N. Ertugrul, Extracting More Power From The Lundell Car Alternator, in Australasian Universities Power Engineering Conference. University of Adelaide, Australia, 2004. W.H. Crouse, Automotive Electrical Equipment, McGraw-Hill, USA. 1976. D. Stoia, M. Cemat and R. Rabinovici, An Electromagnetic Model For Lundell Alternator With Switched-Mode Rectifier, IEEEI, 2008. D. J. Perrault, V. Caliskan, Automotive Power Generation and Control, IEEE Transactions on Power Electronics, Vol 19, No. 3, May 2004. H. Bodur, G Elektroni i Temel Analiz ve Saysal Uygulamalar, Birsen Yaynevi, p. 374-375, Istanbul 2012. C.P. Cho, D.R. Crecelius, Vehicle Alternator/Generator Trends Toward Next Millennium, Vehicle Electronics Conference, (IVEC '99) Proceedings of the IEEE International, Changchun, China, 1999.
Figure 19. Excitation current in load condition
[2] [3]
Load States of The Vehicle 1 0.8 Load States 0.6 0.4 0.2 0 570 571 572 573 574 575 576 Time (Second) 577 Head and Rear Lamps Full Illumination 578 579 580
[4]
[5] [6]
. Figure 20. Load states
A Reliability Study Of Medium Voltage Grid With Private Sector Power Plants
Levent KILI
Trkiye ie ve Cam Fabrikalar A.. Kuleleri Kule 3 34330 4.Levent stanbul lkilic@sisecam.com
Ayen BASA ARSOY

Elektrik Mhendislii Blm Mhendislik Fakltesi Kocaeli niversitesi, 41100 zmit Kocaeli aba@kocaeli.edu.tr
Abstract - The number of private sector power plants connected to the national grid has been increasing recently. Those plants with various in sizes may have a positive contribution to the grid. Numerous calculations are carried out in equipment selection and protection setting to integrate the power plant to the grid. Not only these calculations but also a proper operation of a plant are so important in improving system reliability. This is mostly considered as the base case for the worst scenario. In this paper, a reliability evaluation study of private sector power plants connected to an MV grid is given in terms of plant and grid reliability and reserve margin. Data taken from a real power plant operation is used for this study. Obtained results are compared to show the improvement in grid reliability. Keywords power plants; grid; reliability; reserve margin
production, and to ensure main industrial production continuity under adverse network conditions. In this article, based on the real data from private sector power plants in two different distribution regions, a reliability assessment is carried out by considering 1.) reliability of power plants, 2.) reliability of the network, and 3.) the reserve evaluation of plants connected to the grid at medium voltage (MV) under normal conditions. II. RELIABILITY EVALUATION
I.
INTRODUCTION
Reliability can be defined as the ability of an equipment/system to perform its function under specified conditions and time. The impact of the power plant capacity expansion should be well evaluated in terms of the reliability of the plant and the grid. This section presents the required expressions for power plant reliability , grid reliability and reserve margin. A. Power Plant Reliability The power plant capacity expansion occurs frequently. The increase in load demand as well as the change in conditions being an active player in the market may result in adding more machines in the power plant. The impact of power plants should be analyzed well when integrating them into the grid as they are expensive investments. The results of analyses will become more accurate if complete and correct data from power plant and distribution network are provided. Otherwise, this may cause of improper equipment selection and inapropriate operation of a system. American Reliability Council described the reliability formulas for the power plants as shown in Equation (1-2) [6]:
Increasing number of private sector power plants connected to privatized distribution regions in our country, reveals two important evaluation criteria: The first one is the modeling of the distribution network under normal conditions and the second one is to determine the response of the system against to normal and faulty conditions. Permission for grid integration is given based on a network structure planned for steady state operation with several restrictive criteria [1-4]. In today's competitive market, grid connection criteria must be the same and known by all participants and be standard for all regions. Industrial enterprises have their own investments to increase their energy efficiency and reduce power quality problems from the grid. These investments are mainly on cogeneration and combined cycle power plants. Electricity generated by this type of thermal power plants are used for industrial production and make a contribution to the national grid need [5]. The power plants connected to the grid at MV level are designed to maintain synchronism under normal conditions, to operate in island mode under faulty conditions, to establish load shedding for partial
AV
D ( B A) D
(1)
D ( B A) ( D B)
size and reliability factors. This is expressed in Equation (11). (2)
V
i
2
(11)
Where AV is availability, R is reliability, D is period [Hour, Day, Month, Year], B is maintenance time [Hour, Day, Month, Year], and A is failure time [Hour, Day, Month, Year]. In a power plant, if n is the number of unit with similar size and reliability factor, the probability of (m) number of units being on can be calculated as given in Equation (3) using (P) binomial distribution.
P m n! m! n m!
Each plant variance ( i ) is estimated on the basis of the actual plant operating experience. Power system needs reserve to cover the deficiency because of the unreliable/unsafe power plants. Reserve requirement (RR) can be defined according to the formula given below based on n sigma confidence level (Z):
R 1R
m
n m
(3)
RR 1 Ri .Pui Z .
(12)
The following expressions can be obtained by simplifying the binomial expansion for events a and b [7,8]: Generalized form for n events is given as follows:
In most cases, the largest plant in a power system determines the reserve requirement. The reserve requirement can be shown simply as follows:
P a b 1 (1 P a )(1 P b ).........(1 P n)
(4)
RR 1 Ri .Pui Pmaks
(13)
B. Grid Reliability Reliability is evaluated as the number and duration of the interruptions in the distribution system. Some of the reliability indices can be defined as follows [9]:
Total Number of Customer Interruptions Total Number of Customers Served
Assuming that the highest rated power plant meets 10% of the system load, if the reliability of system without this plant, is 95 %, the reserve requirement will be 15 % ((100% - 95% ) +10% ) [10]. Reserve Margin (RM) taken into consideration when planning the power system can be expressed as follows:
SAIFI
(5)
Customer Interruption Durations SAIDI Total Number of Customers Served

CAIDI Customer Interruption Durations Total Number of Customers Interruptions
RM
(6) (7)
Capacity Peak Load Peak Load
(14)
Reserve power requirement at Z sigma level can be written as in Equation (5):
C. Reserve Margin The power output of a power plant can be calculated as given in Equation (8), where , n is the number of unit, R is the reliability factor and P is the rated power of a single unit. Standard deviation and the variance of the output are also shown in Equations (9) and (10), respectively.
RM
P n.R.Pu
(8) (9) (10)
1 R Z . / n R Z . / n 1 R Z . / n RM R Z . / n
n.Pu n.Pu .R Z . .Pu n.Pu .R Z . .Pu

(15)
V
V N .R.(1 R).Pu2
Reserve margin (RM) depends on the unit number and reliability of each unit, as can be seen from Equation (15). RM yields to
1 R / R ,
when n goes to infinity.
Standard deviation is the square root of the sum of variance of each unit for plants having units with different
Therefore, the value of RM is expected to be 3-11% with
infinite number of units if the reliability of units, R, changes between 90 -97 %. III. REAL CASE STUDY
system are lower before the connection. When a proper coordination and effective management is provided , the reliability of 100 MW busbar can go up to 99.84% from 96.00% . The reserve margin (RM) is dependent of the number of units. Assuming that each generating unit is 97% reliable, while the reserve margin is 14.91 with 10 units, it goes up to 30.58 % with 5 units providing the same amount of load as 10 units. The reserve margin becomes 17.61 % with 10 units and 33.24 % with 5 units when the reliability of each unit is assumed to be 95 %. From these results, it can be recommended that conventional and highly reliable power plants can be considered as a reserve when designing a power system. IV. CONCLUSION AND SUGGESTIONS
A reliability assessment is established for power plants in two regions. In the first region, a power plant is connected to an MV feeder, consisting of 10 generating units with the rated power of 5 50 MW. The generating units have various reliability factors and fuel types [11]. The power plant in the second region has a single generating unit connected to an MV feeder. The impact of additional plants in the second region is evaluated in terms of grid reliability. Power plants can be out of operation in case of malfunction, maintenance, and so on. These unplanned shutdowns are very critical in capacity planning and selection of machine number /size. Table I shows reliability and availability of power plants in two regions based on the 5-year average data of plants with different machine types using Equations (1) and (2).
TABLE I THE VALUES CALCULATED ACCORDING TO THE POWER PLANT DATA Power Engine Reliability Availability Plants Numbers (R) (A) A 1 2 B 1 2 96.93 96.09 98.17 97.40 95.68 94.85 95.32 94.77
In this study, the capacity expansion on the private sector power plants connected to the MV level grid is evaluated from the point of the power plant and the grid. Reliability and reserve margin calculations have been performed based on the real data taken from the power plants in different regions in various times. Selection of the number/size of power plant in a system is important. The adequacy of a power system capacity can be planned by comparing the reliably available capacities. Higher reliability figures can be achieved by installing power plants. Reserves can be evaluated as the unavailable capacity from reliable power plants. This study shows that the grid/network reliability is increased by the connection of power plants to an MV system. Power plants are considered as a reserve since the failure rates decrease and power quality increases in a highly reliable grid/network. This is reflected positively to the energy provider and also the consumers fed by this busbar. REFERENCES [1] [2] Electricity Market Grid Regulation, Official Gazette no. 27418 dated 26.11.2009. Number Electrical Installations Project Regulation, Official Gazzette no. 27434 dated 16.12.2009 Electricity Market Ancillary Service Regulation, Official Gazette no. 27580 dated 13.05.2010. Number Electrical Transmission System Quality Regulation, Official Gazzette no. 25639 dated 10.11.2004. L. Kl, A.B. Arsoy, Cogeneration Contribution for Energy Quality, EVK 07, Kocaeli Standard Definitions for Use in Reporting Electric Generating Unit Reliability, Availability, and Productivity, IEEE Std 762-
By assuming that each generating unit in Power Plant A has a reliability of %96, the probability being at least one unit on is computed as 0.9984 using Equation (5). ( P(1) =1-(1-0.96)(1-0.96)=0.9984) ), while it is 0.9999 with the addition of another unit. That is the criterion taken into consideration when selecting the number and size of units in a power plant and protection expectation. By the study of two different distribution regions, the reliability indices calculated in power plant busbar due to grid failures are given in Table II:
TABLE II RELIABILITY INDICES BASED ON THE POWER PLANT DATA SAIFI [piece/year] Region 1 Region 2 6.0 48.8 SAIDI [min] 489.8 2029.6 CAIDI [min/piece] 81.63 41.59
[3]
[4]
[5] From the power system point of view, approximately 98 MW reserve with 96% reliability is obtained while this reserve is 94 MW with 99.84% reliability. The connection of a power plant into a grid increase the reliability of a power system even the reliability level of both the plant and
[6]
[7]
[8] [9] [10] [11]
2006.. D.J. Smith, Reliability, Maintainability and Risk, 6th Edition 2001, Elsevier ButterworthHeinemann F. Akdeniz, Probability abd Statistics, Adana 2000, Baki Publishment IEEE Guide for Electric Power Distribution Reliability Indices, IEEE1366 Std.-2003. A. Vuorinen, Planning of Optimal Power Systems, 2009 Edition, Finland Trkiye Elektrik letim A.. (TEIAS), Buyukkaristiran Substation Operation Data
Analysis of a Modular Multilevel Inverter Under the Predicted Current Control Based on Finite-Control-Set Strategy
Razieh Nejati Fard, Hamed Nademi and Lars Norum
Department of Electrical Power Engineering Norwegian University of Science and Technology (NTNU) Trondheim-7491, Norway raziehn@stud.ntnu.no, hamed.nademi@ntnu.no, norum@ntnu.no
AbstractThis paper introduces a method for the precise control of (un)balanced load currents in Modular Multilevel Converter (MMC). The algorithm is based on the FiniteControl-Set Model Predictive Control (FCS-MPC) which is a promising predictive control approach for power converters to minimize the circulating currents inside the converter legs, achieve the voltage balance in the capacitors and load current control. The grid-connected MMC topology, its analytical modeling and the controller design procedure are described and design example is studied. Developed predictive control is also employed to reduce sensitivity of the converter output voltage to disturbances in load side and measurement noise. In addition, system stability has been analytically verified in terms of model uncertainties. Simulation results are included to validate the proposed controller especially for good load current reference tracking and disturbance rejection properties despite the variations in the system parameters. KeywordsModel Predictive Control Multilevel Converter (MMC); Current Balancing (MPC); Modular Control; Voltage
I.
INTRODUCTION
Modular Multilevel Converter (MMC) is a new multilevel converter topology that has gained considerable attention of academia and industry in recent times [1]-[3]. There are many advantages of the converter [4]. Modularity, low device rating, and fault tolerant capacity are some of the key features of this converter [5], [6]. The converter has already been introduced in the market for high voltage dc transmission (HVDC) [7]. A considerable research is going on presently for using this converter for dc-ac power conversion e.g. for ac drives [8]-[11]. Basically, three control objectives regarding MMC for the applications such as VSC-HVDC system, STATCOM and medium-voltage motor drive are: output voltage, I/O current and floating capacitors voltage [12]. In recent years, several solutions have been reported in the literature for addressing the mentioned issues [5], [9], [13][15]. The predictive control provides the possibility to
accurate control of inverter output current and voltage with high dynamics in such a way that improves the performance of the system, its ability to consider several terms in cost function, its systematic and implementable without frequency transformations regardless of nonlinear nature of semiconductor power converters [16]-[17]. Among different MPC algorithms, Finite-Control-Set Model Predictive Control (FCS-MPC) is well suited to use in multilevel power converters and motor drive applications because of its several advantages [17-19]. The focus of this work is to minimize the capacitor voltage ripples by controlling the circulating current based on FCS-MPC concept for an MMC. Additionally, the analysis is performed to achieve not only the precise (un)balanced load current tracking but also to obtain well behavior in the case of capacitor voltage balancing. In order to derive the one-step-ahead prediction of the controlled variables, Midpoint Euler method is used as a discretization technique. Furthermore, the sensitivity and robustness of the developed controller to parameter mismatch, measurement noise as well as sudden changes in utility grid voltage are addressed. The complete proposed scheme has been implemented on five-level MMC using MATLAB/Simulink software. Finally, some simulation results are shown. II. OPERATING PRINCIPALS OF THE MMC
For drives application, the main voltage source consists of the dc bus formed by the rectifier system. The schematic diagram of a three-phase (n+1)-level MMC has been depicted in Fig. 1. Each MMC phase consists of two identical arms called upper arm and lower arm. Each arm has been made of n series-connected Sub-modules (SM) that acts as a controllable voltage source, and a current limiting inductor (l). Converter is connected to the load that could be a motor or utility grid which is modeled with a series connected inductor (L), resistor (R) and sinusoidal voltage source (e). Each SM contains a capacitor and two complementary switch combinations that is presented in Fig. 1. All the
capacitors will be charged up to Vdc/n at the beginning of MMC operation [20] and their voltage level should be kept around this value by the controller. When switch S2 is turned on, the capacitor is bypassed and the output voltage of the cell is zero. When S1 is turned on, the capacitor voltage is obtained at the output. With many cells connected in series, the converter output voltage is very smooth and there is usually no need to place filters to improve the output voltage quality.
currents, while it has negative effects on the converter switching losses, capacitor voltage fluctuation and components rating values; it should be minimized by a proper control strategy. According to Fig.1, by employing Kirchhoffs voltage law, the mathematical equations can be described by di di Vdc = v upj + l upj + ri upj + Ri j + L j + e j dt dt 2 (2) di di Vdc = v lowj + l lowj + ri lowj Ri j L j e j dt dt 2 Where vupj and vlowj are the total voltage level of the upper and lower arm SMs and r states the arm inductor and IGBTs resistivity. Phase currents ij can be calculated by subtracting arm currents; also the circulating current can be found as follows
i j = iupj i lowj i zj = 2I 1 iupj + i lowj dc 2 3

(3)
In view of (2), and by some mathematical efforts, two main first order differential equations based on the controlled variables i.e., i and iz can be found as follows
di j dt
Fig. 1: Topology for the three-phase grid-connected system.
( l + 2L )
v lowj v upj (r + 2R )i j 2e j
di zj
Depending on the switching state selected by the controller, the capacitor is inserted or bypassed, in which the SM is called to be ON or OFF respectively. According to the operating rules of MMC, n out of 2n SMs in a phase should be ON at each instant. Therefore, The output pole voltage of MMC, Vao with respect to point O, alternates between -Vdc/2 and +Vdc/2 when all the upper arm SMs are ON and OFF, respectively. The upper arm and lower arm currents can be determined by i I iupj = dc + j + i zj 3 2 (1) Idc i j i lowj = + i zj 3 2 Where j represents the phase (a, b or c) and Idc, ij and iz represent the DC-link current, output phase current and circulating current. The circulating current which is necessary for MMC operation, is pure AC and its frequency is mainly as twice as the output frequency. It has no effect on the output load
1 2 = [Vdc v upj v lowj 2ri zj r Idc ] 2l 3 dt
(4)
It is worth mentioning that the DC-link current is assumed to be constant (
dIdc = 0) . dt
The third controlled variable is the capacitor voltages that can be calculated by
i mj = C
dvCij dt
(5)
Where i=1,2,..,2n is the SM number and im can be zero if SM is off. III. PROPOSED STRUCTURE OF FCS-MPC
In this section, the design procedure of proposed controller based on FCS-MPC concept will be described. FCS-MPC is a powerful control strategy that takes the advantage of discrete nature of power converters in order to reduce the calculation amount of MPC. In this method, a cost function based on the control requirements is defined at
each time instant, then the cost function for all of possible switching states is calculated. Finally, the switching state that minimizes the cost function in the next sampling time will be selected and applied to the converter. In an (n+1)n level MMC, there are N = C2 n possible switching states. As mentioned earlier, there are three desired controlled variables; output AC currents, capacitor voltages and circulating currents. In order to predict the one-step ahead value of these controlled variables, Eq. (4) should be discretized by Midpoint Euler method as follows
Where v upj and v lowj can be obtained by adding voltage of the connected capacitors of upper and lower arm. Again in this case, the DC-link current can be assumed to be constant in one sampling period to simplify the calculations ( Idc ( k + 1) Idc ( k ) ).
2Vdc (v upj ( k + 1) + v upj ( k ) ) i zj ( k + 1) = C.i zj ( k ) + D. 4r + + 1 v k v k I k ( lowj ( ) lowj ( ) ) 3 ( dc ( ) ) (9)
i j ( k + 1) = A.i j ( k ) + v lowj ( k + 1) + v lowj ( k ) B. v upj ( k + 1) + v upj ( k ) 2 e j ( k + 1) + e j ( k )

(6) Where
Backward Euler method gives a satisfying approximation to find v Cij ( k + 1) ,
vCij ( k + 1) = vCij ( k ) +
Ts i mj ( k + 1) C
(10)
A= B=
2 ( l + 2L ) + Ts ( r + 2R ) Ts 2 ( l + 2L ) + Ts ( r + 2R )
2 ( l + 2L ) Ts ( r + 2R )
And Ts is the sampling time. For simplification purposes, ej is assumed to be constant during a sampling period ( e j ( k + 1) e j (k ) ). Therefore, the phase currents can be expressed as
v lowj ( k + 1) + v lowj ( k ) (7) i j ( k + 1) = A.i j ( k ) + B. v upj ( k + 1) + v upj ( k ) 4e j ( k )
In the next step of designing FCS-MPC, a cost function will be defined in order to force the output AC currents to follow their references, keep the capacitor voltages balanced V (i.e. around its nominal value dc ) and minimize the n circulating current.
g = i jref ( k + 1) i j (k + 1) + 1. v Cij ( k + 1)
i =1 2n
Vdc + 2 . i zj (k + 1) n (11)
Where i jref ( k + 1) is the predicted reference current that can be assumed to be equal to i jref ( k ) with a good approximation. The reference current can be determined by user depending on the application requirements. In addition, 1 and 2 are the weighting factors that are necessary to adjust the importance and units of different cost function components. They can be obtained by empirical methods, i.e. try and error. The schematic of the proposed MMC with FCS-MPC controller is shown in Fig. 2.
Another controlled variable that must be predicted is the circulating currents that can be found by
2Vdc (v upj ( k + 1) + v upj ( k ) ) i zj ( k + 1) = C.i zj ( k ) + D. (v lowj ( k + 1) + v lowj ( k ) ) (8) 2r I k + 1 + I k ( ) ( ) ( ) dc dc 3 Where
C= D=
2l rTs 2l + rTs Ts 4l + 2rTs

Fig. 2: Block diagram of the proposed predictive controller for MMC.
TABLE I. PARAMETERS FOR MMC Parameter Number of SMs per arm SM capacitor Arm inductance Arm resistance DC link voltage Load resistance Load inductance Rated line-line voltage Rated load current Rated frequency Sampling time Symbol n C l r Vdc R L el-l In f Ts Value 4 6.6 mF 1.2 mH
The upper/lower arm capacitor voltages are presented in Fig. 5 which are fairly balanced. The capacitors are charged to a voltage of 10kV/4=2.5kV. It is clearly visible in Fig. 5
250
ia
200 150
iupa ilowa iza
44 m
Current (A)
100 50 0 -50 -100 -150
10 kV 5.9 9 mH 6.6 kV 200 A 50 Hz 100s
-200 -250 0.4
0.41
0.42
0.43
0.44
0.45 Time (s)
0.46
0.47
0.48
0.49
0.5
Fig. 4: Upper/lower arm currents and circulating current in phase A.
IV.
SIMULATION RESULTS
2540 lower 4 capacitors 2530 upper 4 capacitors
Two sets of simulation results are presented in this section for three-phase 5-level MMC that operates based on the described control scheme. Some of the circuit parameters are listed in Table I. The first set of results deals with (un)balanced load current reference tracking behavior of the FCS-MPC strategy under steady-state performance. The load currents command is equal to 200 A at 50 Hz. Fig. 3 shows the three-phase load currents where their references are tracked with negligible error.
250
2520 Capacitor Voltages (V)
2510
2500
2490
2480
2470
2460 0.2
0.21
0.22
0.23
0.24
200 150 100

6000
0.25 0.26 Time (S)
0.27
0.28
0.29
0.3
Fig. 5: DC capacitor voltages in upper/lower modules of phase A.
Current (A)
50 0 -50 -100 -150 -200 -250 0.45 0.455

MMC Pole Voltages (V) 2000 4000
-2000
0.46 0.465
0.47 0.475 0.48 0.485 Tim e (s)
0.49 0.495
0.5
-4000
Fig. 3: Reference current tracking of three phases.
The arm currents and circulating current for phase A are illustrated in Fig. 4. The magnitude of iza is reduced.
-6000 0.4
0.41
0.42
0.43
0.44
0.45 Time (s)
0.46
0.47
0.48
0.49
0.5
Fig. 6: Three phase output voltage waveforms of MMC.
that voltage fluctuation is considerably decreased by around 2.5% of reference value. Fig. 6 shows the three phase pole voltage waveforms. It can be seen the presence of 5 levels of voltage, as expected. Moreover, unbalanced load current references (210A, 190A and 170A) have been applied to three phases respectively. This is a common case in three-phase four wire systems where each phase has a different load requirement [21]. From Figs. 7 and 8, we can see that the controller is able to handle each phase current separately and load currents follow the references with less steady-state error. The capacitor voltages have also been kept around 2.5kV. However, the peak to peak value of the capacitors voltages are different that is reasonable due to the difference in current amplitude.
250 Irefa
because of the parameters mismatch, the considered values for R, L are reduced by 20% of nominal values in Table I. For this purpose, when the grid voltage disturbance is also stepped down to 70% of its rated value at t=1sec. The results are given in Figs. 9 to 11. In Fig. 9 comparison is made between reference current and the measured of such signal, highlighting a good tracking performance of the adopted prediction algorithm. As can be seen from Fig. 10 the capacitor voltages remain unaffected and the controller selects the new switching algorithm quickly to balance capacitor voltages.
250 200 150 100 Load Currents (A) 50 0 -50 -100
200 150 100 Load Currents (A) 50 0
Ia(210A) Irefb Ib(190A) Irefc Ic(170A)
-150
-50
-200
-100
-250 0.95
-150 -200 -250 0.4
1 Time (s)
1.05
1.1
Fig. 9: Reference current tracking at grid side when the load voltage source decreases 30% at t=1 sec.
0.41 0.42 0.43 0.44 0.45 Time (s) 0.46 0.47 0.48 0.49 0.5
2540 2530 2520 Capacitors Voltages (V)
Fig. 7: Simulation results of MMC three-phase load currents under unbalanced current references.
2540
2510 2500 2490 2480 2470 2460 2450 0.95
2530
2520 Capacitors Voltages (V)
2510
2500
2490
2480
1.05 Time (s)
1.1
1.15
1.2
2470
Fig. 10: DC capacitor voltages in all modules when the load voltage supply decreases 30% at t=1 sec.
0.41 0.42 0.43 0.44 0.45 Time (s) 0.46 0.47 0.48 0.49 0.5
2460 0.4
Fig. 8: Simulation results of MMC capacitor voltages under unbalanced load current references.
The effects of mentioned step change as well as parameter uncertainty on the MMC output voltages are displayed in Fig. 11. It can be observed that the developed controller can handle system deviations and does not deteriorate the waveforms.
The second part of our simulation study is concerned with the robustness of FCS-MPC control strategy in the presence of measurement errors in load parameters. Therefore,
6000
REFERENCES
[1] [2] [3] [4] [5] [6] A. Lesnicar and R. Marquardt, An innovative modular multilevel converter topology suitable for a wide power range, in Proc. of IEEE Power Tech Conf. 2003, pp. 1-6. M. Glinka and R. Marquardt, A new AC/AC multilevel converter family, IEEE Trans. Ind. Electron., Vol. 52, No. 3, June 2005, pp. 662-669. S. Kouro et al., Recent Advances and Industrial Applications of Multilevel Converters, IEEE Trans. Ind. Electron., Vol. 57, No. 8, Aug 2010, pp. 2553 - 2580. R. Marquardt, Modular Multilevel Converter: An universal concept for HVDC-Networks and extended DC-Bus-applications, in Proc. of IPEC 2010, pp. 502-507. M. Saeedifard and R. Iravani, Dynamic Performance of a Modular Multilevel Back-to-Back HVDC System, IEEE Trans. Power Electron, Vol. 25, No. 4, Oct 2010, pp. 2903-2912. L. Angquist, A. Antonopoulos, D. Siemaszko, K. Ilves, M. Vasiladiotis, and H.-P. Nee, Open-Loop Control of Modular Multilevel Converters Using Estimation of Stored Energy, IEEE Trans. Ind. Appln., Vol. 47, No. 6, Nov/Dec 2011, pp. 2516 - 2524. HVDC Plus- from Siemens AG, available online at http://www.energy.siemens.com/hq/en/powertransmission/hvdc/hvdc-plus/ A. J. Korn, M. Winkelnkemper and P. Steimer, Low output frequency operation of the Modular Multi-Level Converter, in Proc. of ECCE 2010, pp. 3993 - 3997. M. Hagiwara and H. Akagi, Control and experiment of pulse-width modulated modular multilevel converters, IEEE Trans. Power Electron, Vol. 24, No. 7, Jul 2009, pp. 1737-1746. S. Rohner, S. Bernet, M. Hiller and R. Sommer, Modulation, Losses, and Semiconductor Requirements of Modular Multilevel Converters, IEEE Trans. Ind. Electron., Vol. 57, No. 8, Aug 2010, pp. 2633 2642. J. Kolb, F. Kammerer, M. Braun, "Dimensioning and Design of a Modular Multilevel Converter for Drive Applications", in Proc. of IEEE EPE-PEMC 2012, pp. 1-8. H. Nademi, A. Das and L. Norum, An analysis of improved current control strategy for DC-AC modular multilevel converters, Int. Trans. on Electr. Energ. Syst, Wiley, 2013, in press. H. Akagi, Classification, Terminology, and Application of the Modular Multilevel Cascade Converter (MMCC), IEEE Trans. Power Electron, Vol. 26, No. 11, Nov. 2011, pp. 3119-3130. Q. Song, W. Liu et al., A Steady-State Analysis Method for a Modular Multilevel Converter, IEEE Trans. Power Electron, Vol. 28, No. 8, Aug. 2013, pp. 3702-3713. K. Wang, Y. Li, Z. Zheng, L. Xu Voltage Balancing and Fluctuation-Suppression Methods of Floating Capacitors in a New Modular Multilevel Converter, IEEE Trans. Ind. Electron., Vol. 60, No. 5, May 2013, pp. 1943-1954. P. Cortes, M. P. Kazmierkowski, R. Kennel, et al, Predictive Control in Power Electronics and Drives, IEEE Trans. Ind. Electron., Vol. 55, No. 2, Dec. 2008, pp. 4312-4324. M. A. Perez, J. Rodriguez, E. J. Fuentes and F. Kammerer, Predictive Control of AC-AC Modular Multilevel Converters, IEEE Trans. Ind. Electron., Vol. 59, No. 7, July 2012, pp. 2832-2839. S. A. Davari, D. A. Khaburi and R. Kennel, An Improved FCS MPC Algorithm for an Induction Motor With an Imposed Optimized Weighting Factor, IEEE Trans. Power Electron, Vol. 27, No. 3, Mar. 2012, pp. 1540-1551. R. P. Aguilera, P. Lezana and D. E. Quevedo, Finite-Control-Set Model Predictive Control with Improved Steady-State Performance, IEEE Trans. Ind. Informat., Vol. 9, No. 2, May 2013, pp. 658-667. A. Das, H. Nademi and L. Norum, A Method for Charging and Discharging Capacitors in Modular Multilevel Converter, in Proc. of IEEE IECON 2011, pp. 1058-1062. V. Yaramasu, M. Rivera, B. Wu, J. Rodriguez, Model Predictive Current Control of Two-Level Four-Leg Inverters-PartI: Concepts, Algorithm, and Simulation Analysis, IEEE Trans. Power Electron, Vol. 28, No. 7, Jul 2013, pp. 3459-3468.
4000
MMC Pole Voltages (V)
2000
-2000
-4000
-6000 0.95
1 Time (s)
1.05
1.1
Fig. 11: Dynamic responses of MMC output voltages when the load voltage source decreases 30% at t=1 sec.
[7] [8] [9] [10]
Finally, the results indicate that proposed FCS-MPC improves the steady-state and dynamic response of conventional FCS-MPC strategy which is reported in literatures mainly based on constant references for a MMC. Moreover, the accuracy of current reference tracking depends on the sampling frequency directly. Higher sampling frequency reasonably leads to less output current error with its reference and less ripple as a result. The switching frequency of MMC is mainly one fourth to one third of sampling frequency; therefore, there is a tradeoff between having high accuracy and low switching frequency which is directly related to the switching loss.
[11] [12] [13] [14] [15]
V.
CONCLUSION
In this paper, a Finite-Control-Set Model Predictive Control (FCS-MPC) strategy for current control of an MMC has been discussed. Unlike most previous works, the proposed approach here is developed for either balanced or unbalanced reference current tracking. This structure of the controller is actually well adopted since it considers minimization of the circulating currents inside the converter, balancing the capacitor voltages and load current control in the cost function. Simulation results confirm the excellent steady- and transient-state response, which is one of the remarkable characteristics of predictive controllers even when operating with parameter mismatch and load voltage disturbance.
[16] [17] [18]
[19] [20] [21]
New Reliability Index for Power System Protection based on Multi-Agent Technique
Nadheer A. Shalash
Faculty of Electrical and Electronics Engineering University Malaysia Pahang, 26600 Pekan, Malaysia eng_n_a_msc@yahoo.com
Abu Zaharin Ahmad

Faculty of Electrical and Electronics Engineering University Malaysia Pahang, 26600 Pekan, Malaysia zaharin@ump.edu.my
Abstract A new index reliability for power system protection by optimizing the relay time operating using multi agent system (MAS) is presented. In the proposed method, the MAS consists of two agents as follows; first agent is a fault current agent, i.e., to determine the fault current at all points before and after grounding. While, the second agent is the time operating, i.e., to determine the time operating relay before and after modifying the fault current. The simulation implementation is carried out in Iraqi national grid system (132 kV) Najaf city. The results show that the failure rate is decreased to 50% approximation for over current and earth fault relays. Keywords- Multi-agent system; reliability indices; power system protection
I.
INTRODUCTION
Reliability assessment has been studied for many years in the field of transmission and distribution of power systems. Most of the related companies had invested a lot of equipment that required for periodically maintenance. In most of maintenance cases, the procedures are based on the historical experiences of maintenances workers, where they are rarely cases supported by the analytical evidences [1]. However, recently, new concepts to analyze the maintenance procedures have been introduced, such in [2], new models and concepts of Markov model based on the concepts of self-down state and induced-down state are developed that incorporating the effect of protection system failures into power system reliability evaluation. Meanwhile, in [3], a probability based models has been developed to accurately reflect the stochastic nature of a power system behavior and assess its reliability performance. This had improved the common deterministic criteria approach that cannot handle the stochastic nature of power system. Then, in [4], the reliability considerations using synchrophasor monitoring of system conditions that incorporated into protection schemes is discussed, in which the problem stemming of hidden failure modes can be reduced and also can enhance the effectiveness and reliability of protection systems. Nowadays, agent systems have been used in many areas such as computer science, artificial intelligence and variety of applications [5]. A multi-agent system is a system composed of multiple interacting intelligent agents. An agent can be viewed
as perceiving its environment through sensors and acting upon its environment through actuators. Therefore, a multi-agent system can be used for all type systems composed of multiple autonomous components that demonstrating the following features; 1) each agent has incomplete in terms of global control and capabilities for solving problems, 2) the data are decentralized and its computation is asynchronous, 3) they should be designed according to the problems under consideration [6]. In grid power system, the approach of spread local control must be more robust in order to face the extreme contingencies. Practically, power systems are using many agents such as protective relays and automatic voltage regulators [7]. There are two factors of the relay protection system reliability, i.e., dependability and security. Dependability can be defined as the ability of the relay to trip for a fault in its protective area while security is the ability to refrain from tripping relay when there is no fault in the protected area [8]. Therefore, in this paper, new approach of application the multi agents system (MAS) for reliability assessment by modify operating time relay is presented. The reliability improvements show the quick promising results. II.
RELIABILITY INDICES
The most common methods used for reliability evaluation, are based on the loss of load or energy (LOLE) approach. In this method, the suitability index that describes the generation reliability level is LOLE. It indicates the time in which the load is more than available generation. The reliability expresses the proportion time of the component if it is in service or available for. Therefore, the availability (A) of a component can be expressed as [9], (uptime) (1) A= (uptime) + (downtime)
Where, the up time is the total time when the component is in service and down time is the total time when component not in service. Availability (A) may be expressed in terms of mean time to failure (MTTF) and mean time to repair (MTTR) as given by,
A=
MTTF MTTF + MTTR
(2)
Thus, the mean time between failures (MTBF) is the sum of MTTF and MTTR and can be represented by cycle of time T as seen in Fig. 1. Power system components can be characterized by discrete system states with constant transition rates between these states. State 0 represents the healthy component in an operating condition.
(OC) fault curves [11]. The characteristics of each relay have been used by manufacturers as a set of curves conformable to different values of time dial settings. For the time inverse relay characteristics, the relay starts to operate after an intended time delay [12]. A typical inverse time OC relay has two values to be set, i.e., the pick-up current (Ip) and time multiplier setting (TMS). The Ip current value is the minimum current value for which the relay operates. And, for TMS, it should be in the range of 0.1 to 0.9. The tripping currents of the relays are varied according to the tripping time required and the characteristics of other protection devices used in the network. According to the type of relay as given in Table I, the operating time of protection (OTP) can be adjusted using the curve characteristics [13] as shown in Figures 3 and 4 for transformer (90 MVA). These characteristics are calculated by the following Eqn. (7).
TABLE I. TYPE OF RELAY BASED ON TRANSFORMER CAPACITY CT
400/5 300/5 400/5 300/5
Figure 1. Mean time diagram for a two state component
Meanwhile, for state 1 (failed state) when the component cannot perform its intended function. Transition occurs between state 0 and 1 and the transition rates between these states are the failure rate " and the repair rate as shown in Fig. 2.
Transformers relay specifications Transformer 90 MVA ,(OC ) Transformer 90 MVA ,(OC ) Transformer 90 MVA ,(EF ) Transformer 90 MVA ,(EF)
Ip
5 5 1.5 1.5
TMS
0.3 0.3 0.3 0.3
Figure 2. Diagram of state transition rates
Therefore, Eqn. (1) may be re-expressed in terms of failure rate and repair rate such that,
A=
+
1 +
(3)
Figure 3. Characteristics of OC relay for power transformer (90 MVA)
The fault frequency is the reciprocal of the average cycle time [10].
f =
(4)
The other two reliability indices can be computed from the previously defined ones, as follows:
T=
D=
1 f
P f
(5) (6)
Where, T is a time cycle, D is a mean duration and P is an individual probability III. EARTH FAULT AND OVERCURRENT RELAY In protection power system, the time overcurrent devices are used as the primary scheme, the fault signal in this relay is tripped in a time specified by a delay-time versus overcurrent
Figure 4. MVA) Characteristics of earth fault (EF) relay for power transformer (90
If OTP = I CT P
TMS
(7)
That means the value of K should be in the range of 1 to 3. According to Eqn. (8), hence, Eqn. (12) can be rewritten in terms of application value of K such that,
Where, I f = fault current and CT = current transformer ratio IV.

PROPOSED INDEX OTP
0 .6 <
If I 3 ph
<1
(14)
Usually, accords to MTBF. The protective relay manufacturers will describe the reliability of their apparatus with the relays may vary depending on the relay installation and operation over a period of time. The reliability indices determined by using real-time data analysis that provides more accurate or realistic information as it uses the past failure data of the relays [14]. The basis for all points in the system effective grounding or in a specific part of the system is the ratio of zero-sequence reactance (X0) to positive-sequence reactance (X1) is not more than three; meanwhile the ratio of zero-sequence resistance (R0) to X1 is not more than one for any operational condition. The system may be considered as effectively grounded [15] when the following conditions are met;
From Eqn. (7) & Eqn. (14), it can be noted that the proposed new index reliability of OTP can be calculated by using the following equation.
0.6 I 3 ph < I f < I 3 ph OTP = TMS I CT P
(15)
Eqn. (16) is then used to calculate the expected failure rate according to Eqn. (15);
(t ) =
OTP
m =1
(16)
1<
X0 <3 X1
&
0<
R0 <1 X1
(8)
Where, N is the total number of failures. V.

MULTI-AGENT SYSTEM
The expression of the Z0/Z1 ratio can be expressed as:

Z0 X0 R0 . = j Z1 X1 X1
(9)
By expressing the residual current during the term of the three phase current and Z0/Z1 ratio, the single phase to ground fault current can be expressed as;
If =
3V 3V = 2 Z 1 + Z 0 (2 + K )Z 1
(10)
Where K=Z0/Z1
I 3 ph =
Thus
V Z1
(11)
If I 3 ph
3 (2 + K )
(12)
In all simulated cases, the value of the ratio (R0/X1) is not changed. Therefore, the resistance was neglected during the calculations [16].
K=
Z0 X 0 = Z1 X1
(13)
Technically, the agents or providers act autonomously on the platform of collecting information. A number of agents are identified as multi agents, with individual subtasks are synchronized so as to attain a particular common goal. There are generally two configurations of agents, i.e., first the agents with different subtasks and second is agents of the same kind and on the same hierarchical level or order. A multi agent system is a computational program in which several agents unite to accomplish some tasks or projects. In the power system, there could be a different number of multi agent systems structured [17]. A thorough investigation on the various modeling, both protection and reliability indices are necessary for developing an effective detection technique. In order to achieve the reliability assessment, it is essentially necessary to develop each agent the effective communication within the platform and globally build a working relationship as they negotiate towards the laid criterion [18]. Therefore the design starts with the setting of protection system. When a fault occurs in power system such as short circuits, this fault will be cleared by over OC, earth fault (EF) or differential protection (DP). OC relays are used as short circuit protection and some aspects must be considered in the setting of such protection. For example, the current settings shall be high enough to not risk during maloperation. On the other hand, it is also must be setting to low enough to give a secure operation at minimum fault current (tripping requirement). The proposed techniques for protection and grounding based on MAS can be validated on a real practical power system to measure the response of multi-agent, the accuracy of the
developed reliability index and the sensitivity of the protection system. The procedures of the proposed framework are summarized by the flow chart as given in Fig. 5.
Figure 6. Single line diagram of Najaf city.
VII. PROPOSED METHOD The proposed method is employed to the system by first, the positive, negative and zero sequence of admittance matrices are composed from the network including the grounding of transformers and generators. After that, the fault currents are calculated, then, the operating times of OC and EF relays during fault are computed according to the type of ground in and operation characteristic of relays as presented in section III. Then, determine all the possibilities of grounding in and the other substations. Afterwards, the ratios X0/X1 is calculated. The simulation is repeated to calculate the fault current and OTP according to Eqn. (16). The reliability indices can be evaluated by affecting all bus bar faults. The OTP of EF and OC relays are depending on the fault current magnitude. So, the multi agents can be used for recalculated the reliability indices before and after modified fault current between ( 0.6 I 3 ph < I f < I 3 ph ) in the buses. VIII. RESULTS AND DISCUSSION
TABLE III. CASES OF OPERATING OF EF RELAY BEFORE MODIFIED
Figure 5.
Flow chart of agent-based new index with reliability indices.
VI.
DESCRIPTION OF TEST SYSTEM
In order to evaluate the OTP index with indices reliability, the 132 kV transmission system of Najaf City is used which is one sector of Iraqi national grid (132 kV) as shown in Fig. 6. This system consists of 3 generators, 8 lines, 10 transformers and 3 feeders. The data on demand, reliability and generation was registered in the Iraqi national dispatch center during January 2007. Table II depicted the reliability data of Najaf city.
TABLE II. Component Generator Transformer Circuit breaker Relay OC Bus 1 REALY OC Bus 2 REALY OC Bus 3 REALY OC Bus 4 REALY EF Bus 1 REALY EF Bus 2 REALY EF Bus 3 REALY EF Bus 4 LINE Feeder Busbar RELIABILITY DATA OF NAJAF CITY
No. Comp 3 10 16 2 2 2 2 2 2 2 2 8 3 4
(t )
3 2 4 0.053 0.093 0.3
321 595 662 532 532 542 533 631 647 664 647 0.0830 0.0830 22
Location Fault
If Amp.
OTP(EF) sec.
(t )
2.469 3.42 3.813 3.26
Bus 1 Bus 2 Bus 3 Bus 4

TABLE IV.
1503.25 3887.09 5667.87 3305.51
0.8099 0.5830 0.5240 0.6125
CASES OF OPERATING OF OC RELAY BEFORE MODIFIED.
Location Fault
If Amp.
OTP(OC) sec.
(t )
1.57 1.704 2.57 2.056
1503.25 3887.09 5667.87 3305.51
1.2821 0.7989 0.7713 0.9735
Figure 7 shows an evolution of OTP after modifying the fault current for all bus bars using JADE (Java Agent Development Framework) package. The result of failure rates are sent to calculate indices reliability by agent communication language (ACL), that is a message class, which is responsible for sending and receiving the information between the agents. Developing a JADE agents are, by developing a java class that runs (import jade.core.agent) and applying the setup() method. This approach is where the actual agents are inserted .The agent OTP can communicate with other agents from buses for making a decision (refuse / accept) to take a minimum fault current. According to Eqn. (16), the value of reliability indices will decrease and therefore the reliability of the power system will also improve. The indices reliability results are shown in the tables VII and VIII for before and after modifying fault current. The fault frequency is constant in all cases and the cycle time, mean duration is decreed for all relays. For example, OC relay in bus 1, that has been shown in table VII, the f = 0 .0019, D=196. 93 and T=535.14 for failure rate = 1.57. Then, when the limited fault between ( 0.6 I 3 ph < I f < I 3 ph ) the failure rate has been decreased to 0.781 in table VIII, which is the f= 0.0019, D=195.3 and T= 533.56 are computed. Therefore, the reduction in failure rates of the OC and EF relays are leading to the improvement of reliability and protection quality.
TABLE VII.
SIMULATION RESULT OF RELIABILITY INDICES BEFORE MODIFIED FAULT CURRENT
Figure 7. Visualization of multi agent for OTP.
TABLE V.
CASES OF OPERATING OF EF RELAY AFTER MODIFIED.
Location Fault
3 ph f 3 ph
0.6 I
<I <I
OTP(EF) sec.
(t )
1.243 1.7 1.9 1.64
1503.25 - 1498.1 3887.09 - 3880.3 5667.87 - 5658.4 3305.51 - 3312.2
0.8099 - 0.811 0.5830 - 0.59 0.5240 - 0.525 0.6125 - 0.612
Component TABLE VI.

CASES OF OPREATING OF OC RELAY AFTER MODIFIED.
(t)
3 2 4 1.57 1.704 2.57 2.056 2.469 3.42 3.813 3.26 0.053 0.093 0.3
321 595 662 532 532 542 533 631 647 664 647 0.0830 0.0830 22
.0031 0.0016 .0014 0.0019 .0019 .0018 0.0019 0.0016 0.0015 0.0015 0.0015 1.9724 2.7624 0.431
D
120.33 151.9 218.52 196.93 197.03 201.34 197.65 117.01 120.3 123.57 120.24 0.1227 0.13 8.004
T
327 615 726 535.14 535.4 547.14 537.11 635.93 653.84 671.62 653.52 0.507 0.362 23.2
Location Fault
3 ph f 3 ph
0.6 I
<I <I
OTP(OC) sec.
(t )
0.781 0.852 1.293 1.027
1503.25 - 1498.1 3887.09 - 3880.3 5667.87 - 5658.4 3305.51 - 3312.2
1.2821 - 1.2849 0.7989 - 0.7995 0.7713 - 0.7718 0.9735 - 0.9725
In this section, the results of the proposed method are delivered. From Tables (III to VI), shown the decreasing of employ the new index reliability failure rates approximately to 50% when compared to before modified fault current. For example in bus 1, the fault current is 1503.25A at before modifying the fault current. While after modified, the fault current is limited between the range of 1498.1A and 1503.25A. From that, it can be seen that the results of failure rates of EF and OC have been decreased to approximately 50%.
Generator Transformer Circuit Breaker Relay OC Bus 1 Relay OC Bus 2 Relay OC Bus 3 Relay OC Bus 4 Relay EF Bus 1 Relay EF Bus 2 Relay EF Bus 3 Relay EF Bus 4 Line Feeder Busbar
TABLE VIII.
SIMULATION RESULT OF RELIABILITY INDICES AFTER MODIFIED FAULT CURRENT
[9]
[10] Component Generator Transformer Circuit Breaker Relay OC Bus 1 Relay OC Bus 2 Relay OC Bus 3 Relay OC Bus 4 Relay EF Bus 1 Relay EF Bus 2 Relay EF Bus 3 Relay EF Bus 4 Line Feeder Bus bar
(t)
3 2 4 0.781 0.852 1.293 1.027 1.243 1.7 1.9 1.64 0.053 0.093 0.3
321 595 662 532 532 542 533 631 647 664 647 0.0830 0.0830 22
.0031 0.0016 .0014 0.0019 .0019 .0018 0.0019 0.0016 0.0015 0.0015 0.0015 1.9724 2.7624 0.431
D
120.33 151.9 218.52 195.3 196.4 200.4 196.89 116.56 119.67 122.875 119.65 0.1227 0.13 8.004
T
327 615 726 533.56 533.7 544.58 535.05 633.48 650.4 667.8 650.28 0.507 0.362 23.2 [11]
[12]
[13] [14]
[15]
[16]
IX.
CONCLUSION
[17]
In this paper, the use of agents for solving power system problems has been reviewed. New index reliability for power system protection is proposed in order to improve the quality and efficiency of the power system according to failure rates that has been greatly decreased. With the using of MAS, the high speed rate of protection will facilitate the incorporation of smart grid system by increasing the security of power system based on the developed techniques and decreasing the economic loss due to damaged equipment.
REFERENCES
[1] B. Kalinowski and George Anders,A new look at component maintenance practices and their effect on customer, station and system reliability, Electrical Power & Energy Systems, Elsevier,Vol. 28, pp: 679-695, Dec. 2006. Kai Jiang and Chanan Singh. New models and concepts for power system reliability evaluation including protection system failures. IEEE Transactions on Power Systems, Vol. 26, No.4, pp: 1845 1855, Nov. 2011. A. A. Chowdhury and D. O. Koval. Probabilistic assessment of transmission system reliability performance. IEEE Power Engineering Society General Meeting 2006 . Jonathan Sykes, Vahid Madani, John Burger, Mark Adamiak, William Premerlani. Reliability of protection systems. Proceeding of IEEE conference on Protective Relay Engineers 2010; 1 16 T. Nagata and H. Sasaki A Multi-Agent Approach to Power System Restoration, IEEE Transaction On Power Systems, vol. 17, n. 2, May 2002. Kai Cao, Xiao Feng and Hui Wan, Applying agent-based modeling to the evolution of eco-industrial systems, Ecological Economics, Elsevier,Vol. 68, pp. 28682876, 2009. D. A. Panasetsky, P. V. Etingov, and N. I. Voropai, Multi-agent approach to emergency control of power system, DRPT2008 Conf. Proc., Nanjing, China., 6-9 April 2008. S. Ward,T. Dahlin, W. Higinbotha Improving Reliability for Power System Protection, 58th Annual Protective Relay Conference, Atlanta, GA, April 28 - 30, 2004.
[18]
Mario Rios, Keith Bell. Daniel Kirschen and Ron Allan. Computation of the value of security. Manchester center for electrical engineering. final report; 1999. Roy Billinton,Robert J.Ringlee and Allen J.Wood Power System Reliability Calculations, the Massachusetts Institute of Technology., MIT Press (MA), 3/2003. D. Campos, E. Moreno, and D. Torres, Test and evaluation time2 inverse over2current protection algorithm using SIMULINK, Proc. of the 7th WSEAS Int. Conf. on Signal Processing, Istanbul, Turkey, 2008, pp. 6974.. IEEE Committee Report.,Computer representation of overcurrent relay characteristics, IEEE Trans. Of Power Delivery, Vol. 4, No. 3, pp: 1659-1667, July 1989. ANSI/IEEE Std 242,IEEE Buff Book - Protection & Coordination, Issue date: 1986, Current version: 2012. Mohd Iqbal Ridwan, Kerk Lee Yen, Aminuddin Musa and Bahisham Yunus, Application of life data analysis for the reliability assessment of numerical overcurrent relays, World Academy of Science, Engineering & Technology; Vol. 72, p971, Feb2011 Nelson and John P., The grounding of power systems above 600 volts: a practical view point, Petroleum and Chemical Industry Conference, 2003. 15-17 Sept., 2003. Nadheer A.Shalash, Ahmed M. A. Haidar, Kousay Abdul SattarGrounding Locations Assessment of Practical Power System Energy and Power Engineering, January 2012. C. Rehtanz, Autonomous Systems and Intelligent Agents in Power System Control and Operation, Springer-Verlag Berlin Heidelberg New York, 2003. Anant Oonsivilai and Kenedy A. Greyson Power System Contingency Analysis Using Multiagent Systems, World Academy of Science, Engineering and Technology, 2009.
[2]
[3]
[4]
[5]
[6]
[7]
[8]
Transient Model Parameters Identification of Transformer Based on PSO Algorithm

Mohammad Valii1, Mehdi Bigdeli1*, Farid Hojjatiparast2
2
Department of Electrical Engineering, Zanjan Branch, Islamic Azad University, Zanjan, Iran Department of Electrical Engineering, Science and Research Branch, Islamic Azad University, Zanjan, Iran
*
Email: m_bigdeli78@yahoo.com.com
Abstract- In this paper a novel model for analyzing the transient state of the distribution transformers is proposed. The presented model is as simple that the simulation process can be conducted so fast and easy and also its application in the form of a two port element in power network is possible. By considering the complexity of the analytical methods, Particles Swarm Optimization (PSO) algorithm is applied for estimation of the parameters of the transient model of the transformers. In order to do that, related tests were performed on a 2.5 MVA and 6300/420 V distribution transformers, after that, the desired parameters were estimated by the implemented PSO algorithm. Finally, by comparing the experimental and the estimated values, the reliability of the PSO algorithm in this case was evaluated. Also, a comparison between the obtained values in this research and the results of Genetic Algorithm (GA) and the analytical method were carried out. The result reveals the more capabilities and accuracies of the PSO algorithm. Keywords: Transformer; Transient; Measurement; Parameter Estimation; PSO Algorithm.
models), the simulation time is very long and their usage by power network analysis is almost impossible due to their numerous elements. The hybrid model is a combination of physical and black box models and is recommended only for expanding the frequency domain of simulation. This paper presents a model which can set the transformer as a bipolar element in network firstly, and this model should be simple and have enough accuracy secondly, and model's of four experiments in transformer's terminal thirdly. For checking the level of credibility and accuracy of the proposed model, a three phase transformer 2.5MVA and 6300/420V is chosen. Then, model parameters are estimated using PSO algorithm in MATLAB. The comparison between calculated and measured quantities confirms that the accuracy of the proposed method in the middle transient frequency domain is satisfactory. II. THE PROBLEM DEFINITION For verifying the behavior inside the transformer, and also the affects of transformer in power network, on transient and steady state, many various models with different aims have been proposed. One of these goals is the simplicity of the model. The other goal is the simplicity of estimating model parameters; but this simplicity shouldn't cause decrease the accuracy in calculations. Reference [8] presented a model which can set the transformer as a bipolar element in network. The parameters of model calculated through complex mathematical relations based on geometrical dimensions; therefore identification needs many data, which could be unavailable or not precise. The purpose of this paper is presenting a model which has following characteristics: The model is proper for analyzing power network. The proposed model is simple and also has acceptable accuracy in 10-200 kHz frequency range. For estimating model parameters, the results of four experiments are needed; therefore there will be much economy in time and cost. In order to estimating model parameters, we use PSO algorithm in MATLAB, and there is no need to complex mathematical relations.
I.
INTRODUCTION
Transformers are among the most important equipment of power networks. These apparatuses must be able to withstand different types of strengths. Additionally, they must be monitored during their operation in order to prevent any large destructive faults. Hence, the study of transformers in transient states is very important. A variety of different models is proposed by authors, which can be classified as follows: Black box models [1-3] Physical models [4-6] Hybrid model [7] Each one of these models owns its special capabilities and properties. Black box models are dependent on the measured data from the terminal of transformer. Furthermore, it is impossible to consider any internal fault by the model. By physical models, which are based on FEM (Finite Element Method) or consists a lot of RLC elements (so called detailed
III.
THE PROPOSED MODEL
The model owning the previously mentioned properties is shown in Fig. 1
C12
R1
C g1 C g1
C1 L1
M
L2
R2
Rm
Lm
n :1
C2
Cg 2 Cg 2
C12
Figure 1. Proposed model for transient state of transformer
The parameters of the model are defined as follows: 1Ideal transformer with turns ratio n to consider the voltages ratio, 2Rm and Lm: Core loss resistance and magnetization inductance to consider the no load current containing the core loss and magnetization current, 3R1 and R2: The primary and secondary winding resistances including conductors skin effects, 4C1 and C2: Series capacitances of the primary and secondary windings, 5C12: Parallel capacitance between two windings, 6Cg1 and Cg2: Capacitances between each winding and earth, 7L1 and L2: Self inductances of windings, and 8M: Mutual inductance between two windings. The calculation of parameters with some proper experiment and using PSO algorithm, which is an essential step in this model, is discussed in the next section. IV. TEST CIRCUITS For estimating transformer parameters and also checking the validity of this method, a three-phase transformer with power rating of 2.5 MVA is chosen as test object. This transformer consists of a disk-type high-voltage winding with 22 disks, 12 turns in each disk, and a layer-type low-voltage winding with four layers, 36 turns in each layer. In present investigation all measurements were executed in the time domain to determine different transfer functions defined by the terminal conditions of the transformer which is illustrate in Ref. [9]. In the time domain, test objects are excited by low or high impulse voltages. The input and output transients are measured and analyzed. In low voltage measurements the amplitudes are usually 100 V to 2000 V. The shape of the impulse voltage depends on the
test device and the test set-up. The bandwidth of the exciting signal should be as high as possible. Typical parameters of the impulse shapes are front times of 100 to 500 ns and time to half values of 40 to 200 s. The spectral distribution of the time domain signals are calculated using FFT. The quotient of output to input signal represents the transfer function in the frequency domain. In Fig. 2, the necessary circuits for measuring frequency features in different conditions of terminal connections are shown. If the circuit of Fig. 2 is done in proposed model, disregarding the magnetic circuit of equivalent circuits, Fig. 3 will result. Notice that using coupling properties (Fig 3(e)), M is considered in L1p. V. A. ESTIMATING THE MODEL PARAMETERS USING PSO ALGORITHM
PSO Algorithm PSO is an efficient algorithm and developed in recent years. PSO is one of the new algorithms invented by Kennedy and Eberhart in 1995 [10-11]. This algorithm inspired from social behavior of animals such as bird flocking or fish schooling. In comparison with other optimization algorithms, PSO has considerable search for complex optimization problems with faster convergence rate. As an advantage in programming, PSO requires fewer parameters for regulation than other optimization algorithms. Implementation steps of this algorithm are as follows: 1-Random generating of primary population, 2-Particles fitness calculation respect to their current positions, 3-compration current fitness of particles with their best experience: if F ( Pi ) pbesti (1) Then
pbesti = F ( Pi ) , xpbest1 = x1 (t ) (2) 4- Comparison of the current fitness of particles with the best experience of all particles if F ( Pi ) gbesti (3) Then gbesti = F ( Pi ) , xgbest1 = x1 (t ) (4) 5-change in velocity of each particle according to equation (5): v1 (t ) = v1 (t 1) + 1 ( xpbest1 x1 (t )) + 2 ( xgbest1 x1 (t )) 6-The particle position change to new position according to equation (6): x1 (t ) = x1 (t 1) + v1 (t ) (6) 7-Algorithm is iterated from step 2 until the convergence is obtained. Parameters used in the above algorithm are defined as follows: F ( Pi ) : Fitness of the i-th particle.
-
if Vi (t ) Vmax Vi (t ) = Vmax (8) The maximum value of velocity is selected with respect to change of position vector parameters. However, researches have shown that if the equation(5) is applied for updating velocity vector; it doesn't require checking previous conditions [15], [20]. v1 (t ) = k (v1 (t 1) + 1 ( xpbest1 x1 (t )) + 2 ( xgbest1 x1 (t ))) Where: 2 k= , = ( 1 + 2 ) > 4 (10) 2 2 4
Inertia weight: This parameters controls effect of previous velocity on current velocity. Its large certainly causes wide search space and vice versa. Implementation of inertia weight can be done by using of equation (11):
v1 (t ) = v1 (t 1) + 1 ( xpbest1 x1 (t )) + 2 ( xgbest1 x1 (t )) Preliminarily inertia weight is initialized by 1 value and reduce along algorithm. Also, the equation (9) must be true . If not , PSOs behavior becomes oscillatory and maybe divergence [12], [14], [15]: 1 > (C1 + C2 ) 1 (12) 2
x1 (t ) : Position of the i-th particle. pbesti : Best fitness of the i-th particle. xpbest1 Position of the i-th particle related to pbesti gbesti Best fitness of all particle in the population. xgbest1 Position of the particle with the best fitness.
C.
B. General Notes on Implementation of PSO Number of particles: Number of particles in search space is chosen by trial and error method to get better convergence. Since PSO application is relatively based on swarm intelligence, more particles lead to a better response. On the other hand, more particles require more calculations and therefore the method will be time consuming. Velocity limitation: One higher limit for velocity prevents that particles jump with high speed in searching region. Consequently, Space is searched to reach better region accurately. Additionally, this limitation prevents algorithm divergence due to high velocities of particles. After updating velocity vector, following conditions are checked: if Vi (t ) Vmax Vi (t ) = Vmax (7)
VH
VL
L1
H1
IMPLEMENT OF PSO ALGORITHM
If Zsi is the simulated models response to the input Xei and Zxi the output vector got from experimental results, the goal of parameters identification is in this way: Zsi= Zxi According to the noise, numeral errors in simulation and the errors of measurement devices, there is never equality. So the best estimation for parameters is the estimation which decreases the sum of squares of errors for n couple of Ysi and Yei or in other words, increases the standard function to maximum:
f fit =
([real (Z
i =1
xi
Z si )]2 + [imag ( Z xi Z si )]2 )
(13)
VH
L1
H1
VL
L1
H1
H1
L1
H2
L2
IL
H2
IH
L2
H2
L2
H2
L2
I T1 = L VH
a) T1
I T2 = H VL
b) T2
T3 =
I VH
T4 =
I VL
c) T3
d) T4
Figure 2. Measuring circuits in various conditions of terminal connections
T1
2Cg1
2C12
2C12
2Cg 2
T2
a) T1
R1 C1
C12
R2 L2
b) T2
T3
Cg1
C2
C12
Cg 2
T4
L1p
c) T3
d) T4
L1
L2
Leq
L1 p n :1
L2
n :1
e) Convert of transformer with coupling to without coupling Figure 3. The obtained equivalent circuits from running transfer function to circuit of figure 1.
Parameter identification is converted to an optimizing problem and can be solved using PSO algorithm. the parameters which are used For estimating parameters, the PSO algorithm is done by two groups of different parameters. In the first level, T1 and T2 are used (two functions are measured) for estimating Cg1, Cg2 and C12. So the parameters vector is like this: P1=[Cg1 Cg2 C12] In the second level, for estimating of C1, R1 and L1, the transfer function T3 is used. In this case, the parameters vector is like this: P2=[Ceq1 L1 R1] In the above relation, Ceq1 is the sum of Cg1, C12 and C1. The amount of C1 can be obtained by Cg1 and C12. In the third level, like level two, T4 is used for estimating following parameters: P3=[Ceq2 L2 R1]. Ceq2 is the sum of Cg2, C12 and C2. Runs the program According to this fact that in checking of transient states, the amounts of resistances (in ohm), inductances (in mH) and capacitors (microfarad) are measured, it is better for increasing the speed of convergence in PSO algorithm, to determine the first amount of variables personally. D. RESULTS AND EVALUATING THE PROPOSED MODEL The obtained parameters of the model using different methods (analytical, GA and PSO) are listed in Table 1. For verifying the accuracy and validity of proposed method, we substitute the parameters obtained from TFs and get its frequency response by simulation. Thereafter, the results are compared. In Fig. 4 the results of TF estimation in the range of 10-200 kHz are compared. According to this figure it can be say the accuracy of this model is more better than other well-known
methods. Therefore, it can be used as a reliable method for estimation of transformer transient model parameters.
a) T1
b) T2
Admittance [mA/V]
c) T3
Figure 4. Comparison of experimental and modeling results
Admittance [mA/V]
d) T4
TABLE I.
THE AMOUNTS OF ESTIMATED PARAMETERS USING DIFFERENT ALGORITHMS (RESISTANCE (OHM), INDUCTANCE (MH), CAPACITANCES (PF))
Method
Parameter Analytical Method [8] GA Method [16] PSO Method REFERENCES
C1
C2
C12
Cg1
Cg2
L1
L2
R1
R2
14.4 14.2 14.5
98.7 97.9 98.2
225.2 223.5 225.1
25.4 24.8 25.2
43.9 43.1 42.8
18.8 18.2 18.9
0.03 0.03 0.03
0.42 0.44 0.43
175.5 173.9 176.1
10.1 10.3 10.2
[1]
P .T .M .Vaessen, Transformer Model for High Frequencies, IEEE Transactions on Power Delivery, Vol.3, No.4, pp. 1761-1768, 1988. [2] S. Chimklai, and J.R.Marti, Simplified Three-Phase Transformer Model for Electromagnet Transient Studies High Frequencies, IEEE Transactions on Power Delivery, Vol.10, No.3, pp. 1316-1325, 1995. [3] T. Noda, H. Nakamoto, and S. Yokoyama, Accurate Modeling of Core Type Distribution Transformers for Electromagnetic Transient Studies, IEEE Transactions on Power Delivery, Vol.17, No.4, pp. 969-976, 2002. [4] P. I. Fergestad, and V. Henriksen, Transient Oscillations in Multiwinding Transformers, IEEE Transactions on Power Apparatus and Systems, Vol.PAS-93, pp. 500-509, 1974. [5] K. Cornick, B. Filliat, C. Kieny, and W. Mller, Distribution of Very Fast Transient Overvoltages in Transformer Windings, CIGRE, Session, WG 12-204, 1992. [6] Y. Shibuya, S. Fujita and N. Hosokawa, Analysis of Very Fast Transient Overvoltage in Transformer Winding, IEE Proceedings, Part C, Vol.144, No.5, pp. 461-468, 1997. [7] G. B. Gharehpetian, H. Mohseni, and K. Mller, Hybrid Modeling of Inhomogeneous Transformer Windings for Very Fast Transient Overvoltage Studies, IEEE Transactions on Power Delivery, Vol.13, No.1, pp. 157-163, 1998. [8] E. Rahimpour, and M. Bigdeli, Simplified Transient Model of Transformers Based on Geometrical Dimensions Used in Power Network Analysis and Fault Detection Studies, IEEE Conference, Powereng, Portugal, March 18-20, 2009. [9] E. Rahimpour, J. Christian, K. Feser, and H. Mohseni, Transfer Function Method to Diagnose Axial Displacement and Radial Deformation of Transformer Winding, IEEE Transactions on Power Delivery, Vol. 18, pp .493505, 2003. [10] J. Kennedy, R. Eberhart Particle Swarm Optimization, IEEE International Conf. on Neural Networks, Perth, Australia. Vol.4, pp. 1942-1948, 1995.
[11] Y. Shi, R. Eberhart Empirical study of particle swarm optimization, Proceeding of the 1999 Congress on Evolutionnary Computation, CEC 99, Vol. 3, pp. 1945-1950, 1999. [12] M. Clerc and J. Kenndy, The particle swarm-explosion, stability, and convergence in a multidimensional complex space, IEEE Trans. Evol Comput., Vol. 6, pp. 58-73, 2002. [13] R. C. Eberhart and Y. Shi, Comparing inertia weights and constriction factors in particle swarm optimization, In Proc. Congr. Evol. Comput, pp.84-88, 2000. [14] I. Trelea, The particle Swarm optimization algorithm; Convergence analysis and parameter selection, Inf. Process Lett., Vol.85, No.6, pp. 17-325, 2003. [15] Y. Zheng, L. Ma, L. Zhang, and I. Qian, On the Convergence analysis and parameter selection in partical swarm optimization, in Proc. Int. Conf. Machine Learning Cybern., pp.1802-1807, 2003. [16] M. Bigdeli, and E. Rahimpour, Estimation of Simplified Transient Model Parameters Using Genetic Algorithm, IEEE Conference, ICEE, Iran, May, 2010.
Using the Input Characteristic Impedance of a Static Converters for MPPT PV Systems
Jefferson William Zanotti, Walbermark Marques dos Santos, Denizar Cruz Martins
Federal University of Santa Catarina - UFSC Power Electronics Institute - INEP Florianpolis, Brazil jeffersonzanotti@inep.ufsc.br, walbermark@inep.ufsc.br, denizar@inep.ufsc.br
AbstractThis paper introduces a new Maximum Power Point Tracker (MPPT) technique for photovoltaic (PV) systems employing just a single voltage sensor. The MPPT algorithm is based on the variation of the input characteristic impedance of the basic DC-DC converter. Although the input impedance theoretically depends on its load resistance, it is shown that employing the proposed algorithm leads to the maximum available power without previous knowledge of its value. This technique can be applied to any DC-DC converter, mainly to the basic DC-DC converters as Buck, Boost, Buck-Boost, Ck, SEPIC and Zeta. The validation of the proposed MPPT technique is given based on comparison between theoretical and experimental test results using the SEPIC DC-DC converter. KeywordsCurrent sensorless; MPPT; PV panels; SEPIC converter.
proposed MPPT technique and the broadcast perturb and observe method (P&O). Section IV presents the hardware constructed and specifications. Using the SEPIC in Section V, experimental tests show the achieved result with the proposed MPPT technique, and finally in Section VI conclusions of the work are given. II. MATHEMATICAL ANALYSIS
This section shows a simplified PV panel electrical model and the derivation of relevant models for the basic DC-DC converters. These models are then used to study the interactions among them as the basis for the proposed MPPT technique. A. Simplified Modeling of a PV Panel Fig. 1 shows the simplified electrical circuit model of a photovoltaic panel according to [5].
I.
INTRODUCTION
The photovoltaic (PV) panel is one of the main components of the PV system. It generates electric energy in a DC form, and its characteristic curve current versus voltage (I xV) varies according to the incident radiation and the temperature, i.e. the power generated depends on these parameters. As PV panels still have low efficiency, it is essential to extract the maximum power available. Therefore, converters are used for maximum power point tracking (MPPT).Thus, several MPPT techniques have been studied, analyzed and compared in the literature [1, 2, 3, 4] to look for the maximum power point (MPP) at curve I x V. . They measure voltage, current, or both, and sometimes the temperature. Some are expensive and complex which sometimes means they are not worth the energy harvested. This paper introduces a new MPPT technique called Input characteristic impedance method (ICI), which uses a single voltage sensor. It makes use of partial derivatives of the estimated power in relation to the voltage and input characteristic impedance of the basic DC-DC. It will be shown through SEPIC that the accuracy with which the MPP is reached depends only on the variation of the duty cycle. This paper is divided as follows: Section II shows the employed model of a PV panel along with the derivation of the input characteristic impedance of the basic DC-DC converters and the interactions in the system comprising both, panel and converter. Section III exhibits the comparison between the
Fig. 1. Model of the photovoltaic panel according to [5].
This circuit is modeled by the following equations:
q Vpv + I Rs Vpv + I Rs kT 1 I = I pvcel Ir e Rp
(1)
G I pvcel = I sc + (T Tr ) 1000
(2)
T I r = I rr e T r
Where:
q E 1 1 G k Tr T
(3)
V pv , I pv - PV panel output voltage and current; I pvcel - current generated by the junction incident
radiation;
Fig. 3. Modeling of the PV panel as series resistance voltage source.
I r - panel reverse saturation current; Rs , R p - series and parallel panel resistance;

q - elementary charge of the electron, 1.6x10-19 C;
B. Simplified Modeling of the Basic DC-DC Converter An ideal (lossless) basic DC-DC converter operating at steady state with a fixed duty cycle and continuous conduction mode presents the relations between input and output voltages and currents given in (4) and (5),
- quality factor of the p-n junction;

k - Boltzmann constant, 1.38x10-23 J/K; T - environment temperature, [K];
Vo = f ( D) Vi Ii = f ( D) Io
Where:
(4)
I sc - panel's rated short-circuit current;
(5)
- temperature coefficient of the I sc ;

Tr - reference temperature, 298 K;
G - solar radiation intensity, W/m2;
Vi , Vo - input and output voltage, respectively; I i , I o - input and output current, respectively;
f ( D) - is duty cycle function.
I rr - reverse saturation current reference; EG - energy gap, 1.1 eV.
According to the adopted circuit modeling, a photovoltaic panel typical output characteristic (I-V) is shown in Fig. 2. Looking at the I-V characteristic curve, there are two distinct operation regions: As a voltage source and as a current source. When operating in the voltage source region, the PV panel can be modeled by a voltage source ( E pv ) in series with an output resistance ( Rs ). Fig. 3 E pv is the internally generated voltage and V pv is the voltage delivered to the load.
The previous relations are employed in the definition of the input static (steady state) impedance, i.e. the ratio between the input voltage and the input current to derive the input impedance of one DC-DC converter. This is the impedance observed by the input voltage source and is given by (6),
Vi R = Z in ( D ) = 2 o Ii f ( D)
(6)
This equation defines the input impedance of the basic DCDC converter seen by power source load. It is dependent of the load and the function of duty cycle. The input impedance can Zin (min) ( D) Z in ( D) Z in (max) ( D ) for range between
f max ( D ) f ( D) f min ( D) . C. Connecting the PV Panel to the Basic DC-DC Converter Fig. 4 shows the connection of the PV panel to the basic DC-DC converter, where the load resistance Ro is placed outside the converter block, for didactic reasons.
Fig. 2. Operating regions of the photovoltaic panel.
The variation of the basic DC-DC converter input impedance Z in ( D ) due to the variation of its duty cycle defines the voltage V pv and, thus, the transferable power from the PV panel. Considering a lossless converter, then Pin = Pout, and the converter delivers the maximum power available to the load if the proper duty cycle value is chosen. The MPP is found in the transferred power characteristics. Thus, applying the partial derivative with respect to V pv and Z in into (8) leads to (9),
Fig. 4. PV panel connected to the basic DC-DC converter.
Considering the basic DC-DC converter characteristic impedance, obtained in (6), Fig. 4 may be redraw as in Fig. 5, where the PV panel imposes a voltage V pv on Z in ( D ) terminals.
dPin =
2V pv Zin ( D)
dV
2 V pv 2 Zin ( D)
dZin
(9)
It follows that dPin = 0 at the MPP. This point depends solely on the instantaneous values of the converter input voltage, of the load resistance and of the incremental impedance Z in .
Fig. 5. Panel connected to external impedance modeling the basic DC-DC converter input characteristics.
Applying an impedance voltage divider into the circuit given in Fig. 5, and replacing the definitions in (6), the output voltage is given by (7).
To implement this equation in a microcontroller, an Euler method to discretely compute the derivative is employed as shown in the following approaches. The first difference equations are,
V pv =
Ro R o + f ( D) Rs
2
E pv
(7)
dV pv = Vk = V pv ( k ) V pv ( k 1) dZ in = Z k = Zin( k ) Zin ( k 1)
(10)
This equation defines the input voltage of the basic DC-DC converter depending on its duty cycle, the internal dc voltage, the series output resistance of the PV panel, and the load resistance. Equation (8) is employed to compute the power delivered by the panel to the basic DC-DC converter. The according characteristic is depicted in Fig. 6.
2 V pv
Where k is a sampling instant index or the calculation step of the variables. Replacing (10) in (9) and realizing the following algebraic simplifications leads to,
2Vpv(k) Vpv(k) Vpv(k1) Zin(k)
2 ) =Vpv in(k) Z in(k 1) ) (k) ( Z 2 Zin (k)
(11)
Pin = V pv I pv =
Zin ( D)
(8)
2 V pv ( k ) V pv ( k 1) V pv ( k )
) = ( Zin(k ) Zin(k 1) )
Zin ( k )
o Ro Z in(o k ) Z in ( k 1)
(12)
2 V pv ( k ) V pv ( k 1) V pv ( k )
)=
*R
*R
Ro Z in(o k)
*R
(13)
2V pv V pv ( k )
Fig. 6. Power variation in the input of the basic DC-DC converter, depending on duty cycle.
Zin o Zin (o k)
*R
*R
(14)
Where:
2 1 D) ( =
compared with a tolerance band. Next step is to calculate dV pv ,

*R
Zin(o k)
D2
(15)
which is related to dPin to determine if the duty cycle should be increased or decreased. At the end of each computation cycle, the values of D ( k 1) , V ( k 1) and Z ( k 1) are updated. III. COMPARISON WITH THE P&O METHOD
Equation (15) gives the input characteristic impedance at sampling index k or ( k 1) and this is simply dependent on the duty cycle. This feature shows that with the proposed technique the MPP is achieved with no influence from the load resistance value Ro since by measuring the basic DC-DC converter input voltage the according duty cycle value is straightforward to compute and leads to the maximum transferable power being effectively delivered to the converter load at all times. Of course, this is achieved on this simplified model not taking the non-idealities of the basic DC-DC converter, i.e. components losses, into account. However, employing a high efficiency is mandatory in solar power applications and losses are expected to be small leading to very low impact in the MPP tracking. The proposed MPPT algorithm is implemented as given in Fig. 7. The algorithm is based on the computation of the power derivative as explained in the previous section. Two variables must be known, the duty cycle, which is directly taken from a microcontroller register and the converter input voltage, which is the only sensed variable. The initialization of the variables is as follows: D ( k 1) = 0 , V ( k 1) = 0 and Z ( k 1) = large number.
Before verifying the experimental results, a way to validate the Input Characteristic Impedance method (ICI) is to make a fair comparison with the P&O method, keeping both methods with the same fixed step size and with the same algorithm that defines the direction of tracking. The conventional P&O technique, with fixed step size, has its widespread use, because it can track the MPP for different variations of temperature and radiation. However, the P&O technique has some disadvantages, like having to choose between high speed tracking or low steady state power oscillations, and also, it does not gives good results under rapidly changing atmospheric conditions. Nevertheless, as this method is consolidated in the literature, a comparison with it is essential for the validation of the input characteristic impedance technique proposed in this paper. The comparison method is exhibited from the steady state to the standard test conditions (STC), where Radiation = 1000 W / m 2 and Temperature = 25o C . Where, four disturbances were produced in order to show the tracking performance. The first disturbance occurs with a radiation step of 300 W / m 2 for less and soon for more. And the second happens in temperature, first adding 25o C and then removing it. In all disturbances both methods showed the same dynamic tracking, as show in Fig. 8.
Power (W)
1400
P&O ICI
1000
600 7 10 Time (s)
Fig. 8. Comparison between P&O and ICI methods.
Fig. 7. Implemented algorithm.
After this, the voltage across the input terminals of the basic DC-DC converter is measured, the value of Z in ( k ) is computed, and by employing (9), dPin is obtained. This is then
It is verified that both power curves extracted from the PV panels are overlapped, i.e. they are exactly the same. Therefore, the method of the input characteristic impedance is presented as a form of valid tracking, because it has the same performance as the P&O. But it has the advantage of using only a voltage sensor, which is easily made. And it eradicates the use of the current sensor, which is expensive and reduces the robustness of the converter.
IV.
HARDWARE CONSTRUCTED
The basic DC-DC converters that can sweep entire the plan I-V (Fig. 2) of PV panels are Buck-Boost, Ck, SEPIC and Zeta [3]. SEPIC (Single-Ended Primary Inductance Converter) was chosen to verify the input characteristic impedance technique by presenting the advantages, like non-inverting output, isolation between input and output by coupling capacitor, possibility of using transformer, among other. Fig. 9 shows the circuit diagram constructed, used in the simulations and experimental tests. The circuit parameters and PV panel details are given in Tables I and Table II respectively.
modules KC200GT in series, which totals approximately 78 Vmpp in the standard conditions.
TABLE II. KC200GT PV MODULE ELECTRICAL SPECIFICATIONS
Specifications @ S=1000W/m2 and T=25C Maximum power
Pmax Vmpp I mpp
200 W 26.3 V 7.61 A 32.9 V 8.21 A -0.14 C/V
Voltage at maximum power point Current at maximum power point Open circuit voltage Short-circuit current
Voc I SC
Maximum power point voltage coefficient with the temperature uV
V.
EXPERIMENTAL RESULTS
Fig. 9. Complete circuit in closed loop.
The experimental results show the performance of the ICI method when the panel is subjected to radiation variation ( G = [200,900] W / m 2 ). The acquisition of these results happened when a cloud covered the surface of the panel, and thus, the radiation decreases. The temperature remained at 50C and the load resistance at Ro = 160 . Fig. 10 shows the overlap of the experimental curve I x V from the panel and traced by the oscilloscope, with the theoretical curves for different levels of radiation.
In order to perform the control system, the PIC18F1320 microcontroller was used which among other features has a digital/analogical input port and a PWM module. They are the required peripherals for this project. It also has hardware support for connecting an LCD display, which was used to visualize the input voltage, output voltage and duty cycle. As the microcontroller has no current capacity to drive the MOSFET, a driver circuit was developed with transistors in Totem-pole configuration to trigger the switch.
TABLE I.
Input inductor (Lin) Magnetizing inductor (Lm) Input filter capacitor (Cbar) Coupling capacitor (Ca) Output capacitor (Co) Diode (D) MOSFET (S) Switching frequency (fs) Rated power
SEPIC CONVERTER DATA

1.75 mH
1.7 mH 220 F/400 V 47 F/400 V 2 F/1 kV RURG30120 IXFK44N60 50 kHz 1 kW Fig. 10. Superposition of the experimental curve (cross-hatched) and theoretical curves (colorful). Current x Panel voltage.
It is known that in SEPIC converter, the maximum voltage across the main switch is the sum of output voltage and input voltage. MOSFET used (see Table I) withstands the maximum voltage Vdss = 600 V . For safety reasons, voltage on the MOSFET was limited to two thirds of Vdss and resulting in 400 V. Thus, in the experimental tests were used only 3
It can be seen that during the passage of a cloud and constant panel temperature, the operating voltage remained practically unchanged. In turn, current gradually decreased with radiation. But, the system continued holding the panel at the point of maximum power, as it is verified in the dark crosshatched region that lies close to the knees of the panel curves. Fig. 11 shows the curves obtained by experimental test, respectively, when applying a step load and then the load is removed. With G = 850 W / m 2 and T = 45 C .
REFERENCES
Vo(ch3)
[1]
[2]
Vpv(ch1)
[3]
Ipv(ch2)
[4]
[5] Fig. 11. Experimental results of load voltage (Vo (ch3) 50 V/div), panel voltage (Vpv (ch1) 50 V/div) and current in the panel (Ipv (ch2) 5 A/div) for step loads.
de Brito, M.A.G.; Junior, L.G.; Sampaio, L.P.; e Melo, G.A.; Canesin, C.A., "Main maximum power point tracking strategies intended for photovoltaics," Power Electronics Conference (COBEP), 2011 Brazilian , vol., no., pp.524,530, 11-15 Sept. 2011. Pallavee Bhatnagar, R.K. Nema, Maximum power point tracking control techniques: State-of-the-art in photovoltaic applications, Renewable and Sustainable Energy Reviews, Volume 23, July 2013, Pages 224-241, ISSN 1364-0321, http://dx.doi.org/10.1016/j.rser.2013.02.011. R.F. Coelho, Study of Buck and Boost Converters Applied to the Maximum Power Point Tracker of Photovoltaic Solar Systems, Master's Dissertation (in portuguese). INEP-UFSC. Brazil, (2008). R. F. Coelho, F. Concer, D. C. Martins, A Study of the Basic DC-DC Converters Applied in Maximum Power Point Tracking, In: 10 Brazilian Congress of Power Electronics - COBEP 2009, 2009, BonitoMS. 10 Brazilian Congress of Power Electronics, 2009. p. 673-678. R. F. Coelho, F. Concer, D. C. Martins, A Proposed Photovoltaic Module and Array Mathematical Modeling Destined to Simulation, In: ISIE - International Symposium on Industrial Electronics, 2009, Seul. International Symposium on Industrial Electronics, 2009. p. 16241629.
In relation to Fig. 11, the system initially is in steady state, operating in the MPP with a load of 160 , and when the load changes to 51, the output current increases, so that the voltage in the load falls. At this point on the panel side, voltage drops and current increases, taking off the panel from MPP. However, it is observed that the control restores the panel voltage to the point of maximum power. When the load changes from 51 to 160 the panel voltage rises and the current falls. Yet, it is observed the control reestablishes again the system in MPP. It is important to emphasize that the load does not interfere in the proposed MPPT algorithm. VI. CONCLUSION
In this paper a new maximum power point tracker technique for photovoltaic systems employing just a single voltage sensor has been proposed. The MPPT is achieved by taking advantage of the steady state behavior input characteristic impedance of the basic DC-DC converter, and the power derivative calculation based on such impedance. The comparison between the proposed MPPT technique and the perturb and observe method was done and showed an excellent result, since the results were the same but the input characteristic impedance method does not use a current sensor, this potentially leads to less cost for such a PV system. The SEPIC was chosen to validate the proposed technique, and the extension to the Zeta converter is straightforward, since one is dual of the other, and both converters present the same input characteristic impedance, such as Ck and Buck-Boot. In spite of, the Buck and Boost converters do not sweep entire plan I-V of PV panel, this technique can be applied to them. Simulation and experimental results validated the presented MPPT algorithm for several operating conditions. ACKNOWLEDGMENT The authors would like to thank the CAPES (Coordination for the Improvement of Higher Level Personnel), CNPq (National Council for Scientific and Technological Development) and FINEP (Financier of Studies and Projects) by the financial support.
A novel control method in Grid Interconnection of DG Based on Pulse Adaptive VSI in order to Harmonic Compensation and Power Quality
R.Galandary Tazekandi MSc. Student and Member, IEEE
K.N.Toosi University of Technology Tehran, Iran r.galandary@ee.kntu.ac.ir
AbstractDistributed Generation Resources are being increasingly connected in distribution systems utilizing power electronic converters. Having attended to this consideration, a new PWM scheme, suitable for being used in DG interface is proposed at this paper. Using proposed pulse adaptation method in conventional voltage-control-based method of DG, harmonics in network is compensated without adding to the complication of the main power control level, because this method, unlike the conventional harmonic compensation methods, applies pulse shaping process in PWM level of the system to do harmonic depression service, than in main control loops. The result of this method compared with two other different type control methods such as Unit Vector Template and Repetitive control based on discrete Fourier transform. All results achieved in MATLAB/Simulink. Keywords- Pulse Adaptive VSI; Unit Vector Template; Repetitive control; Active Power Filter; DG; Power Quality; Harmonic compensations. .
S.M.T. Bathaee Associate Professor and Member, IEEE

K.N.Toosi University of Technology Tehran, Iran Bathaee@kntu.ac.ir
power control, which according to (1) and (2) is calculated and by using PLL 1coordinates with the network [1].
p q
V DG . V PCC x x
.sin
(1) (2)
V DG
(V DG V PCC )
According to above Equation, active and reactive power are controlled by
and V DG respectively.
B. Islanding operation mode In this mood DG sources are used in order to local load or micro grid supplying. Likewise in this mode current control method is used because of its ability to control PCC2 voltage and frequency [2]. C. Connected to the network and islanding mode simultaneously Because of importance of this mode uses, synthetic method of voltage and current control are used which, require quick islanding detection. Also we can hierarchically method [3]. The paper is arranged as follows: Section II describes the new control method of pulse adaptive VSI. Unit vector template method is presented in Section III. Section IV describes the repetitive based control method and, finally, Section V concludes the paper. II. PULSE ADAPTIVE VSI Most DG systems use VSI as the interface for the coupling to the electrical network. Therefore, many control schemes have been developed. These controllers can be classified as communication-based or droop controllers. Methods such as the central-limit control, master/slave control and distributed control are presented as communication based scheme. In
I. INTRODUCTION Presence of switching device and non-linear loads causes harmonic turbulences in voltage and current waveform. Turbulences increase and network unbalancing have increased the tendency for using power electronic converter. Three power electronic converters which, are used more than the offers are: (1) voltage source inverter, (2) Current source inverter, and (3) Z-source inverter. Over ally control methods in active power filters for harmonic compensation are divided in two section. First section is in time domain and second section in frequency domain, which the first section used more than second one because of its better dynamic response. There are three modes for DG operation, therefore three different mode for converters control exist A. Connected to network mode In this mood purpose using of DG is compensating of load power requirement which, is controlled by current control converter. In fact the role of converter is active and reactive
1 2
Phase Locked Loop Point of Common Coupling
order to avoid communication based failures and to increase the system reliability, droop-based control is widely used [4]. As another classification, the control mechanism of a VSI mainly consists of two types: current control method (CCM) and voltage control method (VCM). CCM has been adopted in active power filters (APFs) to mitigate system harmonics. For micro grid applications, especially considering the possible autonomous islanding operation, the DG units are preferred to operate with VCM as it can provide direct voltage and frequency support for the loads [5]. To facilitate the sharing of load demands, VCM-based Vf droop control is usually considered to control the DG inverter when the ancillary service of system harmonic compensation is not considered. To consider harmonic compensation in V-f droop control strategy, it has been used CCM-based V-f droop control with extra units of harmonic current reference and a current tracking control loop. This units in addition to make control system more complicated, need system models to be designed properly. In this method, voltage control method is used to harmonic current control without any extra units which are common in CCM based method. This method uses pulse shaping in PWM level of system than in main control level and therefore to do harmonic compensation, a new scheme of pulse command generating technique for inverter is proposed. Conventionally, a voltage-command-based open-loop PWM scheme is used in VCM and a current-command-based closed-loop PWM scheme in CCM. Proposed scheme is voltage-command-based closed loop PWM scheme with current shaping ability. Using this scheme, for a given voltage command, pulses produced in inverter output are so arrayed that line currents have the closest shape to arbitrary current reference and at the same time RMS values of reference voltage and produced voltage be the same. Applying this strategy, current harmonics throughout the power system can be compensated. In voltage control method as illustrated in Fig. 1, the fundamental reference voltage is obtained from the power control loop. The real power reference can be produced using the maximum power point tracking like in a wind or a photovoltaic system, the maximum system efficiency control like in a fuel cell system, or the command value from a micro grid energy management center [5]. The reactive power command can be produced from the voltage support requirements or load power factor compensation algorithms. Non-linear load current is automatically divided between DG unit and the grid, based on circuit conditions and so no compensation is performed. To consider ancillary service of harmonic current compensation in DG unit, many strategies, mostly based on current control have proposed, Fig. 2. Basically, current control method functions by controlling the DG unit as a shunt active power filter, where the DG unit absorbs harmonic currents of the nonlinear load, leaving an improved source current and PCC voltage with lower total harmonic distortion. To achieve this function, it is conventional to control DG as a resistive active power filter, where the PCC voltage or its close
voltage is measured and the harmonic components are extracted to produce the reference harmonic current of the DG. As a result, the DG unit acts as a small resistance at harmonic frequencies. To avoid conflicts between the compensation of harmonic currents and the primary function of real power injection of a DG system, the reference harmonic current generation and the current tracking loop need to be designed carefully. The primary service of a DG unit i.e. real power injection and the secondary service of harmonic compensation are of independent natures. These two functions are synthesized in CCM and cause some difficulties such as increasing complication of control system, magnitude and phase errors in the extracted signals, need to resonant filters and etc. So independently control of each function of DG can have of interest. So, here its proposed to do harmonic compensation in PWM level of system to avoid interference with real power injection function.
Figure 1. Voltage control method strategy
Figure 2. Current control method strategy
A. New PWM scheme As illustrated in Fig. 3(a), open-loop scheme of PWM technique, gets voltage command and produces voltage pulses as an uncontrolled manner. Techniques such as sub oscillation method, space vector modulation, synchronized carrier modulation, optimal sub cycle method and selected harmonic elimination are considered as open-loop PWM technique [6], [7].Closed-loop schemes mainly developed based on motor control requirements and their main goal is accurately tracking of stator current or VSI line current like reference input current, Fig. 3(b). PWM Techniques such as hysteresis current control, sub oscillation current control, space vector current control, predictive current control and trajectory tracking control are grouped as closed-loop methods [6], [8].All mentioned techniques try to accurately track the input current reference and at the same time have the least response time, as these characteristics are desired in motor control usages [7][9].These techniques have found many applications in fields other than motor control, such as DG units. As its mentioned, using closed-loop current-command-based PWM scheme in harmonic compensating DG units may produce some interference with other functions of system. Proposed PWM scheme is a closed-loop voltage-command based line-current controllable technique, so that VSI output current can track a reference current, independent of commanded voltage, Fig. 3(c).
changes voltage pulses of VSI such that resulted current shapes have closest similarity to the reference current. The power processing unit generates voltage command of:
v * v * sin(* t)
(3)
Figure 4. Pulse Adaptation strategy
So, the generated voltage in the terminals of VSI must have of the same magnitude (or RMS value here) and the same frequency of commanded voltage. Instead of sending these values directly to SVM unit, control unit produces other equivalent set of values as: v 1 , v 2 ..., v n ..., v N ,so that, this equation to be true:
v * N v 1 v 2 ... v n ... v
commanded ( v
* / 2
(4)
If condition of the equation (2) to be met, RMS values of the ) and the resulted voltages will be same.
SVM technique used here is same as normal Space Vector Modulation (SVM) but all SVM sampling values (the N values) in each cycle of main frequency vary as commanded from control unit, Fig. 4. SVM sampling time is attained from the commanded frequency of w*.
Figure 3. PWM schemes. (a) Open-loop. (b) Closed-loop. (c) Proposed scheme.
B. Pulse Adaptation method As it is seen from Fig. 3, sampling of current is done in terminals of load and so output current is regulated and we feed filtered voltage and current into the load. At first, control unit produces standard and lowest possible harmonic content normal SVM-based voltage pulses and then sampling unit sends resulted current waveform data to control unit. Comparing actual current and reference current, control unit
Figure 5. SVM from power processing loop (v*) and PWM control (v)
Fig. 5 shows VSI output line-to-line voltage pulses. The space vectors sequences can be selected as desired, for example in sector 1, it may be like: u0 (or u7) u2 u1 u0 (or u7), as the choice of Fig. 5.
(b)
Figure 6. Line to line voltage of VSI output
(c)
Figure 7. Before pulse adaptation. (a) VSI output Voltage. (b) Load voltage. (c) Load current.
Basic pulse time (equal to SVM sampling time) can be calculated from this equation:
Tp
*
2 * n
(5)
Where N is the number of the basic pulses per cycle of the main frequency. Pulse durations are calculated as:
T a ,n
T b, n
v v
v v
a ,n max
b, n
T p*
T p*
(6)
max
Fig. 7(a) shows line voltage pulses of one of the three phase set, that is produced by SVM as commanded from power processing unit at first step (before pulse adaptation begins), which is symmetrical and has lowest harmonic content. It goes through LC filter and becomes a smooth sinusoidal as shown in Fig. 7(b). Smooth voltages react with the load and cause generation of a set of three phase currents. Current waveform of one of the phases is shown in Fig. 7(c). Sensing-sampling system begins to sample of the current and corrects its shape without having any information about the load and its nature. The sampling system divides each cycle of current waveform into N equal parts and then samples magnitudes of current in the middle time of each N part. These sampled magnitudes ( i 1 , i 2 ..., i n ..., i N ) enter sine estimator unit so that the closest sine shape to these values to be estimated. The sine estimator unit is a digital low pass filter and extracts fundamental frequency of the current. Sensed value of i n is compared to the estimated value of i
r ,n
T 0, n T p* (T a, n T b, n )
Where, v SVM [9]. is maximum allowable amplitude of vectors in
max
and error en is
(a)
produced. Errors enter controller and the control system determines the best choices for pulses command so that current deviation to be decreased. At the next cycle, sampling will be done again and error signals will be produced again and correction on pulse durations of inverter will be executed again too, to attenuate current deviation. Number of required cycles for pulse adaptation process, to go all errors to zero or to a special predefined range, depends on the nature of the load and also the controller. The voltage pulses of same phase as shown in Fig. 7(a), is illustrated in Fig. 8(a). Voltage waveform of the load terminal
is changed as it is shown in Fig. 8(b) and the modified current waveform is as of Fig. 8(c).
is: I r
i 1,r , i
2 , r ..., i n ,r ..., i N ,r
.So, the error vector is

(7)
defined as: Where error vector is:
e Ir I
e T e1 , e 2 ,..., e n ..., e N .The output
vector of the neural network, be calculated as:
LT l1 ,l2 ,...,ln ...,lN can
L f 3( W 3 *f 2 ( W 2 *f 1 ( W 1 *e B1 ) B 2 ) B 3 ) (8)
(a)
(a) (b)
(b)
Figure 9. Simple three layer, N-N-N, neural network. (a) Structure. (b) Activation functions.
WhereW 1 , W 2 and W 3 are weight matrixes and are of the form
(c)
Figure 8. After pulse adaptation. (a) VSI output voltage. (b) Load voltage. (c) Load current.
of:
C. Control system Because of complicated nature of inverter-filter-load, it is difficult to determine desired width of pulses of inverters output voltage via mathematical tools or predefined look-up tables so that we reach to a set of modified current waveforms in various conditions. Thus, we need to have an adaptive-type flexible control system. Artificial neural network (ANN)-based control system may be a suitable choice, because the problem of finding the best set of values ( v 1 , v 2 ..., v n ..., v N ) so that the set of errors (e1, e2 en eN) lay in a defined range, seems to be a pattern recognition problem. Fig. 9(a) indicates structure of a simple back-propagation based three layer ANN [10], [11]. First layer as the input layer has N neurons and error signals enter there. Hidden and output layers have the same number of neuron as the first layer. The sensed-sampled vector is: I the estimated
T
W 111 W1 a 1N 1
a11N a1NN
And B 1 ,
a21N W 211 W2 a a2 NN 2N 1 a31N W 311 W3 a a3NN 3N 1 B 2 and B 3 are the bias vectors as:
B1T b11
b1n
b2 n
b1N
b2 N
i 1, i
2 ..., i n ..., i N
and
current
B 2T b21
reference
vector
of
B 3T b31
b3 n
b3 N
The activation functions of f , f and f are considered as Fig. 9(b). The vector L enters comparison and modification block and after being processed there, voltage commands set is produced and enter SVM block. The back-propagation method has been used for training of the network [7]. Training set was selected based on this fact that all members of error vector to be in the range of one percent of the sensed current. Training process here, was performed for one special case of non-linear load. One hundred possible working situations of this nonlinear load were considered as training set. Training has done using Levenberg-Marquardt algorithm [10]. For codebased simulation executed in MATLAB environment, pulse adaptation strategy was performed and after 16 cycles of main frequency, the results of Fig. 8 were attained for the same nonlinear load as of the Fig. 7, that is composed of a static three phase rectifier and unbalanced resistive loads. The trained network is not suitable for other non-linear loads and thus to design a general control system, other ANN-based control systems with the ability of on-line training or adaptivetype multi-input multi-output controllers may be needed. D. Harmonic compensation with new scheme Fig. 10 shows voltage-controlled method in DG-connected network with the capability of pulse adaptation in PWM level of the system. Before the start of the pulse adaptation, conditions of the currents and voltages of the system are as illustrated in Fig. 1. Unlike the CCM strategy that harmonic component of the non-linear load is fed by DG, using this strategy, harmonic component of load disappears and sinusoidal current of load is divided between DG unit and the grid (based on the commands from power processing unit of the DG). In fact, the VSI output voltage or approximately the PCC voltage is changed so that the current of the DG becomes sinusoidal and as a result, the current of load becomes sinusoidal, because this is DG unit that feeds a portion of load current. The above situation will be resulted in this fact that grid current is sinusoidal or without harmonic current.
Fig. 11 and Fig. 12 exhibit simulation results for a simple case of DG-connected network feeding a non-linear load. A 50 KW three phase load was selected including resistive loads and a PWM rectifier as the non-linear load. Frequency and voltage of the network were considered 50 Hz and 330 V respectively.
Figure 11. Voltage and current before pulse adaptation. a) DG votage. b) DG current. c) Grid curre
Figure 12. Voltage and current after pulse adaptation. a) DG votage. b) DG current. c) Grid current
III. UNIT VECTOR TEMPLATE (UVT) METHOD For this method 4-leg grid interfacing inverter is proposed which, its schematic have been showed, as in [12]. According to this scheme and by use UVT method, we controlled the DG to operate as on active power filter and we inject DG's generated power to the network and compensates the harmonics. Simulink control block diagram showed in Fig 13, Fig 14 and Fig 15.In this method all three DG modes operation are studied, first mood when DG dont have any generation and compensation, in second mood likewise DG dont have generation but have harmonic compensating, and in third mode DG injects power generation either have harmonic compensation and simulation result listed in Table .
Figure 10. Voltage control based pulse adaptation
IV. REPETITIVE BASED CONTROL BY DFT3 METHOD This method uses a close-loop repetitive control scheme which, operate based on FIR4 filter and DFT. This method based on DFT compensates harmonics selectively. Active power of DG injects to network and by current harmonic injection, network current will be sinusoidal and decreased harmonic distortions decreases [13]. Control scheme structure showed in Fig 16.
Figure 16. Control scheme structure Figure 13. Hysterisis controller structure in UVT
Figure 17. Schematic of converter reference current generated
Figure 14. Simulation structure of PLL and UVT
Fig 17, shows the generation scheme of reference current of converter. In this scheme the load current is transfer abc to frame and after passing the resonant filter, we can harmonic load current by use of difference of load current between main component of current. Likewise the PCC voltage is transfer abc to frame and after passing resonant filter by using instantaneous power theorem and (9) and (10) the active and reactive power injected of DG can be calculated [14], [15].
p i L 1v 1 i L 1v 1
Figure 15. Simulation structure of low-pass filter with PI controller in UVT TABLE I. SIMULATION RESULT OF UVT THD Harmonic spectrum in Phase a b c
Before compensating without DG generated After compensating without DG generated After compensating with DG generated
(9) (10)
q i L 1v 1 i L 1v 1
In this scheme k1 and k2 coefficient are used for coordination between active and reactive power.so the reference current of DG by (11) can be calculated.
15.55% 23.58% 15.54%
1.75% 1.73% 1.74%
3.98% 3.99% 3.96%

3 4
i 1DG 1 i 2 2 1DG v 1 v 1
v 1 v 1
v 1 p * * v 1 q
(11)
Discrete Fourier Transform Finite Impulse Response
With summation of reference current and harmonic load current, we can calculated the reference current of converter. Transfer function of resonant filter in ideal and non-ideal mode are calculated with:
[4]
G h (s)
2s s (h n )
2 2
(12)
[5]
Transactions on Power Delivery, Vol.21, No. 3, July 2006, pp. 15151523. Tine Vandoorn, Bart Meersman, Jeroen De Kooning and Lieven Vandevelde, Controllable harmonic current sharing in islanded microgrids: DG units with programmable resistive behavior toward harmonics IEEE Trans. on power delivery, vol. 27, no. 2, pp. 831-841, April 2012. Jinwei He, YunWei Li and Md Shirajum Munir, A flexible harmonic control approach through voltage-controlled DGgrid interfacing converters, IEEE Trans. on industrial electronics, vol. 59, no. 1, pp.444-455, Jan. 2012.
Joachim Holtz, Pulsewidth modulation for electronic power conversion, Proceedings of the IEEE, vol. 82, no. 8, pp. 1194-2014, August 1994.
[6]
G h (s)
2s s 2n (h n ) 2
2
(13)
[7] [8]
In this above Equation is damping factor and h is harmonic degree. Fourier transform Equation is as follows.
[9]
FDFT (z)
2 N
N 1 i 0
( cos( N h (i N
k N h
[10] [11]
a
)))z i
(14)
[12]
In this above formula N is the coefficients, h is the set of selective harmonics, and Na is the number of steps to the system stability [14].Simulation result listed in Table II.
[13]
[14] TABLE II.

SIMULATION RESULT OF REPETITIVE BASED CONTROL
[15]
Operated Mood Before DG connected to Network After DG connected and selective harmonic compensating After DG connected and total harmonic compensating
THD 26.97% 15.92% 4.64%
Bimal K. Bose, Moderm power electronics and AC drives, the university of Tennessee, Knoxville, prentice Hall PTR, 2002. D. M. Brod and D. W. Novotny, Current control of VSI-PWM inverters, IEEE Trans. Ind. Appl., vol. IA-21, no. 3,pp. 562-570, May/June 1985. Nguyen Phung Quang, Jrg-Andreas Dittrich Vector control of threephase AC machines, system development in the practice, Springer press, 2008. Math Works Inc. Neural Network Toolbox Users Guide, 2008. B. K. Bose, Expert systems, fuzzy logic, and neural network applications in power electronics and motion control, Proc. Of the IEEE, vol. 82, pp. 1303-1323, Aug. 1994. M. Singh et al., Grid Interconnection of Renewable Energy Sources at the Distribution Level With Power-Quality Improvement Features, IEEE Transactions on Power Delivery, Vol. 26, Iss. 1, January 2011. L. R. Limongi, R. Bojoi, A. Tenconi, and L. Clotea, Single-phase Inverter with Power Quality features for Distributed Generation systems, in Proc. IEEE OPTIM Conf. Rec., 2008, pp. 313 318. R. I. Bojoi, L. R. Limongi, D. R. A. Tenconi, Enhanced Power Quality Control Strategy for Single-Phase Inverters in Distributed Generation Systems, IEEE Trans. On power Elec., vol. 26, no. 3, March 2011. L. R. Limongi, R. Bojoi, A. Tenconi, and L. Clotea, Single-phase Inverter with Power Quality features for Distributed Generation systems, in Proc. IEEE OPTIM Conf. Rec., 2008, pp. 313 318.
V. CONCLUSION In this adaptive pulse VSI method two services of real power injection, harmonic compensation in the DG unit are performed without having any interference with each other and as a result, complexity of control system and some other problems like difficulties in proper adjustments are solved. UVT method decreases the harmonics in three modes by the same scale and compare with other methods have fast dynamic response. Repetitive based control method according to use of DFT can compensates the harmonics selectively and totally. VI. REFERENCES
[1] J. Carrasco, L. Franquelo, J. Bialasiewicz, E. Galvan, R. PortilloGuisado, M. Prats, J. Leon, and N.Moreno-Alfonso, Powerelectronic systems for the grid integration of renewable energy sources: A survey, IEEE Trans.Ind. Electron, vol. 53, no. 4, pp. 1002 1016, Jun. 2006. F. Katiraei, M. R. Iravani, and P. W. Lehn, Micro-grid autonomous operation during and subsequent to islanding process, IEEE Trans. Power Del., vol. 20, no. 1, pp. 248257, Jan. 2005. H.H. Zeineldin, E.F. El-Saadany, M.M.A. Salama, Impact of DG Interface Control on Islanding Detection and Nondetection Zones,IEEE
[2]
[3]
Parametric Analysis of Overvoltages Caused by Back-flashover In "Siah-bishe" 400kV GIS Substation

Mehdi Babaei
High Voltage Design Department Faraniroo Engineering Company Tehran, Iran
Mehdi.babaei@faraniroo.net
Maziar Babaei
Electrical & Computer Faculty Semnan University Semnan, Iran
Maziar.babaei@students.semnan.ac.ir
Mohsen Niasati
Electrical & Computer Faculty Semnan University Semnan, Iran
Mniasati@semnan.ac.ir
Abstract This paper presents the way of implementation of transient modeling concepts which are fully analyzed in referred standards, for the purpose of evaluating the lightning performance of studied system caused by back-flashover. In this regard, the impact of several parameters on lightning behavior of the system has been studied. The focus of this paper is on the lightning study of "Siah-bishe" 400kV gas insulated substation (GIS) and practical application of the study to the insulation design of this substation as a substation with a distinctive arrangement of equipment. This practical application has been rarely done before. The study was carried out by modeling all pieces of equipment in EMTP-RV and careful consideration of different modeling guidelines and instructions described in available standards. KeywordsLightning overvoltage, EMTP, gas insulated substation, back flashover, lightning stroke, cable, lightning arrester, contingency condition
The project included establishment of four 400kV incoming lines as 400kV cables with length of about 300m from power plant unit 250MVA transformers and three 400kV outgoing lines all in a double busbar with bypass DS arrangement. Generally, the incoming surge can be determined for both the Shielding Failure and the Back-flashover and the lightning performance of transmission lines is defined as sum of the SFFOR (Shielding Failure Flashover Rate) and BFR (Back Flashover Rate). Shielding failure occurs when phase conductors are struck by lightning surge directly, while Back flashover occurs in arching horn due to the increased tower's voltage by lightning on towers or overhead ground wire of transmission line. Rate of back flashover of the transmission line depends on various factors such as tower footing resistance, height of the tower, the number of flashes/km2/year, etc. Since overhead transmission lines are usually well designed and shielded to avoid direct lightning strokes to the phase conductor and due to perfect shielding of 400kV transmission lines connected to Siah-bishe substation, shielding failure rates are normally negligible in comparison with the back-flashover rates [3], therefore, only the back-flashover event is considered in this study. Flashovers are commonly caused by the back-flashover mechanism. If the stroke current flows through a transmission tower, it will produce a voltage difference across the grounding resistance of the tower and elevate the tower to a potential above that of the remote earth. The potential of the phase conductors remains the same as it was before the contact of the lightning flash with the tower. As a result, a voltage develops between the tower and the phase conductor. The current passing through the tower also produces a magnetic field in the region between the tower and the phase conductor. This magnetic field changes with time, as does the lightning current, producing a voltage between the tower and phase conductor that is proportional to the rate of rise of lightning current and if it exceeds the insulation strength of the insulator, flashover occurs. The impact of an overvoltage in GIS more severe compared with air insulated substation. Moreover, the insulation coordination of GIS connected to an overhead line through
I.
INTRODUCTION
The insulation design of devices and equipment in power system highly depends on lightning overvoltage in power network, thus for the optimum system design and operation of the power system, these transient overvoltages have to be analyzed and investigated inevitably. This imposes a highly qualified analysis for overvoltage phenomenon in power systems to accomplish a rationalization of insulation designs. To meet these demands, digital calculations, i.e. analyzing methods for overvoltage by computer tools, e.g. EMTP (Electro Magnetic Transient Program) became inevitable. Siah-bishe substation is situated in a mountainous region of Iran at about 125km north of Tehran (Capital) and 100km south of Caspian Sea in north of Iran. This geographical feature of the site location at 2500m above the sea level became the major concern over design of insulation withstand levels for the equipment due to low air density. The substation is connected to a pumped-storage 1000MW power plant via a 400kV cable led through a tunnel (Fig.1). The project was performed by Faraniroo Co. (Iran, Tehran) as one of the leading EPC contractors in the Middle East affiliated with Hyosung Corporation (South Korea, Seoul) as GIS supplier.
cable poses a particular problem because of several discontinuities in network (i.e., due to multiple reflections). In addition, due to significant lightning overvoltage appear in GIS-Cable-Transmission line configuration, to limit the number of surge arresters required to be installed in GIS configuration is itself a major concern.
Fig. 2. CIGRE concave waveform (tf is the front time and If the crest value of the current) Fig. 1. Siah-bisheh GIS-Cable-Transmission line configuration
The power-frequency voltage, at the inception time of a lightning stroke is represented as a power-frequency voltage source and an adaptation resistor in series [2]. B. Overhead Line Three 400kV transmission lines are connected to Siah-bishe substation with the span length of 300m except four last spans which are of 150m length. The frequency dependent model is used to represent 2km of line with conductors and towers accurately [6]. Five single circuit towers from the substation gantry are modeled for each overhead line connected to the substation. Phase conductors are ACSR Curlew (3 bundle) type for incoming line and also 7AWG5 for shield wires. Single circuit multi-story towers are modeled as a lossless propagation line with constant characteristics with 150 and 220 surge impedances (for lower and upper sections, respectively) in series with a 10 footing resistance [2]. The flashover mechanism of the insulators is represented by volt-time curves and insulator length which is based on the application of the Leader Progression method [2]. In EMTP, the volt-time characteristics are modeled using Transient Analysis of Control Systems (TACS) routine, for the insulators that are likely to flashover at each tower. For simulation of the air gap, the leader method is utilized as a program adapted and implemented into the EMTP built-in program in order to produce experimentally flashover characteristics and also to predict the performance of insulation as a function of impulse voltage waveform, time to flashover, gap configuration and others. Regarding this, the air gap distance of 400kV line is set to 2.9m. C. Insulated Cable Four 315MVA transformers of power plant are connected to GIS substation by 400kV cables. Cable spans are 300m length and are run inside a tunnel. However, for short lengths, a simple model with constant parameters and without crossbonding can be considered, in this study, the frequencydependent distributed parameter model is applied as the most accurate representation of an insulated cable in lightning
II.
SIMULATION
A. Lightning Stroke In this study, it has been assumed that the lightning stroke hits the first tower which is nearest to the substation. Lightning stroke is modeled using CIGRE characteristics due to its more realistic results [1]. Generally Subsequent strokes create more insulator voltage but at shorter times where the insulator strength is higher. Since the severity of voltage across the insulators is concerned, subsequent strokes in the same flash are no worse than the first stroke. Therefore the subsequent strokes are ignored in this study. Fig.2 shows the concave wave shape of injected current source with the following characteristics which was done for the purpose of simulating the lightning stroke hits the tower: Crest amplitude = 150kA Front time = 3s Time to tail = 77.5s The amplitude is obtained from the annual reports of National Meteorological Organization and the lightning path impedance of 1k is obtained from [7]. The time to crest is a probabilistic value, and further, it is dependent on the stroke current, i.e. the time to crest is conditioned on the magnitude of the stroke current as well as the steepness of waveform. In this study, Tf is assumed to be constant and the maximum rate-of-rise Sm of the current has been so adjusted such that the ratio of Im/Tf to Sm corresponds to the average values of a first stroke. However, since the steepness of the surge is of primary importance in determining the voltage across the terminals of the station equipment, the surge with the highest steepness is most frequently chosen. In some cases, the surge with the highest steepness has a crest voltage such that the surge voltage at the station would be less than the lightning impulse insulation level (BIL) of the equipment.
studies [2]. The characteristics of the cable considered in this study are as follows: Conductor cross section= 630mm2 XLPE insulation thickness= 28.6mm XLPE relative permittivity= 2.7 Conductor diameter= 30.3mm Overall diameter= 118mm Cable sheath is bonded to the main grounding grid at both ends. D. Lightning Arrester To entail the nonlinear behavior of lightning arrester in our study, the model recommended by IEEE Working Group 3.4.11 is used which is composed of two nonlinear elements separated by a resistance-inductance network and is based on the V-I characteristic of lightning arrester (Fig.3) [4]. Moreover, the parameters are obtained from [1] according to dimensional specification of modeled arrester: L1=66.6H, R1=286, L0=0.88H R0=440, C0=0.02272 nF
selection depends upon the steepness of the surge, the minimum length of traveling wave models, plus the use of flashover gaps and surge arresters with significant lead lengths. In this study, time step and simulation time are set to 10ns and 100s, respectively.
III.
ANALYSIS OF INFLUENCING FACTORS
A. Effect of surge conditions Different standards recommend various wave shapes to represent lightning return stroke currents including ramp, double exponential, concave waveforms. A double exponential wave shape usually results in greater over-voltages than other waveforms and the maximum voltage appears at t=2 (: Traveling time along the tower). Presently, it is assumed that a concave waveform of the first stroke is a better representation since it does not show a discontinuity at t=0. It is noticed that the concave wave shape causes far smaller voltage rather than other waveforms and provides more realistic results while other waveforms present more conservative results [7], [9].
TABLE I. BASIC ANALYSIS CONDITION FOR LIGHTNING OVERVOLTAGE
STUDY
Parameter Magnitude Wave shape Superposition of power frequency Stroke point T/L model Conductor type Tower model/span Footing resistance Circuit condition AIS or GIS LA characteristics 150kA 3/77.5s Yes
Base model
Fig. 3. IEEE model for lightning arrester
Lightning Condition
E. GIS Components Due to the high frequency nature of the lightning phenomenon, all pieces of GIS substation equipment were modeled as appropriate phase-to-ground capacitances and surge impedances with associated length and wave propagation speed. Values used in the analysis are as follows which are obtained from their physical dimensions and manufacturer's data: - Gas insulated bus duct (GIB): Lossless line with Zs=72.5 - SF6/air Bushing: Lossless line with Zs=250 and a 200 pF capacitance to ground - Current Transformers: A capacitance of 680pF - Capacitive Voltage Transformers: A capacitance of 500pF - Disconnector with Earthing switch: A capacitance of 200pF - Circuit Breaker: A capacitance of 150pF The GIS model includes most substation equipment (ducts, elbows, spacers, arresters, etc.). If GIS is coupled to overhead lines through relatively short cables, many of these components will have a negligible effect on lightning transients. Gas insulated switchgear is consist of single-phase ducts. Therefore, the Constant Parameter (CP) line model is used for representing GIB [2]. Table I shows the simulation conditions for analyzing the lightning overvoltages induced by back-flashover. The accuracy of the digital simulation can be affected by the use of a time step that is too large or too small. The time step
Transmission line
Top of last tower Line with constant parameter ACSR Curlew/3 Bundle Single circuit/300m, 150m 10 4 incoming from unit P.Tr with cable 3 outgoing with 3 bundle line GIS Yes
Substation & Power plant
1) Peak of lightning current: In our study, peak current varied within the range 50-150kA, started from 50kA increasing the amplitude in 20kA steps. As shown in Fig.4, the peak voltage at the Line-GIS termination will be dependent on the peak of strike current. The peak current is statistically related to the steepness or time to crest of the current wave form. The steepness and the front time increase with peak current. 2) Front Time of Lightning Stroke: Fig.5 compares induced voltage at the Line-GIS termination with various front times of lightning strokes: 1.2, 2, 3, 4, 5s with magnitude of 150kA. As shown in this figure, it is noticed that the shorter front wave time increases the induced voltage. As the front wave time increases, the induced overvoltage decreases and in turn the time to clear the overvoltage increase.
3) Tail Time of Lightning Stroke: Fig.6 compares induced voltage wave form at the Line-GIS termination with various tail times of lightning strokes: 40, 50, 60, 70s with magnitude of 150kA. As shown in this figure, it is noticed that the shorter tail wave time decreases the induced voltage. As the tail wave time increases the induced overvoltage increases and in turn the time to clear the overvoltage decreases.
effect on the severity of the overvoltages. In this study different configurations are chosen regarding different possible contingency conditions based on the single line diagram of Siah-bishe substation shown in Fig.7. Multiple lines in a station provide the benefit of decreasing the surge crest voltage and front steepness. However, these lines also collect more surges, and therefore an incoming surge with a larger steepness is required. Thus, it can be concluded
from this example that the case of all lines in service is the most critical case and dictates the required BILs. This may not always be true. It depends on the assumed probability of the contingency, which in turn produces the steepness of the incoming surge.
Fig.8 shows the results of simulation with different contingency conditions described in Table II. It can be seen that: The severest case for transformer is the case in which lightning surge enters into the circuit under the condition that a single line and a single transformer are connected (Config.-4)
Fig. 4. Voltage at the Line-GIS termination with different peak currents
The severest case for GIS at the line entrance is the case in which lightning surge enters into the circuit under the condition that a single line and all transformer feeders are connected. (Config.-1) The severest case for GIS main bus is the case in which lightning surge enters into the circuit under the conditions that a single line is connected. (Config.-1) The voltages at the transformer tend to increase slightly for multiline stations, and the voltages at other locations tend to decrease. Generally, the all-lines-in-service condition requires higher BILs than for contingency conditions [8].
Fig. 5. Voltage at Line-GIS termination with different front times
Fig. 7. Single line diagram of Siah-bishe 400kV GIS substation
Fig. 6. Voltage at Line-GIS termination with different tail times
TABLE II. ANALYZED CONTINGENCY CONDITIONS (O: OPERATING BAY, N: NON-OPERATING BAY, L: LINE FEEDER, T: TRANS. FEEDER, C: BUS
COUPLER
(BFR) and the distance between the station and the stroketerminating point [1]. Since back-flashover rate as a statistical
parameter is derived from mean time between failure (MTBF) of the equipment and MTBF depends on the probability of line outages, different contingency conditions have considerable
B. Effect of contingency conditions The number of surges that arrive at the substation is a function of magnitude of strike current, back-flashover rate
Arr. Config.1 Config.2 Config.3 Config.4
=C01 (L)
=C02 (L)
=C03 (T)
=C04 (T)
=C05 (C)
=C06 (T)
=C07 (T)
=C08 (L)
O O O O
N O O N
O O O O
O O O N
O O O O
O O O N
O O O N
N N O N
sheath of the cable against undesirable overvoltages. In Bothends methods, cable sheaths are grounded at both ends of cable spans. Fig.11 compares the results in two sheath bonding methods. It is noticed that Both-ends method provides the safer situation regarding the lightning overvoltages appeared on measuring points. On the contrary, Single Point method causes more severe lightning overvoltages.
Fig. 8. Voltage at some equipment of line feeder (=C01) in different contingency conditions
C. Effect of overhead line parameters 1) Effect of tower model: The surge impedance of each part of tower including legs and cross arms is expressed by the functions of their dimension and geometry [2]. Fig.9 compares the induced over-voltage occurred at different locations in 400kV substation between two different single-circuit and double-circuit tower models. It is understood that the coupling between circuits of double circuit towers feeding Siah-bisheh substation contributed significantly in the sense of reducing overvoltages due to lightning at the entrance of the station. The effect of the coupling capacitor is a tendency to reduce the rise time of the incoming wave [12]. 2) Effect of tower footing impedance: The peak overvoltage occurring at the top of the tower is partly determined by the apparent tower footing resistance. BFR is reduced by decreasing the footing resistance [12]. Lightning design of the line usually begins with the specification of the desired or design BFR. To reduce the severity of the surge, the obvious method is to decrease the BFR, that is, increase the line insulation level, or more practically, decrease the footing resistance. This decrease of footing resistance only needs to be made for towers adjacent to the station. Maximum surge impedance and the frequency-dependent effect of a tower are heavily dependent on the modeling of a tower footing impedance. Frequency dependent behavior of wave propagation along the tower is negligible unless the footing resistance is represented by resistive model [7]. According to standard recommendations, the footing impedance can be resistive and reactive depending on the soil resistivity and permittivity i.e. the soil temperature and moisture which show seasonal variation and weather dependence [7]. In base case of this study, the footing resistance was set to 10; however the derived results from its application may be more conservative rather than reactive models. Fig.10 shows that the voltage appeared at air bushing, increases with footing resistance. D. Effect of cable shielding Three different shielding practices were modeled in this study. In Single Point method, cable sheaths are solidly grounded at only one end while Sheath Voltage Limiter (SVL) is installed at another side of the cable to avoid damaging
Fig. 9. Voltage at different locations of feeder (=C01) with different tower models
Fig. 10. Voltage at GIS-Line termination with different footing resistances
Fig. 11. Voltage at Cable-GIS termination with different shielding methods
E. Effect of Location of Surge arrester Lightning arresters of the system were designed to be installed at GIS-line termination for all the line feeders and also at GIS-cable termination for transformer feeders. Meanwhile, we were requested to evaluate the necessity of installing additional arresters adjacent to the transformers. Consequently, it is found that additional arresters are not required for optimal insulation design. Generally, the number and location of arresters are determined by the balance between performance
and economical restrictions, although different case studies defined in Table III are examined regarding different locations of surge arrester in equipment arrangement. The likelihood of a lightning strike at the last tower is very low. Having a back-flashover at the last tower is even lower and the arresters majority of lightning strikes. Therefore, it is recommended to install arresters at the line entrance as well as transformer terminal. Fig.12 shows the voltage appeared at several locations comparing different locations of surge arrester. It is noticed that: The most effective case is Case-I with arresters installed at line side, cable termination and also adjacent to power transformer The least effective case is Case-II with arresters installed at cable termination only
TABLE III. Location of arrester Case I Case II Case III Case IV Case V Case VI DEFINED CASE STUDIES
At GIS-Cable termination At Power Transformer
Double exponential wave form results in greater induced voltage than other wave forms, while concave wave form results in far smaller and more realistic voltages. The shorter front wave times result in higher induced voltage, while the shorter tail wave times results in lower induced voltage. The case of all lines in service is the most critical case and dictates the required BILs. Coupling between circuits of double circuit lines has a remarkable effect on reducing lightning over-voltages The longer grounding conductors, for the purpose of grounding the towers, result in higher induced voltage. Among all the parameters analyzed, it has been found that contingency conditions and front time have the major effects on BIL of high voltage equipment. REFERENCES
[1] [2] [3] [4] A.R Hileman. "Insulation Coordination for Power Systems. Marcel Dekker. 1999. IEC TR 60071-4. "Insulation Co-ordination Part 4: Computational guide to insulation co-ordination and modeling of electrical networks". 2004. IEEE Std 1313.2-1999. "IEEE Guide for the Application of Insulation Coordination" IEEE Working Group on Surge Arrester Modeling. "Modeling of metal oxide surge arresters". IEEE Trans. on Power Delivery. Vol. 7 No 1, pp. 302-309 January 1992. IEEE Std 1243-1997. "IEEE Guide for improving the lightning performance of transmission lines", 1997. Ali F. Imece, Daniel W. Durbak, Hamid Elahi, Sharma Kolluri, Andre Lux, Doug Mader, Thomas E. McDermott, Atef Morched, Abdul M. Mousa, Ramasamy Natarajan, Luis Rugeles, and Eva Tarasiewicz "Modeling Guideline For Fast Front Transients", Report Prepared by the Fast Front Transients Task Force of the IEEE Modeling and Analysis of System Transients Working Group. A. Ametani and T. Kawamura, "A Method of a Lightning Surge Analysis Recommended in Japan Using EMTP", IEEE Transaction on Power Delivery, Vol.20, No. 2, April 2005 T.Kawamur, Y.Ichihara, Y.Takagi, M.Fuji, T.Suzuki, "Pursuing Reduced Insulation Coordiantion For GIS Substation by Application of high performance metal oxide surge arrester" CIGRE, International Conference on Large High Voltage Electric Systems, 1988 Juan A. Martinez-Velasco , Ferley Castro-Aranda, " Parametric Analysis of the Lightning Performance of Overhead Transmission Lines Using an Electromagnetic Transients Program", International Conference on Power Systems Transients IPST 2003 in New Orleans, USA Juan A. Martinez-Velasco , Ferley Castro-Aranda, "modeling of overhead transmission line for lightning overvoltage calculations" Ingeniare. Revista chilena de ingenieria, vol. 18 No 1, 2010 P.C.v. Esmeraldo, F.M. Salgado, Carvalho, "surge propagation analysis: an application to the GRAJAU 500 kV SF6 Gas Insulated Substation", CIGRE, International Conference on Large High Voltage Electric Systems, 1988 Transmission Line Reference Book, 345kV and above", second edition, by General Electric Co. Projec
At Line-GIS termination
[5] [6]
[7]
[8] Fig. 12. Voltage at some equipment in different case studies
IV.
CONCLUSION
[9]
This paper has described a parametric analysis of lightning overvoltages caused by back-flashover in a power plant 400kV gas insulated substation. For this purpose, all components of the system under study were simulated by utilizing EMTP-RV software. Also such a practical application of the study to the insulation design of such substation with a distinctive arrangement of equipment has been rarely done before. Parameters of the study include the surge model, contingency conditions, transmission line parameters, cable characteristics and location of surge arresters. Findings of this paper are as follows:
[10] [11]
[12]
[13]
Fuzzy-Based Control of On-Load Tap Changers under High Penetration of Distributed Generators
Maher A. Azzouz, Hany E. Farag and E. F. El-Saadany
Electrical & Computer Engineering University of Waterloo Waterloo, Canada mazzouz@uwaterloo.ca, ehab@uwaterloo.ca
Abstract Voltage regulation in the distribution networks becomes more challenging when the distribution generation (DG) units are introduced. The reasons behind that are: the reverse power flow which is introduced by the DG units, in addition to the probabilistic nature of the generated power due to renewablebased DG units. In this paper, a fuzzy algorithm is proposed to provide an adaptive reference of the OLTC controller, such that the OLTC, which feeds a multiple feeders, can mitigate the effect of the high penetration of the DG units. The motivations behind using the fuzzy logic are: 1) it can map nonlinear relations behind its inputs and output; and 2) it can provide a smooth transition, which lead to a more relaxed tap operation compared to the conventional logic control. The proposed fuzzy algorithm is tested using a radial structure distribution network which is modeled using SIMULINK/MATLAB.
Electrical and Computer Engineering York University Toronto, Canada hany.farag@lassonde.yorku.ca
conventional OLTC, which assumes a descending voltage trend from the substation to the feeders terminals, cant regulate the voltage based on local measurements. A fuzzy based control of the OLTC is presented in [1]. The fuzzy controller applies Mamdanis fuzzy inference system; hence the controller is built based on the experience. The controller has designed taking into account a variable time delay to limit the tap operation in case of transient load changes. However, the DG effects havens been considered in this proposed method. In [2], the differences between conventional and electronic tap-changers are presented. The authors of [2] suggested modifications which can be applied on the conventional OLTC controllers, so that the controllers can fit the electronic tap changers. The conventional OLTCs have a discrete nature control which means that the taps are changed by steps. While the electronic tap changers approximately have continuous control behavior. Moreover, the conventional discrete control of OLTC is illustrated in [3]. Although, the discrete control emulates the real behavior of the OLTC, it sophisticates the analysis of DEA models. Hence, the continuous control of the OLTC is preferred for stability studies [4]-[5]. In [6], a hybrid control model of the OLTC is introduced. The proposed method is a hybrid combination of the discrete and continuous control models of the OLTC. The author introduces a static voltage error which avoids singular Jacobian matrices in case of multiple OLTC regulating a certain bus voltage. An adaptive fuzzy reference is suggested in [7] to allow DGs operate at difference power factor, such that they can provide a decentralized voltage control. The problem of this method is that it does not take into account the coordination between the OLTC and the DG, and reactive power limits of the DGs. In additional to that, the reactive power support always depends on the active power; hence the DG cannot support the network while the generated active power is minimal. In [8], the authors proposed two local controllers to regulate the voltage profiles at buses where wind power distributed generators are connected. The first method relies on the sensitivity analysis, and the other suggests a fuzzy based reactive power compensation to support the DG bus voltage. These proposed methods do not take into account the interference between the DG and the OLTC. A coordinated
I.
INTRODUCTION
The main function of the on-load tap changer (OLTC) is to regulate the voltage of the secondary side of the distribution transformer so that the voltage of the regulation bus does not exceed the limits [1]. This regulation bus could be far away from the distribution transformer. Usually this bus is considered the critical bus from the voltage regulation point of view. The critical bus means that a bus which is expected to violate the voltage limit. Commonly, it is located at the center of loads or at the end of the distribution feeder. Nowadays, the global trend is to use renewable energy sources which are considered the optimal solution for environmental pollution, global warming, and high cost of the fossil fuels. The intermittent nature of the renewable energy sources such as wind and solar energy introduces fast and significant deviations of the distribution network voltage. Actually, the distribution generators (DGs) results in an active distribution network. So the distribution network does not have a uni-directional power flow, as in the passive networks, but it has a reverse power flow. The reverse power flow has a considerable effect on the voltage regulation at the distribution level, because R/X is not as small as in the transmission networks. Hence, the voltage regulation problem in the distribution network becomes harder when the DGs are considered. The intermittency and the reverse power flow can result in excessive tap operation of the OLTC. The problem will be worse if the OLTC feeds multiple feeders, because some feeders may suffer from overvoltage due to high DG penetration, while others may suffer from undervoltage during high loading. In this scenario, the
control between distributed energy storage systems (ESS) and OLTC is proposed in [9]. The proposed method relies on minimizing the reverse power flow during the lightly loaded network by activating the charging controllers of ESS. This method assumes that there is an ESS attached to every DG, however it is not the common case. Also, it does not take into consideration the case when the ESS is fully charged. In this paper, an adaptive fuzzy reference for OLTC controller is proposed such that OLTCs, which feed multiple feeders, can mitigate the effect of the DGs. The fuzzy algorithm relies on the systems maximum and minimum voltage to determine the incremental change in the OLTC reference voltage. The rest of this paper is organized as follow. In Section II, the basics of the conventional OLTC control are discussed. Section III illustrates the effect of DGs on the voltage regulation. The adaptive fuzzy reference algorithm is explained in Section IV. The simulation results are provided in Section V. Finally, the conclusion is given in Section V. II. CONVENTIONAL CONTROL OF OLTC
Fig. 1 The OLTC on the primary side
step, and b is the control signal that is applied to the tapchanging mechanism, see Fig. . It is expressed as follow: 0 b=1 1 for t Td for t > Td for t > Td : e = arbitrary e =1 : . e = 1 : (5)
The schematic diagram of the mechanical OLTC is shown in Error! Reference source not found.. The process of tap adjustment is considered a very slow process which takes 3 to 20 sec for each single tap step up or down. The OLTC can vary the taps from zero (no voltage compensation) to Nmax (maximum voltage compensation). The OLTC has a reverse switch which can change the polarity of the tap winding for positive and negative voltage compensation. The system model can be represented in the steady state as follow [1]:
where Td is the time delay introduced by the OLTC controller, and e is the output of the hysteresis controller. It is defined as
0 e=1 1 for for V DB V > DB . (6)
for V < DB
V2 =
V1 I 2 ZT ( a ) a
(1)
The mechanical time delay (Tm) has a constant value, usually it varies from 3 to 10 seconds. In some models of the OLTC [10][11], the controller time delay (Td) is considered constant: Td = 0 (7)
where V1, V2, I2, and a are primary voltage, secondary voltage, secondary current and the transformer tap ratio, respectively. ZT(a) is the transformer series impedance referred to the secondary side. The taps are located on the primary side (high voltage side), and changed linearly; hence it can be expressed as follow:
However, Td is commonly considered as a variable time delay which depends on the voltage error and the dead band [2]:
Td = 0 DB V
(8)
a = a0 + ni a
(2)
where DB is the dead band which is used to avoid hunting. For stability of the controller DB must to be greater than a [1], [2]. It is obvious from (8) that the controller time delay is inversely proportional to the voltage deviation, so when the voltage is close to the reference voltage, Td increases. III. DG EFFECT ON VOLTAGE REGULATION The Traditional distribution networks are passive network which means that the power is transferred from the substation to the customers, i.e. uni-directional power flow. However, in the presence of DG the distribution network is subjected to bidirectional power flow. Fig. (a) represents a simplified distribution network where a DG is connected at the load bus.
where a0 is the rated turn ratio (usually equals 1.0 p.u.), a is the step change of a and ni is the tap position, which can be expressed as follow:
ni = ni 1 + n
(3)
where ni is the present tap position, ni-1 is the previous tap position, and n is the change in the tap position which is defined as follow:
0 n = 1 1
for t Tm : b = arbitrary for t > Tm : b = 1 , for t > Tm : b =1
(4)
where Tm represents mechanical time delay which is required by the motor driver unit to change the tap position by only one
Fig. 2 The schematic diagram of the conventional discrete control of the OLTC
(a)
VG
' VG
To mitigate this problem, there are only two solutions, utilizing the DG unit: (1) reactive power control and (2) curtailment of active power. Off course, it is not desirable to curtail active power because it represents economic waste. On the other hand, the reactive power control of DG is limited by the power rating of the DG. In this paper, the DG is assumed to be either wind- or solar-based, because they have probabilistic generated power; hence the voltage estimation based on local measurements becomes worse. A better control of the OLTC can solve the voltage violation problem and avoid the DG active power curtailment, which in turns allows higher DG penetration. IV. PROPOSED ADAPTIVE FUZZY REFERENCE OF OLTC In this section an adaptive fuzzy reference of the OLTC is proposed to mitigate the effect of the active power injection by the DGs. From (11), it is noted that new installed DG introduces a voltage rise problems in the distribution network. The OLTC control based on local measurement cannot solve this problem. Also the voltage estimation fails under the highly intermittent nature of the wind and solar power. The only way to solve this problem is to introduce remote voltage measurements to the OLTC controller. An adaptive fuzzy reference is proposed which takes into account the remote measurements of the critical buses. The critical buses are the buses which have high probability of voltage violation. Such buses are either at the end of the distribution feeders or the DG buses. The block diagram of the proposed fuzzy reference is shown in Fig. . The remote voltage measurements are collected and the minimum and the maximum voltage are extracted. The adaptive fuzzy reference algorithm utilizes the systems minimum and the maximum voltage in order to adapt the reference voltage of the OLTC control loop. The reference is adapted online such that the minimum and maximum voltages lie within the acceptable range. Whenever the minimum and the maximum voltage are within the safe limits, other busvoltages will be within the safe limit. The membership functions which assigned for Vmin and Vmax are shown in Fig. . The rule base of the adaptive fuzzy reference is summarized in Table II. Actually, this adaptive fuzzy reference imitates the behavior of the power system operator (PSO), i.e., it increases/decreases the voltage at the substation in order to keep the bus voltages within the acceptable limits; hence this fuzzy reference is designed based on the experience.
(b) Fig. 3 Simplified distribution network with DG; (a) single line diagram, (b) the phasor diagram
The phasor diagram of the simplified distribution network is shown in Fig. (b), where: R, X are the feeder resistance and reactance, respectively, IR is the feeder current, V1, V2 are the primary and secondary voltage of the distribution transformer, respectively, VG is the DG voltage, is the power angle, and is phase shift between VG and IR. Using the phasor diagram, the DG voltage can be calculated as follow:
VG' = [V2 + I R R cos( ) + I R X sin( )] = VG cos( )
(9)
The power angle () is very small; hence (9) can be approximated as

VG V2 + I R R cos( ) + I R X sin( ) V = = I R R cos( ) + I R X sin( ) P R R + QR X VG
(10)
(11)
( P G PL ) R + (QG QL ) X VG
where PG and QG are the DG active and reactive power, respectively, while PL and QL are the load active and reactive power, respectively. It is obvious from (11), the most server case happens when the DG generates its maximum power during lightly loaded network. In this case a voltage rise happens due to the excessive reverse power flow.
dVref
Fig. 4 The proposed adaptive fuzzy reference of OLTC
...
-1
-2 1
1 (a) MFs of Vmin 0.8 0.6 0.4 0.2 0 0.9 0.91 0.92 0.93 0.94 0.95 Vmin 0.96 0.97 0.98 0.99 1 V.Low Low Normal
1.05 V max (pu)
1.1
0.98
0.96
0.94 V min (pu)
0.92
0.9
Fig. 6 The surface of the adaptive fuzzy reference
1 (b) MFs of Vmax 0.8 0.6 0.4 0.2 0 1 1.01
Normal
High
V.High
IF Vmin is normal and Vmax is high then decrease the reference voltage. IF Vmin is low and Vmax is normal then increase the reference voltage. IF Vmin is low and Vmax is high then hold. V. SIMULATION RESULTS
1.02
1.03
1.04
1.05 Vmax
1.06
1.07
1.08
1.09
1.1
Fig. 5 The input membership functions; (a) the membership functions of Vmin, (b) the membership functions of Vmax Table II The rule base of the adaptive fuzzy reference Vmax Vmin Normal High V. High V. Low ++ + 0 Low + 0 Normal 0 --
Each input, Vmin and Vmax, has been assigned three membership functions, namely, V. Low, Low, and Normal for Vmin and Normal, High, V. High for Vmax, where letter V stands for very. The consequent part, output, has assigned five membership functions, namely, ++, +, 0, -, and --, where ++ and -- stand for highly positive and highly negative, respectively. The singleton values of the output membership functions are 2, 1, 0, -1, and -2 for ++, +, 0, -, and --, respectively. The output of the proposed fuzzy algorithm is calculated using the weighted average defuzzification method. The proposed rule base is better clarified using the surface of the adaptive fuzzy reference, see Fig. . The shaded rule in Table II is read as fellow: IF the Vmin is V. Low AND Vmax is Normal THEN dVref = ++ Generally, the rule base of the adaptive fuzzy reference can be summarized as follow:
In this section, the proposed adaptive fuzzy reference is examined. The simulation results are performed using SIMULINK/MATLAB. To test the proposed fuzzy algorithm, the distribution system, shown in Fig. , is used [7]. This distribution network consists of two radial feeders, namely, Feeder I and Feeder II. Feeder I has a high DG penetration; hence this feeder has a higher probability of voltage swell due to reverse power flow. However Feeder II has a higher load power without any DG; hence it is more probable to have undervoltage. This multi-feeders configuration complicates the decision made by the OLTC; hence a control loop with a fixed target point is invalid. The critical voltages are measured and the minimum and the maximum voltages are calculated. For this distribution system, there are five critical voltages, i.e., V2, V3, V4, V5, and V9. It is worth noting that V9 can be easily estimated using local measurements, because there is no intermittent power installed at Feeder II. The adaptive fuzzy reference algorithm modifies the reference such that the OLTC can deal with the under/over voltages. In this study, the power curves of the loads are assumed to be residential [12], while the DGs power curves follow the typical wind and PV powers in [13], see Appendix. For simulation purpose, the typical 24 hours day is represented by 2400 seconds. Based on simulation studies, KF (the scaling factor of the adaptive fuzzy reference) is chosen to equal 0.01. Two scenarios are considered to examine the effectiveness of the proposed algorithm: 1) when DG units are wind-based; and 2) when the DG units are PV-based. Fig. shows the response of the OLTC with/without the adaptive fuzzy reference algorithm, in case of wind-based DG units. As shown in the figure, without adaptive fuzzy reference, the distribution system suffers from overvoltage and under voltage, i.e., the voltage violates the standard limits (0.95 ~1.05). The
reference algorithm is applied, the minimum and the maximum voltages lie within the accepted zone. That it is because the proposed algorithm continuously adapts the reference of the OLTC control loop to cope with the DG effect. Next, the proposed fuzzy algorithm is examined when the DG units are PV-based. As shown in Fig. 9a, without the adaptive fuzzy reference, the system suffers from overvoltage and undervoltage during the PV-peak power and max load, respectively. Conversely, the proposed fuzzy algorithm can solve the voltage violation problem, see Fig, 9b. VI. CONCLUSION
Fig. 7 Two-feeders distribution network
1.1 Vmax 1.05 V (p.u.) 1 0.95 0.9 Vmin
10
12 14 Time (hr)
16
18
20
22
24
(a)
1.1 Vmax
In this paper an adaptive fuzzy reference for the OLTC controller is proposed to provide a proper voltage regulation in case of high penetration of renewable-based DG units. The proposed fuzzy algorithm relies on the systems maximum and minimum voltages in order to calculate the incremental change of the OLTC reference voltage. The proposed OLTC control scheme has the following advantages over the conventional OLTC control: 1) it can mitigate the high DG effect on the voltage regulation; 2) it can deal with OLTC feeding multiple feeders; 3) it requires less remote date measurements compared to the centralized approaches. The simulation results show the effectiveness of the proposed algorithms to mitigate the voltage violation introduced by the high penetration of renewablebased DG units. APPENDIX
A. OLTC Parameters
1.05 V (p.u.) 1 0.95 Vmin 0.9 2 4 6 8
A. 1. Tap Changer Parameters

10 12 14 Time (hr) 16 18 20 22 24
(b) Fig. 8 The minimum and maximum voltage response, Wind case; (a) without adaptive fuzzy reference, (b) with adaptive fuzzy reference
1.1 1.05 V (p.u.) 1 0.95 0.9 2 4 6 8 10 12 14 Time (hr) 16 18 20 22 24 Vmin Vmax
a Nmax DB
= = =
0.00625 pu 16 0.01 pu
A. 2. Transformer Parameters Rating = 30 MVA f = 60 Hz r1, r2 = 0.03669 pu x1, x2 = 0.02932 pu Primary winding is 69 kV, delta connected secondary winding is 11 kV, star connected
B. Feeders Parameters
(a)
1.1 Vmax 1.05 V (p.u.) 1 Vmin 0.95 0.9
R1 = 0.05 pu R2 = 0.13 pu R3 = 0.26 pu R4 = 0.60 pu C. Power Curve
X1 = 0.04 pu X2 = 0.10 pu X3 = 0.18 pu X4 = 0.40 pu
C. 1. Residential Daily Power Curve

6 8 10 12 14 Time (hr) 16 18 20 22 24
(b) Fig. 9 The minimum and maximum voltage response, PV case; (a) without adaptive fuzzy reference, (b) with adaptive fuzzy reference
overvoltage is introduced when the wind-based DG units have maximum power, while the undervoltage occurs during the maximum load. However when the proposed adaptive fuzzy
REFERENCES
1
[1]
R e s id e n tia lP o w e r( p .u .) 0.8
0.6
[2]
0.4
0.2
[3]
0 2 4 6 8 10 12 14 Time (hr) 16 18 20 22 24
[4]
C. 2. Daily Wind Power Curve

[5]
1
0.8 W in dP o w e r( p .u .)
[6]
0.6
0.4
[7]
0.2
[8]
0 0 2 4 6 8 10 12 14 Time (hr) 16 18 20 22 24
C. 3. Daily PV Power Curve

1
[9]
0.8 P VP o w e r( p .u .)
[10]
0.6
0.4
[11]
0.2
10
12 14 Time (hr)
16
18
20
22
24
[12]
[13]
B. Kasztenny, E. Rosolowski, J. Izykhin owski, M. M. Saha, and B. Hillstrom, Fuzzy logic controller for on-load transformer tap-changer, IEEE Trans. Power Del., vol. 13, no. 1, pp. 164170, Jan. 1998. J. Faiz and B. Siahkolah, Differences between conventinal and electronic tap-changers and modifications of controllers, IEEE Trans. Power Del., vol. 21, no. 3, pp. 13421349, Jul. 2006. M. S. Calovic, Modeling and analysis of under load tap changing transformer control systems, IEEE Trans. Power App. Syst., vol. PSA 103, pp. 19091915, 1984. H. Ohtsuki, A. Yokoyama, and Y. Sekine, Reverse action of on-load tap changer in association with voltage collapse, IEEE Trans. Power Syst., vol. 6, no. 1, pp. 300306, Jan. 1991. N. Yorino, M. Danyoshi, and M. Kitagawa, Interaction among multiple controls in tap changer under load transformers, IEEE Trans. Power Syst., vol. 12, no. 1, pp. 430436, Feb. 1997. Federico Milano, Hybrid Control Model of Under Load Tap Changers, IEEE Transactions on Power Delivery, vol. 26, no. 4, pp. 2837-2844, Oct. 2011. A.E. Kiprakis and A.R. Wallace, Maximising energy capture from distributed generators in weak network, IEE Proc. Gener. Transm. Distrib., vol. 151, no. 5, pp. 661-618, Septembr 2004. V. Calderaro, G. Conio, V. Galdi, and A. Piccolo, Reactive power Control for improving voltage profile: A comparison between two decentralized approaches, Electric Power System Reseach, vol. 83, pp. 247-254, 2012. Xiaohu Liu, Andreas Aichhorn, Liming Liu, and Hui Li, Coordinated Control of Distributed Energy Storage system With Tap Changer Transformers for Voltage Rise Mitigation Under High Photovoltaic Penetration, IEEE Trans. on Smart Grid, 2012. N. Yorino, M. Danyoshi, and M. Kitagawa, Interaction among multiple controls in tap changer under load transformers, IEEE Trans. Power Syst., vol. 12, no. 1, pp. 430436, Feb. 1997. V.Venkatasubramanian, H. Schttler, and J. Zaborsky, Analysis of the tap changer related voltage collapse phenomena for the large electric power system, IEEE Conf. Decision and Control, pp. 29202927, Dec. 1992. J. A. Jardini, Carlos Tahan, M. R. Gouvea, Se Un Ahn, and F. M. Figueiredo ,Daily Load Profiles for Residential, Commercial and Industrial Low Voltage Consumers, IEEE Transactions on Power Delivery, vol. 15, no. 1, pp. 375-380, Jan. 2000. Fabio Bignucolo , Roberto Caldon , Valter Prandoni, Radial MV networks voltage regulation with distribution management system coordinated controller, Electric Power Systems Research, vol. 78, pp. 634-645, 2008.
Insulator Contour Optimization Using Intelligent Systems & Soft Computational Methods
Aytu Font
Istanbul Technical University Electrical Engineering Department Istanbul, Turkey font@itu.edu.tr
Ozcan Kalenderli
Istanbul Technical University Electrical Engineering Department Istanbul, Turkey kalenderli@itu.edu.tr objective function as well. Also it is very time consuming method because it requires too much iteration [7]. Soft computational methods are easy to apply nonlinear systems for modeling. With the improvement of the computer technology, many algorithms related to soft computing were developed. Two of the important ones are artificial neural networks (ANN) and adaptive neuro-fuzzy inference system (ANFIS). There are lots of studies in the literature about optimization method for insulator design [8-12]. In this study both ANN and ANFIS were used to optimize geometrical form of a 1 kV shuttle insulator. Top radius of the insulator was calculated according to desired input electrical field value obtained from finite element analysis. The obtained numerical results were compared with the real values. Besides, different configurations (number of neurons, various membership functions. etc.) of the optimization methods were also applied to take into account the system in order to minimize the computational error. II. ELECTRICAL PROPERTIES OF SHUTTLE INSULATOR AND ITS DESIGN PARAMETER
Abstract The purpose of this study is to give basic concepts of insulator contour optimization by using intelligent systems and soft computational methods. In this sense, artificial neural networks (ANN) and adaptive neuro-fuzzy inference system (ANFIS) are used as an optimization method. First training and validation data were collected from electric field distribution of an insulator by using finite element method analysis. Data contains coordinates and electrical field values of the middle point and the top point of the insulator. Then the data sets were used as inputs for ANN and ANFIS to train the algorithms. After the training process, any desired output that the top radius of the insulator was calculated for given input data using both methods. Finally results were compared from each other. The methods successfully are used for insulator contour optimization. Keywords-insulator contour optimization; ANN; ANFIS
I.
INTRODUCTION
Insulators are one of very important elements of the high voltage power systems. They insulate the live lines from other live parts or grounded parts in order to protect the human beings and system apparatus. Due to its reliability importance, insulators must be well designed in terms of the electrical and mechanical properties. Sometimes the proper design need to be optimized but it is not always easy to make because there are lots of insulator types and proper design parameters should be used for a specific insulator type. Therefore, there is no such a formula that can be used either for design or optimization. Optimized insulator contour should give approximately uniform stress distribution along the insulator surface, and keeping the electric field as low as possible. Obtaining uniform electric field distribution in any insulation is important for the reliability and life of electrical system. Otherwise, electric field is non-uniform and breakdown or partial discharge phenomena early become in the insulation. The most important design parameter for an insulator is its nominal voltage which is creating electrical stress on it. The voltage level gives general information about shape and size of the insulator. There are many numerical optimization methods widely used but they require objective functions. Sometimes it is hard to define such an objective functions because there are many design parameters and there is no specific formulas related to shape and size values for every single insulator [1-6]. Genetic algorithm is another well known method but it needs an
Figure 1 shows a typical 1 kV shuttle insulator. The insulator is made of resin, and is use as a post insulator under the busbar. Dimensions of the insulator are also shown in same figure. Some of the dimensions which are height of the insulator (h), middle radius (rm), screw socket height (hs) and radius (rs) were kept constant during the optimization. rT top radius of the insulator was chosen as the design parameter for the optimization problem.
rT
rM
hs rs
Figure 1. Shuttle insulator
In order to calculate top radius, two optimization parameters which are electrical field value at the middle point (EMIDDLE) and tangential electric field (ETOP). These values were calculated using finite element method analysis program. For this purpose, it is assumed that insulator was placed between two plane electrodes and 1 kV AC voltage was applied to the top electrode, and the bottom electrode was grounded. Such a case can easily be simulated. Figures 2 and 3 show 2D and 3D simulations done with different top radius values. ETOP and EMIDDLE were calculated for every 0.25 mm position increments starting from 9 mm to 20 mm. ETOP and EMIDDLE values at 45 different points were used as the input data for the optimization methods.
For the calculations, 30 data points that cover the whole workspace and reflect the all characteristic of the maximum value of tangential electric field distribution of the insulator were used to train the NN. 15 data were also chosen among the total of 45 data points for validation as well. On the other hand equal amount of data points for training and validation were chosen for tangential electric field value at mid-point (EMIDDLE) of the insulator.
radius
Figure 4. ANN configuration
Figure 2. Insulator 2D electrical field simulation
Figure 4 shows the general representation of feed forward NN architecture used for electrical field calculations. Gradient descent with adaptive learning rate (GDX) and LevenbergMarquardt (LM) back-propagation algorithms were used as learning algorithm. During the simulations, for both training algorithms, training data size, number of hidden layers and number of neurons in layers, activation function types of layers are changed and compared. Mean square error is used for performance comparison and its level is kept constant for this purpose. In addition to that validation data is not changed to provide a clear view for comparison. Network configuration related to system training sequence consists two hidden layer with different hidden neurons. 0.0000001 was chosen as the error goal for the network in order to high accuracy. IV. ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM (ANFIS) ANFIS is a widely used modelling and optimization technique and it contains the best properties of neural networks and fuzzy logic. For a given data and initial membership functions, ANFIS deduces trained membership functions by training the input data [15]. Data from ANN section were used to train and validate ANFIS. Input data includes tangential electrical field along the outer surface and electrical field at the middle point of the electrode, so ANFIS become two input and single output system shown in figure 5.
Figure 3. Insulator 3D electrical field simulation
III.
ARTIFICIAL NEURAL NETWORKS

Figure 5. ANFIS structure
The electrical field calculations for unconventional geometric insulators contain nonlinearity. Hence it is very difficult to solve these kinds of problems with numerical approaches. ANN is an alternative approach for optimization. Also it is very reliably and adaptive [13-14].
In Figure 6 a, 3D representation of training and validation data can be seen. Several simulations were made for different membership functions. Besides, simulations were repeated with
various number of membership functions (mf) for each type of mf in order to see the effect of number of mfs.
training data
20 18 16 top radius 14 12 10 8 250 200 150 electrical filed in the middle point 100 230 250 240 tangential electrical filed 270 260
Figure 7. Control surface for a given rule base
V.
SIMULATION RESULTS
The ANFIS results for training data given for four triangular membership functions can be seen in figure 8.
ANFIS performance for training data
(a)
25
validation data
20 top radius
20 18 16 top radius 14 12 10 8 250 200 150 electrical filed in the middle point 100 230 250 240 tangential electrical filed 270 260
15 Training Data ANFIS Out
10
5 250 200 150 electrical filed in the middle point 100 230 250 240 tangential electrical filed 270 260
Figure 8. ANFIS performance for training data
(b) Figure 6. Input data for ANFIS a-training data b-validation data
The training data and the ANFIS output data fit very well, so the configuration is suitable for the system. Validation data results are also shown in figure 9. Both training and validation data points have better fitting rates rather than ANN results.
ANFIS performance for validation data
Two inputs that each of them contains four membership functions (mf) and single output FIS system has 16 fuzzy rules (output mf) given below in Table I.
TABLE I.
CALCULATED RULE BASE FOR TWO INPUTS AND SINGLE OUTPUT FIS SYSTEM.
20 18 16 top radius 14 12 10 8 250 200 150 electrical filed in the middle point 100 230 250 240 tangential electrical filed 270 260 Training Data ANFIS Out
Electrical field in the middle (Emiddle)
mf 1 mf 2 mf 3 mf 4
mf 1 Outmf1 Outmf5 Outmf9 Outmf13
The control surface of the system according to rules was also given in Figure 7. It can be seen that the sharpness of the surface near the input 2 (electrical field in the middle point) is dominant for the counter optimization, so we have to take into account this input basically.
Tangential Electrical Field (Etop)
Figure 9. ANFIS performance for validation data
The ANFIS simulations were made with various membership functions and tested different numbers of membership function to find the optimum configuration for the system. Five different membership functions such as triangular,
Gaussian, bell, sigmoid and trapezoidal used for input fuzzyfication. For all membership functions, simulations were repeated five different numbers of membership functions from 2 to 6, respectively. After each simulation mean square errors (MSE) were calculated for both training and validation data. The error rates of the simulations are shown in Table II.
TABLE II. # of MF. 2 3 4 5 6 2 3 4 5 6 2 3 4 5 6 2 3 4 5 6 2 3 4 5 6 ERROR RATES OF THE ANFIS SIMULATIONS WITH DIFFERENT MFS AND DIFFERENT NUMBER OF MFS. Training error 8.7502e-07 8.9334e-08 2.1956e-08 3.1797e-09 4.5085e-11 1.0012e-05 7.2627e-07 3.9236e-08 1.7649e-08 1.5449e-11 3.1589e-05 1.0030e-06 8.8702e-08 1.5171e-09 1.0098e-09 2.9839e-05 6.2918e-06 2.0381e-07 1.3854e-08 7.6957e-10 5.2451e-05 1.2666e-05 7.0024e-06 5.1164e-06 1.0150e-06 Validation error 3.3569e-07 2.0362e-07 3.1540e-07 2.5081e-05 9.2114e-07 5.7825e-06 1.3997e-06 8.8440e-07 1.4149e-06 1.4036e-05 1.2564e-05 1.3233e-06 1.5504e-06 8.3073e-07 4.0456e-06 1.2312e-05 1.8375e-05 3.3033e-06 9.8433e-07 7.5130e-06 3.8418e-05 1.8817e-05 5.8467e-05 1.5638e-04 5.5310e-04
propagation is a fast algorithm, it needs much more memory in simulation

ANN performance for training data
20 18 16 top radius 14 12 10 8 250 200 270 260 150 electrical filed in the middle point 100 230 250 240 tangential electrical filed Training data NN output
MF
Triangular
Gaussian curve
a Generalized bell curve

20 18 ANN performance for validation data
top radius
Sigmoid curve
16 14 12 10 Validation data NN output
Trapezoidal
8 250 200 150 250 240 100 230 tangential electrical filed 270 260
According to Table II, it can be said that triangular membership function has minimum training and validation error. Also four membership functions for each input give best error rates seen from Table II as well. In all cases the increase in number of MFs in input layer provides better performance but after 4 in trapezoidal and generalized bell MFs and after 5, in triangular MF overfitting and memorizing occurs. In Figure 10, neural networks output of training and validation data are given for Levenberg-Marquardt learning algorithm. It can be said that for given data LM learning algorithm fits more accurately rather than GDX. This case easily be seen from validation data cross check. The LM algorithm is more reliable than GDX for this kind of low data point optimization problem. Table III shows the performance comparison of learning algorithms with respect to training and validation errors. The best results were obtained by using Levenberg-Marquardt back-propagation algorithm. Also mean square errors of both data sets state that, choosing gradient descent with adaptive learning rate as learning algorithms yields bigger error. Also LM converges with less epoch number that means faster than GDX. On the other hand, it must be taken care that, although Levenberg-Marquardt back-
electrical filed in the middle point
b Figure 10. ANN performance for training data (a) and validation data (b) TABLE III. Layers [2 [2 [2 [2 [2 [3 [3 [3 [3 [3 [4 [4 [4 [4 [4 1] 2] 3] 4] 5] 1] 2] 3] 4] 5] 1] 2] 3] 4] 5] ERROR RATES OF THE ANN SIMULATIONS FOR DIFFERENT LEARNING ALLGORITHMS. Training error LM GDX 6.5619e-06 0.0176 2.1464e-07 0.0032 1.0621e-06 0.0680 6.4434e-06 0.0372 1.5654e-07 0.0667 4.2815e-07 0.0756 1.7639e-06 0.1845 3.4919e-07 0.0145 3.0668e-06 0.0261 5.3221e-06 0.0245 1.2917e-07 0.0326 9.0372e-07 0.0977 7.4101e-06 0.0301 3.8437e-05 0.0048 2.8586e-07 0.0093 Validation error LM GDX 3.2317e-06 0.0097 1.2546e-07 0.0014 2.6927e-07 0.0358 2.3877e-06 0.0142 8.0458e-08 0.0329 1.2614e-07 0.0394 1.0349e-06 0.0775 1.1136e-07 0.0070 8.5517e-07 0.0136 4.1304e-06 0.0162 6.7251e-08 0.0149 2.7771e-07 0.0329 2.8234e-06 0.0124 2.9607e-05 0.0014 2.7167e-07 0.0029
The top radius values of the insulator obtained with two optimization techniques (ANN & ANFIS) can be seen in Table IV. Real radius values were taken from FEM simulations and they were compared with the proposed methods above. The radius values for the calculations seem very similar to real ones.
TABLE IV. ERROR RATES OF THE SIMULATIONS WITH DIFFERENT MFS AND DIFFERENT NUMBER OF MFS. REAL VALUES 9.25 10.25 11.50 12.25 13.25 14.00 14.75 15.25 16.25 17 17.50 18 18.50 19 19.75 ANFIS OUTPUTS 9.2504 10.2500 11.4996 12.2482 13.2492 14.0011 14.7488 15.2502 16.2487 16.9994 17.5002 17.9998 18.4998 19.0002 19.7506 ANN OUTPUTS 9.2477 10.2470 11.5001 12.2491 13.2492 14.0008 14.7505 15.2496 16.2492 16.9994 17.5005 17.9997 18.4995 19.0003 19.7510
REFERENCES
[1] S. Bolat, . Kalenderli, "Determination of optimized insulator geometry using artificial neural network", Eleco'2004 Symposium on ElectricalElectronics and Computer Engineering, Bursa, pp. 243-247, December 8-12, 2004. [2] S. Bolat, . Kalenderli, "Insulator contour optimization by artificial neural network and fuzzy inference system", 11th National ElectricalElectronics-Computer Engineering Congress, Istanbul, pp. 117-120, September 22-25, 2005. [3] S. Bolat, . Kalenderli, "Insulator contour optimization by artificial neural network", MedPower 2004, 4th Mediterranean IEE Conference on Power Generation, Transmission, Distribution and Energy Conversion, Lemesos, Cyprus, November 14-17, 2004. [4] S. Bolat, . Kalenderli, "Insulator contour optimization using by fuzzy inference system", Eleco 2005 4th International Conference on Electrical and Electronics Engineering, Bursa, Turkey, pp. 255-258, December 7-11, 2005. [5] Z. Stih, "High Voltage Insulating System Design by Application of Electrode and Insulator Contour Optimization", IEEE Transactions on Electrical Insulation, Vol. EI-21, No. 4, pp. 579-584, 1986. [6] M. Abdel-Salam, E. K. Stanek, "Field Optimization of High Voltage Insulators", IEEE Transactions on Electrical Insulation, Vol. 22, No. 1, pp. 47-56, 1987. [7] W. S. Chen, H. T. Yang, H. Y. Huang, "Optimal Design of Support Insulators Using Hashing Integrated Genetic Algorithm and Optimized Charge Simulation Method", IEEE Transactions on Dielectrics and Electrical Insulation, Vol. 15, No. 2, pp. 426-433, April 2008. [8] S. Chakravorti, P. K. Mukherjee, "Application of Artificial Neural Networks for Optimization of Electrode Contour", IEEE Transactions on Dielectrics and Electrical Insulation, Vol. 1, No. 2, pp. 254- 263, 1994. [9] P. K. Mukherjee, "Optimization of HV Electrode Systems by Neural Network using a New Learning Method", IEEE Transaction on Dielectric and Electrical Insulation, Vol. 3, No. 6, pp. 737-742, 1996. [10] J. Liu, J. Sheng, "The Optimization of the High Voltage Axisymmetrical Electrode Contour", IEEE Transactions on Magnetics, Vol. 24, No. 1, pp. 39-42, 1998. [11] K. Bhattacharya, S. Chakravorti, P. K. Mukherjee, "Insulator Contour Optimization by a Neural Network, IEEE Transactions on Dielectrics and Electrical Insulation, Vol. 8, No. 2, pp. 157-161, 2001. [12] A. Chatterjee, A. Rakshit, P. K. Mukherjee, "A Self-Organizing Fuzzy Inference System for Electric Field Optimization of HV Electrode Systems", IEEE Transactions on Dielectrics and Electrical Insulation, Vol. 8, No. 6, pp. 995-1002, 2001. [13] M. T. Hagan, H. B. Demuth and O. D. Jesus, "An introduction to the use of neural networks in control systems", Int. J. Robust Nonlinear Control Vol. 12, pp. 959985, 2002. [14] K. Narendra, K. Parthasarathy, "Identification and Control of Dynamical Systems Using Neural Networks", IEEE Transactions on Neural Networks, Vol. 1, No. 1, March 1990. [15] J. R. Jang, "ANFIS: Adaptive-Network-Based Fuzzy Inference System", IEEE Transactions on Systems, Man, and Cybernetics, Vol. 23, No. 3, pp. 665-684, 1993.
VI.
CONCLUSION
In this study, two of optimization techniques ANN and ANFIS were applied to a contour optimization problem for a shuttle insulator and their performances were examined. For this purpose first of all, neural network toolbox of Matlab is used for simulation. At first gradient descent with adaptive learning rate back-propagation is used as training algorithm in neural network, but it didnt give satisfactory results in validation. For this reason Levenberg-Marquardt backpropagation is used next and successfully results gained in validation. Then, in order to establish a scientific link to the ANN part of the study, which was realized by a feed-forward back-propagation network, same training data and validation data were used in ANFIS model of the system. Different membership functions with different number of MF configurations were also used to show the system performance Levenberg-Marquardt back-propagation algorithm is highly faster and reliable than gradient descent with adaptive learning rate back-propagation algorithm. It must be expressed that performance level is very important and efficient for optimizing, small performance level which means small mean square error, provides both better training and validation. The study carries out that, an ANFIS structure with triangular MFs is more suitable for the problem Consequently, it has been seen that used algorithmic methods are appropriate to determine the insulator contour. Also these methods can be used to determine shed number, shed position, shed form, shed size of an insulator having sheds.
Research on Electric Energy Harvesting from High-voltage Transmission Line

Min Yao
College of Automation Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing, China ym_nuaa@163.com
AbstractIn order to solve the problem of the power supply for the equipment or function module installed upon high-voltage transmission line, an electric power acquisition method from high-voltage transmission line based on electric current sensor is presented. Firstly, magnetic permeability, hysteresis loss and other performance parameters of the materials of silicon steel and nanocrystal are compared. Then the magnetic material of nanocrystal is selected to be made as the iron core of the sensor and the number of windings of the electric current sensor is calculated. In order that the collected power can be converted to a stable DC voltage source, a signal processing circuit is designed, which includes rectifier circuit, signal shaping circuit, extra energy consuming circuit and voltage stabilizing circuit. The results of tests show that when the bus current of transmission line is in the range of 5.0A-40.0A, a stable DC voltage source can be supplied to the function module installed upon the transmission line. Keywords-high-voltage transmission line; nanocrystal; signal processing circuit; energy harvesting
Min Zhao
College of Automation Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing, China xymzhao@126.com
Figure 1. The design diagram of electric energy harvesting module.
II.
PRINCIPLE OF ENERGY HARVESTING FROM HIGHVOLTAGE TRANSMISSION LINE
An electric current mutual inductor is designed to collect current from the alternating magnetic field of high-voltage transmission line. Then the rectification circuit and the AC-DC converter circuit are designed to process the collected current and to supply the stable voltage power. However, with the change of the electric load, the bus current of transmission line changes in a wide range. So the energy harvesting model should have the low start current to ensure to work well when bus current is small and have the energy release function to ensure supply stable power when bus current is large. The current mutual inductor is the key part that collects electric energy from alternating magnetic field. It is composed of iron core and induction coil that both determine the properties of current mutual inductor. Based on the theory of electromagnetism, iron core materials with high permeability, low magnetic loss is helpful to improve energy transfer efficiency and reduce the starting current. So the selection of iron core will influence the start current of current mutual inductor. A current mutual inductor with low start current can ensure the power supply circuit work well even when bus current of transmission line is small. On the other hand the iron core material itself has the property of saturation. When the iron core is saturated, the magnetic flow will stay unchanged, that helps to suppress the power output of the current transformer under a large current. However, when the iron core has been saturated, if the current of the bus keeps increasing, the iron core would come into the state of deep saturation and keep produce heating. Then it cannot keep collecting the energy from the bus current steadily. Therefore, in order to solve this problem, the iron core material with higher saturate induction density is preferred.
I.
INTRODUCTION
The running state of the power transmission line directly determines the safety and efficiency of power supply for an area. The most effective monitoring method is to monitor signal from the transmission line directly[1]-[3]. But the problem of continuous power supply to the equipment or function module installed upon high-voltage transmission line is hard to resolve. The commonly used energy to such function module is solar energy[4]-[6]. Such method is easy to be influenced by climate impact especially when it is in a long rainy season or snowy. In this paper a new power supply method is presented which can acquire power from highvoltage line itself and never influenced by climate and then convert to stable direct voltage by voltage stabilizing circuit[7][9]. Electric energy harvesting module consists of several parts, which is showed in Fig.1. Firstly, electric current mutual inductor collects current which is induced by the changes of magnetic field from the high-voltage transmission line. Then the current is converted to AC voltage, rectified to DC voltage and stabilized to 5V DC voltage which is supplied to the monitoring module installed upon the transmission line.
The typical parameters of several kinds of iron core materials have been shown in table 1 below. It shows that the Silicon steel material has a high saturation magnetic induction, but its magnetic conductivity is low. Permalloy has a higher initial permeability, a lower coercivity and magnetic loss, a stable magnetic performance, but its saturate induction density is not high enough and the processing and heat treatment is rather complex. The heat treatment process of Nanocrystalline alloy is simple and it is a soft-magnetic material with ideal lowcost and high-performance. Although it has a low saturate induction density when compared with the Silicon steel, its high frequency magnetic loss is low and it has a better corrosion resistance and magnetic stability and its initial permeability is high. After comparing, Nanocrystalline alloy is the best choice for iron core material.
TABLE I. THE TYPICAL PARAMETERS FOR SOME IRON CORE MATERIALS Silicon steel 2.03 102 103 4104 7.65 0.95 0.3 Permalloy 0.75 5 8104 60104 8.75 0.9 0.15 Nanocrystalline alloy 1.24 5 10104 60104 7.25 0.7 0.03 Figure 2. The schematic circuit diagram of energy harvesting.
Core materials saturation magnetic induction Bs (T) initial permeability () Maximum permeability m Density ( g/cm Lamination factor Strip thick (mm)
In order to make full use of the ring-shaped magnetic flux around the bus and be easy to meet the demands of the actual installation, Nanocrystalline core of O specification is selected. The specification parameters are: the inner diameter d=40mm, the external diameter D=64mm, the breadth W=20mm. From the parameters we get to know the cross-sectional area is 2.4cm2. Because the lamination factor of Nanocrystalline alloy material is 0.7,which is showed in table 1, cross-sectional area of the core is set as S=2.4 0.7=1.68cm2 to remove the influence of laminated core film. III. DESIGN OF THE ENERGY HARVESTING CIRCUIT
The energy harvesting circuit consists of voltage stabilizing circuit, battery storage circuit and electric consume circuit. This circuit converts alternating voltage which induced by current transformer into direct voltage with Full-Bridge rectification circuit, changes the direct voltage into a stable 5V direct voltage as power-supply voltage for the loads with voltage transformation circuit. In order to prevent the sudden increase bus current exceeding the protective capability of control coil circuit and enhance the stability of power, an electric energy leakage circuit is designed to work as the voltage regulatory module. When rectified voltage exceeds a certain range, the optical coupler MOC3041 is quickly triggered to control two-way SCR BT131 to open electric consuming circuit. IV. EXPERIMENTS
On the basis of theoretical analysis, nanocrystal magnetic material is selected to produce ring core which is shown in figure 3(a). Then the number of turns of the coil is calculated and the current mutual inductor is made which is shown in figure3(b). Signal processing circuit is made and installed with the monitoring module which is shown in figure 4. Tests are carried out in the high-voltage test platform which is showed in figure 5.
The schematic circuit diagram of energy harvesting is shown in figure 2, I is the actual current of high-voltage line, n1 and n2 represent the turns of the coils which provide power and the coils that play important role in feedback control respectively. L is the filter inductance, which can filter burrs in signal effectively, improve the wave form of input voltage and stabilize current. C1 is voltage stabilizing capacitance. Using transient state to restrain diode TVS can avoid high instant voltage.
iron core
(b) induction coil
Figure 3. Current mutual inductor.
130.3 140.6 150.3 160.0 170.3 180.7 189.8 199.4 Figure 4. Electric energy harvesting circuit. 12.0 12.1 12.1 12.2 12.2 12.3 12.6 12.5 14.4 14.5 14.7 14.8 15.1 16.4 17.5 17.6
Figure 6. The curve of rectified output voltage with bus current.
V.
CONCLUSIONS
Experiments show that the current mutual inductor made by electromagnetic materials of Nanocrystal can obtain the energy from high-voltage transmission line. When the bus current of transmission line is in the range of 5.0-40.0A, stable DC voltage source can be supplied to the monitoring module.
Figure 5. Experiment platform.
Some relevant experimental data are given in table 2. The curve of output voltage with bus current is shown in figure 6.
TABLE II. Bus current(A) 0.4 10.2 13.3 19.8 30.0 39.8 50.3 60.7 70.4 80.3 90.1 100.8 110.2 120.2
OUTPUT VOLTAGE OF ELECTRIC ENERGY HARVESTING MODULE
ACKNOWLEDGMENT This research has been funded by the Aeronautical Science Foundation of China (Grant No. 2012ZC52042) and Nanjing University of Aeronautics and Astronautics Research Funding (Grant No. NS 2013030) REFERENCES
[1] Cui Haoyang; Fang Yong; Tang Zhong; Liu Jun; Ye Bo: A Fault Diagnosis Method for Power Transformer Using Bayesian Data Analysis, Computer Science and Information Technology (ICCSIT), 2010 3rd IEEE International Conference on Volume: 6, 2010 , Page(s):567 570 Wenqing Zhao, Yongli Zhu, power transformer fault diagnosis based on rough set theory and support vector machine, Proc. of Int. Conf. on Fuzzy Systems and Knowledge Discovery, pp389-393, 2007 H. Hu, S. Qian, J. Wang, and Z. Shi, Application of Information Fusion Technology in the Remote State On-line Monitoring and Fault Diagnosing System for Power Transformer in Proceedings of The Intern. Conf. on Electronic Meas. and Instr. - ICEMI 2007, pp. 550-555 G.Yi,Sun Guiling,L.Weixiang,and P.Yong,Wireless sensor node design based on solar energy supply,in Power Electronics and Intelligent Transportation System(PEITS),2009 2nd International Conference on,2009,Vol.3,pp.203-207. Ruan Yue, Tang Ying, A water quality monitoring system based on wireless sensor network and solar power supply, IEEE International Conference on Cyber technology in Automation, Control and Intelligent systems, pp.126-129, March 2011.
Input voltage(V) 1.9 7.0 8.6 10.7 11.1 11.4 11.5 11.6 11.7 11.8 11.8 11.9 11.9 12.0
Rectified voltage(V) 1.9 6.1 8.2 12.3 12.9 13.2 13.4 13.6 13.8 13.9 14.0 14.1 14.2 14.3 [5] [4] [3] [2]
[6] L Fraas, L.Minkin, J.Avery, H. X. Huang, J. Fraas, and P. Uppal, Portable concentrating solar power supplies, in Proc. 35th IEEE Photovoltaic Spec. Conf., Honolulu, HI, Jun. 2010, pp. 003025003029. M. Takahashi, K. Nishijima, M. Nagao, T. Sato and T. Nabeshima, Two Phase PFC Circuit to Keep Constant Output Voltage even under Momentary Power Outage8th International Conference on Power Electronics - ECCE Asia 2011 (ICPE'2011-ECCE Asia), May 30-June 3, 2011, The Shilla Jeju, Korea. V. Kinnares and P. Hothongkham, Circuit analysis and modeling of a phase-shifted pulsewidth modulation full-bridge-inverter-fed ozone generator with constant applied electrode voltage, IEEE Trans. Power Electron.,vol. 25, no. 7, pp. 17391752, Jul. 2010. I. Nissinen and J. Kostamovaara, A low voltage CMOS constant currentvoltage reference circuit, in Proc. ISCAS, Vancouver, BC, Canada, 2004, vol. I, pp. 381384.
[7]
[8]
[9]
Linear Quadratic Gaussian (LQG) Control of Wind Turbines

Abdulrahman Kalbat Electrical Engineering Department Columbia University in the City of New York New York, NY 10027 Email: ak3369@columbia.edu
AbstractWith the increasing deployment of wind energy technologies, innovative modern control theories, which utilize linear time invariant models, are being developed. The potential for the success of utilizing these theories is better with the availability of concepts such as the variable speed wind trubines. Such control theories are not only capable of increasing the energy capture efciency of the wind turbines, but also reducing the number of measurements needed which in turn reduces the operation and maintenance costs associated with the system. In this paper, the general formulation of the Linear Quadratic Gaussian Control (LQG) along with the design procedures and the assumptions needed will be stated. Then, a numerical model for the Controls Advacned Research Turbine (CART) that was created and eld tested by the National Renewable Energy Laboratory (NREL) will be used to simulate LQG regulator using MATLab script. At the end, the results obtained from the simulation will be presented and the limitations of the LQG controller will be discussed. Index TermsLinear Quadratic Gaussian, LQG, Kalman Filter, Estimation, Wind Turbine.
presented and the limitations of the LQG controller will be discussed. II. LQG G ENERAL F ORM Linear Quadratic Gaussian (LQG) design problem is rooted in optimal stochastic control theory and has many applications in the modern world which ranges from ight and missile navigation control systems, medical porocesses controllers and even nuclear power plants. It combines both concepts of Linear Quadratic Regulators (LQR) for full state feedback and Kalman Filters for state estimation. In this section, the state space equations dening the LQG regulator along with the statistical assumptions will be presented rst. Then, the design steps for nding the optimal state feedback gain (Kf ) and optimal state estimation gain (Kk ) will be listed. Finally, the state space equations of the augmented system which closes the loop will be presented. The state space equations of the open loop plant for a standard LQG problem is shown in 1. x (t) = Ax(t) + Bu(t) + Gw(t) y (t) = Cx(t) + v (t) (1)
I. I NTRODCUTION The future plans for smart grid deployment , which mainly depend on the concept of distributed generation, make wind turbines more attractive solution, especially in regions with the high average wind speed. The control methods of wind turbines started by PI controllers which imposed many limitations and operation and maintenance costs to the system. With the advancement of modern control theory, it is now possible to use linear state space models of the wind turbines to implement control systems which are more powerful than PI controllers. These modern control methods also reduced the maintenance and operation cost by the utilization of Kalman Filters for state estimation that reduced the number of the sensors necessary to implement full state feedback controllers. In section II of this paper, the general formulation of the Linear Quadratic Gaussian Control (LQG) along with the design procedures and the assumptions needed will be stated. In section III, National Renewable Energy Laboratorys (NREL) Controls Advacned Research Turbine (CART) features and capabilities will be presented. In section IV, a numerical model for CART that was created and eld tested by NREL will be used to simulate LQG regulator using MATLab script. In section V, the results obtained from the simulation will be
where x(t) is the state vector, u(t) is the control input vector, y (t) is the measured output vector, w(t) and v (t) are stochastic white noise processes assoctiated with the process and the measurement, respectively. The matrices A(state matrix), B(control input gain matrix), G(plant noise gain matrix), C(measured state matrix) are all Linear Time Invariant (LTI). As shown in 2, both w(t) and v (t) are assumed to be white gaussian noises with zero mean and the expected values of the initial values of the states x(0) are assumed to be equal to x o . From 3, w(t) and v (t) have covariance matrices of W and V, respectively, and the cross covariance R12 is assumed to be zero (w(t) and v (t) are uncorrelated) in order to simplify many expressions and derivations. Finally, the initial values of the states x(0) are assumed to be uncorrelated with both w(t) and v (t), as shown in 4. E [x(0)] = x o E [w(t)] = 0 E [v (t)] = 0 (2)
E [w(t)w( ) ] = E [v (t)v ( ) ] = E [w(t)v ( ) ] =

T T
W, if t = 0, if t = V, if t = 0, if t = R12 , if t = 0,
T
(3)
if t = (4)
E [x(0)w(t) ] = 0 E [x(0)v (t) ] = 0

T
engineering sense, or by using the method shown in 9 or by using Brysons Rule shown in 10. Brysons Rule related the reciprocal of the maximum squared values of the states with Qf and the reciprocal of the maximum squared values of the control inputs with Rf . After nding the optimal gains, the closed loop system could be created by augmenting the system with Kalman Filter resulting in the new model shown in 11. The augmented system model which include the dynamics of the system and the state estimation error e(t) [2]. Step 1: Optimal gain existance criteria: (A, B ) is Controllable (A, C ) is Observable Step 2: Optimal State Estimation Gain Calculation: )T (x x ) Jk = E (x x APk + Pk A + GW G Pk C V Kk = Pk C V
T 1 T T T 1
In real world control design problems, it is rarely possible to have access to all states of the system which are needed for full state feedback. Instead, access is only possible to specic measured outputs of the system. If these measurements carry enough information about the states of the system, then a state observer using Kalman Filter could be implemented to estimate all states of the system. This observer is capable of rejecting disturbances of the system by acting as a low pass lter. The main inputs to the observer are the control input (u(t)) and the system output (y (t)). The state space equations of the Kalman Filter are shown in 5. It should be noticed that it uses the same state space matrices (A, B and C) as the main system and the estimated states (x (t)) are used as the system states [1]. (t) = (A Kf C ) x x(t) + Bu(t) + Kf y (t) y (t) = C x (t)
(6)
(7a) CPk = 0 (7b) (7c)
Step 3: Optimal State Feedback Gain Calculation and Weighing Matrices Selection:
T
Jf = (5)
0
(z T Qf z + uT Rf u)dt
1 T Rf B Pf
(8a) (8b) (8c)
1 T AT Pf + Pf A Pf BRf B Pf + Qf = 0
From Fig.1, it should be noticed that LQG is formed by connecting the system and the Kalman Filter through the optimal state estimation gain (Kk ) and then creating full state feedback by using the estimated states (x (t)) which passed through the optimal feedback gain (Kf ). Because of the stochastic seperation principle, the previoulsy mentioned gain could be designed individually. The design process starts with checking controllability and observability of the pairs (A, B) and (A, C), respectively, as shown in 6. These criterias are necessary for the existance of the solutions for the equations used to nd the optimal gains. Then, the optimal state estimation gain (Kk ) is calculated as shown in 7c, where Pk is a positive semi-denite mtarix and the solution of the Filter Algebraic Riccati Equation (FARE) shown in 7b. This solution ensures a minimum value of the cost function shown in 7a. After that, the optimal state feedback gain (Kf ) is calculated as shown in 8c, where Pf is a positive semi-denite matrix and the solution of the Control Algebraic Riccati Equation (CARE) shown in 8b. This solution ensures a minimum value of the cost function shown in 8a. Unlike FARE, which requires only noise covariances W and V, CARE requires two weighing matrices: Qf (symmetric positive semi-denite) and Rf (symmetric positive denite). These weighing matrices provide a means to trade-off opposing objectives: state regulation and control usage. They could be selected based on trial and error approach with some
K=
Q = CT C R = I
(9)
1 M ax (x2 ii ) 1 Rii = M ax (u2 ii ) Qii =
(10)
Step 4: Linear Quadratic Gaussian Regulator by combining Optimal State Estimation and Optimal State Feedback: x (t) A BKf = e (t) 0 + G 0 G Kk BKf A Kk C w(t) v (t) w(t) v (t) x(t) e(t) (11)
y (t) = C
x(t) + 0 e(t)
w(t)
Noisy System
v (t) G u(t) B x (t) x(t) C

+ +
y (t)
implement the individual pitching technique. In the collective pitching technique each of the blades are picthed by the same amount and this controlls the symmetric loading on the rotor. But in the individual pitching technique, each blade is individually pitched and this offers a control over both the symmetric and asymmetric loading on the rotor [6]. Because of the access to all sensors data and actuators and the exible capabilities of CART, researchers at NWTC could implement innite combinations of control systems. IV. N UMERICAL E XAMPLE U SING MATLAB
A
+
Kk (t) x x (t) C
+
y (t) y (t) y (t)
A Kf
Determinisric Optimal Controller
Kalman-Bucy Filter
Fig. 1. LQG regulator block diagram
III. C ONTROLS A DVACNED R ESEARCH T URBINE (CART) The numerical example that will be discussed in the next section is based on a linear model of the Controls Advacned Research Turbine (CART) which is used by the National Wind Technology Center (NWTC) that is operated by the National Renewable Energy Laboratory (NREL) and located at Boulder, Colorado. Researchers at NWTC tackles wind industry engineering challenges ranging from atmospheric uid mechanics and aerodynamics, dynamics, structures, fatigue, power systems and electronics and wind turbine engineering applications [3]. CART is a modied Westinghouse WGT-600, 600-KW, 2-bladed, horizontal-axis research wind turbine capable of operating in constant speed or variable speed mode [4]. It is used by the researchers at NRELs NWTC as a test bed for exploring potential control innovations and eld test advanced control systems. These control designs could be implemented by writing an ANSI-C based program on a dedicated personal computer that is connected to the turbine sensors and actuators. The turbine is instrumneted with more than 80 sensors with measurements being recorded at a rate of 100 Hz. These sensors are used to record the dynamic state of the turbine and they fall in one of three categories: Performance (torque and power sensors), loads (strain gauges, accelerometers and position sensors) or meteorological (wind speed, direction, temperature and pressure) [5]. One of the unique capabilities of CART is that beside the conventional collective pitching technique, it is possible to
In [7], several Linear Time Invariant (LTI) state space models of CART were created which varied in their complexities between 1-state model to 9-state model. In this paper, the 3-state model was considered because of its simplicity and because this model is both controllable and observable (minimal) which are both conditions needed for designing LQG controllers. The numerical model shown below was created for CART by NWTC after linearizing the motion equations of the turbines considered states at a control design point of 18 m/s for wind speed, 12 degress for rotor collective pitch and 42 RPM for rotor speed. The main objective of the controller is to operate the machine as a variable speed wind turbine in region 3 by applying constant torque to the generator by maintaining a constant rotor speed through the collective rotor blade pitching. The control input u(t) of the considered model is the collective blade pitch angle and the states are the rotor speed x1 (t), drive train (shaft) torsional spring force x2 (t) and the generator speed x3 (t). The disturbances are the turbine system noise w(t) and the measurement noise v (t). For the sake of simplicity, both turbine and measurement noises were assumed to have the same magnitude as the wind but in the case of turbine system noise, it was scaled by the matrix G. The non-zero rst row of the control input gain B indicates that the control input (rotor collective pitch) affects the system by changing the rotor speed which affetcs the other states through the coupling caused by the state gain matrix A. Also, from matrix C, it should be noticed that the only measurement taken is the generator speed. The simualtion of LQG controller for CART was done using MATLab by following the design steps shown in section II and by using the numerical models matrices shown below. The design started by checking the controllability and the observability of the pairs (A, B) and (A, C), respectively. Then, the optimal state estimation gain (Kk ) was calculated using the noise covariances shown below. Sizing the optimal state feedback gain was found using a trial and error approach by changing the weighing matrices (Qf and Rf ). The best response of the generator speed at which it stabilized at steady state when perturbed by step changes in wind speed was found at the values of the weighing matrices shown below. If the maximum limits of all states and control inputs were available, which was not the case for CART, then Brysons rule could be used to initialize the weighting matrices as shown below [8]. After that, the closed loop system of the LQG regulator
was created by augmenting the turbine system with the state estimator. The inputs to the closed loop regulating system are turbine system noise w(t) and the measurement noise v (t) which were generated as a normally distributed random number by MATLab to represent the white gaussian noise. The peroformance of the closed loop LQG system will be discussed next. MATLab Simulation inputs for CART x1 (t) x1 (t) : is rotor speed x2 (t) : is drive train torsion x(t) = x2 (t) where x3 (t) x3 (t) : is generator speed w(t) u(t) = v (t) w(t) : is turbine system noise v (t) : is measurement noise 3.1078x106 0.0 1.5601x105 0.0 0.0
T
Wind Speed (m/sec) 20 15 0 10 20 30 40 50 60 70
Generator Speed (RPM) 50 40 30 20 0 10 20 30 40 50 60 70
Pitch Angle (degrees) 15 10 5 0 10 20 30 40 Time sec 50 60 70
Fig. 2. Wind turbines response to different wind speeds

()*+,+-)./01 %# %' &#
where
78095:0-;.3456
1.4454x101 2.6910x107 A= 0.0 B = 3.4559 C= 0 0 1 0.0
0.0 (12) 2.6910x107 0.0
&' # ' # &' &#
G = 7.8938x102
T T
0.0
T
%' %# !"# ! $"# $ %"# % 2+0*.3456 &"# & '"# '
W = E [w(t)w( ) ] = 0.1 (Turbine system noise covariance) V = E [v (t)v ( ) ] = 0.1 (Measurement noise covariance)
Fig. 3. Poles/Zeros plot of the closed loop system
Results 1 0 Q = 0 1x1013 0 0 Kk = 7.6282x10 Kf = 2.0336

3
0 0 1
R=1 (13)
2
1.2663x10
6.2859x10
2 T
2.1225x107
6.6055x101
V. R ESULTS AND D ISCUSSION Fig. 2 shows the generator speed and the pitch angle when the wind speed is varying between 14 m/s to 20 m/s. As the wind speed is increasing, the pitch angle is also increasing to keep the generator speed at a constnt value of 42 RPM. From the gure, it is clear that the generator is operating at 42 RPM only at 18 m/s and when the pitch angle is 12 degrees, the control design point. So, LQG regulator is not able to keep the generator speed at 42 RPM even when the wind speed is varying. This example clearly shows the robustness problem of LQG controllers. According to [9] , control system robustness is dened as the ability to maintain satisfactory stability or performance characteristics in the presence of all conceivable system parameter variations. In other words,a robust control system works not only for the linear system which serves as the plant model but it also works for the real physical system with minor performance degradation [10]. Another
reason might be models simplicity and uncertainty because of the neglected high frequecny dynamics of the turbine. In order to improve the robustness of the LQG regulator, one could go further and design a Loop Transfer Recovery (LTR) which recovers the robustness of Linear Quadratic Regulators (LQR) and Kalman Filters. Fig. 3 shows the pole/zero map of the closed loop system. Since all poles lies in the left half plane, then the closed loop system is stable. From Fig. 4, one should notice how the LQG regulator successfully reduced the effect of the system white gaussian noise w(t) (input 1) on the closed loop system. Another observation is that the LQG at high frequencies the effect of the measurement white gaussian noise v (t) is higher in the closed loop system than in open loop system. One explanation might be because the v (t) is being multiplied by the optimal state estimation gain Kk in the nal augmented model shown in 11. VI. C ONCLUSION In this paper, the general formulation of the Linear Quadratic Gaussian Control (LQG) along with the design procedures and the assumptions needed were stated. Then, a numerical model for the Controls Advacned Research Turbine (CART) was used to simulate LQG controller using MATLab script. Then, the peroformance of the LQG regulator was evaluated by plotting the time response of the system and the
67&89:)7&&;$%&'()*+,!"# $%&'()*+,2-
# .&()/01,!-
A?B+C10:9),:D-)E)FG?89),:9B-
"#
!##
!"# !4# 5# .&()/01,!# 5# !4# 23# !#

#
)
/;9+ =7&89:
)
$%9<09+=>)),%?:@89=-
!#
Fig. 4. bode plot of the open and closed loop systems
pole/zero map of the closed loop system. Finally, the bode plot of both the open loop and the closed loop systems were compared. R EFERENCES
[1] C.-H. Won and K. Gunaratne, Performance study of lqg, mcv, and risksensitive control methods for satellite structure control, in American Control Conference, 2002. Proceedings of the 2002, vol. 3, pp. 2481 2486 vol.3. [2] E. E. Ostertag, Optimal stochastic control, in Mono- and Multivariable Control and Estimation, ser. Mathematical Engineering. Springer Berlin Heidelberg, 2011, vol. 2, pp. 259266. [Online]. Available: http://dx.doi.org/10.1007/978-3-642-13734-1-5 [3] National wind technology center (nwtc). [Online]. Available: http://www.nrel.gov/wind/pdfs/53405.pdf [4] K. Johnson, L. Fingersh, and A. Wright, Controls advanced research turbine: Lessons learned during advanced controls testing, National Renewable Energy Laboratory, NREL/TP 500-38130, June 2005. [5] L. Fingersh and K. Johnson, Controls advanced research turbine (cart) commissioning and baseline data collection, National Renewable Energy Laboratory, NREL/TP 500-32879, October 2002. [6] K. A. Stol, W. Zhao, and A. D. Wright, Individual blade pitch control for the controls advanced research turbine (cart), Journal of solar energy engineering, vol. 128, no. 4, pp. 498 505, 2006. [7] A. D. Wright, Modern control design for exible wind turbines, National Renewable Energy Laboratory, NREL/TP 500-35816, July 2004. [8] . Oral, L. etin, and E. Uyar, A novel method on selection of q and r matrices in the theory of optimal control, International Journal of Systems Control, vol. 1, no. 2, pp. 8492, 2010. [9] R. Stengel and L. Ryan, Stochastic robustness of linear time-invariant control systems, Automatic Control, IEEE Transactions on, vol. 36, no. 1, pp. 8287, 1991. [10] U. Mackenroth, Control Systems: Basic Denitions and Concepts. Springer Berlin Heidelberg, 2004, pp. 45.
Residential Load Harmonics in Distribution Grid

J. Niitsoo, I. Palu, J. Kilter, P. Taklaja, T. Vaimann
Department of Electrical Power Engineering Tallinn University of Technology Tallinn, Estonia jaan.niitsoo@ttu.ee
AbstractPurpose of the present paper is to provide the results of the study performed in order to assess current distortion of domestic loads and corresponding voltage distortion in the residential distribution networks. For the study harmonic current amplitudes and phase angles of different home appliances were measured and afterwards the corresponding models were composed in DIgSILENT Power Factory. The focus of the study was to analyse the characteristics of the residential distribution network with different amounts of real nonlinear loads. The results of the study showed an extensive harmonic distortion in residential load current and considerable voltage distortion at the substations busbar. The results presented in this paper are especially important when dispersed generation (DG) and electrical vehicles (EV) are going to be connected in bulk into the distribution networks. Keywords-distribution grid, load modelling, power quality, power system harmonics, power system measurements.
Distorted voltage and current in the distribution system may bring along unwanted effects, e.g. overloading, over-voltages, mechanical stress, malfunction of critical control and protection equipment and degradation of efficiency of appliances. Distortion affects all customers fed through this point of common coupling (PCC). Share of nonlinear loads is growing rapidly worldwide. It has been estimated that for 2012, 60% of the loads on power system in USA were nonlinear loads [2]. Therefore, due to increase of this number the power quality research observing the corresponding challenges is becoming more vital than ever before. The lack of enforceable harmonic standards has not encouraged the use of expensive monitoring, but in the situation where energy markets are turning more open and amount of power electronic in the grid is growing, it is getting more important to keep an eye on the harmonics [5]. To estimate the harmonics it is very important to know the harmonic components and their magnitudes for designing the harmonic mitigation devices and other controllers. Power system is highly dynamic in nature, hence, network topology and parameters keep on changing rapidly [6]. For example, the adoption of compact fluorescent lamps (CFL) based on market trend could result in an increase of voltage total harmonic distortion by about 10% per year over the next couple of years [7]. However, even now the levels of distortions are significant. Monitoring of 100 distribution network feeders in USA revealed that the average voltage THD at PCC was 4.73% [8]. Results are seen at Fig. 1. It can be presumed that the stated percentage is quite common for many distribution grids. Even more severe cases have been revealed where distortion has been higher than 17% [9]. Nevertheless, it should be considered that the power supply standards (EN50160) have set the upper voltage THD limit for public networks at 8% [10]. Predicting the possible future harmonic distortion levels in details for the distribution networks is a very complex assignment. For example, many nonlinear loads often have very low power consumption and for very low power levels their quality requirements are not strictly regulated by standards. The effect of such nonlinear load may be hazardous to network operation if the amount of such devices is extensive. The total impact depends on the number of appliances, their power ratings, and characteristics of their current waveforms.
I.
INTRODUCTION
The appliances and electric devices connected to the public mains network are designed to operate with a sinusoidal voltage at rated power. However, many of the connected loads are of nonlinear type, meaning that they draw current with a distorted sine waveform. This also causes distorted sine voltage drop, thus resulting in distorted network voltage waveform. The overall distortion level is described by total harmonic distortion (THD). Some loads have the current with total harmonic distortion over 100%, but their active power consumption is not so significant compared to other harmonics generating apparatuses [1]. Therefore their quality requirements are not so precisely regulated, but harmonic load may increase highly when numerous harmonics emitting devices are utilized in bulk. The total impact depends on the number of appliances, their power ratings, and their harmonic diversity. Harmonic angle diversity is relevant when multiple appliances are operating simultaneously, creating either reinforcement or cancellation of harmonic magnitudes [2]. The attenuation effect is dependent only on the phase angle, but the effects severity is dependent on the magnitude of the harmonic voltage [3]. Nonlinear loads absorb distorted currents which owing through the impedances of the power distribution system and result in a distortion of system bus voltage [4]. Harmonics go upstream within the network and affect the voltage waveform, which may become very distorted, deviating far from a proper sinusoidal signal.
corresponding measured real and reactive power level values, harmonic currents magnitudes and harmonic current phase shift angle values of measured devices are presented in Table 1. Based on the measurements the corresponding load models, including the amplitudes and angles of harmonics up to 50th order, were composed. In order to correspond to the actual network configuration, similar real power consumption in each phase was arranged. III. METHOD The residential distribution network and the loads were modelled in DIgSILENT Power Factory. The model consisted three phase residential load at 0.4 kV line voltage level composed of single phase loads. The composed residential network and loads were connected to distribution network substation via 1 km long overhead line. Distribution network substation was connected to 10 kV network with short-circuit power of 100 MVA. Transformer used in the distribution substation was with parameters 16 kV, uk% = 4% and vector group Yyn. Scheme of constructed grid is shown in Fig. 3. For the purpose of assessing the effect of adding loads to corresponding harmonic emissions the current distortions at residential load 0.4 kV feeders and voltage distortions at distribution network substation low-voltage busbars were observed. The number and configuration of residential loads were changed for different scenarios. In case of increasing the number of nonlinear loads it is presumed to be worst case scenario, no coincidence factors were taken into account. Amount of active power were also added in the scheme to model the situation where not only nonlinear loads are connected. IV. RESULTS Modelled harmonic content and current curves of one residential load are depicted in Fig. 2 and Fig. 4, Figure 4respectively. High distortions for phase currents can be seen at odd orders and especially at 3rd and 5th order. More extensive are distortion levels in phase C. Due to remarkable distortion in three phases and unbalance between them, vast current in neutral conductor was produced. Especially high was the 3rd harmonic current level in neutral.
Figure 1. Histogram of the voltage THD of 134 (13.2 kV) feeders [8]
The purpose of the present study is to demonstrate and analyse possible power quality situations in the observable distribution network. This paper examines the impact of nonlinear domestic load in distribution grid. For the analysis, measured power consumption and current waveforms of different home appliances have been used. Main idea of this paper is in the use of actual measurements data of different real appliances for modelling the effects on the residential distribution network with different type of residential loads. II. DISTORTION ANALYSIS Distorted sine wave can be observed, dividing it into numerous components, each having an integer-multiple frequency of the mains frequency. Such harmonic components are used to describe the distortions in the electric networks. For different waveforms, there is a different harmonic content present, referring to individual harmonic magnitudes and phase shift compared to the mains frequency component. The measurements of loads presented later in this paper are indicated as magnitudes and phase shift of each individual harmonic. The distortions can be observed individually by comparing different harmonic components. More general approach to quantify the distortions is the use of total harmonic distortion level (THD). The total harmonic distortion can be observed as a current harmonic distortion THDI and/or voltage distortion THDU. Another option is to observe the harmonic distortion as a geometrical sum of positive, negative and zero sequence component distortions as total harmonic distortion THDtot. These harmonic distortion indicators can be calculated as in (1), (2) and (3), respectively. = = 1 2 2 1 100 1 1 2 2 1 100 1 (1) (2) (3)
2 2 2 = 1 + 2 + 0
where THD1 is the distortion caused by positive sequence, THD2 by negative sequence and THD0 by zero sequence. In order to model the network response of nonlinear loads, 14 different home appliances were measured. The results of the
Figure 2. Modelled harmonic currents (% of fundamental) on logarithmic scale in case of residential load
TABLE 1. POWER AND HARMONIC CURRENT VALUES OF MEASURED DEVICES device fridge 1 lamp set 2 lamp set 1 P [W] Q [VAr] P [W] Q [VAr] P [W] Q [VAr] fridge 2 TV 2 laptop 2 TV 1 laptop 1 printer desk lamp monitor air purifier TV-tuner 1 TV-tuner 2 P [W] Q [VAr] P [W] Q [VAr] P [W] Q [VAr] P [W] Q [VAr] P [W] Q [VAr] P [W] Q [VAr] P [W] Q [VAr] P [W] Q [VAr] P [W] Q [VAr] P [W] Q [VAr] P [W] Q [VAr] 119 77.9 91.4 -122.1 82.7 -106.6 80.7 19.3 80.3 68.5 69.9 -130.4 58.6 -66.6 47.8 -29.3 42.3 -31.6 30.7 67.4 29.9 -40.3 11.9 10.3 6.6 -10.1 3.9 -5.4 harmonic order [n] current [%] angle [] current [%] angle [] current [%] angle [] current [%] angle [] current [%] angle [] current [%] angle [] current [%] angle [] current [%] angle [] current [%] angle [] current [%] angle [] current [%] angle [] current [%] angle [] current [%] angle [] current [%] angle [] 1 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 3 14 314 78 358 83 358 16 246 70 339 90 346 83 342 42 301 17 290 6 305 85 318 29 110 81 294 82 347 5 6 42 52 343 59 355 7 224 31 310 76 339 60 328 5 148 13 243 2 104 64 292 22 118 65 252 70 327 7 4 139 27 11 35 8 7 157 13 236 62 338 33 313 7 57 10 198 1 298 41 272 5 57 49 215 54 249 9 1 324 13 52 18 37 6 156 10 176 53 341 12 281 2 120 6 147 0 136 21 269 6 66 31 181 37 50 11 2 341 11 105 11 88 2 161 4 117 49 344 6 149 2 308 3 100 0 322 17 294 4 95 20 164 23 29 13 1 168 11 130 9 124 2 206 3 14 45 344 11 115 2 247 1 208 0 143 20 292 2 48 16 151 13 75 15 1 264 14 156 9 201 0 294 2 291 39 344 10 90 2 102 2 183 0 276 17 277 2 51 15 128 11 106 17 0 315 15 196 13 253 1 224 2 194 32 346 5 51 1 187 2 141 0 171 12 263 1 50 14 103 13 121 19 0 92 14 222 15 262 0 145 2 134 27 351 3 305 1 262 2 95 0 206 7 270 1 84 11 76 13 135 21 0 243 10 231 15 265 0 194 1 44 23 329 5 261 1 140 2 47 0 197 6 305 1 197 8 62 11 154 23 0 277 7 235 11 268 0 195 1 342 20 133 5 233 1 98 1 272 0 131 8 304 1 254 6 50 9 184 25 0 311 7 245 8 279 0 207 1 250 17 38 3 185 1 128 1 244 0 192 7 291 1 267 6 38 7 215
Figure 3. Scheme of the composed model in DIgSILENT Power Factory
busbar the changes were 199%, 165% and 162% respectively to phase A, B and C. At second scenario voltage distortions at 0.4 kV substation busbar increased 190%, 155% and 140% respectively to phase A, B and C. At residential load busbar the changes were 187%, 122% and 109% accordingly. Voltage distortions for both scenarios at substation 0.4 kV busbar are also seen in Table 2 and at residential load busbar in Table 3. When total voltage THD values are compared, approximately 1% THD difference is seen at substation 0.4 kV busbar (5.81% and 4.80%). This approximately 17% decrease can be considered as harmonic cancellation. In the case of not transposed loads ninth harmonic at phase C exceeded 1.5% limit by 0.69 points. In case of 15th and 21st harmonic values in all phases were higher than permitted 0.5%. For phase C the excesses were 1.87 and 1.55 points respectively. Compared to the previous case voltage at phase C exceeded the limits also at 17th, 19th and 25th harmonic order. Voltage curves at substation 0.4 kV busbar are seen in Fig. 6 and distortions are given in Fig. 7.
TABLE 2. VOLTAGE THD VALUES AT SUBSTATION 0.4 KV BUSBAR THDA [%] One load Three loads Transposed loads 1.05 3.17 3.05 THDB [%] 1.68 5.07 4.28 THDC [%] 2.69 8.08 6.44
Figure 4. Current (A) versus time (s) characteristic of residential load
TABLE 3. VOLTAGE THD VALUES AT RESIDENTIAL LOAD BUSBAR THDA [%] One load Figure 5. Modelled harmonic voltages (% of fundamental) at substation in case of residential load Three loads Transposed loads 1.06 3.18 3.05 THDB [%] 2.05 5.43 4.57 THDC [%] 3.44 8.83 7.17
In the case of only one residential load, voltage distortions at substation 0.4 kV busbar are shown at diagram in Fig. 5. Individual harmonic voltages at phase C exceed the 0.5% limit for 15th and 21st order, others remained under the fixed limits of EN 50160 standard. It can be said that high distortions for phase currents were at odd orders and as expected, high values at 3rd and 5th order were seen. Voltage distortion limits were exceeded at higher orders, but it occurred even in case of one residential load. A. Effect of adding nonlinear load Influence of installing more nonlinear loads was compared in two scenarios where upper described residential load was modelled as base load and extra nonlinear loads were added in two cases. Firstly two loads, same as base load, were added and in second scenario order of phases (ABC) of those loads were transposed as ACB and BAC respectively. Idea of rearranging order of phases is done for determining quantity of the harmonic cancellation effect. At first scenario voltage distortions at 0.4 kV substation busbar increased around 200% at all phases. At residential load
Figure 6. Modelled voltages (V) versus time (s) at substation in case of three not transposed loads
At second scenario voltage distortions at substation 0.4 kV busbar decreased roughly 10 times more, about 0.1% in actual values at phases. At residential load busbar the changes were not considerably higher. It has to highlight that voltage also decreased about three volts at costumers busbar. Current distortions decreased in same numbers like in first case. Voltage distortions at substation 0.4 kV busbar for both scenarios are given in Table 4. Changes in current curve of residential load are seen in Fig. 9 and harmonic content of phase currents are given at Fig. 10. Comparable results were given in Fig. 2 and Fig. 4. Because of different nonlinear loads in the phases, harmonic distortions are not suppressed equally in the phases, even though added active load was balanced.
TABLE 4. VOLTAGE THD VALUES FOR LOAD ADDING CASE AT SUBSTATION THDA [%] Residential load Residential+active load Three residential loads Three residential+active loads 1.05 1.03 3.17 2.98 THDB [%] 1.68 1.66 5.07 4.93 THDC [%] 2.69 2.67 8.08 7.95
Figure 7. Modelled harmonic voltages (% of fundamental) at substation in case of not transposed loads
Figure 8. Modelled harmonic voltages (% of fundamental) at substation busbar in case of transposed loads
In case of transposed loads several voltage harmonics were out of the limit set by EN 50160 standard for individual harmonics. For example ninth harmonic at phase C exceeded 1.5% limit by 0.28 points. In case of 15th and 21st harmonic values in all phases were higher than permitted 0.5%. For phase C the excesses were 1.37 and 1.2 points respectively. Results are seen also in Fig. 8. It can be concluded that notable harmonic cancellation can be achieved when similar nonlinear loads are connected to different phases. This kind of arrangements needs a respectable overview of installed loads. B. Effect of adding active load Influence of adding active load was modelled in two scenarios. In first case only one residential load was used and in second scenario three same residential loads were connected. In both cases three phase active loads were added to all residential loads. 2.4 kW of active load was chosen as it makes three times higher power than initial apparent power of one residential load. At first case voltage distortions at substation 0.4 kV busbar decreased only about 0.01% in actual values at phases. At residential load busbar the changes were about 0.02%. Current distortions contrary decreased greatly, 72%, 59% and 57% respectively to phase A, B and C. No change in neutral current was registered.
Figure 9. Modelled current (A) versus time (s) characteristics of residential load with active load added
Figure 10. Modelled harmonic currents (% of fundamental) on logarithmic scale in case of residential load with additive active load
It can be concluded that reducing voltage harmonic with adding active power need numerous purely active loads. On another hand current distortion in conductors can be reduced notably by adding active loads. V. DISCUSSION The findings of present research are important when distributed generation and electrical vehicles are going to be connected into distribution networks. When grid strengthening and load connecting are arranged wisely investments could be postponed or reduced. In Estonia, such weak distribution grids with residential loads as described in this paper with predictable new nonlinear loads and EVs and DG units are very common. Besides that vast amount of heat-pumps with extensive power consumption are connected to residential houses and are not considered in this study. The combination of the unbalanced nature of residential loads and usage of wyewye service transformers means that a continuous increase in third harmonic levels in distribution grids can be expected [11]. Harmonic voltage increases are also more likely in longer feeders in rural areas due to the higher line impedances [12]. It has to be noticed that loads behind long feeders (even couple of kilometres in 0.4 kV grids) which are connected to wye-wye transformers are seen numerously in Estonia. The results can be applied to other types of networks as well. Considering above mentioned it may presume that possible problems in protection systems and errors in readings of revenue meters may occur due to high neutral currents and higher harmonics. It also has to be careful when grid improvements are targeted, because compensation of harmonics and reactive power, with capacitor banks, can worsen the situation with a harmonic resonances [11]. Transformer changes in meaning of power and vector group has to be considered economically and compared possible savings from increasing the life expectancy of devices and reducing the losses due to harmonics. One measure for reducing the distortion is to arrange the loads between phases wisely. It can be done when harmonic content of these loads are known. Also transposing the phases in substation in case of similarly loaded feeders may be considered beneficial in future. All aspects covered here should be considered when notable changes in loads are planned e.g. building comprehensive charging networks for EVs or installing nonlinear DG units all over the distribution network. Measuring harmonic content of current of different loads and modelling loads together in networks should be done beforehand. VI. CONCLUSION Due to nonlinear loads with significantly distorted currents whole residential load is also distorted severely. It concludes with distorted voltages at substation even in case where only one residential household is connected. Even more serious problem
may be the high current in neutral conductor due to unbalanced nonlinear loads. Voltage quality is going worse when more loads are connected, especially when exactly the same loads are in the grid. Limits for voltage harmonics established in standards are exceeded already with few loads connected. Notable cancellation of harmonics is achievable when phases are transposed in case of similar loads. Effect of adding purely active load next to nonlinear load is seen when proportion of them is remarkable. In case where three times more active power was added than initial nonlinear load, notable current distortion decrease is seen. On another hand it does not reduce voltage distortion at substation busbar particularly. VII. REFERENCES
[1] J. Niitsoo and I. Palu, "Distorted load impacts on distribution grid," in Proceedings of the 12th International Scientific Conference Electric Power Engineering 2011, Kouty nad Desnou, Czech Republic, 2011. [2] M. Rawa, D. Thomas and M. Sumner, "Simulation of non-linear loads for harmonic studies," in 11th International Conference on Electrical Power Quality and Utilisation (EPQU), Lisbon, Portugal, 2011. [3] J. Yong, L. Chen and S. Chen, "Modeling of Home Appliances for Power Distribution System Harmonic Analysis," IEEE Transactions on Power Delivery, vol. 25, no. 4, pp. 3147 - 3155 , October 2010. [4] M. Aiello, A. Cataliotti, S. Favuzza and G. Graditi, "Theoretical and experimental comparison of total harmonic distortion factors for the evaluation of harmonic and interharmonic pollution of grid-connected photovoltaic systems," IEEE Transactions on Power Delivery, vol. 21, no. 3, pp. 1390 - 1397 , July 2006. [5] J. Arrilaga and N. R. Watson, Power System Harmonics, 2nd ed., John Wiley & Sons, Ltd, 2003. [6] S. K. Jain and S. Singh, "Harmonics estimation in emerging power system: Key issues and challenges," Electric Power Systems Research, vol. 81, pp. 1754-1766, 2011. [7] C. Jiang, D. Salles, W. Xu and W. Freitas, "Assessing the Collective Harmonic Impact of Modern Residential LoadsPart II: Applications," IEEE Transactions on Power Delivery, vol. 27, no. 4, pp. 1947-1955 , October 2012. [8] K. McBee and M. Simoes, "Evaluating the long-term impact of a continuously increasing harmonic load demand on feeder level voltage distortion," in 2012 IEEE Industry Applications Society Annual Meeting (IAS), Las Vegas, NV, USA, 2012. [9] M. T. Au, J. Navamany and E. C. Yeoh, "Impact of medium power inverter-based residential harmonic loads and PFCs on low voltage lines," in 2010 IEEE International Conference on Power and Energy (PECon), Kuala Lumpur, Malaisia, 2010. [10] EN 50160:2010, "Voltage Characteristics in Public Distribution Systems. European standard.," 2010. [11] H. Sharma, P. Keebler, F. Sharp, M. Josef and J. Price, "Handling excessive third harmonics in utility systems due to proliferation of nonlinear loads," in 2012 IEEE Power and Energy Society General Meeting, San Diego, CA, USA, 2012. [12] S. Hardie and N. Watson, "The effect of new residential appliances on Power Quality," in 20th Australasian Universities Power Engineering Conference (AUPEC), Christchurch, New-Zealand, 2010.
Novel Diverse Tariff Scheme to Enhance Demand Response

Afan Bahadur Khan1,2, T. P. Imthias Ahamed2, S. Q. Ali2, Essam A. Al-Ammar1,2 Department of Electrical Engineering, King Saud University, Riyadh 11421, Saudi Arabia 2 Saudi Aramco Chair in Electrical Power, EE Department, College of Engineering, King Saud University, Riyadh 11421, Saudi Arabia (E-mail: afan246@hotmail.com, imthiasa@gmail.com, sqali@ksu.edu.sa, essam@ksu.edu.sa)
1
Abstract With the development of smart grid technology, price based demand response (PBDR) scheme will gain momentum and utilities are expected to enforce time dependent tariff. In near future, more and more domestic consumers will use different algorithms to schedule their flexible demand in order to reduce their energy costs. It is well known that with a single time of use tariff most of the customers schedule their load in the lower price intervals, hence their electricity cost will reduce but maximum demand (MD) on the power grid will not decrease. This paper proposes novel tariff scheme in which domestic consumers are split into n load groups and each group will have a separate tariff scheme. It is shown through simulations that with such a tariff scheme residential customers will be able to cut down their electricity bill while at the same time the utilities can bring down the maximum demand (MD). Index Terms Demand Response (DR), Load Scheduling Algorithm, Maximum Memand (MD), Tariff Scheme, Time of Use Pricing (ToUP).
I.
INTRODUCTION
As the nations develop economically and world population increases, the demand of electrical power is increasing and consequently the maximum demand (MD) on the grid is also increasing [1]. To meet with this challenge, one needs to increase the electrical energy capacity or to schedule the demand or both. Adding additional capacity is quite expensive; hence utilities prefer the second option because it refers to changes in power consumption pattern by end-use customers from their routine usage pattern in response to changes in the price of electricity or other parameters. This phenomenon is known as Demand Response (DR). The DR programs have been active in different regions since a decade, when extremely hot weather combined with energy shortages or caused major power brokers to default resulting in a blackout during a critical period. One of the DR programs i.e. Demand Dispatch (DD) [2], is a new way of thinking regarding the balancing of power generation and its consumption. It has the ability to aggregate and precisely control individual loads by two way communication between the power grid and the electrical load device. It involves switching ON and OFF loads not only at critical peak times, but instead it is employed all the time. Power grid must be operated in such a way so as to maintain the constant frequency by scheduling some types of flexible loads (i.e. dishwashers, dryers, water heater). There are two major techniques which can be used to implement DR, i.e. either by direct load control (DLC) or by indirect demand control through price based demand response schemes.
Direct load control (DLC) is one of the techniques used to control the demand but these changes are made in response to some incentive payments (e.g. direct payment, bill credit). In [3] a strategy is used to aggregate electric loads to balance power generation and its usage by means of load switches. These load switches can turn OFF the load directly by using the two way communication with internet protocol (IP) as the tool for the purpose of alleviating forecast errors. One ordinary example is refrigerator thermostat which acts both as temperature sensor and as actuation switch. Though DLC can be helpful in curtailment and shedding of electrical load but it will depend on the will of the consumer to grant access for a third party to directly control his/her individual loads such as appliances. The tracking of such appliances is also a difficult and a tough task. Another concept of reducing the maximum load by motivating the customers to limit their own electrical consumption is known as price-based demand response (PBDR). Numerous papers have discussed different types of price-based schemes such as time of use pricing (ToUP), critical peak pricing (CPP), real time pricing (RTP), and the use of automatic meter readings (AMR) for a demand based tariff [4-7]. All such tariff schemes use one of the three DR plans (i.e. fully automated, semi-automated, manual). An intelligent structure encourages fully automated DR plan which doesnt involve human intervention. Though PBDR can be beneficial to customers, it mainly focuses on the reduction of electrical energy consumption charges rather than the curtailment of maximum demand (MD). It also requires the cooperation between the utility operator and the end-user customers. The focus of this article is to understand the general philosophy of time dependent tariff schemes that reflect the production cost. It is well known that peaking plants are expensive; so electricity price will coincide with critical peak load time [8]. Hence if all customers schedule their flexible loads at the same time in the lower price intervals, they will benefit financially but the MD on the system might not get reduced, but will only get shifted or even may be increased [4]. This paper introduces a novel diverse tariff scheme where the domestic consumers are divided into n groups such that each group will have a different tariff in a way that the lower cost period is different for each group. The primary objective of such a tariff scheme is to motivate and encourage the customers to shift their demand to different lower price periods.
Rest of the paper is arranged as follows. The next section discusses the proposed system model along with its parameters. It is followed by simulation results in section . In section V, analysis, discussions and remarks are highlighted. Finally the paper is concluded in section V. II. SYSTEM MODEL Power generation, transmission and distribution facilities are designed to meet maximum power demand which occur for a short duration of time. Dimensioning the grid for the purpose of balancing the peak demand load is therefore expensive [9]. Hence, it is a need of time to shape an efficient model which simultaneously copes with maximum demand and offers economical as well. Since the objective of the paper is to curtail the maximum demand (MD) and to reduce the energy consumption charges through a proposed tariff scheme, we are proposing a system model as shown in Fig. 1, which has the capability to achieve these objectives.
this scheme, the consumers are divided into three groups; and each group will have a different tariff scheme. The main idea of the proposed tariff scheme is twofold; first allowing a separate tariff scheme for different type of customers to reflect fluctuating prices so that they are charged what the electricity is worth at different times of the day; second motivate the customers to shift their flexible high-load appliances to proposed timing or off-peak timing in order to not only reduce their electrical energy consumption cost but also to help the grid to curtailed the maximum demand.
Fig. 1. Block diagram of system model
A. Un-Scheduled Loads In [4][7][10] each individual unscheduled load is modeled by a 4-tuple represented by ? ? = (?? , ?? , ?? , ?? ) where [?? ?? ] is the interval during which the load j may be switched ON; ?? represents the length of time for which the load should remain ON and ?? represents the power rating of the load in kW. For example, ? ? = (13 19 5 8) means that load j can be switched ON anywhere between the 13th and the 19th interval and once turned ON, it has to remain ON for 5 intervals and the load rating is 8 kW. The un-scheduled loads are of two categories i.e. flexible load and non-flexible load. Flexible loads include home appliances such as chargeable equipment (e.g. laptops, cell phones, PHEVs) dishwashers, dryers, water heaters etc which can be switched ON at a fixed time or can wait for a later time to be switched ON. Non-flexible loads include air-conditioning system, and refrigerators etc which are assumed to occur at any time. B. Overview of Proposed Tariff Rates Numerous studies have discussed that households respond to time-based pricing schemes between peak and off-peak hours and hence according to time of the day, day of the week and season of the year are commonly charged within these predefined periods [9]. Maximum demand during peak periods from the viewpoint of the grid operator is more crucial than any individual peak demand. Hence keeping all things in consideration we propose a different diverse tariff scheme. In
Fig. 2. Overview of Propose Tariff Scheme (a) Conventional Price of Electricity vs. Time (b) Price of Electricity vs. Time for Consumer group A (c) Price of Electricity vs. Time for Consumer group B (d) Price of Electricity vs. Time for Consumer group C
To illustrate the idea, consider a utility employing conventional ToUP for all consumers as shown in Fig. 2(a). The motivation behind this TT tariff is the fact that peak demand needs to be reduced during intervals 11-13 and 20-22. In these intervals, tariff rate is double of the off-peak period. However we argue that all consumers may shift the load and maximum demand (MD) on the system is only shifted time wise but may not decrease. Therefore we propose that instead of tariff TT, employ three different tariffs TA, TB and TC to the three groups of consumers for loads as shown in Fig. 2(b-d).
It is interesting to note that the proposed tariff schemes contains three cost levels i.e. lower, intermediate and higher in which higher electricity cost is almost the same during the 11-13 and 20-22 intervals. However for group A, lower price period is from 1-8 and 16-17 intervals and intermediate price period occurs during 9-10, 14-15, 18-19 and 23-24 intervals. The idea of introducing an intermediate cost level is to encourage the consumers to make sure that in contingencies they avoid the peak critical pricing intervals and on the other hand it will give benefits to utilities operator in terms of cost incurred function. Similarly for group B, lower price period is from 1-6, 9-10 and 18-19 intervals and intermediate cost level occurs during 7-8, 14-17 and 23-24 intervals. For group C, 16, 14-15 and 23-24 is the lower cost intervals while 7-10 and 16-19 are the intermediate cost intervals. Thus this scheme will motivate consumers to shift the loads to different low price period. We assume that each residential customer is equipped with a smart meter connected to a smart power grid distribution system with a two-way communication through the internet protocol (IP) networking system [11]. The benefit of differentiated pricing scheme is that it reflects the conditions in the electricity market. It has been observed that the results of differentiated pricing scheme are modest and variable. It is reported that the load reduction of a large number of households participating in the scheme does not coincide with the system peak [5]. C. Pursuit Algorithm In [4][12-13] the authors use the single conventional tariff for a total of 100 unscheduled loads using pursuit algorithm which shifted the maximum demand from the peak hours to some other low tariff interval. What make our work different from the typical ToUP tariff [8] is that here we are implementing the separate tariff rates for each of the different category customers. The tariffs used are TA, TB and TC where; TA is the tariff used to schedule loads of consumer group A i.e. LA TB is the tariff used to schedule loads of consumer group B i.e. LB TC is the tariff used to schedule loads of consumer group C i.e. LC D. Scheduled Loads The intention of the model is by using such an algorithm, we must shift the flexible high power loads to lower cost intervals with the help of a novel distribution tariff. In the propose model, we divide the residential customers into three type categories namely A, B, and C. We tested this model by using the pursuit algorithm as such; ResA contains the first set of scheduled loads using TA ResB contains the second set of scheduled loads using TB ResC contains the third set of scheduled loads using TC
III. SIMULATIONS RESULTS To test the efficacy and scalability of the system model, we generated a variable L that contains a total number of 99 unscheduled flexible and non-flexible loads. As mentioned in [4], flexible loads are further divided in four regions i.e. peak day, peak night, off-peak day and off-peak night. The proportion of the unscheduled loads and their proposed timings are as follows; i. Non-flexible load (10 % during 8-15 intervals) ii. Flexible peak day (40 % during 8-15 intervals) iii. Flexible peak night (30 % during 17-20 intervals) iv. Flexible off-peak day (10 % during 1-8 intervals) v. Flexible off-peak night (10 % during 13-15 intervals) The variable L is further divided into three groups of 33 unscheduled loads as such; LA = first set of 33 unscheduled loads as shown in table LB = second set of 33 unscheduled loads as shown in table LC = third set of 33 unscheduled loads as shown in table The simulation results are given in Fig. 3;
Fig. 3. Simulation Results for Dissected 99 Loads (a) Scheduled Demand vs. Time using TA (b) Scheduled Demand vs. Time using TB (c) Scheduled Demand vs. Time using TC
It can be seen that the algorithm schedules the flexible loads in the low price intervals. These loads were randomly picked such that the schedule produced by the algorithm could be manually validated.
It is interesting to note that in all of proposed tariffs, all the customers shift their demand to the cheaper and lower cost intervals in order to save the consumption cost as well as to curtail the maximum demand (MD) from the power grid as seen in Fig. 3(a-c). Whereas in conventional tariff, all the customers might shift their load to the lesser price interval to
save their electricity cost but in that case MD will only switch from peak critical to off-peak period.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
Table I Loads of Group A S f l 2 5 4 7 8 2 13 16 4 2 5 4 9 13 2 9 13 2 8 13 2 12 16 3 9 15 3 11 13 1 10 11 1 12 18 3 11 13 2 14 17 1 11 15 2 8 12 2 9 15 2 20 22 2 21 22 1 21 22 1 21 24 1 20 23 2 21 25 2 21 23 2 20 21 1 20 23 1 20 24 2 21 24 1 4 6 2 7 12 3 1 3 2 13 17 3 13 14 1
r 1 4 5 4 5 5 1 4 1 3 3 4 4 3 4 3 3 4 5 2 4 4 1 5 2 3 5 1 5 2 3 2 1
j 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
Table II Loads of Group B S f l 23 23 1 13 16 4 15 15 1 10 12 1 9 14 1 10 13 2 8 13 3 14 18 1 11 15 2 11 16 1 11 13 1 9 14 3 8 13 2 10 16 3 9 15 3 9 13 1 12 16 3 21 23 1 20 23 2 21 25 2 20 23 2 20 22 1 21 25 2 20 23 2 20 23 2 21 24 1 21 24 2 6 9 3 6 11 3 1 3 1 3 4 1 14 18 3 15 17 2
r 4 4 5 5 3 5 2 2 4 2 4 2 1 4 5 5 2 1 1 4 3 3 2 4 2 2 5 2 5 3 3 1 2
j 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
Table III Loads of Group C S f l 16 18 3 6 7 2 5 8 4 8 12 3 12 14 2 10 15 2 11 14 3 13 15 1 13 15 2 13 16 2 8 10 2 12 13 1 14 16 1 8 13 1 10 17 3 2 16 2 12 16 2 20 22 2 20 23 2 20 23 2 20 21 1 21 23 1 21 25 2 20 21 1 21 24 2 21 23 2 21 25 2 1 4 3 2 7 3 3 6 2 15 19 3 14 15 1 15 17 2
r 3 1 3 4 5 3 1 4 2 5 4 2 3 5 3 2 2 2 2 4 3 1 5 1 4 2 4 4 3 1 5 1 5
V. ANALYSIS & DISCUSSIONS
Table V Effects of Separate Tariff Scheme on Maximum Demand (MD) & Cost Incurred for 99 Loads UnScheduled Scheduled Scheduled ResTT Res(A+B+C) Maximum 79 74 60 Demand (MD) Cost Incurred 5180 4291 4913 It can be seen in table V, that the algorithm shifted the demand into lower cost intervals; hence cost incurred will get reduced as compared to un-scheduled load. But the introduction of diverse tariff scheme reduces the even MD to a larger extent as can be seen in third column of the above table. Table V Effects of Intermediate Price Levels on Individual Category Consumers Intermediate ResA ResB ResC Total Cost Level 6 1476 1402 1613 4491 8 1520 1777 1616 4913 10 1941 1756 1638 5335 Different levels for intermediate tariff were tested to ascertain the level which the algorithm considers a higher priced period and at the same time it does not penalize the consumers by a large margin. Costs incurred by the different consumer groups and the total cost are presented in Table V. It can be observed that the cost incurred by all the groups is within reasonable agreement with each other. Furthermore the tariffs can be rotated among the user groups daily to keep the incurred cost with in a similar range for all the consumer groups. It can also be observed that for the intermediate cost being 6 the consumers still save a lot of money (5180 - 4491 = 689 cost units) while keeping the MD at 60. Also, the total is just 200 cost units higher with the proposed diverse tariffs when compared to the cost incurred by the tariff 'TT'. Therefore the intermediate cost level was selected as 6. In general, few things should be considered before the objective of tariff making policy. First, it is desired to avoid load synchronization among different electrical appliances in each household to prevent a sharp spike in the residential load. Second, it is desired to forecast the demand load by the end use customers. IV. CONCLUSIONS This paper illustrated that if same ToUP is employed for all consumers and if all consumers use intelligent scheduling algorithm to shift their flexible loads, MD will not get reduced drastically but will only get shifted to low price periods. We proposed a novel diverse tariff scheme where total consumers are split into n groups and different groups have different low priced regions. The proposed tariff scheme was illustrated considering three groups and using pursuit algorithm for scheduling. Simulation results showed that the novel scheme resulted in reduction of MD. However, the total cost incurred for consumers was more than the single tariff case. Thus this
Fig. 4. Comparison of Tariff for 100 Loads (a) Un-scheduled Demand vs. Time (b) Scheduled Demand vs. Time using TT Tariff (c) Scheduled Demand vs. Time using Propose Tariff (TA, TB, TC)
Fig. 4(a) presents the 100 un-scheduled loads with the highest MD peak as 79. It should be noted that during the 8th to 16th interval, the average MD peak is about 50. It is interesting to note in Fig. 4(b) that the actions taken in terms of conventional tariff to schedule the demands equally in all intervals results in a total shift of electricity consumption from peak to off-peak hours as can be seen from the 10th to 12th interval. Though a decrease in maximum demand at peak periods only does not curtailed it as the highest MD peak is at 74. It is apparent from Fig. 4(c) that the customer as a whole should benefit economically from being charged according to the different proposed tariff schemes as compared to the conventional tariff. The highest MD peak in the proposed tariff scheme is calculated on 15th interval which is about 60. It should also be noted that the average MD peak during the 8th to 16th intervals is about 44.87. Table V presents the effect of novel separate tariff scheme on MD and the total cost incurred for 99 loads.
paper of Diverse Tariff scheme will be beneficial for utility as well as for consumers. ACKNOWLEDGEMENT The authors are thankful to Saudi Aramco Chair in Electrical Power, Electrical Engineering Department, King Saud University, Riyadh for supporting this work. REFERENCES [1] Pazheri F. R.; N. H. Malik; M. F. Othman; M. Babar; Optimum Power Dispatch Problems: An Overview, International Journal of Engineering Inventions, vol. 1, no. 6, pp. 49-54. [2] Brooks, A.; Lu, E.; Reicher, D.; Spirakis, C.; Weihl, B., "Demand Dispatch," Power and Energy Magazine, IEEE , vol.8, no.3, pp.20,29, May-June 2010 [3] Keep, T.M.; Sifuentes, F.E.; Auslander, D.M.; Callaway, D.S., "Using load switches to control aggregated electricity demand for load following and regulation," Power and Energy Society General Meeting, 2011 IEEE , vol., no., pp.1,7, 24-29 July 2011 [4] S. Q. Ali, I. A. T. Parambath, and N. H. Malik, Learning automata algorithms for load scheduling, Electric Power Components and Systems, vol. 41, no. 3, pp. 286303, 2013. [5] Wallin, F.; Bartusch, C.; Thorin, E.; Backstrom, T.; Dahlquist, E., "The Use of Automatic Meter Readings for a Demand-Based Tariff," Transmission and Distribution Conference and Exhibition: Asia and Pacific, 2005 IEEE/PES , vol., no., pp.1,6, 2005 [6] Jin Xiao; Jae Yoon Chung; Jian Li; Boutaba, R.; Hong, J.W.-K., "Near optimal demand-side energy management under real-time demand-response pricing," Network and Service Management (CNSM), 2010 International Conference on , vol., no., pp.527,532, 25-29 Oct. 2010 [7] Maqbool, S.D.; Ahamed, T.P.I.; Malik, N. H.; Optimized load scheduling considering maximum demand and real time pricing, STEM 2012 program, paper # 78 [8] T. P. I. Ahamed, S. D. Maqbool, and N. Malik A reinforcement learning approach to demand response, in Centenary Conference EE -Sc, (Bangalore) 2011. [9] Wallin, F.; Bartusch, C.; Monica, O.; Iana, V.; Lars, W.; Introducing a demand based electricity distribution tariff in the residential sector: Demand response and customer perception, Energy Policy Jul. 2011 [10] Ahamed, T.P.I.; Maqbool, S.D.; Al-Ammar, E.A.; Malik, N. H., "A simulated annealing algorithm for demand response," Innovative Smart Grid Technologies (ISGT Europe), 2011 2nd IEEE PES International Conference and Exhibition on , vol., no., pp.1,4, 5-7 Dec. 2011 [11] Mohsenian-Rad, A.-H.; Leon-Garcia, A., "Optimal Residential Load Control With Price Prediction in RealTime Electricity Pricing Environments," Smart Grid, IEEE Transactions on, vol.1, no.2, pp.120, 133, Sept. 2010
[12] Parambath, I.A.T.; Jasmin, E. A.; Pazheri, F.R.; AlAmmar, E.A., "Reinforcement learning solution to economic dispatch using pursuit algorithm," GCC Conference and Exhibition (GCC), 2011 IEEE , vol., no., pp.263,266, 19-22 Feb. 2011 [13] Ali, S.Q.; Maqbool, S.D.; Ahamed, T.P.I.; Malik, N.H., "Pursuit Algorithm for optimized load scheduling," Power Engineering and Optimization Conference (PEDCO) Melaka, Malaysia, 2012 Ieee International , vol., no., pp.193,198, 6-7 June 2012
Online Voltage Stability Monitoring Based on PMU Measurements and System Topology
Dinh Thuc Duong and Kjetil Uhlen Department of Electric Power Engineering
Norwegian University of Science and Technology Trondheim, Norway
AbstractThis paper presents a new approach to estimate the Thevenin impedance and the maximum power transfer at load buses, which are criteria for voltage stability assessment. The proposed algorithm utilizes system topology, load impedances and phasors from phasor measurement unit (PMU) as means of estimation, in which load impedances are integrated into the admittance matrix of the power system to compute Thevenin equivalent parameters and phasors at monitored load buses are used to identify the maximum loadability point. The proposed method shows good performance on monitoring voltage stability. It is also applicable for online applications since PMUs are required at buses of concern and other information is available from existing state estimator. In case study section, the proposed methodology is demonstrated and validated by simulation on IEEE test systems. Index TermsAdmittance matrix, load impedance, maximum power transfer, Thevenin impedance, voltage stability.
calculation no longer converge while power of the other load is unchanged. These tools are useful for operation planning and contingency analysis. Recently, more and more phasor measurement units (PMUs) are deployed in power systems. With the capability of measuring synchronously phasors of voltage and current across the network, along with present communication infrastructure, PMUs pave the way for online monitoring applications, including dynamic voltage stability assessment. The essence of this is that the maximum power transfer or voltage instability point is the point where load impedance at one load bus is equal to the Thevenin equivalent impedance of the rest of the system seen at that bus. While the load impedance can be obtained by local voltage and current phasors, the estimation of the Thevenin equivalent poses challenges and remains the object of ongoing researches. There are presently two main approaches described in the literature to estimate the Thevenin impedance. The first one is based on a local measurement, while the other pursues a method that considers both system topology and wide area measurements. In [2], the estimation is an adaptive process with a user-defined tuning factor. In addition, the Thevenin parameters, i.e. voltage ETh and impedance XTh, are assumed constant during the short period of their identification. Their values are completely determined by local load voltage and current phasors and variation directions of both load impedance and XTh. The performance of this approach is significantly influenced by the tuning factor and quality of measurements such as noise. On the other hand, [3] introduces the concept of coupled single-port circuit by pulling all generators and loads out of the grid, and formulating impacts of other load currents (Ecoupled) on the considered load. Here, the virtual impedance that is equal to the coupling term Ecoupled divided by the examined load current is modeled as an impedance in series with the equivalent impedance obtained from inversion of the admittance matrix; and the sum of the two impedances is considered as the Thevenin equivalent impedance seen at the study bus. However, it is noted that this methodology is tested with assumption that all loads increase simultaneously by the
I.
INTRODUCTION
Voltage stability remains one of the major concerns in power system operation due to its severity, especially for those systems heavily loaded or highly penetrated by fluctuating renewable generation. Voltage stability limits cannot be found by normal load flow calculation because of singularity of the Jacobian matrix when the system is close to the limits. However, this problem can be solved by the continuation power flow method [1], whose main idea is that load flow calculation is run with increasing load until it does not converge. Then, the prediction of state variables is made based on the tangent vector at current state. Next, the state variables are corrected under the condition of fixed power or voltage, depending on the current location on the V-P curve. These two steps are iterated until the end of the V-P curve. Overall, the process imposes high demand of computation and availability of a good system-wide state estimator for use in on-line operation. Therefore, this approach is preferable in offline analysis. Likewise, there are some functions to determine voltage stability limits in some commercial simulation software for power system with similar features. To determine the V-P or V-Q curves of a specific load bus, its active or reactive loads are kept increasing until the load flow
This research is supported by the Nordic Energy Research project STRONgrid, hereby kindly acknowledged.
same scaling factor. Therefore, the virtual impedance remains quite constant during the test. The method presented in this article is to approximate Thevenin equivalence and the maximum power transfer, which functions as the criterion to assess voltage stability at load nodes. The methodology is based on a combination system topology and PMU measurements. In this paper, the structure is as follows. Section II presents an overview of voltage stability assessment based on impedance matching condition and maximum power transfer, followed by the proposed algorithm to determine Thevenin parameters in Section III. Section IV shows the results of test cases, and conclusion is finally drawn in Section V. II. REVIEW OF VOLTAGE STABILITY ASSESSMENT BASED ON IMPEDANCE MATCHING CRITERION To examine voltage stability of a particular load bus in power system, let us consider a common example of the twonode grid as shown in Fig. 1, where ETh is a voltage source and ZTh is an impedance. The load is represented by the impedance ZL, which is the ratio of the bus voltage VL to the current IL. It has been proven that the maximum power transferred to the load is reached when the magnitude of the load impedance is equal to that of the impedance ZTh,
Consequently, the power margin of the present operation point is [3] margin = Smax S L 100%, SL (5)
where SL is the current apparent power of the load. In summary, real-time voltage stability monitoring presented in the literature is the process based on continuous comparison of impedance and apparent power of load to the estimated Thevenin impedance and maximum power transfer, which are proposed and presented in the next section. III. DESRIPTION OF PROPOSED METHOD
Z L = ZTh .
(1)
A. Estimation of Thevenin equivalent paramters State estimators have been widely used by many TSOs for long period of time. In these estimators, system topology and its changes after events like breaker operation, change of tap position of on-load tap changer are available and updated. With recent development of communication infrastructure, refresh rate of state estimators has been improved significantly. Obviously, state estimators are working well and would not be superseded in near future. On the other hand, regarding voltage stability issue, it is noticed that only some load buses are vulnerable to this problem, and need closer monitoring. Solutions that require PMU installation only at these concerned buses would be cost effective and viable for practical implementation. Therefore, the main idea of the proposed approach is to estimate equivalent Thevenin parameters at load buses, and consequently maximum power transfer as means to evaluate voltage stability, based on the existing information of system topology extracted from state estimator and PMU measurements at load nodes. Given a simple grid in Fig. 2, the generator G1 is connected to bus B1 through the step-up transformer T1. There are two loads, namely L1 and L2, at bus B1 and B2 respectively; meanwhile Line 1 is the connection between B1 and B2. Assume that G1 is capable of keeping its terminal voltage unchanged. At one instant of time, the electric equivalent circuit of this system is depicted in Fig. 3, where T1, Line 1, L1 and L2 are represented by impedances ZT, ZL, ZL1, and ZL2 respectively. For the sake of illustration, shunt admittances of Line 1 are neglected.
Figure 1. Thevenin equivalent circuit.
If the ZTh is known, the ETh is formulated as

ETh = VL + I L ZTh . (2) where VL and IL are voltage and current phasors of the load. Then we can form the Impedance Stability Index (ISI) at the load bus as [5]
ISI = ZTh ZL , (3)
and estimate the maximum power transfer at the load node by [3]
Smax
2 ETh ZTh ( imag ( ZTh ) sin + real ( ZTh ) cos ) , = 2 2 imag ( ZTh ) cos real ( ZTh ) sin (4)
Figure 2. A simple two-node grid.
where is the load power angle.
It is noticed that to calculate the Thevenin impedance seen at bus B1, one can repeat the above procedure, including inversion of the admittance matrix. However, matrix inversion takes tremendous efforts of computation, especially in large power system. Let us make a small modification of the matrix Yeq in (8) by adding the admittance of load L2 into bus B2, forming
Figure 3. Equivalent circuit of the two-node grid.
Obviously, the Thevenin equivalent impedance of the system seen at bus B2 is ZTh Z Z = Z L + L1 T Z L1 + ZT (6)
1 1 1 Z +Z +Z L T L1 Yeq = 1 ZL
. 1 1 + ZL ZL2 1 ZL
(12)
After calculating the impedance matrix by (9), we obtain the equivalent circuit for bus B2 as shown in Fig. 4 where E'Th and Z'Th are new Thevenin voltage and impedance respectively. B1 ZTh B2
On the other hand, from Fig. 2, we can classically form the admittance matrix of the grid as
1 Z L Y = 1 ZL 1 ZL 1 ZL
(7)
ETh
Let us modify the diagonal elements of the matrix Y by adding the two admittances of T1 and L1 to establish a new matrix 1 1 1 Z +Z +Z L T L1 Yeq = 1 ZL
1 Z eq = Yeq .
Figure 4. Thevenin equivalent circuit, including all load impedances.
1 ZL 1 ZL
(8)
However, the parameter we intend to find is ZTh, not Z'Th. By contrasting the two equivalent circuits in Fig. 1 and Fig. 4, one can recognize that the Z'Th is equal to the equivalent impedance of the two parallel impedances ZTh and ZL2. Therefore, the ZTh is formulated as
ZTh =
' Z L 2 ZTh . ' Z L 2 ZTh
(13)
Define the impedance matrix Zeq as the inversion of Yeq, (9)
is
The diagonal element in the second row and column of Zeq Z eq ( 2, 2 ) = Z L + Z L1 ZT , Z L1 + ZT (10)
With this small adjustment, we can estimate all Thevenin impedances seen by load buses with only one inversion the modified admittance matrix. B. Network reduction In large power systems, inversion of the modified admittance matrix of the whole grid as mentioned above requires tremendous computation effort. However, one can notice that there are many parts of the system that are linked to other areas by single connection as depicted in Fig. 5. In this case, in order to estimate Thevenin parameters at buses in the study area, we can follow these steps: First, consider each external area as a subsystem, and then, use the above proposed method to compute Thevenin impedance seen at the boundary with respect to the study area. Here they are B1, B2 and B3. Second, integrate newly estimated Thevenin impedances shown in Fig. 6 into the admittance matrix of the study grid and continue calculation process.
which is exactly equal to the Thevenin impedance seen at bus B2 in (6). Indeed, it is generally valid that the diagonal element indexed (i,i) of the impedance matrix is the Thevenin impedance seen at the equivalent ith bus in the grid [4]. Once the Thevenin impedance at the load bus is determined, the Thevenin equivalent circuit of this system is equivalent the circuit in Fig. 1 where load L2 is the considered as load. Then, the equivalent Thevenin voltage can be computed by ETh = VL + I L ZTh . (11)
where VL and IL are phasors of voltage and current at observed bus obtained from PMU measurements.
This feature also makes it possible to investigate voltage stability of a small part of the grid, for example branch 3, which has weak connection to the rest of the system.
study area Area 2 Area 1 B1 B2
B3 Area 3
Figure 5. Simplification of system with radial branches.
study area Area 1 ZTh1 B1 B2 Area 2 ZTh2
B3 Area 3 ZTh3
Figure 6. Equivalence of radial connections.
C. Effects of generator excitation limit When the system is approaching voltage collapse, demand of reactive is very high and hence it may lead to the situation that nearby generators hit maximum reactive power limit imposed by overexcitation limiter, resulting in acceleration of voltage collapse phenomenon. In this case, the internal voltage of generator cannot continue increasing; hence the terminal voltage is no longer constant and drops significantly due to the voltage drop on the synchronous reactance. Therefore, this synchronous reactance must be added in series with the stepup transformer impedance into the admittance matrix. Now, the electric circuit of the two-node grid in Fig. 3 is modified as depicted in Fig. 7.
must be updated to represent the true configuration of the grid. Breaker statuses could be communicated from SCADA or EMS/state estimators at regular intervals or preferrably be event based. 2) Calculate all load impedances and admittances: For buses where PMUs are installed, load impedance is the ratio between phasors of bus voltage and load current. Instead, voltage and power injections obtained from state estimation are utilized. PMUs are required at load buses involved in voltage stability monitoring scheme. 3) Modify the admittance matrix for the observable part of the grid: Add load admittances to corresponding elements on the diagonal of the admittance matrix by (12). Note that terminal bus of generators is considered as the reference bus in Thevenin equivalent. Thus, impedance of step-up transformer, if any, must be added to the bus at the high voltage side. For those generators that hit excitation limits, their synchronous reactances must be added in series of stepup transformer impedance. 4) Inverse the admittance matrix and compute Thevenin impedance at concerned load buses: Conduct inversion of the admittance matrix to achieve the impedance matrix. Note that elements on the diagonal of this matrix are not Thevenin impedances at each load buses. Use (13) to obtain the equivalent impedance for every bus of the observed network. 5) Establish ISI for each load bus: Form ISI for each load bus by (3). The ISIs are dynamic and depend on operating point of power system; the load bus with maximum ISI is the weakest one in the grid. In addition, the maximum power transfer - the voltage instability, condition is the point at which ISI is equal to 1. 6) Determine maximum power transfer and margin: It is more convenient, especially for system operators, to express the distance to the limit in power. By applying (4) and (5), one can determine proximity to limits in terms of power. Those buses with small margin are the most critical ones. IV. TEST CASES To validate the proposed approach, we use the RMS Simulation function of the commercial software DIgSILENT PowerFactory. All generators have both governor and automatic voltage regulator. Voltage collapse is emulated by gradually ramping up one load while the others are modeled as mixture of constant impedance (60%), constant current (20%) and constant power (20%). The first test is carried out on the IEEE 9-bus test system as shown in Fig. 8. All generators do not have limits in excitation system. The load at bus #7 is arbitrarily chosen for projection to voltage collapse although the procedure is definitely applicable to other loads. First, from the base operation point, the load at bus #5 increases in both active and reactive power by 200% during the period from 2s to 7s. This variation is seen at bus #7 by a slight drop of the Thevenin impedance in Fig. 9.
Figure 7. Equivalent circuit as generator reaches excitation limit.
D. Summary of proposed methodology Overall, the proposed voltage stability assessment is summarized as following: 1) Update the admittance matrix of the grid: When system topology changes are detected, the admittance matrix
Figure 8. The IEEE 9-bus test system.
In the second case study, we use the IEEE 39-bus test system shown in Fig. 12. The load at bus #15 is arbitrarily chosen for emulation of voltage instability. Again, excitation limit is not considered. The Thevenin impedance at the bus #15 is estimated by two approaches. The first one compute equivalence parameters based on the admittance matrix of the entire system. Meanwhile the second treats the dashed area as external system and identifies its Thevenin impedance seen at bus #16. This impedance is then integrated to the admittance matrix of rest of system to compute the Thevenin impedance seen at bus #15. As can be seen in Fig. 13, the impedances estimated by two methods show the same performance. Regarding Fig. 14, the maximum apparent power of load #15 matches the estimated maximum power transfer at the instant of time the two impedances are equal as seen in Fig. 13.
In next step, the load at bus #7 is projected to voltage collapse. In Fig. 10, the load impedance is plotted versus the estimated Thevenin impedance while Fig. 11 depicts the load apparent power and the estimated maximum power transfer at the bus #7. It can be seen that when the two impedances are equal, the actual maximum of the load roughly matches the estimated power transfer computed from proposed method. In other words, the method has successfully estimated the power limit at the bus.
Thevenin impedance (Ohm) 37.075 37.07 37.065 37.06 37.055 2 3 4 Time (s) 5 6 7
Figure 9. Thevenin impedance seen at bus #7 during load #5 variation.

80 Thevenin impedance Impedance (Ohm) Load impedance 60
Figure 12. The IEEE 39-bus test system.

40 Estimated Zth1 Impedance (Ohm) 30 20 10 0 Estimated Zth2 Load
40
20 10
11
12 Time (s)
13
14
15
500
1000 Time (s)
1500
2000
Figure 10. Estimated Thevenin impedance and the load #7.

520 Apparent power (MVA) 500 480 460 440 10 Estimated Smax Load 11 12 Time (s) 13 14 15
Figure 13. Estimated Thevenin impedance versus the load at bus #15.
3000 Apparent power (MVA)
2000 Load Estimated Smax1 1000 Estimated Samx2
500
1000 Time (s)
1500
2000
Figure 11. Estimated maximum power transfer and apparent power at bus #7.
Figure 14. Estimated maximum power transfer and apparent power at bus #15.
The third study examines the impact of excitation limit of generators on maximum power transfer. During the ramp of load #15, most of generators, except G7, hit excitation limits as seen in Fig. 15. The effect of these limits on Thevenin impedance and maximum power transfer are presented in Fig. 16 and Fig. 17. Obviously, excitation constraints result in increase of the Thevenin impedance seen at the load bus, and consequently drop of the maximum power that the source can supply the load, which is well estimated by the proposed method as depicted in Fig. 17.
6 5 Internal voltage (pu) 4 3 2 1 0 E1 E2 E3 E4 E5 E6 E8 E9 E10 0 200 400 600 800 1000 Time (s) 1200 1400 1600
DISCUSSION Estimation of Thevenin equivalent parameters based on system topology and local phasor measurements at load bus shows stable performance and good approximation of the maximum power transfer. The method is feasible for practical implementation since information of system topology is available from state estimation, and PMU is only needed at those buses where voltage stability is of concern. Since changes of topology affect the estimated equivalent parameters, a fast track of system structure is essential in this approach. Regarding load impedances at buses that are out of concern, their values are normally much larger than that of power lines and transformers. Variations of these impedances have small impact on the estimation of Thevenin equivalence and maximum power transfer. Thus, it does not require fast refresh rate. Update frequency of existing state estimators can be sufficient. It can be noticed that voltage stability is rather a local problem, and impact of remote generators is small. It is not necessary to take the entire system into calculation, especially in large interconnected system. Here network reduction promises an appropriate tool to facilitate the proposed method. This topic is under ongoing research. V.
CONCLUSION
Figure 15. Internal voltage of generators under overexcitation limiter

60 Thevenin impedance Impedance (Ohm) Load 40
20
0 200
400
600
800 1000 Time (s)
1200
1400
1600
Figure 16. Estimated Thevenin impedance versus the load at bus #15 in case of lacking reactive support from generators.
3000 Apparent power (MVA) 2500 2000 1500 1000 500 0 Estimated Smax Load
The paper has presented the new approach to estimate the Thevenin equivalent parameters and the maximum power transfer based on the combination of system topology and PMU measurements. Based on available information from existing state estimation and minimum requirements of PMU installation at monitored buses, the algorithm can be implemented in real time without high demand of resources. The results from simulations have, in addition, validated the performance of this approach. REFERENCES
[1] [2] [3] V. Ajjarapu and C. Christy, The continuation power flow: a tool for steady state voltage stability analysis, IEEE Trans. Power Syst., vol. 7, pp. 416-423, Feb. 1992. S. Corsi and G. N. Taranto, "A real-time voltage instability identification algorithm based on local phasor measurements," IEEE Trans. Power Syst., vol. 23, pp. 1271-1279, Aug. 2008. Y. Wang, I. R. Pordanjani, W. Li, W. Xu, T. Chen, E. Vaahedi and J. Gurney, "Voltage stability monitoring based on the concept of coupled single-port circuit," IEEE Trans. Power Syst., vol. 26, pp. 2154-2163, Nov. 2011. J. J. Grainger and W. D. Stevenson, Power System Analysis, Singapore: McGraw-Hill, 1994, pp. 283-294. M. Begovic, B. Milosevic and D. Novosel, A novel method for voltage instability protection, in Proc. 2002 IEEE Computer Society System Sciences Conf., pp. 802-811.
200
400
600
800 1000 Time (s)
1200
1400
1600
[4] [5]
Figure 17. Estimated maximum power transfer and load power at bus #15 in case of lacking reactive support from generators.
Impedance Based Fault Location Method for HVDC Transmission Line

Mohammad-Mehdi Kamali-faz Javad Sadeh Yaser Damchi
Electrical Engineering Department, Faculty of Engineering Ferdowsi University of Mashhad Mashhad, Iran mehdi.kamali1986@yahoo.com, sadeh@um.ac.ir, damchi_pe@yahoo.com
Abstract This paper proposes a novel impedance-based fault location algorithm for high voltage DC (HVDC) transmission lines. The fault is located using voltage and current measurements at two terminals of the faulted line during the fault occurrence. This algorithm depends on the low sampling frequency and few numbers of data for fault location. The performance of the proposed approach is evaluated through simulations in PSCAD/MTDC and analysis in MATLAB. The obtained results confirm that the proposed approach is able to accurately locate faults in the HVDC transmission lines and is not susceptible to fault resistance. Keywords; impedance-based method; fault location; HVDC transmission line.
where synchronized and unsynchronized phasors of the fundamental harmonic contents of the voltage and current at all terminals are used for fault location. In [4], a two-terminal impedance-based technique, considering distributed model of AC transmission line, is developed. This technique uses unsynchronized measurements of voltages and currents at both ends and works based on the fundamental frequency phasors of the symmetrical components of the measured signals. In [5], an impedance-based fault location technique, independent of the line parameters has been proposed. In this technique, voltage and current measurements at both ends are used and fast communication channels are installed at the line terminals for the purpose of measurement synchronization. The travelling wave techniques are commonly applied to HVDC lines. They estimate location of faults according to the propagation time of the voltage and current waveforms in the transmission line, and are classified into one- or two- terminal methods whether measurements are carried out in one or two terminals, respectively. In one-terminal techniques, the fault is located by measuring the travel time of the fault surge from the terminal to the location of fault and back again. On the other hand, the time difference between the fault surge arrivals at both ends is used in two-terminal techniques, where measurements at two ends are synchronized by a GPS [6]-[10]. In [11], discrete wavelet transform (DWT) has been applied to estimate arrival of surge forehead and locate fault. Application of continuous wavelet transform (CWT) for estimation surge forehead arrival has been proposed in [12] and [13]. The obtained results of these papers show that the CWT is better than the DWT. In [12], a two-terminal travelling-wavebased technique has been developed for HVDC transmission including both overhead lines and cable connections, while a fault location algorithm for multi-terminal HVDC systems with star topology has been presented in [13]. Calculated voltage along transmission line with the synchronized and unsynchronized measurements at both ends have been employed for fault location in [14]. A fault location algorithm based on cable sheath current has been proposed in [15] for HVDC system with DC cables. Based on the reviewed literature, impedance based technique have not already been applied to locating fault in HVDC transmission lines. Moreover, the impedance-based technique has favorable properties such as simplicity, speed and it no need to high sampling frequency. Hence, in this paper
I.
INTRODUCTION
High voltage DC (HVDC) transmission lines, like other equipment in power systems, are subject to different types of faults. Short-circuit faults are among the common faults occurring in the HVDC lines. Quick restoration of the faulted line after the disconnection by protective devices is highly important. Since HVDC lines are usually long, traditional fault location based on line inspection will be time consuming, and in some cases, extremely difficult. Hence, effective fault location algorithms not only help fast locating of fault and restoration of the faulted line, but also they bring about other advantages such as time, economic and human resources savings. Fault location in transmission lines has been comprehensively studies to date. Various techniques have already been developed for locating fault in transmission line, such as travelling wave based and impedance based methods. In [1] an impedance based fault location algorithm has been proposed for AC systems, which does not require line parameters. This algorithm uses unsynchronized current and voltage measurements at both ends and locates faults based on the main harmonic contents of the measured signals. Unsynchronized voltage and current measurements have been also employed in [2] for fault location in AC systems. To determine fault distance and synchronization angle, pre- and post-fault signals have been processed in [1]. In [2], in contrast to the classical approaches, only the amplitude of the fault voltage is used and phase angle information is not required. An impedance-based technique has been proposed in [3] for fault location in two- and three-terminal AC transmission lines,
an impedance-based technique is developed for fault location in HVDC transmission lines. II. PROPOSED FAULT LOCATION ALGORITHM
respectively. Also Vrec and Vinv denote rectifier and inverter voltages, respectively. According to (2) and (4), location of fault is estimated as:
In the proposed impedance-based algorithm, the voltage and current measurements at rectifier and inverter stations during the fault are used for locating the fault. To estimate location of the fault, lumped model is used for transmission lines. The estimation of fault location is a function of voltages and currents at inverter and rectifier stations. In order to perform this estimation, initially, fault point voltage (Vf) is calculated using measurements of two terminals. The procedure of this calculation is shown in the following: Rectifier side:
di dt
RC dV dV R LC d 2V I d 2V R 2 I + 2 x dt dt 2 2 dt 2 dt dV I dI d 2V I dI + R ( I R I I ) + L R I Cl R L dt dt dt 2 dt + (V I V R ) + RlI I +
x (5)
dI RCl 2 dV I LCl 2 d 2V I + Ll I + =0 2 2 dt 2 dt dt
V f = V rec Rxi Lx
(1)
C dV rec V f =V rec Rx i rec x 2 dt d C dV rec Lx i rec x 2 dt dt

Inverter side:
(2)
This equation has two solutions that only one of them is acceptable. To calculate these solutions, four consecutive samples are used. Regarding to sampling window, there are several acceptable solutions for fault location. There are three steps to determine the final fault location. Based on the proposed method, at first, acceptable solutions are sorted from minimum value to maximum value. Then 50% of central data are extracted from these solutions. Finally, fault location is calculated by averaging these data. III. SIMULATION RESULTS
V f =V inv + R ( l x ) i + L ( l x )
di dt
(3)
To assess capabilities of the proposed algorithm, a monopolar, 12-pulse HVDC transmission system with a 400 km DC link, shown in Fig. 1, is employed. Nominal power, voltage level and current of the DC link are 1000 MW, 500 kV and 2 kA, respectively. The sampling frequency and time steps are set to 20 kHz and 50 s, respectively. The simulations are carried out in PSCAD/EMTDC while fault location algorithm is implemented in MATLAB software. Performance of the proposed algorithm is evaluated and analyzed under different conditions of the system and for various fault resistances and locations. The proposed algorithm uses 10 msec information for fault location. Based on (5), location of fault is computed for every 4 voltage and current samples. Then, the final location of the fault is determined by averaging over the values computed during this time 10 msec data window. The estimation error is calculated using (6).
dV C V f =V inv + R ( l x ) i inv + ( l x ) inv 2 dt dV d C + L ( l x ) i inv + ( l x ) inv 2 dt dt
(4)
where, x represents fault distance from the rectifier station and l indicates transmission line length. R, L and C are resistance, inductance and capacitance per-length of the line,
Figure 1. Mono-polar HVDC transmission system
TABLE I. Fault distance (from rectifier, km) SIMULATION RESULTS Location result (km) 49.8999 49.9503 50.2129 50.3747 99.6511 100.1475 99.4901 100.1272 149.3045 149.0140 149.6255 149.3171 200.0552 199.8603 200.1344 199.8727 249.9335 249.8721 250.1068 249.4237 299.8534 299.7982 299.1435 299.1923 350.1145 350.0337 349.7429 348.0638 Error (%) 0.0250 0.0124 0.0532 0.0937 0.0872 0.0369 0.1275 0.0318 0.1739 0.2465 0.0936 0.1707 0.0138 0.0349 0.0336 0.0318 0.0166 0.0320 0.0267 0.1441 0.0367 0.0504 0.2141 0.2019 0.0286 0.0084 0.0643 0.4840
Fault resistance () 0.01 30 60 90 0.01 30 60 90 0.01
50
100
Figure 2. Rectifier voltage and current waveforms
150
30 60 90 0.01
%error =
x actual x calculated l
100
(6)
200
30 60 90 0.01 30 60 90 0.01 30 60 90 0.01 30 60 90
where, xactual and xcalculated indicate the actual and calculated distance of fault from the rectifier station. Fig. 2 depicts voltage and current waveforms of the rectifier side for a fault with 90 resistance, occurring 30 km away from the rectifier station. The DC link voltage drops to zero and exhibits slight oscillations due to the stored energy in the capacitor and the DC current shows an abrupt transition. The simulation results for DC short-circuit fault at different location when fault resistance is equal to 90 ohm are summarized in Fig. 3. Furthermore, Table I shows more comprehensive results of different situations. Based on the results presented in Table I, variations in fault resistance and distance do not significantly influence accuracy of the proposed algorithm. The maximum error for different fault locations and resistances is below 0.5%.
250
300
350
Figure 3. Results of the fault location when a fault occurs at different location from the rectifier side through 90- resistance
IV.
CONCLUSION
[6]
In this paper, a novel algorithm for fault location in HVDC transmission lines is developed. The proposed algorithm is based on the impedance method because this method needs a few data and simple implementation. The proposed algorithm employed voltage and current measurements at both ends during the fault for locating it and take advantage of low amount of the required data (10 msec data window). This algorithm needs a sampling rate far lower than in the travelling wave methods, although they need a dedicated high sampling frequency device. The results of simulation revealed that the proposed algorithm is not susceptible to fault resistance and distance. Furthermore, it is able to provide accurate and reliable estimation of the fault location. REFERENCES
[7]
[8]
[9]
[10]
[11] [1] C.A. Apostolopoulos and G.N. Korres, A novel algorithm for locating faults on transposed/untransposed transmission lines without utilizing line parameters, IEEE Trans. Power Del., vol. 25, no. 4, pp. 2328 2338, Oct. 2010. J. Clerk Maxwell E.G. Silveira and C. Pereira, Transmission line fault location using two-terminal data without time synchronization, IEEE Trans. Power Syst., vol. 22, no. 1, pp. 498499, Feb. 2007. I. S. Jacobs A. A. Girgis, D. G. Hart and W. L. Peterson, A New Fault Location Technique for Two and Three Terminal Lines, IEEE Trans. Power Del., vol. 7, no. 1, pp. 98-107, Jan 1992. J. Izykowski, R. Molag, E. Rosolowski and M.M. Saha, Accurate location of faults on power transmission lines with use of two-end unsynchronized measurements, IEEE Trans. Power Del., vol. 21, no. 2, pp. 627633, Apr. 2006. G. Preston, Z. M. Radojevi, C. H. Kim and V. Terzija, New settingsfree fault location algorithm based on synchronised sampling, IET Generation, Transmission & Distribution, vol. 5, no 3, pp. 376 - 383, March 2011.
[12]
[2]
[3]
[13]
[4]
[14]
[5]
[15]
M. Ando, E.O. Schweitzer and R. A. Baker, Development and elddata evaluation of single-end fault locator for two-terminal HVDC transmission lines Part I: Data collection system and led data, IEEE Trans. Power App. Syst., vol. PAS-104, no. 12, pp. 35243530, Dec. 1985. K. Nanayakkara, A.D. Rajapakse and R. Wachal Fault location in extra long HVDC transmission lines using continuous wavelet transform, in International Conference on Power Systems Transients, Netherlands, June 2011. M.B. Dewe, S. Sankar and J. Arrillaga, Application of satellite time references to HVDC fault location, IEEE Trans. Power Del., vol. 8, no. 3, pp. 12951302, Jul. 1993. Z. Yi-ning, L. Yong-hao, X. Min and C. Ze-xiang, A novel algorithm for HVDC line fault location based on variant travelling wave speed, in 4th Electric Utility Deregulation and Restructuring and Power Technologies Int. Conf, pp. 1459-1463, 2011. P. Chen; B. Xu and J. Li. A traveling wave based fault locating system for HVDC transmission lines, Int. Conf. Power System Technology, 2006. P.K. Murthy, J. Amarnath, S. Kamakshiah, and B.P. Singh, Wavelet transform approach for detection and location of faults in HVDC system, in Proc. IEEE Region 10 3rd Int. Conf. Industrial and Information Systems, Kharagpur, India, pp. 16, 2008. O.M.K.K. Nanayakkara, A.D. Rajapakse and R. Wachal, Location of DC line faults in conventional HVDC systems with segments of cables and overhead lines using terminal measurements, IEEE Trans. Power Del., vol. 27, no. 1, pp. 279288, Jan. 2012. O.M.K.K. Nanayakkara, A.D. Rajapakse and R. Wachal, Traveling wave based line fault location in star connected multiterminal HVDC systems, IEEE Trans. Power Del., vol. 27, no 4, pp. 22862294, Oct. 2012. J. Suonan, G. Shuping, S. Guobing, J. Zaibin and K. Xiaoning, A novel fault location method for HVDC transmission lines, IEEE Trans. Power Del., vol. 25, no 2, pp. 12031209, 2010. S.H. Ashrafi Niaki, H.Kazemi Karregar and M. Ghalei Monfared, A novel fault location technique for VSC-HVDC cable of offshore wind farm, in 7th Power Systems Protection and Control Conf, Jan 2013.
The Application of Simple Thermal Models to Improve Dynamic Load Models

Colin Christy, Yongli Zhu
Department of Electrical and Computer Engineering Iowa State University Ames, Iowa, USA colin@iastate.edu, yongliz@iastate.edu
AbstractThe classical approach to load modeling is to divide the load by customer class and then identify and model the load components corresponding to each customer class. This work explores an alternative way to break down the load using readily available data. Namely, using historical usage data and historical weather data the load can be first divided into temperaturedependent and temperature-independent portions. A simple thermal model of the structure(s) served by the electrical circuit forms the theoretical and mathematical basis of this separation. During summer peaking conditions, the temperature-dependent load is generally composed of air conditioners, which must be modeled properly in order to analyze delayed voltage recovery or other dynamic phenomenon. Once the temperature-dependent fraction of the load has been characterized, the size of compressor, fan, and pump motors can be inferred by the customer class. Keywords-load modeling; induction motor; thermal model; thermal resistance; thermal capacity; change-point curves
that is temperature dependent. In other words, since the summertime temperature-dependent load is generally comprised of air conditioning, we can identify the air conditioning load. The data needed to complete this task is power versus time and outdoor ambient temperature versus time; the former can be obtained by substation recorders and the latter can be obtained from nearby weather stations. In addition, it will be demonstrated that monthly energy usage data (versus average monthly temperature) is needed to identify the proper indoor temperature in the thermal model. II. DEVELOPMENT OF THE THERMAL MODEL CONCEPT
I.
INTRODUCTION
In creating a dynamic equivalent load model for a substation, the load is typically divided into various classes and then the components corresponding to each class are identified and modeled [1]. The customer classes typically consist of industrial, commercial, and residential [1,2]. Of course, there are many different load components within each customer class. One of the most important components in a dynamic load model is air conditioning, because it requires induction motors for compressors, fans, and pumps. In a summer peaking situation, air conditioning load is at its highest, and the associated induction motors may stall during a low voltage event. Consequently, the accurate assessment and modeling of air conditioner loads is essential for the accurate analysis of delayed voltage recovery or other dynamic phenomenon [3]. Generally speaking, air conditioner load is temperature dependent. When the outdoor ambient temperature goes up, air conditioning load goes up, and vice-versa. However, the effect is not immediate because it takes some time for heat to travel through the walls and ceilings. In other words, there is a time constant associated with the flow of heat from the outdoors into a structure. The premise of this work is that the amount and time constant of heat flowing into the structures served by an electric distribution circuit can be modeled by means of a simple thermal circuit. Furthermore, by creating and identifying this model, we can identify the power consumption
Thermal circuits are analogous to electrical circuits in that they contain thermal resistance and thermal capacitance, which is usually called heat capacity or thermal capacity. Furthermore, the driving force to heat flow is temperature, so temperature in a thermal circuit is analogous to voltage in an electrical circuit. However, thermal circuits and electrical circuits are not exactly analogous because heat flow (q) is energy per unit time. Thus, heat flow corresponds directly to power. Keeping that in mind, we desire to create a thermal model to represent the wall or ceiling of a structure (for the following explanation, we will just refer to it as a wall). Such a circuit is shown in Figure 1 and will now be explained. First of all, one side of the wall is exposed to the outdoor ambient temperature, TO, and the other is exposed to the indoor temperature, TI. Thus TO and TI are shown at each end of the thermal circuit shown in Figure 1. Secondly, the wall provides resistance to the flow of heat, which is divided into two and represented as R1 in the circuit. The thermal resistance is divided into two so that we can represent the average internal temperature of the wall, Tm1. The wall holds heat energy which means that it has thermal capacity, represented by C1.
R1 2
Tm1 ( s )
R1
q1 ( s) TI ( s)
1 sC1
TO ( s )
Figure 1. The thermal circuit.
As the temperature of the wall rises, the heat energy held in the wall also increases. While some of the heat energy is held in the wall, other heat energy flows into the interior of the structure, and is designated as q1. Heat energy flow, q1, is very important because any heat energy that flows into the structure must be removed from the structure by the air conditioning system if the interior is to be kept cool. Furthermore, heat energy flow has units of energy per unit of time so it can be expressed in J/s or Watts, which are the same units as electric power. To further elaborate; let us assume that the indoor temperature, TI, is to be held constant. Then the cooling power of the air conditioner must match the heat flow q1. Thus, the electric power required to remove the heat energy can be calculated by considering the efficiency of the air conditioning system:
temperature by the air conditioning system and the outdoor temperature fluctuates on a daily basis. III. IDENTIFYING THE THERMAL MODEL FOR A SINGLE STRUCTURE
The parameters of the thermal model can be identified by first analyzing long term energy usage data and then considering temperature dependence on a shorter time frame. Long term energy usage data is often discussed in the literature on energy efficiency [4]. Monthly electrical energy usage is plotted against the average monthly temperature to create what are called change-point (CP) curves, like the one shown in Figure 2.
(1)
The foregoing discussion was conceptually based on a single wall or ceiling and that is why the subscript 1 was used on all circuit parameters. If it represented a single wall, the thermal model of the whole structure could be created by creating a set of simple thermal models that would describe the heat flow into the structure. The total cooling power required would be the sum of heat energy flow into all interior surfaces and that could be found by adding the heat energy flow of the set of thermal circuits. However, it would be very useful if a single circuit like Figure 1 could represent the thermal system of the entire structure. Still further, it would be even more useful if a single circuit could be used to represent the thermal system of an entire set of structures that is connected to a feeder or substation. Both of these applications are further described later in the paper. Now that the thermal circuit elements and topology have been described, we can write the associated equations. First, the internal temperature of the wall can be expressed as:
Figure 2. A change-point (CP) curve showing monthly energy usage versus the monthly average temperature.
(2)

From this we can define the time constant = and see that under steady state conditions (when TI and TO are constant), = + . In words, under steady state conditions, the internal temperature of the wall is simply the average of the interior and outdoor temperatures. This is a logical result that fits our intuitive expectation. The second equation of interest is the expression for the heat energy flow into the interior, which is:
An important point on the CP curve is TCP, which is the temperature above which energy consumption increases with an increase in average monthly temperature. Another important feature of the CP curve is WC, which is a constant term and represents energy usage that is independent of variations in the average monthly temperature. Points with average monthly temperatures above TCP represent summer months when the cooling system in the structure is in service. In these months, the energy usage rises above WC because extra energy is needed for cooling. It turns out that the CP curve can be used to estimate the effective thermal resistance R in the thermal circuit developed earlier in this paper. To calculate R, we start by calculating the mean thermal heat flow for a month:
Energy Usage
(3)
is the extent to which the mean temperature was where above TCP during the month. Then, the average power consumed by the air conditioner can be calculated as:
(4)
Again, as a check, we see that under steady state conditions, this expression reduces to = . Thus, thermal capacity acts just like an electrical capacitor under steady state conditions in that it becomes an open circuit to heat energy flow. That being said, the case of interest in this work is when the indoor temperature, TI, is held at a constant
(5)
from which the monthly cooling energy usage can be calculated as = , where tmo is the number of hours in a month and WAC is measured in kilowatt-hours (kWh). Combining these equations and solving for R results in:
(6)
The calculation for R can then be completed by recognizing that the slope of the cooling portion of the CP curve is .
Note that the above calculation implies that TCP will be used for TI in the thermal model. In other words, we are assuming that heat flows into the structure whenever the outdoor ambient temperature goes above TCP and extra energy is consumed in order fo for r the cooling cooling system to remove the heat. Now that R and TI have been identified, the only unknown parameter is C, the thermal capacity. This parameter can be identified by remembering that = and using system identification techniques to find the pa parameters rameters of
Figure ure 3. A change-point point (CP) curve for an industrial facility in North America.
(7)
The implications of the above equation are significant. Namely, time series data of power consumption and outdoor temperature can be used to find the time constant (an (and d thermal resistance) of the thermal circuit. These data are readily available; time series of power consumption can be obtained from modern meters and outdoor temperature data can be obtained from weather station data, which is often available on the internet. int The assumptio assumption n made here is that the indoor temperature, TI, is held constant. Note that the transfer function H(s) is simply a first order desc description ription of time and temperature dependent power consumption. It should also be pointed out that the ap appropriate propriate value of TI cannot cannot be found by identifying H(s). More discussion of this important caveat will be included in the examples below. IV. APPLICATION EXAMPLES OF THE THERMAL CIRCUIT CONCEPT
The second thermal circuit parameter that can be obtained from the change point curve is TI, which should be equal to TCP discussed earlier. It is the intersection of the two lines of the change point curve and in this case is 15.2 15.2C. C. At this point, all we lack to complete the thermal circuit is the thermal capacity, C. As mentioned earlier, C can be obtained by finding the time constant of the thermal circuit and for this we require time series data of power consumption and outdoo outdoor r temperature. Unfortunately, we could not obtain time series power consumption for this facility, so we chose to assume a time constant and create time series power data. As we will demonstrate, it turns out that this is an instructive exercise. First, we assumed that the time constant of the thermal circuit is 2 hours. Subsequently, the thermal capacity can be calculated as follows:
= 100.1 100
The thermal circuit concept is a work in progress. The basic concepts and equations have been presented above and two application examples will now be provided. Mor More e case studies are now underway, but these two examples help to demonstrate the potential utility of the thermal model. A. Application 1: An Industrial Fa Facility The first example is an industrial facility in North America. A CP curve for the facility is shown in Figure 3. Using Equation quation (6) developed earlier and the slope of the cooling portion of the CP curve, the thermal resistance of this facility is:
This represents t the he thermal energy stored in the walls and ceiling of the structure for every degree of temperature rise. Of course, in this context, we are expressing the thermal capacitance in units that are meaningful for electric power systems calculations. Once C was established, we calculated the total power consu consumption mption of the plant. First, we generated the time series data for the cooling power consumed by the facility by calculating the heat energy flow using equation (3) above and then applying an assumed efficiency of 90%. Th This is was added to the temperature-independent temperature dependent power consumption of the facility; the latter calculated by dividing the change change-point energy (WC) by the number of hours in a month that the facility is in operation. Finally, a zero zero-mean mean random component was added to simulate the frequent varia variations tions in power consumption that would be typical of a manufacturing facility. These calculations were performed using the Simulink diagram shown in Figure 4.
= 0.0799 0799
where the efficiency of the cooling system was assumed to be 0.9. The physical physical meaning of this thermal resistance can be obtained by considering its inverse. Namely, for every degree Celsius of temperature difference between the outdoor temperature and the indoor temperature, heat energy will flow into the structure at a rate o of f 12.5kW (or 12.5kJ/s).
Figure 4. Simulink diagram used to calculate the total power.
B. Application 2: An Entire Distribution Feeder In this application, a simple thermal model is made to represent the group of structures served by an entire distribution feeder located in the Eastern Interconnection of North America. The data available for this example is the apparent power (MVA) from the substation recordi recording equipment and outdoor temperature recordings from a local weather station [5]. [5]. Power factor information was not available so it was assumed to be a constant 0.9. Figure 6 shows the temperature and power versus time for a 24 hour period, which was the amount of data available. Again, the delayed response of power consumption to changes in temperature is apparent.
Figure 5. The first 50 ho hours of f temperature and calculated total power.
Figure 6. Outdoor temperature and substation power for a summer day.
Figure 5 shows the resulting total power consumption for the facility, facility graphed along with the outdoor temperature. The time delay between outdoor temperature and the resulting variation in power consumption is readily readily apparent. Subsequently, Subsequently, system identification was performed using the Matlab system identification toolbox. The process model solution was used to determine the values of and for the first rst order model in E Equation (7) ( above. The value of was found to be 13.98 13.98, from which R is calculated to be 0.0795 C/kW. C/kW The value of was found to be 1.995 hr. hr., , which is very close to the known value of 2 hours. Of course, these were ideal simulated conditions so we would expect that our identification entification of R and would be very accurate. An interesting aspect of the total power consumption is the contribution of the indoor temperature, TI. When the indoor temperature is held constant, TI contributes a c constant onstant term to the temperature-depende temperature dependent nt power consumption (see Equation 3 and Figure 4). At the same time, the temperature-independent temperature independent power consumption also has a constant term and probably will in most circumstances. These two constants combine in the total power and cant be identified (separated) separated) by applying system identification techniques to the time series data. It appears that the only way to identify TI (and therefore the actual cooling load) is through analysis of the change change-point point curves, which provide a longer term view of energy consumption.
System identification techniques were used to find the parameters of the first order model of the form of Equation 7. The value of was found to be 306.31 kW/C and when the efficiency of the cooling systems was assumed to be 0.9, the calculated thermal resistance is 3.63 x 10-3 C/kW. This means that for every degree of outdoor temperature rise, an additional 276 kJ/s (kW) of heat power will flow into the structures served by the feeder. In addition, by system identification, the time constant, , was found to be 3.5 hours, which is longer than the 2 hour time constant we assumed for the industrial facility. Knowing , the thermal capa capacity city could then be calculated and was found to be 3834 kWh/C. Finally, the model with this gain, , and time constant, , was plotted alongside the original data used for system identification. The plot, in Figure 7, shows that the model describes a very ver significant amount of the variation in power supplied by the feeder. It is important to note that an appropriate constant term had to be added to the model, but that TI could not specifically be identified because changechange-point point data is not available for this example. Again, referring to Equation 3 3, we understand that the thermal energy transfer into the structures and corresponding power flow required to cool the structures can not be fully characterized without identification of TI.
fraction of the load will be useful in properly specifying dynamic load models, wherein the proper quantity and size of induction motors must be determined. The two applications of the thermal model demonstrated The that certain parameters of the thermal model can be identified identif using ng change-point change point curves or that certain parameters can be identified using time series data of power and temperature. However, to fully identify the model, both the change change-point curve a and nd time series data must be used. This work is preliminary, and future research will expand to applications where both types of data are available for the same facility. In fact, many such applica applications tions are available because the required data can be obtained from AMI meters or substation data recorders recorders, , both of which are commonly used. The high availabili availability ty of the required data is one of the potential strengths of a thermal model approach. REFERENCES
[1] [2] [3] Prabha Kundur, Power System Stability and Control, Control 1st st ed., ed. US: McGraw-Hill Hill Professiona Professional, 1994. EPRI, Load Modeling for Power Flow and Transient Stability Computer Studies, Volume 1: Summary Report, Report 1987. A. Meklin, J. Undrill, B. Lesieutre, W. Price, D. Chassin, R. Bravo, S. Yang, Load Modeling in Power System Studies: WECC Progress Update, IEEE Power Power and Energy Society General Meeting, Pittsburgh PA, 2008. S. Doty and W.C. Turner, Energy Management Handbook, 7th ed., GA: The Fairmont Press, 2009. www.ncdc.noaa.gov Michel Verhaegen, Vincent Verdult, Filtering and system identification : a least squares approach, 1st ed., London: Cambridge University Press, 2007. Lennart Ljung, System Identification: Theory for the Us User, er, 2nd ed., US: Prentice Hall, 1999. F.P.Incropera, D.P.DeWitt, T.L.Bergmann, A.S.Lavine, Fundamentals of Heat and Mass Transfer, 6th 6th ed., ed. US: John Wiley & Sons Inc., 2006. Michael J. Moran, Howard N. Shapiro, Fundamentals of Engineeri Engineering Thermodynamics, NJ: John Wiley & Sons, 2004.
Figure 7. The actual power consumption consumption of a feeder is plotted along with the power consumption described by a simple thermal model.
Of course, the thermal model does not fully describe the variations in power. Other variations in power consumption are apparent and these may be due to to time dependent customer behavior or other tempe temperature rature-independent pendent loads. But this result shows that the simple thermal model does a goo good d job of describing temperature temperature-dependent dependent variations in power consumption and f furthermore urthermore that the temperature-depend temperature dependent variation is a significant part of the overall load variation for the feeder. V. CONCLUSIONS
[4] [5] [6]
In this work, a simple thermal model was developed to describe the thermal energy entering a structure or set of structures. During summer power system operation operation, , the thermal energy is removed by cooling equipment within the structures. Thus, the development of the model provides a means of identifying the temperature temperature-dependent dependent portion of power consumption. This portion of the load is generally due to the fans, compressors, and pumps used in cooling equipment to remove heat. Thus, characterization of this
[7] [8] [9]
Self-healing Restoration of a Distribution System Using Hybrid Fuzzy Control/Ant-Colony Optimization Algorithm
Mohamed Zaki Abd El-Hamed
Arab academy for science and technology Cairo, Egypt eng.mzaki@hotmail.com
Walid El-Khattam
Ain Shams University Cairo, Egypt walid_el_khattam@hotmail.com
Rania El-Sharkawy
Arab academy for science and technology Cairo, Egypt raniasula@ hotmail.com
Abstract The restoration process for distribution system grid needs the operation of line switching to restore as many loads as possible for the faulted area. In this paper a multi objective, multi constraint combinatorial optimization problem is formulated to solve the self-healing restoration problem. A hybrid Fuzzy Control (FC)-Ant Colony Optimization Algorithm (ACOA) is proposed using values for P, Q from a SCADA system. Two case studies without/with distributed generation are carried out to evaluate the effectiveness and speed of the proposed algorithm. The obtained results are compared with the conventional ACO to illustrate the accuracy of the proposed hybrid algorithm. Finally, conclusions are discussed.
Index Terms Ant Colony Optimization Algorithm, Fuzzy Control, and Power System Restoration.
I.
INTRODUCTION
Over the past few decades, electrical distribution systems have received considerably less attention than transmission and generation systems. This is mainly due to the fact that transmission and generation systems are usually very capital intensive. Furthermore, any inadequacies in either, often lead to widespread catastrophic consequences. On the other hand, distribution systems are relatively inexpensive. However, large sums of money are spent collectively on such systems as they serve a huge number of customers. Recently, engineers have been equipped with the facilities to cope with the computational burden of the distribution system in performing accurate modeling and simulation to achieve a reliable continuous supply during steady state and system restoration after occurrence of disturbance events. There are many artificial intelligence methods used to restore the radial distributed system. In [1], results show that Artificial Neural Networking (ANN) can be used to solve the restoration problem with high processing speed and can be considered for real time application. Genetic Algorithm in [2] provided a list of switches that must be operated (open or close) in order to give the best solution for the system reconfiguration. The criterion used in the judgment is based on unsupplied loads, overload factor, overload transmission line number, and number of created loops. However, this procedure takes time. In [3], a Heuristic-based method was
used as a prototype for real-time knowledge-based restoration system integrating data acquisition, user interface, power system simulator and foreign function interface to a series of power system simulation programs. The hierarchical restoration plan generation procedure and the consideration of various system constraints make the system potentially useful for the bulk power system restoration. In [4], a Multi agent method is used to realize self-healing operation due to its parallel processing and autonomous decision capabilities. In [5], a Petri-net model is used where the knowledge/ configuration of the concern system in the form of graphic representation is expressed through a structured model. An expert system method has been developed to quickly determine restoration plans and build look-up tables for distribution personals [6]. Ant Colony Optimization Algorithm (ACOA) was introduced in [7] and [8], where the combination of the flexible interaction of agents and the heuristic search of ACOA makes the restoration strategy fast and a global optimal one. Simulation showed that the proposed model is robust for the changes of electric network topology and has the flexibility of treating various problems during restoration. In [9] and [10], Fuzzy Control (FC) was used to present not only the fault section candidates but also the section's possibility of being a fault as the form of degree of membership. Therefore an operator of power systems can diagnose a fault section taking into account uncertainties. Through applying threshold to the degree of membership, it can be easily implemented to substation automation problems. Therefore, ACO and FC methods were used separately to restore the radial power system failure efficiently where they can provide an effective assistant in the real time restoration and dealing with a complicated system as well. This paper proposes a new online strategy model for the distribution system restoration. The model combines two intelligent methods; ACO and FC. Each method worked perfectly alone with some disadvantages. Therefore, combining the two methods can make use of their advantages in improving accuracy of the restoration process and avoiding their disadvantages. Two case studies will be examined. The obtained results from the proposed algorithm will be compared with the ACO algorithm alone. Finally, conclusions are discussed.
II. DISTRIBUTION SYSTEM RESTORATION MODEL A distribution system restoration problem can be formulated as a multi-objective, multi-constrained, combinatorial optimization problem as follows: A. Objective functions The main objective functions in the restoration are maximizing the restored load to be severed (1) by minimizing the number of switching operation in the breakers (2). Stage-I: Maximize the capacity of served loads [7]:
Where; : The total reactive power from supply : The total reactive power demanded at the load : The total reactive power loss in the transmission line. 5) Feeder line current limits should not be changed. 6 Where; and : Minimum and maximum current level of the line respectively : The current level of the line j 6) Higher priority customers should always be supplied. III. DISTRIBUTION SYSTEM RESTORATION BASED ON ACOA AND FC A. ANT COLONY OPTIMIZATION ALGORITHM (ACOA) The ant colony optimization is a stochastic population based heuristic algorithm which is equivalent to the behavior of the ant or bee in there colony in the real life [11]. In the distribution system restoration ACO is used widely where the outage buses find its own feeder bus as follows [7] and [8]: 7
Where; : The load at bus no. M : The binary decision whether the load at bus M is restored or not ( = 1: resorted, = 0: not restored) D: The set of all de-energized loads. Stage-II: Minimize the number of switching [7]: min 2
Where; : Representation of the switching operation in bus M = 0: no change ( = 1: the switch state is changed, occurs in the switch state) B. Constrains The restoration problem constraints are; 1) Radial network structure should be maintained. 2) The voltage level for each bus should not be changed. 3 Where; and : Minimum and maximum voltage levels at bus M respectively : The voltage level at bus M 3) The Active power balance between supply and demand: 0 4
Where; : The probability of the ant to move from point i to j : The quantity of remnant pheromone on the trail from i to j : The desirability of the trail which is 1 distance : The parameters that control the relative importance of the trail pheromone versus the desirability of the trail The pheromone update for each iteration is used according to: 1 8
Where; : The evaporation of trail pheromone between i and j : The pheromone left on trail ij by current optimal solution 9 Where; is a constant and whole trip. is the tour length of the ant s to the
Where; : The total active power from supply : The total active power demanded at the load : The total active power loss in the transmission line. 4) The Reactive power balance between supply and demand: 0 5
The optimal solution is obtained; after a number of iterations, comparing the probability of each path with each other, and by choosing the one with the highest probability. This method has many advanced features than the other method [8] such as: 1. Easy to deploy and implement to solve many optimal problems. 2. Flexible for many problem formulations.
3. It can achieve the global optimal solutions not a local solution only. On the other hand, it has some drawbacks such as [12]: 1. Theoretical analysis is difficult 2. Sequence of random decisions (not independent) 3. Probability distribution changes by iteration 4. Research is experimental rather than theoretical B. FUZZY LOGIC CONTROL The Fuzzy logic control is developed to address the uncertainty of any system. The fuzzy logic control relations deal with uncertainties imposed on fault section diagnosis of power systems [13]. The fuzzy modeling uses fuzzy interfered rules to describe and express the problem of a complex non-linear system, where the rules for the fuzzy system are obtained either from domain experts or by observing the developer control of the actions, the membership functions takes its sets from the demined actions to be taken. The fuzzy logic controller consists [14-19]: fuzzification of the input data and mapping them on the scale rang which can convert these data into a suitable linguistic values; identification of the linguistic rules for the system recovery; the execution process of the rules is deployed which is equivalent to the developer decisions; and finally the defuzzification of the inputs to map them to a corresponding universe of discourse. IV.THE PROPOSED FUZZY CONTROL-ANT COLONY OPTIMIZATION ALGORITHM According to the advantages obtained from using the two methods (FC and ACO) individually, the two methods can be hybridized together in the power system restoration process to increase the accuracy of the operation. The proposed method is applied using MATLAB. Fig. 1 shows the proposed method where the FC can be used to deal with the uncertainty of the P and Q for all the buses surrounding the fault. FC is used to classify the P and Q in the membership function for each faulted bus, as shown in Fig. 2. Each one of the membership function for (P, Q, and S) in the FC consists of 3 trapezoidal sets (LOW, MEDIUM, and HIGH) to increase the accuracy. The obtained classified P and Q values are used to rank the buses based on the belonging degree of accepted limits to be the supplier for the fault. As shown in Fig. 1, the FC takes place in the pheromone detection in the ACO. Therefore, in this case the membership function in the FC takes its inputs from a SCADA system for P and Q in the faulted loads considering the impedance and the state of the switches between the faulted loads and there possible feeders. Then the FC takes the action for each nominated bus in restoring the faulted loads giving each feeder a percentage of acceptances according to the membership for
each input. The action of the FC replaces the action of each ant to produce a pheromone for each feeder. Thus, the iterations time of the ant to produce the pheromone is reduced in this action. The percentage of acceptance is equivalent to the pheromone and then ACO is used to make another reconfiguration for the acceptance in all the available buses of restoration. This proposed algorithm increases the percentage of the restoration, reduces the time of the restoration, and prevents the system to go in any blackout. The proposed hybrid FC-ACO algorithm is not only more robust, but can also achieve a better static and dynamic performance of the system.
Fig. 1 Flow chart for the FC Program Execution in the Proposed FC-ACO Algorithm
Fig. 2 Membership function of the fuzzy control
V.SIMULATIONS AND RESULTS To evaluate the proposed hybrid FC-ACO algorithm, it is applied on two systems. It is worth mentioning that in this paper, the load restoration is about 80% of the outage load, the binding constraints are (3) to (5). However, it is assumed that the feeders thermal capacity is not violated as in (6). In the mean while, no priorities for the load restoration priorities are given, assuming equal priorities. A- System-1 under study: The IEEE 14-Bus distributed system shown in Fig. 3 is used. The details of the system are given in Tables I-III. The faulted area takes place between Buses 3 and 4, where L4 is the out of service load. The possible buses to restore this fault were (Buses 2, 5, 7, or 9). In the mean while, a DG installed at Bus 8 can be used to help in the restoration process. Table I Brief summary of System-1 No. of Switches Total P (KW) Total Q (KVAR) 17 259,000 73,500 Bus No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 Table II Bus load data of System-1 KW KVAR 0 0 21700 12700 94200 19000 47800 3900 7600 1600 11200 7500 0 0 0 0 29500 16600 9000 5800 3500 1800 6100 1600 13500 5800 14900 5000 Line 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Table III Feeder data for System-1 From To Resistance Bus Bus (ohm) 1 2 0.19 1 5 0.54 2 5 0.57 5 6 0.14 5 4 0.13 2 4 0.58 2 3 0.47 6 11 0.95 6 13 0.66 6 12 1.23 12 13 2.21 13 14 1.71 11 10 0.82 14 9 1.27 10 9 0.32 9 4 0.14 9 7 0.001 4 7 0.14 4 3 0.67 7 8 0.001 Table IV Probability results for System-1 ACO ACO (1th iteration) (4th iteration) 9.31% 78.34% 39.25% 16.90% 25.72% 4.70% 25.72% 0%
Reactance (ohm) 0.59 2.23 1.74 0.75 0.42 1.76 1.98 1.99 1.3 2.56 2 3.48 1.92 2.7 0.84 0.75 1.1 0.75 1.71 1.76 Proposed FC-ACO 85.31% 20.46% 5.80% 0%
Feeding buses Bus 2 Bus 5 Bus 7 Bus 9
Table IV shows the results obtained from solving the restoration process using ACO alone and with the proposed hybrid FC-ACO algorithm. The results show the probability for each bus to make the restoration process for the out of service load (L 4). Buses 2 and 5 can be used to help in the restoration process with a high probability. Bus 7 can help in the restoration process by using the DG installed at Bus 8. Bus 9 cannot contribute to the restoration as its power is used to satisfy its load. According to the obtained results, it can be deduced that the restoration process by using only ACO will converge after 4 iterations. In the 1st iteration, the probability results for the nominated buses had large divergences which lead to wrong decisions from the dispatcher in the restoration. In the 4th iteration the saturation of the final result can be done by updating the pheromone for each path in the proposed system. On the other hand, the proposed FC-ACO reaches the required results in one iteration. B- System-2 under study: The proposed algorithm is applied to a system, Fig. 4, given by [8] and the obtained results were compared with applying ACO only. Tables V, VI, and VIII provide the system under
Fig. 3 System-1: Network after fault
study details. A fault is assumed to take place between Buses 9 and 10 and the out of service loads are the loads at Buses 10 and 11. The possible bus to restore Bus 10 is only Bus 5. The faulted Bus 11 has Buses 7 and 13 to be restored from them. Similar to System-1, the proposed FC-ACO reached the required result in one iteration with more accuracy than applying ACO alone. Furthermore, the proposed FC-ACO considers more than one input (KW, KVAR, Impedance between buses and the switching operation) and increased number of constrains which lead to more accurate results [20]. No. of buses 13 Table V Brief summary for System-2 No. of Nominal Total P switches voltage (KV) (KW) 10 11 2652 Total Q (KVAR) 866
Fig. 4 System-2: Network after fault
Bus no. 1 2 3 4 5 6 7 8 9 10 11 12 13 Line no. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Feeding buses Bus 7 Bus 13
Table VI Bus load data for System-2 KWATT KVAR 0.00 0.00 473 155 127 41 35 11 438 144 211 69 42 13 473 155 127 41 35 11 438 144 211 69 42 13 Table VII Feeder data for System-2 From To Resistance Bus Bus (Ohm) 1 2 0.148 2 3 0.044 3 4 0.028 4 8 0.16 8 9 0.029 9 10 0.053 10 11 0.059 9 12 0.038 12 13 0.037 4 5 0.06 5 6 0.034 6 7 0.032 7 11 0.047 5 10 0.042 11 13 0.056 Table VIII Results for System-2 ACO ACO (1th iteration) (3th iteration) 51.1% 47.33% 48.9% 52.67% Reactance (Ohm) 0.287 0.124 0.078 0.31 0.083 0.151 0.166 0.107 0.104 0.167 0.097 0.092 0.101 0.083 0.114 FC-ACO 47.21% 58.49%
VI. CONCLUSION All in all in this paper, a novel online hybrid FC-ACO algorithm is proposed to solve the distribution system restoration problem. The multi-function multi-constraint problem is solved using MatLab taking into consideration more than one input (KW, KVAR, Impedance between buses, and the switching operation) and increased number of constrains that result in accurate results. Two case studies were examined without and with DG in the system. The output response confirms the suitability of the proposed algorithm, compared with ACO alone, as a high-performance control system, and fast execution time as there are no iterations in producing the pheromone for each feeder to the faulted area. Finally, the proposed algorithm is characterized by easy and simple in modification to accommodate any distribution system configuration. REFERENCES
[1] A.S. Bretas and A.G. Phadke, Artificial neural networks in power system restoration, IEEE Transactions on Power Delivery, vol. 18, Issue: 4, Oct. 2003, pp. 1181-1186. Li Xiao, Xie Zhong, and Qiu Yuhui, Adaptivity implementation based on multi-agent technology, Proceeding of IEEE International Conference on Advanced Learning Technologies, 2004, 30 Aug.-1 Sept. 2004, pp. 146-150. W. Jiang and C.Y. Teo, A knowledge-based approach for bulk power system restoration, International Conference on Energy Management and Power Delivery, 1995, vol. 1, 21-23 Nov. 1995, pp. 108111. A. Zidan , E.F. El-Saadany , and A. El Chaar L., Cooperative AgentBased Architecture for Self-Healing Distributed Power Systems, IEEE Conference on Innovations in Information Technology (IIT) , 2011, pp. 100105. N.A. Fountas, N.D. Hatziargyriou, K.P. Valavanis, Hierachical timeextended Petri nets as a generic tool for power system restoration, IEEE Transactions on Power Systems, vol. 12, Issue: 2, May 1997, pp. 837843. L. Helal, H. Fathalla, M.M.S. Mansour, S. Al-Debieky Power system Restoration using expert system technique, International conference on Electrical, Electronic and Computer Engineering, 5-7 Sept. 2004, pp. 747-752. Li Ling, Liao Zhiwei, Huang Shaoxian, and Wang Gang, A distributed model for power system restoration based on ant colony optimization,
[2]
[3]
[4]
[5]
[6]
[7]
Transmission and Distribution Conference and Exhibition: Asia and Pacific, 2005. Rajneesh Karn, Yogendra Kumar, and Gayatri Agnihotrim, Development of ACO Algorithm for Service Restoration in Distribution system, International Journal on Emerging Technologies 2(1), 2011, pp. 71-77. H.J. Cho and J. K. Park, An expert system for fault section Diagnosis of power systems using fuzzy relations, IEEE Transactions Power Systems, vol. 12, no. 1, Feb. 1997, pp. 342-348. S. W. Min, J. M. Sohn, J. K. Park, and K. H. Kim, Adaptive fault section estimation using matrix representation with fuzzy relations, IEEE Transactions on Power System, vol. 19, no. 2, May 2004, pp. 842 848. Xin-she Yang,Nature-Inspired Metaheuristic Algorithms, 2nd edition, 2010, Luniver Press. V.Selvi and Dr.R.Umarani, Comparative Analysis of Ant Colony and Particle Swarm Optimization Techniques, International Journal of Computer Applications (0975 8887) vol. 5, no.4, August 2010. Hyun-Joon Cho and Jong-Keun Park, An Expert System for Fault Section Diagnosis of power system using fuzzy relation, IEEE Transactions on Power Systems, vol. 12, no. 1, February 1997, pp. 342 348. Ahmed M. El-Fallah Ismail and Rajiv Ranjan Tewari, Design a hybrid intelligent controller (fuzzy-based ant colony algorithm) for improving a tracking performance of actual output response of SEDC motor under the effect of the external disturbances, International Journal of Electrical, Electronics and Data Communication, vol. 1, Issue: 2, 2013, pp. 2320-2084. [15] Z.A. Zadeh, Fuzzy sets, Information and Control, vol. 8, 1965, pp. 338-353. [16] S.K. Pal et al., Fuzzy logic and approximate reasoning: An Overview, Journal of Institution of Electronics and Telecommunication Engineers, 1991, pp. 548- 559. [17] Y.H. Song and A.T. Johns, Application of fuzzy logic in power systems: Part 1 General Introduction to fuzzy logic, IEE power Engineering Journal, vol. 11, no. 5, 1997, pp. 219-222. [18] Y.H. Song and A.T. Johns, Application of fuzzy logic in power systems: Part 2 Comparison and Integration with expert systems, neural networks, and genetic algorithms, IEE power Engineering Journal, vol. 12, no. 4, 1998, pp. 185-190. [19] J. Nanda and B.L. Kaul, Automatic Generation Control of An Interconnected Power System, Proceeding of IEE, vol. 125, Issue: 5, May 1978, pp. 385-390. [20] Hamid Falaghi, Mahmood-Reza Haghifam and Chanan Singh, Antcolony optimization-Based method for placement of sectionalizing switches in distributed networks using a fuzzy multiobjective approach, IEEE Transactions on Power Delivery, vol. 24, no. 1, January 2009, pp. 268-276.
[8]
[9]
[10]
[11] [12]
[13]
[14]
Overhead Lines Maintenance Scheduling in Power Distribution System Considering Feeders Ranking
Hadi Ghasemabadi
Department of Electrical Engineering, Faculty of Engineering
Ali Peiravi
Department of Electrical Engineering, Faculty of Engineering
Ferdowsi University of Mashhad Mashhad, Iran Ghasemabadi.hadi@gmail.com

Abstract This paper presents a methodology including mixedinteger linear programming formulation for long term maintenance schedule of overhead lines considering feeders ranking. Important feeders are extracted from a ranking list by basic indices such as the type of costumers, loads, length, extent as well as failure rates of feeders. It is necessary to consider importance of feeder in overhead lines maintenance scheduling. In other words, feeders ranking would enable distribution planners and asset managers to have a high accuracy investment decision which can be made effectively towards maintenance scheduling. The proposed methodology is tested on 11-feeder the Roy Billinton Test System which revealed the accuracy and efficiency of the approach. Distribution KeywordsMaintenance Scheduling System; Feeders Ranking;
Ferdowsi University of Mashhad Mashhad, Iran peiravi@um.ac.ir
Nif Pi Extent
Number of intervals of the stair wise failure-rate. Points obtained by feeder i due to extent index between others feeders. Points obtained by feeder i due to failure rate index between others feeders. Points obtained by feeder i due to length index between others feeders. Points obtained by feeder i due to peak load index between others feeders. Points obtained by feeder i due to type of Costumers index between others feeders. Total points obtained by feeder i due to several indices. Penalty cost due to planned undelivered energy on account of the nth maintenance task of feeder i. (e.g., minor and major maintenance and tree trimming). Mandatory reliability index for important feeders (j), extracted from feeders ranking list in year t. Mandatory reliability index for unimportant feeders (z), extracted from feeders ranking list in year t. Mandatory reliability index for important feeders (j), extracted from feeders ranking list in year t. Mandatory reliability index for unimportant feeders (z), extracted from feeders ranking list in year t. Period of maintenance scheduling.
Pi FR
Pi Length
Pi Load
AL i , k
NOMENCLATURE Average load of the nth customer Average selling price of energy.
PiType
APi DFR (m ) DFR (Initial )

dr I in (t )
n i n i
mth stair wise step of the nth decoupled failure rate of feeder i
Initial value of the nth decoupled failure rate of feeder i. Discount rate. Binary decision variable; equal to 1 if the nth maintenance activity preformed on feeder i in year t and 0 otherwise. Load increase rate Maintenance budget in year t.
PiTotal PC in
SAIFIimp (t)
LI MB (t ) N imp N uimp NC Ndf NF
SAIFIuimp (t)
Number of important feeders Number of unimportant feeders Total number of customers. Number of decupled failure rates. Number of feeders.
SAIDIimp (t)
SAIDI uimp (t)
WH (t )
Working hours available for minor and major maintenance or repair in year t.
I.
INTRODUCTION
WH tt (t )
Working hours available for tree trimming- or vegetation-related failure repair in year t. nth decoupled failure rate of feeder i in year t . Average planned and unplanned outage durations of kth customers due to nth maintenance task or repair of the feeder i, respectively. Total cost of labor and material for the nth maintenance activity and nth fault occurred on feeder i, respectively. Materials price required for the nth maintenance activity and repairing the nth fault occurred on feeder i, respectively. Total cost associated with nth maintenance activity and nth fault occurred on feeder i in period t, respectively. One working hour needed for the nth maintenance activity and for repairing the nth fault occurred on feeder i, respectively. Number of customers affected due to the nth maintenance activity and the nth fault occurred on feeder i, respectively. Planned and unplanned energy cost for feeder i, respectively. Revenue lost due to planned and unplanned undelivered energy on account of the nth maintenance task or repairing the nth fault occurred on feeder i, respectively. Undelivered energy due to the nth maintenance activity and fault occurring on feeder i, respectively. Working hours needed for the nth maintenance activity and repairing the nth fault occurred on feeder i, respectively. ABBREVIATION
in (t )
,UO , PO AD in, k , AD in, k
C in , M , C in , R
C in , MMat , C in , RMat
MC in (t ) , FC in (t )
LC n , M , LC n , R
NC in , M , NC in , F
PC i , UC i PC in , M , UC in , R
UD n , M , UD n , F
W H in , M , W H in , R
MILP SAIDI SAIFI TBM
Mixed Integer Linear Programming; System Average Interruption Duration Index; System Average Interruption Frequency Index Time Based Maintenance
The frequency of service interruptions and the many undesirable consequences of such interruptions can be reduced with effective maintenance policies. Probability reduction of failure equipment and extension of its lifetime is the purpose of maintenance activity [1]. One of the most maintenance intensive parts of power systems is distribution networks. However, generation and transmission have received more attention than distribution systems, considering the published paper in the literature [2]. Type, target as well as timing of the maintenance activities have a significant role in allocating maintenance scheduling resource. Strategy of maintenance can be classified into corrective and preventive maintenance [3]. Corrective maintenance repairs or replaces failed components and known as run to failure, while preventive maintenance is a proactive effort to improve the condition of equipment that performed before a failure occurs or may be deteriorated to some degree. The state-of-the-art in preventive maintenance offers at least four basic approaches: 1) time-based maintenance; 2) conditionbased maintenance; 3) reliability-centered maintenance; 4) risk-based maintenance [4]. Risk-based maintenance is the latest approach utilized for distribution network maintenance scheduling in which considers the risk of each equipment failure and its consequences and provides a more realistic model [5], [6]. Experience has shown that distribution feeders have an important role in improving the reliability of the distribution system. Therefore, maintaining the desired level of system reliability of feeders needs to be periodically inspected and repaired, but performing such plans are costly and budget is limited for doing this. According to the privatization of the electricity industry, competitiveness between distribution utilities leads to improve the power quality and to provide better service to customers and their satisfaction. Feeders ranking which consists of type of costumers, loads of feeders, length of feeders and extent and failure rates of feeders are necessary to achieve the optimal long term overhead line maintenance scheduling in power distribution system. The methodology followed here is based on [7] but include considering importance of feeders in satisfying the mandatory reliability constraints. In this paper we apply a risk-based maintenance scheduling in a way that important feeders which could be identified by the use of feeders ranking in distribution system have the least frequency and time duration of outages. The paper is organized as follows. Section II explains how to ranks feeders and which feeder has a top rated in indices of feeders ranking. Section III presents a description of the riskbased preventive maintenance in the presence feeders ranking, while display of the MILP formulation considering importance of feeders which extracted from feeders ranking list is discussed in section IV. Numerical results are given in section
V too. Ultimately, Section VI presents some conclusions that can be drawn from this study. II. FEEDERS RANKING
Optimal operation and increased reliability of distribution networks requires regular periodic maintenance and networks with new technologies. There is no possibility to implement new methods such as automation for all the feeders on the network at one stage due to wide distribution networks, large number of feeders, existing economic as well as operation constraints. Identifying the important feeders is the result of logical feeders ranking which enable us to enhance the accuracy of decision making and implementation of automation. It is clear that important feeders which are extracted from the ranking list have higher priority than other feeders from the standpoint of sensitivity, operation, plans to reduce outages, improve reliability and implementation of automation. Hence, it is essential to obtain a ranking list of feeders [8][10]. Type of costumers, loads, length, extent as well as failure rates of feeders is most important indices in the ranking list of them [11].
A. Type of Costumers Some feeders supply sensitive locations such as political, security and government places, hospitals, industrial centers, factories and or commercial centers. Outages are important issues in these feeders, too. Therefore, type of costumers is one of operation indices for formation feeders ranking list. B. Load of Feeders Operators have always tried to hold balance the loads of feeders because the load is also an important factor in the operation. Here, feeder Number 8 with 6.518MW has the highest point among other 11 feeders; hence they will be normalized by it. C. Length of Feeders Feeders with large length have always created the most problems in the operation. In addition to large length these feeders, they are eventful and full loaded. According to the previous efforts, it can be seen that the ones maximum volume operation often is for this type of costumers. Feeder Number 8 with 3.8km in this index has the highest rank too. D. Extent of Feeders Extent of feeder is important due to the length of feeders or numbers of customer of each feeder into their consumption. For example, it may be the case that two feeders have the same load but numbers of customer of each feeder are more different according to its extent. Since regular service to the most of the customers is expected, extent of feeder and its point have direct relation which means greater extent causes arising in point and reverse.
E. Failure Rate of Feeders Random and deteriorating failures are two general categories in component failures. Aging process, such as chronological aging, thermal aging, and mechanical wear can cause deteriorating failures. However, the severe weather, harmful animals, etc. is failures which are named random failures. Exponential distribution and constant failure rate models random failures while Weibull or normal distribution model deteriorating failures. Since the rate of failure is constant in random failures, maintenance activities cannot produce any improvement to random failures. Anyway, failure rates originated from deterioration are time varying and maintenance activities will bring about improvement [1].
It should be noted that another importance index could be considered in feeders ranking list such as restructuring of network, but it has not effect on total point of feeders here because of the network topology. III.
A.
RISK BASED PREVENTIVE MAINTENANCE CONSIDERING FEEDERS RANKING
Decision tree Equipment in distribution system has a finite life because failure rate of them increase continuously which could be prevented by use of risk based maintenance. Inspection, eliminating minor faults, major line overhaul and tree trimming are maintenance activities on overhead distribution in this research work.
An appropriate mathematical model linking distribution system cost, maintenance and reliability is developed in the fourth step. The markov and tree decision are two common models in linking reliability and maintenance among others. In this article providing a dynamic overview of long term period besides, displaying appropriate prone to aging is needed, since markov model is not able to supply our requirements, tree decision model is used. In addition, it has more advantages which as explained in references [6], [7]. Fig. 1 is shown the decision tree model. Maintenance and repair costs are basis the transition rate between different states in this figure. Number of maintenance tasks determines the branches number of decision tree.
tt = tree trimming, veg = vegetation Fig. 1. Decision tree of one overhead line for maintenance task [7].
If n represents the number of possible maintenance tasks (i.e., minor, major and vegetation maintenance), the number of decision tree branches is equal to 2n. Total costs of maintenance tasks and equipment failure are costs that can be imposed to the system. Sum of these costs is the transition rates between different states of maintenance decision tree model. In Fig.1, it is assumed that one, two and three years are elapsed from the last time maintenances activity. In this figure, Minor, Major and Veg are set of decoupled failure rates of the overhead line states. They could be reduced to their initial value if related maintenance is performed and could be increased if related maintenance is not performed. IMinor , IMajor and Itt are binary decision variables and determine which maintenance scenario has been selected for equipment in each period. It should be noted that binary variable is equal to 1 if maintenance activity is performed and zero otherwise.
B. The Costs Imposed to Distribution System Location of overhead lines and topology of network have the main role in imposed costs to distribution system. Undelivered energy due to failure or maintenance and labor and material cost of maintenance or repair are two terms of costs which imposed to distribution system.
because does not require switching of the line. The total costs related to the nth maintenance tasks and fault occurring on feeder i in period t can be determined as follows due to costs that are mentioned above, respectively.
MC in (t ) = ((C in , M ) + (PC in + PC in , M )
(1 + LI )t 1 )
(11) (12)
FC in , F (t ) = ((C in , R ) + (UC in + UC in ,R )
(1 + LI )t 1 )
IV. CONSIDERING FEEDERS RANKING LIST IN LONG TERM OVERHEAD LINES MAINTENANCE SCHEDULING FORMULATION BY USE OF MILP MILP guaranties convergence to the optimal solution in a finite number of steps. The objective function of long-term overhead line maintenance scheduling problem is formulated as follows [7]: Minimize
T NF Ndf
Revenue lost for distribution utilities and penalty costs due to customer damage are several types of costs undelivered energy to customers during an outage and either distribution utilities should be bears them. The penalty that should be paid by distribution utilities is different for type of customers (e.g., residential, small user and commercial, etc.) and planned and unplanned undelivered energy. Planned and unplanned undelivered energy costs and penalty cost that should be paid by distribution utilities due to nth maintenance activity or fault occurring on feeder i and planned and unplanned revenue lost by distribution utilities can be formulated as follows, respectively:
((MC
t =1 i =1 n =1
n i
(t ) I in (t ) + FC in (t ) in (t ))
(13)
(1 + dr ) )
(MC
i =1 n =1
NF Ndf
n i
(t ) I in (t ) + FC in (t ) in (t )
(14)
MB(t)
NF
NF
Ndf
i =1 n = Minor , Major
(W H in , M I in (t ) +W H in , R in (t ))
(15)
W H (t )
UD
n ,M i
= =
NC in ,M
k =1
, PO ALi ,k AD in, k
(3)
i =1 n =TreeTri min g
(W H in , M I in (t ) +W H in , R in (t ))
(16)
UD
n ,F i
NC in ,F
k =1
WHtt (t )
,UO ALi ,k AD in,k
n ,M i n ,F i
(4) (5) (6) (7) (8)
n = Minor , Major
I in (t ) 1
(17) (18)
(19)
PC = PC i UD
n i
n i
UC = UC i UD UC
n ,R i
in (t ) DFR in (Initial ) I in (t )
in (t ) DFR in (m ) 1 I in (t r )
r =0
PC in , M = APi UD in , M
= APi UD
n ,F i
m 1
The labor and material costs are two main costs of repair and maintenance activities at each period that can be determined as follows.
C in , R = C in , RMat + LC n , R W H in , R C in , M = C in , MMat + LC n , M W H in , M
i = 1,..., NF
t = 1,...,T .
(9) (10)
It should be note that planned undelivered energy cost is not considered for tree trimming in maintenance activity cost
As it is presented in the above formulations, the goal of the objective function is to minimize the total cost, which is defined as the sum of the maintenance task costs and fault occurring cost within some constraint. In Eq. 13 binary decision variable (Iin(t)) is multiplied by maintenance task cost. This binary variable determines whether the nth maintenance
task should be performed on feeder i in period t or not and continuous variable (in(t)) that represents the nth decoupled failure rate of feeder i in each period multiplied by fault occurring cost. The blocks of constraints (14), (15) and (16) represent the budget and labor constraints in period t, respectively. Different maintenance tasks cause labor constraints of tree trimming maintenance tasks are considered individually. Constraint (17) depends on decision maintenance policy of distribution utility. If they dont want minor and major maintenance to perform on each feeder in the same year, this constraint is satisfied. MILP constraints (18)(19) are proposed to link failure-rate variables and maintenance tasks binary decision variables. These constraints restrict the failure-rate decision variable according to the last time corresponding maintenance activity was performed. If maintenance task preformed in a special year, its failure-rate decreases. For more information, refer to [7]. In section II feeders are divided in two categories being important and unimportant. The aim of this division is considering higher reliability for important feeders than other feeders because of the outage of important feeders' imposed more cost to distribution utility. Applying importance rate of feeders which arise from ranking list for satisfying the mandatory reliability in long term overhead line maintenance scheduling formulation is novel idea in this paper. Impact feeders ranking has been considered on two mandatory reliability indices, SAIFI and SAIDI. It is noteworthy that reducing SAIFI and SAIDI leads to increase reliability. For this purpose, parameters SAIFI and SAIDI have specified less value than unimportant feeders whereby obtain higher amount of reliability of important feeders. SAIFI and SAIDI related to important feeders and other feeders are formulated as follows:
N uimp Ndf
z =1
NC AD n , PO z ,k NC n =1
n ,M i
k =1
I zn (t )
UO NC AD zn,,k + NC
n ,F i
k =1
zn (t ) SAIDIuimp (t)
(23)
z = Number of unimportant feeders
V.
NUMERICAL RESULTS
The proposed approach is applied to a case study based on the 11-feeder RBTS distribution network [12]. Moreover for simplicity just overhead lines are considered in this approach while all other devices can be considered. All data in this paper are taken from [7], [12]. Indices presented in sections A to E dont have the same effect in formation of ranking list. The weakness and severity of indices are corrected with the mentioned coefficients. If these coefficients which are relative and dependent completely on each other change logically, the same result will be achieved, while there are some exceptions. This formulation is obtained from empirical and industrial experts as following [11]. Total point of feeders = 0.15 PiType + 0.15 Pi Load + 0.1 Pi Length + 0.1 Pi Extent + 0.4 Pi FR + 0.1 restructure points Points of each feeder regarding to different indices and total point of them considering coefficients are summarized in Table 1. In this Table, feeders' number 8, 5, 7 and 11 with total points 82.34, 78.61, 74.7 and 74.17 have specified higher points between other feeders, respectively. These feeders are identified as important feeders in current distribution system. Hence, they have most priority than other feeders in operation, maintenance scheduling, applying new methods for reduce outages, etc. Identification of important feeders is performed by ranking list and using it in the maintenance scheduling problem that total cost is minimized leads to achieve more efficient and accurate maintenance scheduling, whereas constraints of labor and budget and reliability are satisfied. (2)
N imp Ndf

j =1 n =1
NC jn , M NC
I jn (t ) +
jn (t ) NC SAIFIimp (t ) NC jn , F
I jn (t )
(20)
N imp Ndf
NC AD n , PO j ,k NC j =1 n =1
n ,M i
k =1
UO NC A D jn,,k + NC
n ,F i
k =1
jn (t ) SAIDIimp (t)
(21)
TABLE I.
FEEDERS RANKIN LIST WITH THEIR INDICES POINTS
NF
8 5 7 11 4 1 3 10 6 9 2
PiType
55.65 52.17 51.83 38.96 99.48 100 83.13 44.52 44.52 44.52 29.68
PiLoad
100 87.51 86.4 89.7 84.52 91.04 77.58 87.52 87.52 75.02 53.7
PiLength
100 97.37 93.42 94.74 76.32 75 76.32 61.84 57.89 56.58 35.53
PiExtent
89.9 100 97.19 95 81.15 73.99 88.57 63.69 59.63 67.99 59.57
PiFR
100 94.81 87.27 89.75 58.24 56.25 58.24 38.24 33.52 32.01 12.62
PiTotal
82.34 78.61 74.7 74.17 66.64 66.06 63.89 47.66 44.97 43.19 27.07
j = Number of important feeders
N uimp Ndf z =1
NC zn , M NC zn , F I zn (t ) + zn (t ) NC NC n =1
(22)
SAIFIuimp (t )
Due to the ranking list, feeders' number 8, 5, 7, 11 are important feeders (the feeders which line drawn around them in table 1) and other feeders have less importance and are called unimportant feeders here. The objective function in this case study optimizes maintenance strategy for a 5-year-time horizon. The optimal minor, major and tree trimming maintenance task schedules obtained are presented in Table II. When maintenance task is performed in t year, IMinor, IMajor and IVeg equal 1. It should be noted that SAIFIimp and SAIDIimp constraints are assumed to 0.5 f/cus./yr and 1.25 hr/cus./yr, respectively. SAIFIuimp and SAIDIuimp constraints are assumed to 1 f/cus./yr and 2.5 hr/cus./yr per 5 year time horizons. Amount of reliability indices of important feeders are halved compared to other feeders. This work has been done in order to increase the reliability of important feeders where following it increases the reliability of the system.
TABLE II. NF 1 2 3 4 5 6 7 8 9 10 11 year A B C + + + + + + + + + + 1 2 + + 3 4 5 + + + + + + MAINTENANCE SCHEDULES OF THE FEEDERS FOR FIVE YEARS A B C A B C A B C A B C + +
TABLE IV. COMPARING TBM METHOD COST AND PERPOSE MILP METHOD CONSIDERING FEEDERS RANKING COST Maintenance 1/1 1/2 1/3 1/4 1/5 method TBM ($) MILP ($) 506170 368736 335541 326593 338714 351528
1/n denotes one maintenance activity performed per n years
VI.
CONCLUSION
Incurring a lower amount of cost and better network performance is the main aim of maintenance in the distribution systems. On the other hand, some feeders are more important than others from the point of several indices such as type of costumers, loads, length and extent as well as failure rates of feeders. In this study, importance of the feeders were considered in the overhead lines maintenance scheduling so that, important feeders were specified a higher amount of reliability according to their importance ranking. Cost of this methodology was compared with TBM approach. It showed that, not only the higher amount of reliability was considered for the system, but also fewer amounts of costs achieved in comparison with TBM method. REFERENCES
[1] J. Endrenyi et al., The present status of maintenance strategies and the impact of maintenance on reliability, IEEE Trans. Power Syst., vol. 16, no. 4, pp. 638646, Nov. 2001. [2] R. Billinton, M. Fotuhi-Firuzabad, and L. Bertling, Bibliography on the application of probability methods in power system reliability evaluation 19961999, IEEE Trans. Power Syst., vol. 16, no. 4, pp. 595602, Nov. 2001. [3] R. Billinton and R.N. Allan, Reliability Evaluation of Engineering Systems. New York: Plenum, 1992. [4] Risk based resource allocation for distribution system maintenance, Final project rep., 2006. [Online]. Available: www.pserc.org. [5] Y. Li, S. Yeddanapudi, J.D. McCalley, A.A. Chowdhury, and W. Jewell, Resource management for distribution system maintenance using optimized risk reduction, presented at the 9th Int. Conf. Probabilistic Methods Applied to Power Systems, Stockholm, Sweden, Jun. 2006, KTH. [6] A.D. Janjic and D.S. Popovic, Selective maintenance schedule of distribution networks based on risk management approach, IEEE Trans. Power Syst., vol. 22, no. 2, pp. 597604, May 2007. [7] A. Abiri-Jahromi, M. Fotuhi-Firuzabad, and E. Abbasi, An efficient mixed-integer linear formulation for long-term overhead lines maintenance scheduling in power distribution systems, IEEE Trans. Power Del., vol. 24, no. 4, pp. 20432053, Oct. 2009. [8] N. Ozay, A.N. Guven, Design and implementation of a feeder automation systems for distribution, International Conference on power tech. Budapest 99, No. 99EX376, pp. 122-126, 1999. [9] S.R. Southerlin, New distribution automation switchgear and its contribution to overhead distribution reliability, IEE Conference on Electricity Distribution, vol. 4, pp. 32/1-4, 1997. [10] T. Newman, System automation in power distribution, IEE North Eastern center power section symposium, pp.9/1-9/6, 1995. [11] M. Sadeghian, A. Afshar, G. Gharehpatian and S.O. mousavi, 20kv feeders ranking distribution network city of sari in order to optimal operation and automation, 9th Electric Power Distribution Conference., Apr. 2004 (in Persian). [12] R.N. Allan, R. Billinton, I. Sjarief, L. Goel, and K.S. So, A reliability test system for educational purposes basic distribution system data and results, IEEE Trans. Power Syst., vol. 6, no. 2, pp. 813820, May 1991.
A = IMinor , B = IMajor , C = IVeg + = Maintenance task is preformed
The budget, labors, and reliability constraints for important and unimportant feeders' solution for the five-year time horizon with the 0.07% duality gap are also shown in Table III. The best solution of total cost over the five years' time horizon is U.S.$322376.8 based on the MILP approach considering importance of feeders in maintenance scheduling. To show the efficiency of this approach, its cost compared to TBM strategy cost is listed in Table IV.
TABLE III. Constraint WH (hr) WHtt (hr) SAIFIimp (f/cus./yr) SAIDIimp (f/cus./yr) SAIFIuimp (f/cus./yr) SAIDIuimp (f/cus./yr) CONSTRAINTS FOR IMPORTANT AND UNIMPORTANT FEEDERS year
1 2 3 4 5
Limit
110 100 0.5 1.25 1 2.5
84.76 87.25 0.422 1.244 0.087 0.291
90.76 96.81 0.435 1.246 0.182 0.399
22.51 72.27 0.48 1.173 0.092 0.319
23.56 96.24 0.472 1.199 0.091 0.327
12.74 87.76 0.305 1.059 0.099 0.342
$ = dollar, hr = hour, f = failure, cus = customer, yr = year
Speed Control of Induction Motor Based on Model Reference Adaptive Control Using Experimental Approach
Hakan Ackgoz
Dept. of Electrical Science Kilis 7 Aralik University Kilis, Turkey Email: hakanacikgoz@kilis.edu.tr
O.Fatih Kececioglu
Dept. of Electrical and Electronics Kahramanmaras Sutcu Imam University Kahramanmaras, Turkey Email: fkececioglu@ksu.edu.tr
Ceyhun Yildiz
Electrical Power Resources Survey and Development Ankara, Turkey Email: cyildiz@yegm.gov.tr
Mustafa Sekkeli
Dept. of Electrical and Electronics Kahramanmaras Sutcu Imam University Kahramanmaras, Turkey Email: msekkeli@ksu.edu.tr
Abstract In recent years, adaptive and intelligent control schemes have begun to take the place of conventional control methods. In this study, a model reference adaptive control (MRAC) scheme based on genetic algorithm (GA) and Fuzzy Logic (FL) has been used by the help of Vector Control type driving method based on Space Vector Pulse Width Modulation (SVPWM). In this control technique, the conventional PI controller and the proposed control strategy has been applied to induction motor in real conditions by using an appropriate digital signal processor and peripheral units. As digital processor, DSPIC 30F6010 which is produced by Microchip Co. has been used. Experimental studies have been implemented for both of the control techniques under the same conditions. The results obtained from conventional PI controller and the proposed control strategy have been analysed and compared. Keywords- Induction motor, Fuzzy adaptive control, Fuzzy logic controller, Genetic algorithms, Model reference adaptive control, Vector control
Armature and excited winding currents of self-excited DC motors can be independently controlled because they are vertical to each other. This is not the case in induction motors. In the FOC, induction motor can be controlled as a separately excited DC motor [1-6]. Since FOC was introduced in 1971 [6], a great number of articles have appeared in the literature to improve the high performance. These methods are fuzzy logic controller (FLC) and various adaptive controllers such as MRAC, sliding-mode controller, variable-structure controller, etc [7-9]. Before control studies, genetic algorithms, which are the stochastic search methods based on the natural process of evolution, were carried out and implemented in Matlab environment. Afterwards, the fuzzy logic based adaptive control mechanisms have been developed. Classical fuzzy logic controller design is based on the trial-error methods and the expert information. Therefore, this method takes a longer time and the design is not always feasible to reach best outcomes. This method can be applied to the simple fuzzy logic controllers, but it is not convenient to complex parameters. In this case, different methods are used for the fuzzy logic controller design [10-14]. In recent years, GA has been used to design the fuzzy logic controllers. Thus, the ability of finding the best solutions is combined with effectively process structure of the fuzzy logic controllers [1012]. Here, an adaptive control system is preferred in order to realize the speed control. By using fuzzy logic controller based on the model structure, a speed control system is established. The adaptive control system is an effective solution to solve speed control for vector controlled induction machine. Required control performance is represented with a reference model in the control system.
I. INTRODUCTION Induction motors have many applications in the industry. Because they are simple, cheap, robust, reliable, durable and they need little maintenance. However, induction motors have some disadvantages such as mainly controllability due to its complex mathematical model and its nonlinear behavior [1-4]. In recent years, by the rapid development of microprocessor and power semiconductor technology, controlling methods of induction motors have been improved. In order to achieve high performance and better controllability, the field oriented control (FOC) has been investigated. DC motors have high performance in terms of dynamic behaviour and their control is simple.
Design of FLC based on the comparison of actual system output and reference model is quite difficult. GA is preferred to solve this problem. The best solutions are obtained in Matlab by characterizing the problem. This paper deals with MRAC based on GA and fuzzy logic controller (GA+FMRAC) in speed control of vector controlled induction motor. An inverter-controlled induction motor is realized in the laboratory. Then, the comparison of the drive response under conventional PI and proposed control technique has been demonstrated by several graphics on a number of different reference speeds. After that, the performance of proposed control technique has been researched and compared with conventional PI controller. II. VECTOR CONTROL AND MATHEMATICAL MODEL The d-q equations of 3-phase induction motor expressed in the stationary reference frame are [1-2]:
Vector control methods which are done transform of axis have been developed. Thus, flux and torque of induction motors can be independently controlled [15-17]. III. FUZZY LOGIC CONTROLLER AND GENETIC ALGORITHMS
A. Fuzzy Logic Controller Fuzzy Logic Control is an appropriate method for designing nonlinear controllers via the use of heuristic information [18]. A FLC system allows changing the control laws in order to deal with parameter variations and disturbances [19-23]. The block diagram of Fuzzy PD system for vector controlled induction motor is given in Fig. 1. The fuzzy PD controller has two inputs and one output. The inputs of FLC are speed error e(k) and change in speed error e(k). These inputs are normalized to obtain error e(k) and its change e(k) in the range of -1 to +1. The fuzzy membership functions consist of seven fuzzy sets: NB, NM, NS, Z, PS, PM, PB as shown in Fig. 2. The rule base of FLC system is given in table 1. There may be 77=49 possible rules in the matrix. The fuzzy PD controller action can be written in the form of K-domain equations as follows:
V ds Rs ids p ds
V qs Rs iqs p qs 0 Rr idr p dr wr qr 0 Rr iqr p qr wr dr
The flux linkage equations:
(1) (2) (3) (4)
u (t ) Kp.e(t ) Kd
de(t ) dt
(e(k ) e(k 1)) Ts
(1)
qs Ls iqs Lm iqr
(5) (6) (7) (8)
u (k ) Kp.e(k ) Kd
(2)
ds Ls ids Lm idr
qr Lr iqr Lm iqs
Where; Kp and Ki are gain factors, Ts is sample time.
dr Lr idr Lm ids
e(k) z -1
-+
G1 G2 Fuzzy PD Controller G3
Electromagnetic torque in the stationary reference frame is given as:
z -1
++
3 P Lm ds iqs qs ids Te 2 2 Lr
(9)
e(k)-e(k-1)
u(k)
Figure 1. Fuzzy PD controller system

NB 1 NM NS Z PS PM PB
Where; p= (d/dt), Rs, Rr are stator and rotor resistances; Ls, Lr, Lm are stator, rotor and mutual inductances; ds, qs are stator flux in d-q frame; dr, qr are rotor flux in d-q frame; ids, iqs , iqr, iqr are stator and rotor currents in d-q frame, wr is rotor speed and P is the pairs of poles. Vector control of induction motors are based on control principle of DC motors. DC motors have high performance in terms of dynamic behaviour. Armature and excited winding currents of self-excited DC motors can be independently controlled because they are vertical to each other. However, there is not such case in induction motors [1-3]. Made studies on induction motors showed that these motors could be controlled such as DC motors if three-phase variables are converted to dq-axis and dq-axis components are controlled.
0.8
0.6
0.4
0.2
0 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
Figure 2. Membership functions for inputs and output
TABLE I.
RULE BASE
IV. PM NS Z PS PM PM PB PB PB Z PS PM PB PB PB PB
FUZZY ADAPTIVE CONTROL
de e NB NM NS Z PS PM PB
NB NB NB NB NB NM NS Z
NM NB NB NM NM NS Z PS
NS NB NM NS NS Z PS PM
Z NB NM NS Z PS PM PB
PS NM NS Z PS PS PM PB
Adaptive control takes an important part in modern control systems. While control system is running, it works against to various disturbances, parameter variations and changes in the operating conditions. The success of the first adaptive control has appeared with the realization of a time-varying mass control of the rocket and spacecraft. One of the direct adaptive control methods is MRAC. This mechanism has been proposed to be performed in aircraft control. The adaptation of these control systems is performed by comparing the system output and reference based model output. Thus, an error signal (em) is obtained. The signal from the output of the adaptation mechanism is used for the parameter or signal adaptation. The block diagram of a MRAC system is given in Fig. 4.
Reference Model
B. Genetic Algorithms GAs are a stochastic global search method that mimics the process of natural evolution. It is one of the methods used for optimization. GAs are reliable and robust methods for searching solution spaces. GAs were introduced by John Holland in 1970. GAs have three main stages such as reproduction, crossover and mutation. GAs starts with no knowledge of the correct solution and depends wholly on responses from its environment and evolution operators such as reproduction, crossover and mutation to succeed at the best solution. By starting at several independent points and searching in parallel, the algorithm avoids local minimum and converging to sub optimal solutions [11-14]. Genetic algorithm process flowchart is shown in Fig.3.
ym
Adaptive Mechanism
uA
em
+ -
ur
y
Controller Plant
Figure 4. MRAC system
Create/Initialize Population
Measure/Evaluate fitness
Select Fittest
Mutation
Crossover/ Production
A fuzzy adaptive control system is designed in Fig. 5, which is used as a suitable structure such as PD-type FLC with feedback mechanism. The inputs of the adaptation mechanism are the difference between the reference model and system outputs (error), and variation for a sampling time of this difference (error variation). The output is used to adjust the gain of the fuzzy logic. Fuzzy model reference adaptive control (FMRAC) system has six gains, which are occurred with two gains in the fuzzy logic and adaptation mechanism, and one gain in the outputs. The gain of the FLC output is adjusted by using adaptation mechanism and this setting affects directly the signal of the control system. The other five gains are parameters that affect the performance of the control system [24-27]. The algorithm aims to determine the gains affecting the desired system performance. Each individual of the GA consists of combination of the five gains. Each member of the community is decomposed to achieve the necessary gain values. Each individuals' performance is determined by evaluating the degree of compatibility. It impacts the probability of survival of individuals and provides the desired values passed to new generations [10,11,26].
Non Optimum Solution
Optimum Solution
Figure 3. Genetic algorithm process flowchart
Design of a MRAC scheme based on GA is especially complex due to have more parameters affecting the control system performance. Therefore, GA is used to solve this complex problem. As it seen in figure 5, G1, G2, G3, G4 G5, G6 gains are selected as parameters and created society of individuals, which represent all of these parameters. Thus, the problem is optimized to be solved by GA.
Reference Model
ym uA
G4 Fuzzy Adaptation Mechanism G5
em
G6
em
+ -
ur
G1
G2
Fuzzy Logic Controller
X
G3
Plant
Figure 5. The Proposed Control Technique
Figure 7. Snapshot of the experimental setup TABLE II. INDUCTION MOTOR PARAMETERS
The conformity assessment function is established by sum of the absolute value which obtained from the error of the reference speed and actual speed. By using Matlab/Genetic Algorithm, the conclusions are given as shown in Fig. 6.
1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 x 10
5
Best: 28440.4373 Mean: 28455.0683

Best fitness Mean fitness
Fitness value
Parameters Stator resistance (Rs) Rotor resistance (Rr) Number of poles (P) Stator self-inductance (Ls) Rotor self-inductance (Lr) Moment of inertia (J) Mutual inductance (Lm) Friction factor (B) Rated speed(Nn) Rated power(Pmek)
Values 1.45 1.93 2 0.4H 0.599H 0.03kg-m2 0.18787H 0.03 3450Rpm 3kW
10
15
20
25 30 Generation
35
40
45
50
Fig. 8 shows the performance of conventional PI controller. In the figure 8, unloaded induction motor has been accelerated. Induction motor starts to operate at a steady state at 2000 rpm reference speed. Then, a sudden step speed command increasing, form 2000 rpm to 3800 rpm is applied. Then, the reference speed is reduced from 3800 rpm to 1000 rpm at 6 seconds. As seen in Fig. 8, the PI controller response reaches the reference speed after nearly 4 sec. with overshoot.
4500
Figure 6. Optimized result of genetic algorithm
Reference Speed
V. EXPERIMENTAL RESULTS In this section, in order to validate the performance of the proposed control technique, an experimental setup was designed as it was shown in Fig.7. The experimental setup consists of an induction motor, encoder, dsPIC 30f6010, measurement and IGBT card. The proposed control technique for vector controlled induction motor is experimentally implemented using digital signal processor board dsPIC 30f6010 through both hardware and software. Several experimental results for speed control of vector controlled induction motor drive using both PI and the proposed control technique are realized. The experimental results are performed for different reference speeds with noload during 10 sec. The parameters of the induction motor are given in table II.
3800
Speed (Rpm)
Actual Speed
2000
1000
0 1 2 3 4 5 6 7 8 9 10
t(s)
Figure 8. Speed response of PI controller
Fig. 9 demonstrates the dynamic response of the proposed control technique. As it seen in Fig. 9, the reference speed is chosen 2000 rpm. At 2 seconds, the reference speed is increased to 3800 rpm. Then, the reference speed is changed from 3800 rpm to 1000 rpm at 6 seconds. As can be seen from figure 9, the proposed control technique response reaches to reference speed after nearly 2.5 seconds without overshoot.
4500
While the conventional PI controller follows the reference speed with overshoot.
4500
Reference Speed
3400
Reference Speed
3800
Speed (Rpm)
GA + FMRAC
2000
PI
1000
Speed (Rpm)
2000
Actual Speed
0 1 2 3 4 5 6 7 8 9 10
1000
t(s)
Figure 11. Comparison of the speed response of the controllers

0 1 2 3 4 5 6 7 8 9 10
t(s)
Figure 9. Speed response of proposed control technique
In Fig. 10, the reference speed is chosen as 2000 rpm. Unloaded induction motor is required to accelerate from 2000 rpm to 3800 rpm in 2 seconds. The reference speed is kept constant at 3800 rpm from time, t=2 sec to 6 sec. After that, the reference speed is changed from 3800 rpm to 1000 rpm in 6 seconds. As it can be seen, the proposed control technique gives better performance against the conventional PI controller with respect to overshoot and response time.
4500
VI. CONCLUSIONS In this work, a MRAC scheme based on genetic algorithm and fuzzy logic controller has been used for vector controlled induction motor. The motor drive system is performed using both conventional PI and the proposed control technique. Both of the control systems are compared and the effectiveness of the proposed control technique against PI control performance is illustrated. The running of the induction motor in real conditions has demonstrated that the reference speed values have been obtained with smooth convergence regime and rather small error values being left under the 2 per cent of set values in a good response time. As a conclusion, Considering the overshoot and the response time, the proposed control technique gives obviously better performance than the conventional PI controller. REFERENCES
[1] Sen, P. C., 1990. Electric Motor Drives and Control Past Present, and Future, IEEE Transactions on Industrial Electronics, Vol.37, No.6, 562575. Bose, B. K., 2002. Modern Power Electronics and AC Drivers, Prentice Hall, New Jersey. Ho, E.Y.Y. and Sen, P. C., 1988. Decoupling Control of Induction Motor Driver, IEEE Transactions on Industrial Electronics, Vol.35, No.2, 253-262. Santisteban, J. A. and Stephan, R. M., 2001. Vector Control Methods for Induction Machines: An Overview, IEEE Transactions on Education, Vol.44, No.2, 170-175. Novotny, D. W. and Lorenz, R. D., 1986. Introduction to Field Orientation and High Performance AC Drives, IEEE, Tutorial Course Notes. Blaschke, F., 1971. A New Method for the Structural Decoupling of AC Induction Machines, In Conf. Rec. IFAC, Dusseldorf, Germany, Oct. pp. 1-15. B. Yoo and W. Ham, Adaptive fuzzy sliding mode control of nonlinear system, IEEE Transactions on Fuzzy Systems, vol. 6, no. 2, pp. 315 321, 1998.
Reference Speed
3800
Speed (Rpm)
[2]
2000
GA + FMRAC
1000
[3] PI
[4]
6 7 8 9 10
0 1 2 3 4 5
t(s)
[5]
Figure 10. Comparison of the speed response of the controllers
In the Fig. 11, the reference speed is 1000 rpm. Then, a sudden step command increasing, from 1000 rpm to 3400 rpm is applied. After that, the reference speed is decreased from 3400 rpm to 2000 rpm. As can be seen from Fig. 11, the proposed control technique follows the reference speed without overshoot and steady state error.
[6]
[7]
Y. Xia and Y. Jia, Robust sliding -mode control for uncertain time-delay systems: an LMI approach, IEEE Transactions on Automatic Control, vol. 48, no. 6, pp. 1086 1092, 2003. V. I. Utkin, Sliding Modes and Their Application in Variable Structure Systems, MIR Publishers, Moscow, Russia, 1978. Yildiz, C., Vector Control of AC Induction Machine Using Genetic Algorithm Based Fuzzy Controller, MSc Thesis, Department of Electrical and Electronics Engineering, K.Maras Stc mam University, K.Maras, Turkey, 2008. Mansfield, R.A., 1990. Genetic Algorithms. University of Wales College of Cardiff. Goldberg, David E. Genetic Algorithms in Search, Optimization and Machine LearningAddison -Wesley Pub. Co. 1989. Gotshall,S. and Rylander, B., Optimal Population Size And The Genetic Algoithm., Proc On Genetic And Evolutionary Computation Conference, 2000. Goldberg. D.E. Genetic Algorithms in Search, Optimisation and Machine Learning Addison Wesley, 1989 Bose, B. K., 1997. High Performance Control and Estimation in AC Drives, IECON 97, 23rd International Conference on Industrial Electronics, Control and Instrumentation, Vol.2, 377-385. Harnefors, L., 2001. Design and Analysis of General Rotor Flux Oriented Vector Control Systems, IEEE Transactions on Industrial Electronics, Vol.48, No.2, 383390. Trzynadlowski, A. M., 2001. Control of Induction Motors, Academic Press, San Diego, USA. Zadeh, L. A. 1965. Fuzzy Sets. Informat and Control. Vol. 8: 338-353 Lin, C. T. and Lee, C. S., 1996. Neural Fuzzy Systems, Prentice Hall PTR, New Jersey. Ross, T. J., 1995. Fuzzy Logic with Engineering Applications, McGrawHill Inc., ISBN 0-07-053917-0. Babuska, R., 1998. Fuzzy Modeling for Control Kluwer Academic Publishers, Boston. Dubois, D., Lang, J. and Prade, H., 1991. Fuzzy Sets in approximate Reasoning, Part II: Logical Approaches, ELSEVIER, Fuzzy Sets and Syst., Vol.40, No.1, 203-244. Sekkeli, M., Yildiz, C., Ozcalik, H.R. , Fuzzy Logic Based Intelligent Speed Control of Induction Motor Using Experimental Approach International Symposium on 1Nnovaitons in intelligent SysTems and Applications, Trabzon/Turkey, June 29, July 1, 2009 INISTA Draper, C. S., LI, Y.T. 1951. Principles of Optimalizing Control Systems and an Application to the Internel Combustion Engine. ASME Publications, New York. Gizolme, O., Thollon, F. 1998. Shape Optimization of Synchronous Machine Rotor. International Journal of Applied Electromagnetics and Machines. Vol 9, Issue 3. Whitaker, H.P., Yamron, J., Kezer, A. 1958. Design of Model Reference Adaptive System for Aircraft. R-164, Instrumentation Laboratory, MIT, Cambridge. USA. S. Maiti and al., Model referen ce adaptive controller based rotor resistance and speed estimation techniques for vector induction motor drive utilizing reactive power, IEEE Trans. on Industry Electronics, vol. 55, pp. 594601, Feb. 2008.
[8]
[9] [10]
[11] [12] [13]
[14] [15]
[16]
[17] [18] [19] [20] [21] [22]
[23]
[24]
[25]
[26]
[27]
Optimization of a PV/microturbine hybrid system for tropical climates

M.S. Ismail1,2,3, M. Moghavvemi1,3,4, T.M.I. Mahlia5
Department of Electrical Engineering, University of Malaya, Kuala Lumpur, Malaysia Electrical Engineering Department, Palestine Technical University, Tulkarm, Palestine 3 Center of Research in Applied Electronics (CRAE), University of Malaya, 50603 Kuala Lumpur, Malaysia 4 Faculty of Electrical and computer Engineering, University of Tehran, Tehran, Iran 5 Department of Mechanical Engineering, Universiti Tenaga Nasional, Kajang, Selangor, Malaysia mahmoud_kafa@yahoo.com; mahmoud@um.edu.my
2 1
Abstract Hybrid system based on photovoltaic is considered an effective option to electrify remote and isolated areas far from grid. This is true for areas that receive high averages of solar radiation annually. Using microturbine as a standby source will make utilization of hybrid systems more attractive. A sizing optimization of the hybrid system components, an economic feasibility study and a complete design of the hybrid system consisting of photovoltaic (PV) panels, a microturbine as a backup power source and a battery system supplying a small community in a tropical climate area were presented in this paper. A scenario depending on PV standalone and another scenario depending on microturbine alone were also studied and analyzed in this paper. This is to select the most appropriate considering cost and pollutant emissions for these areas. A simulation program depending on an iterative approach was used to optimize the sizes of PV system and battery bank. Specifications of the hybrid system components are then determined according to the optimized values. Solar radiation data is firstly analyzed and the tilted angle of the PV panels is also optimized. It is found that electrifying rural small community using this hybrid system is very beneficial and competitive with other types of conventional sources as it decreases both operating costs and pollutant emissions. Keywords Photovoltaic; Microturbine; Optimization Hybrid system; Economics;
In this paper, a hybrid PV/microturbine/battery system for generation of electric energy for a small community in tropical climate was analyzed. The block diagram of this system is shown in Figure 1. In this hybrid system, there are two voltage buses: the DC voltage bus that combines the battery bank with the DC output of the charge regulator and the AC voltage bus that combines each of the output of the bidirectional converter, the AC output of the microturbine and the load. The type of the hybrid system and its configuration depend mainly on the availability of the renewable source in the location selected for installing this hybrid system. For the tropical climate areas, the average daily solar radiation intensity on a horizontal surface ranges from about 4.2 kWh/m2 to about 5.6 kWh/m2, while the total annual sunshine hours amounts to about 3000 hours. These values are relatively considered high and encourages for using PV technology. The chosen site for this study is at Langkawi Island (Lat.: 6 20' N and Long.: 99 44' E). Hourly data for both solar radiation and ambient temperatures for duration of one year were used in this study. In this study the energy balance was carried out based on hourly data rather than daily data. Many remote areas that are located in different parts of Malaysia, mainly in Sarawak state, Sabah state (in eastern part of Malaysia) and in many Malaysian islands are still without permanent electricity source. People in these remote areas are normally settled in small villages located far away from grids. Some of them use diesel generators to electrify their homes. These generators usually operate for a limited period at night. Using microturbines as backup sources in PV hybrid systems to electrify these remote locations is a viable attractive solution. In comparison to the diesel generator, the microturbine is more reliable, more flexible in terms of fuel, emits less pollutant gases, requires less maintenance, and is less noisy. Investigation of the feasibility of using microturbines as backup sources in the PV renewable hybrid systems is actually one of the purposes of this study. Sizing optimization of different components constructing any hybrid system is one of the important issues that should be considered while designing this system. Maximizing utilization of the renewable source, minimizing the cost of generating energy and minimizing the pollutant emissions are objective functions of this optimization.
I.
INTRODUCTION
Different types of renewable energy sources are nowadays used to supply different applications in rural and urban areas [1]. Increased reliability and energy security issues are of the most benefits that can be achieved by using hybrid renewable systems [2-6]. A hybrid system is a system consisted of more than one energy source and usually combines conventional and renewable energy resources. Hybrid systems are appropriate to supply loads in isolated areas due to different benefits achieved in comparison with a single renewable system. Hybrid systems that depend on photovoltaic (PV) are considered the most popular among other types of renewable systems. The main advantages of this technology are their low maintenance costs and low pollutant emissions [7-11].
Khatib et al. [12] has concluded that a PV/diesel generator hybrid system is the more feasible system compared to a diesel generator system or standalone PV system for Malaysia case. Ismail et al. [7] carried out a techno economic analysis of a PV/diesel generator hybrid system dedicated for remote houses in tropical climate areas. They found that the hybrid system which consists of a combination of the PV panels, the battery bank and the diesel generator is the most optimal scenario. They concluded that powering remote areas using this hybrid system is advantageous as it decreases both the cost of energy and the pollutant emissions. Rajkumar et al. [13] used an adaptive neuro-fuzzy inference system to develop a model for both the PV and the wind sources in a PV/wind/battery hybrid system. Lowering the cost of energy production and the excess energy are the objective functions in this study where the Malaysian metrological data was used to conduct this study. Their results showed that the developed model was able to produce accurate values for the output power. Furthermore, the selected configuration had the ability to meet the load requirement at the minimum cost and excess energy. In another study carried out by Haidar et al. [14],the configuration of a PV-wind-diesel generator hybrid system in Malaysia that satisfies the least cost was determined. Homer program was used to carry out the simulation analysis. It was concluded that in the Malaysian context, a PV-diesel generator hybrid system was the most viable solution considering the economic and pollution aspects. A hybrid system that consists of a microturbine and a wind turbine was studied and analyzed by Caisheng et al. [15]. For their study, a real wind data and an actual residential load profile were considered. Their simulation results indicated the suitability of these types of hybrid systems to supply the residential loads with a high level of reliability. This paper has introduced a methodology to optimize sizes of a hybrid system with microturbine as a backup source and taking into account tropical climate as a case study. In fact, a detailed review has indicated that the literature lacks a detailed analysis for the hybrid systems with a microturbine as a backup source. In this paper, different components making the hybrid system should be modeled, studied, specified, and appropriately selected to minimize the system cost. The available data of solar radiation has been also analyzed in this paper. It is assumed that the optimized configuration should fulfill the load requirement for each hour in the year (no interruption of power supply). For this purpose, a simulation program was developed to accomplish an energy balance for each hour in the year. Satisfying the load requirement under all circumstances (no interruption of load is assumed) should be fulfilled by the obtained design of the hybrid system. The effect of utilizing the microturbine cogeneration feature on the energy production cost was also analyzed. The models of the components constructing the hybrid system adopted in this paper are previously used. Actually, these models take into account the effect of solar radiation and temperature on the PV panels energy production. Moreover, the effect of load variation on the fuel consumption of the microturbine has been considered and a mathematical model has been developed for this purpose as it will be later illustrated.
Fig. 1. Block diagram of the hybrid system
II.
COMPONENTS MODELING AND SIMULATION APPROACH
As it is stated before, a reliability of 100% is assumed in this study. An energy balance is calculated for each hour in a year to achieve this. This is done by a developed simulation program where hourly data for solar radiation, ambient temperature and load are inputted to it. A power management strategy is needed in order to control the flow of power between the various sources in the hybrid system. According to this strategy, the priority to supply the load is from the energy generated by the PV panels and the battery bank. When this energy is not enough, the microturbine as a backup source is switched on. For each hour step, the developed simulation program compares the load requirement and the available generated energy by the PV system. A decision to charge the battery, discharge it or operate the microturbine will be taken according to this comparison. In certain cases where the generated energy exceeds the load requirement and the battery bank is fully charged, this excess energy is consumed by a dump load. As it mentioned before, a decision to operate the microturbine is taken when the battery is discharged to its depth of discharge level and there is no sufficient generated energy by PV system to supply the load. This case continues until the battery is fully recharged where the bidirectional inverter works as a rectifier and permits to charge the battery. In this paper, an iterative approach has been used to obtain the optimized sizes of the components constructing the hybrid system. In this approach, both the size of the PV system in a form of PV contribution, and the size of the battery in a form of number of autonomy days have been iteratively changed. The combination with the lowest COE production has been selected as the optimized solution. In this paper, two parameters only have to be optimized, so the iterative approach has an ability to give the optimized sizes without any complexity. Other optimization techniques depending on novel algorithms can be used for this purpose. These recent techniques have to be used when multi-objective functions with different parameters have to be optimized. The power generated by the PV panels is given by (1) [16] : PPV-out = PVcont PN-PV (G/Gref)[1+ KT(Tc - Tref)] (1) where PPV-out is the output power generated from the PV panel, PVcont is a certain percentage value used to express the optimization procedure of the PV system, PN-PV is the PV nominal power at reference conditions , G is solar radiation ( W/m) , Gref is the solar radiation at reference conditions (Gref =1000 W/ m), Tref is the cell temperature at reference conditions (Tref =25C), KT is the temperature coefficient of the
PV panel (KT = - 3.7 x 10-3 (1/C) ) for mono and poly crystalline silicon. Equation (2) is used to calculate the cell temperature Tc such that [7]: Tc = Tamb + (((NOCT-20)/800) G) (2) where Tamb is the ambient temperature in C, and NOCT is the normal operating cell temperature in C (PV specification and given by manufacturer). Its value is around 45 C for most of the PV panels from different manufacturers (in this study it is considered 45 C). The rated power PN-PV can be calculated using (3) [7]. (3) PN-PV = (ELSF) / (RVPSH) where EL is daily load energy, SF is stacking factor considered to compensate for resistive and PV-temperature losses , R,V are efficiencies of solar charging regulator and bidirectional inverter respectively, and PSH is the peak sun shine hours (numerically equals to daily average of solar radiation at the specified location). The value Ppv system given in (4) represents the optimized PV panel size where: Ppv system= PVcont PN-PV (4) The role of the battery system is essential in maintaining continuity in supplying loads during the absence of a solar source. The capacity of the battery mainly depends on the daily load energy and the duration of time required in supplying this load from the battery bank, in the absence of the solar source. The storage capacity of the battery (CWh) is calculated using (5) [12]. CWh = (EL AD)/ (VB DOD) (5) where DOD is allowable depth of discharge of the battery, AD is number of autonomy days, and B is battery efficiency. The fuel consumption of the microturbine depends on the actual output power supplied by it. Manufacturers of 30 kW Capstone microturbines provide values of fuel flow in (BTU/h) for different values of loads (from 2 to 30 kW) [17, 18]. Fuel flow in (m3/h) is calculated for each load power. Equation (6) represents a second order equation fitting that relates natural gas fuel flow of the microturbine in (m3/h), with load power in (kW). This equation is used to calculate the natural gas fuel flow consumption of the microturbine (FFMT ) in (m3/h). FFMT = 0.0007(PMT-out)2+ 0.29(PMT-out) + 1.684 where PMT-out in the microturbine output power in (kW). The output of the PV panels varies as a result of varying the environmental conditions. So, the solar charge converter is used in the hybrid system to regulate the battery charging under different varying cases. It is designed to protect the PV panels from the reversal of power, and the battery against deep discharge and excessive overcharge. The charge regulator ratings are selected to suit both the PV panels at its input and the battery bank at its output. The charge regulator is part of the hybrid system, so its efficiency affects the overall efficiency of the system. The bidirectional inverter shown in Fig. 1 works as an inverter to convert DC voltage to AC voltage to supply the AC (6)
loads, and as a rectifier to convert the AC voltage from the microturbine to DC voltage in order to charge the battery. The bidirectional inverter ratings are selected to suit the battery bank at its input and the load at its output. The efficiency of the bidirectional inverter appears in the equations used for calculation of the power generated from the PV panels, and for the calculation of the capacity of the battery. It is obvious from the two equations that the required power from the PV panels and the battery capacity increase as the efficiency of the bidirectional inverter decreases. Table I includes different inputs and part of outputs of the simulation program. Other inputs like metrological data and location specifications are inputted to the simulation program directly or through EXCEL files. In tropical areas, the consumption behavior is relatively the same throughout the year since the climate is categorized as equatorial with no seasonality. The climate is being hot and humid throughout the year. The load profile used in our study is for a small community and it is based on weekly variations. Two categories are taken into account. The first category is for working days where number of working days is five while the other category is for the weekend days. Fig. 2 shows the details of these two categories. This load profile is also inputted to the simulation program. In this paper the economic analysis is carried out based on total life cycle costing method. In this method, all types of costs for different components (initial costs, maintenance costs, fuel and operation costs, replacement costs, and salvage value) are taken into account. Cost of energy (COE) which is the cost required to produce a 1 kWh is the indicator used in the optimization process. It is the ratio between the total annual cost and the total annual energy consumed by the load. Different economic factors that affect the value of money over the life time of the project are also considered. These rates are: inflation rate and discount rate. The project life cycle time is considered 25 years. It is the lifetime of the component that possesses maximum lifetime. TABLE I.
Inputs Quantity PV cost($/kW) PV installation cost ($/kW) Bi-directional inverter cost ($/kW) PV regulator cost ($/A) Battery cost ($/kWh) PV regulator efficiency Bi-directional inverter efficiency Battery Wh efficiency PV life (year) Battery life (year) Value 2290 300 711 5.92 213 95 92 85 25 6
DIFFERENT INPUTS AND PART OF OUTPUTS OF THE

SIMULATION PROGRAM
Interest (%)
(discount)
rate
General inflation rate (%) Fuel inflation rate (%) Project life cycle period (year) Outputs Quantity PV panel size (kW) Battery capacity (kWh) Yearly load demand (kWh) Yearly energy generated by PV panel (kWh) Yearly energy generated by Microturbine (kWh)
4 5 25
Value 55 198 139240 85980 88462
Microturbine rated power (kW) Microturbine capital cost ($/kW) Natural gas price ($/m3) Air and fuel filters replacement (h) Fuel injectors, igniters and thermocouples inspection and replacement (h) Battery replacement (h) Major overhaul, core turbine replacement (h) Microturbine replacement (h) Microturbine annual maintenance cost ($/kWh generated) 30 2970 1.73 8,000 16,000 20,000 Yearly dump energy (kWh) Microturbine operating hours (hour) Yearly fuel consumption (m3) COE ($/kWh) Yearly (Ton) CO2 produced 9720 2949 32478 0.267 69.66
contribution and 0.3 autonomy days (AD). Table II illustrates this result and other results considering different values of PV contribution and different values for autonomy days. Figure 3 shows the contribution of the PV modules and the microturbine in supplying the load in each month in the year. The annual contribution of the PV panels is about 49.3%. Since the AD of the battery considering the optimum case is 0.3 day, the microturbine is running every night for a certain period after the battery discharges to its DOD. This is actually what makes the number of the operating hours high for the microturbine. The COE considering the microturbine alone scenario and the other combinations of the PV hybrid system scenario approaches the COE for the optimum case. This implies that any small variation in the prices of different components or natural gas fuel may result in new configurations for the optimum case. As given in Table I, a limited value of dump energy is produced. This dump energy can be used for different load applications such as heaters where the heated water is stored in isolated tanks to be used later, charging batteries of emergency lights, pumping water to tanks, or in refrigeration applications. For hybrid systems connected to the grid, this dump energy can be sold to the grid especially if a feed in tariff is present for this purpose. The percentages of the initial costs of the hybrid system components are shown in Fig. 4. It is obvious from the figure that the initial cost of the PV panels forms the highest percentage. TABLE II.
AD (day) 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
20000 40000 80000 0.02
Fig. 2. Hourly typical load profiles for working and weekend days.
III.
A. PV data analysis results Analysis of the available hourly PV data for a one year shows that yearly average daily solar radiation on a horizontal surface amounts to 5.0 kWh/m2. For tilt angle optimization, the yearly energy generated by a 1 kW PV array has been calculated for each tilt angle varied from 0 to 90 . To calculate the energy generated by the PV panels, equations (12) have been used, while in order to obtain the values of the optimum tilt angles of the PV panels, the value of the solar radiation on the tilted surface from global radiation on a horizontal surface needs to be calculated. The anisotropic model was used for the purpose of this calculation. This model is used to obtain the solar radiation on tilted surfaces in terms of components of solar radiation reached to a horizontal surface. To calculate the diffused solar component, the Orgill and Holand correlation has been adopted and used in this paper. The results obtained using this approach are closer to the measured values, so it is recommended by many authors, especially for surfaces directed towards the equator [16]. From this analysis it is found that the optimum tilt angle is at 19 . The yearly average daily solar radiation at this optimum tilted angle is 5.1 kWh/m2. B. Simulation results The simulation program results indicate that the lowest COE obtained is 0.258 $/kWh achieved at 50% PV
COE ($/KWH) FOR DIFFERENT PV CONTRIBUTIONS AND AUTONOMY DAYS (AD)

PV contribution (%)
10 0.270 0.296 0.309 0.325 0.342 0.360 0.376 0.397
20 0.267 0.289 0.303 0.320 0.338 0.355 0.372 0.392
30 0.265 0.282 0.297 0.315 0.332 0.349 0.368 0.385
40 0.262 0.269 0.287 0.304 0.322 0.339 0.357 0.375
50 0.258 0.269 0.285 0.301 0.318 0.335 0.353 0.371
60 0.263 0.270 0.284 0.299 0.315 0.333 0.350 0.368
70 0.271 0.268 0.282 0.296 0.312 0.329 0.346 0.366
80 0.279 0.267 0.279 0.295 0.311 0.328 0.345 0.363
90 0.288 0.270 0.283 0.298 0.314 0.332 0.349 0.365
100 0.298 0.276 0.290 0.305 0.321 0.336 0.354 0.370
Fig. 3. Energy contribution of PV and microturbine in each month
the microturbine (2970 $/kW) includes the cost of the heat exchanger required to extract the heat of the exhaust gases [19]. To utilize the cogeneration feature, the only requirement is to modify the internal domestic water or steam installations. This is easily can be done by connecting a water supply to the inlet of the heat exchanger and connecting the outlet of the heat exchanger to the hot water distribution system.
Fig. 4. Percentages of the initial costs of the components making up the hybrid system.
Other scenarios are also analyzed in addition to the main scenario. It is found that the COE for a PV standalone system (i.e. without microturbine) is 0.381 $/kWh and occurs at 180% PV contribution and 0.8 AD. The rated power of the PV modules in this case is 198.3 kWp and the required storage capacity is 528 kWh. The other scenario is the microturbine only scenario. For this scenario the COE is 0.262 $/kWh and the amount of yearly produced CO2 is 109.6 Ton. The value of CO2 generated by the microturbine is calculated based on the fact that the generation of 1 MWh electrical energy from a 30 kW microturbine using natural gas produces 1736 lb ( 787.44kg) of CO2 [19, 20]. Table III summarizes the results of the main scenario as well as the other scenarios. TABLE III. RESULTS OF THE 3 SCENARIOS
Scenario Main scenariosystem scenario hybrid System configuration 50% PV contribution and 0.3 AD with 30 kW microturbine 180% PV contribution and 0.8 AD without microturbine Microturbine alone COE ($/kWh) 0.258 0.381 0.262
Fig. 5. Hourly heat load profiles for work and weekend days
D. Hybrid system design The recommended voltage for the DC bus for this hybrid system is 220 V. Other specifications and design issues of the main components making this hybrid system are shown in Table IV. TABLE IV. SPECIFICATIONS OF DIFFERENT COMPONENTS
CONSTRUCTING THE HYBRID SYSTEM
Component PV Modules
PV standalone scenariowithout microturbine Microturbine scenario alone
Specification Each PV module has 200 Wp,26.3 V as Vmp, and 32.9 V as Voc
Battery system
Each battery has 2 V DC and 450 Ah
C. Cogeneration operation of the microturbine As previously mentioned, one of the important features of the microturbine is its generation of utilized heat while it is running to generate electricity. In the cogeneration mode of operation, the electrical load is supplied by the electrical output of the microturbine, while the heat load is supplied directly by the microturbine (through a heat exchanger) instead of supplying this heat load from an electrical output of the microturbine. The excess electrical power is used to charge the battery. The heat load profiles for both working day and weekend day are shown in Fig. 5. The electrical load is the difference between the total load given in Fig.2 and the heat load given in Fig. 5. The COE taking into account the cogeneration feature of the microturbine decreases to 0.239 $/kWh. The configuration of the hybrid system for this case changes. The PV contribution and the AD become 30% and 0.3 day respectively. The number of running hours of the microturbine in this case decreases to 3197 hours. Microturbines are usually provided with the additional equipment in order to utilize the useful heat. The capital cost of
Microturbine Charger Converter Bidirectional Inverter 30 kW, 3ph, 400 V, 50Hz 60 kW, 360 V input, 220 V output 35 kW, 400 V, 3-ph ac output, 220 V DC input
Number Number of series PV modules in each string Number of parallel strings Number of series batteries in each string Number of parallel strings 1 1 1
11 25 110 2
IV.
CONCLUSIONS
This paper has introduced a methodology to optimize sizes of a hybrid system with microturbine as a backup source and taking into account tropical climate as a case study. In fact, a detailed review has indicated that the literature lacks a detailed analysis for the hybrid systems with a microturbine as a backup source. Firstly, solar radiation data analysis for the location of the study shows that the yearly average daily solar radiation is about 5.0 kWh/m2 while tilting the PV modules at the optimized angle results in yearly average equals to 5.1 kWh/m2. This optimized tilt angle is found to be 19 . Three scenarios have been analyzed in this paper. The most economic scenario is the one that includes in addition to the PV panels, the battery system and the microturbine. The COE for this scenario is found to be 0.258 $/kWh and happens at 50% PV
contribution and 0.3 AD. Other scenarios dependent on standalone PV and microturbine only give results of COE greater than this value. For the microturbine only scenario, both the COE and amount of produced CO2 are greater. The amount of the produced CO2 is about 160% greater compared to the hybrid one. Taking into account locations far from the grid and the environmental considerations make the implementation of such hybrid systems more acceptable. Moreover, as the prices of the hybrid system components decrease, this will encourage more acceptances of these systems in near future. ACKNOWLEGEMENT The authors would like to acknowledge the University of Malaya, Kuala Lumpur, Malaysia for the financial support under POSTGRADUATE RESEARCH GRANT with project account number PS022-2012A. The authors also would like to acknowledge the Ministry of Higher Education of Malaysia and The University of Malaya, Kuala Lumpur, Malaysia for the financial support under UM.C/HIR/MOHE/ENG/21. REFERENCES [1] [2] R. Bhandari and I. Stadler, "Electrification using solar photovoltaic systems in Nepal," Applied Energy, vol. 88, pp. 458-465, 2011. G. D. Kamalapur and R. Y. Udaykumar, "Rural electrification in India and feasibility of Photovoltaic Solar Home Systems," International Journal of Electrical Power & Energy Systems, vol. 33, pp. 594-599, 2011. A. Daud and M. S. Ismail, "Design of isolated hybrid systems minimizing costs and pollutant emissions," Renewable Energy, vol. 44, pp. 215-24, 2012. T. Senjyu, et al., "Optimum Configuration for Renewable Generating Systems in Residence Using Genetic Algorithm," IEEE TRANSACTIONS ON ENERGY CONVERSION, vol. 21, pp. 459-466, 2006. Y.-Y. Hong and R.-C. Lian, "Optimal Sizing of HybridWind/PV/Diesel Generation in a Stand-Alone Power System Using Markov-Based Genetic Algorithm," IEEE TRANSACTIONS ON POWER DELIVERY, vol. 27, pp. 640-647, 2012. D.-J. Lee and L. Wang, "Small-Signal Stability Analysis of an Autonomous Hybrid Renewable Energy Power Generation/Energy Storage System Part I: Time-Domain Simulations," IEEE TRANSACTIONS ON ENERGY CONVERSION, vol. 23, pp. 311-320, 2008. M. S. Ismail, et al., "Techno-economic analysis of an optimized photovoltaic and diesel generator hybrid power system for remote houses in a tropical climate," Energy Conversion & Management, vol. 69, pp. 163-173, 2013. S. Twaha, et al., "Applying grid-connected photovoltaic system as alternative source of
[9]
[10]
[11]
[12]
[13] [14] [15]
[3] [4]
[16]
[5]
[17]
[6]
[18]
[7]
[19] [20]
[8]
electricity to supplement hydro power instead of using diesel in Uganda," Energy vol. 37, pp. 185-194, 2012. S. A. Pourmousavi, et al., "Real-Time Energy Management of a Stand-Alone Hybrid WindMicroturbine Energy System Using Particle Swarm Optimization," IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, vol. 1, pp. 193-201, 2010. M. Alsayed, et al., "Multicriteria Optimal Sizing of Photovoltaic-Wind Turbine Grid Connected Systems," IEEE TRANSACTIONS ON ENERGY CONVERSION, vol. 28, pp. 370-379, 2013. M. H. Nehrir, et al., "A Review of Hybrid Renewable/Alternative Energy Systems for Electric Power Generation: Configurations, Control, and Applications," IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, vol. 2, pp. 392-403, 2011. T. Khatib, et al., "Optimal sizing of building integrated hybrid PV/diesel generator system for zero load rejection for Malaysia," Energy and Buildings, vol. 43, pp. 34303435, 2011. R. K. Rajkumar, et al., "Techno-economical optimization of hybrid pv/wind/battery system using Neuro-Fuzzy," Energy, vol. 36, pp. 5148-5153, 2011. A. M. H. Haidar, et al., "Optimal configuration assessment of renewable energy in Malaysia," Renewable Energy, vol. 36, pp. 881-888, 2011. W. Caisheng, et al., "Power management of a standalone hybrid wind-microturbine distributed generation system," in Power Electronics and Machines in Wind Applications, 2009. PEMWA 2009. IEEE, 2009, pp. 1-7. M. S. Ismail, et al., "Analysis and evaluation of various aspects of solar radiation in the Palestinian Territories," Energy Conversion and Management, vol. 73, pp. 57-68, 2013. F. Alkhalil, et al., " Secant minimizes the cost of fuel and emissions in a power station of small cogeneration multi-machines," presented at the 1re Confrence Franco-Syrienne sur les nergies renouvelables (CFSER 2010), 2010. M. S. Ismail, et al., "Current utilization of microturbines as a part of a hybrid system in distributed generation technology," Renewable and Sustainable Energy Reviews vol. 21, pp. 142-152, 2013. "Catalog Of CHP technologies," U.S. Environmental Protection Agency2008. M. Shekarchian, et al., "A review on the pattern of electricity generation and emission in Malaysia from 1976 to 2008," Renewable and Sustainable Energy Reviews, vol. 15, pp. 2629-2642, 2011.
Sensitivity Analysis for the IEEE 30 Bus System using Load-Flow Studies
Ibrahem Totonchi, Hussain Al Akash, Abdelhadi Al Akash and Ayman Faza
Ibrahem_tot@yahoo.com, h-u-s-s-a-i-n-@hotmail.com, Abdelhadi.akash@students.psut.edu.jo, a.faza@psut.edu.jo. King Abdullah II School for Electrical Engineering Princess Sumaya University for Technology Amman-Jordan
Abstract Load flow analysis is the backbone of the power system studies and design, and through it the voltage magnitude and phase angle at each bus and the complex power flowing in each transmission line can be obtained. In this paper, we use the load-flow to perform a sensitivity analysis of the IEEE 30 bus system. We find the maximum complex power flowing in each transmission line in case of no fault and in case of a single transmission line fault in the steady state condition. The results of this analysis helps identify the most critical lines in the system, which can help better plan the capacities of such lines, and minimize the probability of potential cascading failures. Keywords: load flow; reliability; sensitivity analysis ; cascading failures
I.
INTRODUCTION
Capacity planning for power systems is very important to ensure proper and reliable operation of the grid. Proper planning ensures that a system is capable of generating, transmitting and delivering electric power to everyone, without interruption. However, it is also desirable that the system be capable of providing those services when one or more lines in the system have failed. Many of the failures that randomly occur in a transmission line can have a catastrophic effect. A failure of one line forces the flow of power to be rerouted through the other lines in the system, potentially causing some of them to overload, in order to compensate for the loss of the initial line. The overloads that occur in the other transmission lines cause them to be disconnected from the system by the action of the protective devices. This in turn can cause similar overloads and disconnections in other lines in the system in a cascading fashion, eventually leading to a system blackout. In this paper, we present a sensitivity analysis of the IEEE 30 bus system, which aims to reduce the probability of a cascading effect in the system. In order to do that, we perform a load flow analysis on the initial fault free system, and additionally perform a similar analysis when at least one transmission line has failed, enumerating all the possible cases. The goal of this study is to find the worst case flow scenario; that is, the maximum flow in the systems remaining transmission lines, for each transmission line failure, and look for potential cascading failures in the system. The overarching objective of this work is to develop ideas and methodologies
to improve the reliability of the IEEE30 bus system, and further generalize it to any other power system, using methodologies such as capacity planning, the use flow control devices such as FACTS devices, and the use of distributed generation (DG) methods, including renewable energy methods such as solar and wind to gradually convert the power system from a radially connected network to a meshed network with generation occurring at both sides of the system. Some of the suggestions that we will present at the end of this paper will include raising the capacities of some of the transmission lines in the system. Raising the capacity of all the transmissions lines in the system may not be very efficient or may be a bit hard to accomplish, but we could raise the capacity of some critical transmission lines to minimize the occurrence of cascading failure. The main contribution in this paper is to determine the most critical transmission lines in the IEEE 30 bus system, whose failure can have catastrophic effect on the system. We further provide suggestions on how to improve the operation of the grid by providing suggestions on improving the operation of such lines. Those critical transmission lines will be revealed after studying the results obtained from the load flow analysis. The rest of the paper is organized as follows. Section II provides a literature review of the studies performed on the IEEE30 bus system, in addition to other relevant studies. Section III presents the IEEE 30 bus system in more details. Section IV introduces the load flow analysis and the initial results obtained in a fault free system. Section V presents the sensitivity analysis performed and shows the results, and also presents a few examples of cascading failures that can occur in the system. Section VI concludes the paper and provides a vision of the expected future work in this project. II. LITERATURE REVIEW
Many studies has been conducted on the IEEE 30 bus system tackling various subjects. In this section we present a number of such studies that are relevant to this work. The authors of [2] present a basic load flow analysis, which aims to find the converged bus voltages using both the Gauss-Siedel and Newton-Raphson methods and compare the two numerical methods in terms of complexity and convergence rate among
others. In another study [3], the Total Transfer Capability (TTC) and Available Transfer Capability (ATC) have been studied [3] which give a general feeling of the amount of complex power generated into the system. According to the authors, with the recent trend towards deregulating power systems around the world, transfer capability computation emerges as the key issue to a smoothly running power market with multiple transactions. A key concept in the restructuring of the electric power industry is the ability to accurately and rapidly quantify the capabilities of the transmission system. Total Transfer Capability (TTC) is defined as the amount of electric power that can be transferred over the interconnected transmission network in a reliable manner while meeting all of a specific set of defined pre- and post-contingency system conditions. Available Transfer Capability (ATC) is a measure of the transfer capability remaining in the physical transmission network for further commercial activity over and above already committed uses. In [3], MATLAB software is used to determine the TTC of power transfers between different control areas in deregulated power systems without violating system constraints such as thermal, voltage, and stability constraints. The main aim of the paper is to calculate total transfer capability and capacity benefit margin of IEEE 30-bus Reliability Test System (RTS) for calculation of ATC. In another project [4], the authors study the performance of indirectly controlled STATCOM; a Flexible AC Transmission System (FACTS) device primarily used for reactive power compensation and improvement of voltage profile in the system, is used with IEEE 30-bus system. In this article a Cascade Multilevel Inverter (CMLI) based on the indirectly controlled STATCOM has been placed in IEEE30-bus system for its performance evaluation in regulating the voltages of different buses. Other studies such as [5] tackled the voltage
collapse prediction problem for interconnected power systems which is a steady state analysis applied to study the voltage collapse problems. In this paper, a modal analysis method is used to investigate the stability of the power system. Q-V curves are used to confirm the obtained results and to predict the stability margin or distance to voltage collapse based on reactive power load demand. The analysis in this paper is performed for three well-known systems; the Western System Coordinating Council (WSCC) 3-Machines, 9-Bus system, the IEEE14 Bus system, and the IEEE30-bus system. Our work is different from these studies in that it focuses on understanding the effect of a single transmission line failure on the overall operation of the system. We study the sensitivity of the system to the occurrence of a single line fault, and provide a capacity planning method for such a scenario. In addition, we explore the notion of a cascading failure in the IEEE 30 bus system, in preparation of a further investigation into methods to mitigate such failures and reduce the probability of their occurrence. III. LOAD FLOW AND THE IEEE 30 BUS SYSTEM The IEEE 30 bus system consists of 30 buses, 41 transmission lines and a number of transformers and synchronous condensers. It represents a portion of the American electrical power system in the Midwest region. Figure 1 below depicts the system. The bus generation and loads data were obtained from [6]. For the analysis of this paper, we performed load flow for the IEEE30 bus system using the Gauss-Seidel method. The load flow was performed for the initial case where all transmission lines are functioning in the system. Additionally, we run load flow after removing one transmission line at a time for the total of 41 transmission lines, and we documented
Figure 1 The IEEE 30 Bus System
the results. The main purpose of this study was to determine the worst case flow values in each transmission line in the system, when one line fails; that is a sensitivity analysis of the IEEE30 bus system to a failure in a single transmission line. In addition, if we know the line capacities in the IEEE30 bus system, we can simulate the effect of cascading failures, by taking out the lines that overload when a single line failure occurs, and repeating the process. To run the simulations, we used a MATLAB based load flow simulator, and we simulated a total of 41 failure cases, in addition to the case were all lines are working. In the following section, we present the results of the load flow analysis and further discuss their implications. IV.
LOAD FLOW ANALYSIS RESULTS
Figure 2 A Snapshot of Bus i and its Parameters. 2) Constructing the Admittance Matrix(the Y-bus Matrix) The second important step in performing the load flow is to construct the Y-bus matrix. The Y-bus matrix is an N N matrix, where N is the total number of buses, and it can generally be constructed by inspection, by systematically evaluating its parameters based on the transmission line data given in the problem. In general, and entry in the matrix at row i and column j; Yij is equal to the negative sum of the parallel admittances connected between buses i and j. The diagonal elements; Yii is the sum of the admittances connected to bus i, regardless of which bus the other side of the admittance is connected to . With that in mind, finding all the elements in the Y-bus matrix becomes straightforward. After constructing the Y-bus matrix, and using the GaussSeidel method, the voltage at each bus can be found by iteratively solving the following equations: For a load bus, the kth iteration of the voltage at bus i can be found as follows:
1) The Load Flow Problem [1] The load flow problem or the power flow problem is the computation of voltage magnitude and phase angle at each bus in a power system under balanced three-phase steady state conditions. As a byproduct of this calculation, real and reactive power flows in equipment such as transmission lines and transformers, as well as equipment losses, can be computed. The starting point for a power-flow problem is a single line diagram of the power system, from which the input data for computer solutions can be obtained. Input data consist of bus voltage, real and reactive power generated, and real and reactive power demand at each bus, in addition to the transmission line parameters. As shown in Figure 2, for any bus i the following eight variables are associated with each bus; Vi : The voltage at bus i
i : The phase angle at bus i,

Pig : The generated real power Qgi : The generated reactive power
Pid : The real power demand Qid : The reactive power demand
From these parameters, the scheduled real and reactive powers are calculated as follows:
Vi ( k ) =
(3)
N 1 Pisch jQisch i 1 (k ) Y V YinVn( k 1) + in n *( k 1) Yii Vi n = i +1 n =1
For a generator bus, we first need to get an estimate of
Qi to be used in place of Qisch . We use the following

equation:
N i 1 Qi( k ) = Im Vi*( k 1) YinVi ( k ) + YinVi ( k 1) (4) n =i n=1
Pisch = Pig Pid (1) Qisch = Qig Qid (2)

Depending on the type of bus; slack, generator, or load bus, the known variables and the unknown could be anything among the voltage magnitude, the voltage angle, and the real power generated, and the reactive power generated.
Then the voltage at bus I can be found by using a modified version of equation (3) as follows:
Vi ( k ) =
(5)
1 Yii
N Pisch jQi( k ) i 1 (k ) Y V YinVn( k 1) + in n *( k 1) n = i +1 n =1 Vi
3.Results of The Load Flow Analysis, for a normally running system with no failures. Running the load flow simulation for the normal fault free IEEE 30 bus system resulted in the following bus voltage results as shown in Table 1. Simulation was performed using Matlab and the results were confirmed using the Power World Simulator (PWS). The simulation parameters were set such that the error tolerance in the voltage was set to 0.01 per unit. Table 1 - Load Flow Results for the Fault-Free IEEE118 Bus System Bus Voltage Bus Voltage 1 2 1.060 1.04 5.2 3 5 7 9 11 13 15 17 19 21 23 25 27 29
is the currents carried by the transmission line connecting bus i and bus j. Given the previous analysis, we removed one line from the system at a time, and found the amount of power flow in each transmission line. Then we took the maximum power flow in each line that could occur due to the failure of one transmission line. The results of this simulation are documented in Table 2 below. Table 2 - Maximum Power Flow in Each Transmission Line, Due to a Single Transmission Line Failure Max Max TL TL TL TL Flow Flow number (from number (from bus to bus to bus) bus) 1 1-2 2.6428 22 15-18 0.1582 2 1-3 2.9309 23 18-19 0.1221 3 2-4 0.8224 24 19-20 0.1726 4 3-4 2.7255 25 10-20 0.1986 5 2-5 1.2383 26 10-17 0.2585 6 2-6 0.9997 27 10-21 0.2653 7 4-6 1.7064 28 10-22 0.2412 8 5-7 1.0603 29 21-22 0.2034 9 6-7 1.2994 30 15-23 0.1215 10 6-8 0.3160 31 22-24 0.2099 11 6-9 0.5078 32 23-24 0.1103 12 6-10 0.2810 33 24-25 0.1940 13 9-11 0.1771 34 25-26 0.0428 14 9-10 0.4983 35 25-27 0.1415 15 4-12 0.6202 36 27-28 0.2480 16 12-13 0.3310 37 27-29 0.1443 17 12-14 0.1714 38 27-30 0.1424 18 12-15 0.2579 39 29-30 0.1122 19 12-16 0.1584 40 8-28 0.3083 20 14-15 0.1029 41 6-28 0.4734 21 16-17 0.1779 Note that the above values are in per unit with 100MVA base complex power. For the sensitivity analysis the resultant maximum complex power flowing in each transmission line was caused by removing or a failure in a specific transmission line. For example, the maximum power flowing in transmission lines number 2, 4, 7 and 10 is caused by failure in transmission line number 1, and the maximum power flowing in transmission lines number 11, 12, 14, 16, 21, 24, 25, 26, 32 and 36 is caused by the failure in transmission line number 15. In addition, the maximum power flowing in transmission lines number 27, 30, 31, 33, 34 and 35 is caused by the failure in transmission line number 36. It is obvious from this analysis that transmission line 1, 15 and 36 are the most critical lines in the grid; therefore, raising the capacity of those three lines only we can help reduce or the number of cascading failures (to be explained in the following section) that can occur in the system.
1.03 7.4 1.01 13.8 1.01 12.6 1.04 13.8 1.08 13.8 1.07 14.7 1.02 15.7 1.02 15.5 1.01 16.4 1.02 15.8 1.01 16 1.00 15.8 1.00 15.3 0.98 16.5
4 6 8
10 12 14 16 18 20 22 24 26 28 30
1.02 9.1 1.02 10.8 1.01 11.3 1.03 15.4 1.04 14.7 1.02 15.6 1.03 15.3 1.01 16.3 1.01 16.2 1.02 15.8 1.00 16.2 0.98 16.2 1.02 11.4 0.97 17.5
The system needed 123 iterations to converge. Note that all the above values are in per unit with 100MVA base complex power. V.
SENSITIVITY ANALYSIS AND THE CASCADING FAILURE EXAMPLES
1) The Sensitivity Analysis The main purpose of the sensitivity analysis in this paper, is to determine the worst case scenario that could occur due to the failure of one transmission line. To do that, we obtain the maximum complex power going through each transmission lines in the case of no faults and in case of a one transmission line fault. This is performed for all 41 transmission lines. The complex power between bus i and bus j can be found as follows:
* Sij = Vi I ij (6)
Where,
I ij = Yij (Vi V j ) (7)
2) Cascading Failure Examples In addition to the sensitivity analysis presented above, it is of great value to try and discover potential cascading failures in the system. A cascading failure occurs when the failure of one transmission line causes the other lines to overload, which in turn causes these overloaded lines to be disconnected from the system using protective devices, which then causes further overloads and failures leading eventually to a complete system blackout. To simulate cascading failures, we set the capacity of each transmission line to 120% of the rated complex power value flowing in the line, which was obtained from the initial case of no transmission line failures presented in Section IV above. We analyzed three interesting cascading failure cases in this system. In the first case, we remove transmission line number 1 which connects bus 1 and bus 2. This will cause an overload at transmission lines 2, 3, 4, 7, 8, 9, 15, 16, 19, 20, 21, 22, 23, 30 and 32. The failure of these transmission lines causes an overload at lines 5, 6, 13, 14, 26, 27, 29, 31, 33 and 35, and after that all of these failures cause an over load at transmission lines 26, 25, 15, 14, 13 and 12, eventually leading to a total blackout. In the second case, we remove transmission line number 8. This will cause a failure in transmission line number 5 and those two failures will then isolate bus number 5 with bus voltage magnitude and angle of 0. In the third case, we remove transmission line number 39, which will cause an overload and then failure in transmission line number 38, then that will cause an overload at transmission lines number 40, 35, 33, and 29. In this case bus number 30 will be isolated (islanding) with no power flowing towards it. 3) Further Analysis In addition to the previous analysis, we have two buses in the system which generate real power; that is bus number 1 and 2 with four transmission lines connecting them to the rest of the system; transmission lines number 3, 5, 6 and 8. This shows that these lines can be considered critical transmission lines and raising their power capacity would help improve the systems reliability and reduce the possibility of system blackouts. The rest of the transmission line failure effects can be summarized in Table 3 below. Table III shows the effect of a failure in each line, and specifies which transmission line gets the maximum amount of flow due to this failure. The results shown in Table III above help grasp the causeeffect scenarios for transmission lien failures, which can further help us in better planning for the line capacities in the IEEE 30 bus system.
Table 3 - Transmission Line Failures, and Worst Case Scenarios TL in which Initial failure TL at which maximum power occurs flows due to this failure 2 1 3 4 9 5 5 6, 8, 9 11 13 18 17, 20 7 15, 18, 19 25 22, 23 27 28, 29 38 37, 39 37 38 10 40, 41 VI.
COCLUSION AND FUTURE WORK
Power flow or load-flow studies are important for planning future expansion of power systems as well as in determining the best operation of existing systems. The principal information obtained from the power flow study is the magnitude and phase angle of the voltage at each bus, and the real and reactive power flowing in each line. We performed a sensitivity analysis which aims to have a general feeling of the system by finding the maximum complex power flowing in each transmission line in the case of no faults and in the case of one transmission line fault in the steady state condition. Results show that a few of the lines are significantly more critical than the others when it comes to their failure. Our study helped us identify the lines that are most critical to the operation of the system, which can help better plan for their capacities to reduce the effects of failures in the system, and decrease the probability of occurrence of a cascading failure. Future work in this project will include implementing methods to improve the reliability of the system by using intelligent devices such as FACTS devices, in addition to implementing distributed generation and renewable energy sources into the grid. Our objective is to make the grid more intelligent, and convert the traditional IEEE 30 bus system from the conventional radial scheme to a more intelligent meshed network as we gradually move the existing power systems from the conventional power grid to the more advanced Smart Grid. REFERENCES [1] Duncan. G, Mulukutla. S and Thomas.O, Power flows in Power System Analysis And Design 5th Ed. Stanford, California, USA, Cengage Learning 2008, Ch6. [2] Dharamjit and D.K.Tanti, Load Flow Analysis on IEEE 30 bus System , International Journal of Scientific and Research Publications, Vol 2, No 11, ISSN 2250-3153, 2012. [3] Priti.K, and Paladurkra. M, TTC and CBM Calculation of IEEE-30 Bus System, Washington DC, USA, In Proceedings of the 2009 Second International Conference
on Emerging Trends in Engineering & Technology ICETET '09. [4] Jagdish Kumar, Performance of Indirectly Controlled STATCOM with IEEE 30-bus System, International Journal of Advanced Electrical and Electronics Engineering (IJAEEE) ISSN 2278-8948, Vol. 1, No. 2, 2012
[5] Amer. H, Thesis "Voltage Collapse Prediction for Interconnected Power Systems, West Virginia University, (2000). [6] Power Systems Test Case Archive, Universitiy of Washington. Website:
http://www.ee.washington.edu/research/pstca
Multi-Gene Genetic Programming for Short Term Load Forecasting

W. T. Ghareeb, Student Member, IEEE and E. F. El Saadany, Senior Member, IEEE
Department of Electrical and Computer Engineering University of Waterloo Waterloo, Canada
AbstractThe Short Term Load Forecasting (STLF) plays a critical role in power system operation. The accuracy of the STLF is very important since it affects the generation scheduling and the electricity prices and hence an accurate STLF method should be used. This paper presents a new variant of genetic programming namely: Multi-Gene Genetic Programming (MGGP) for the problem of STLF. In order to demonstrate this technique capability, the MGGP has been compared with the RBF network and the standard single-gene Genetic Programming (GP) in terms of the forecasting accuracy. The data used in this study is a real data set of the Egyptian electrical network. The weather factors represented by the minimum and the maximum daily temperature have been included in this study. The MGGP has successfully forecasted the future load with high accuracy compared to that of the Radial Basis Function (RBF) network and that of the standard single-gene Genetic Programming (GP). Keywords- Short-term load forecasting; radial basis function; genetic programming; multi-gene genetic programming.
networks. In this paper, the use of Multi-Gene Genetic Programming (MGGP) for the problem of short term load forecasting has been presented and studied. The MGGP has been compared with the RBF network and the standard singlegene genetic programming in terms of the forecasting accuracy. The performance index used in this study is the Mean Absolute Percentage Error (MAPE). This paper is organized as follows: A literature review is presented in section II. Section III discusses the forecasting techniques used in this study. The results are presented and discussed in section IV followed by a conclusion in section V and finally an acknowledgement is presented. II. LITERATURE REVIEW
I.
INTRODUCTION
The balance between the generated power and the actual demand is very important for the reliability of the power system. The generation scheduling is affected by the short term load forecasting and hence the short term load forecasting affects the reliability of the power system. The short term load forecasting also affects the economic operation of the power system. During the peak loads, the energy prices may be increased by a factor of ten or more [1]. For these reasons and since several countries have moved from the regulated energy markets to the deregulated energy markets, obtaining an accurate load forecasting became very important issue for the decision making processes of any electric utility [2]. However, having an accurate load forecasting is not an easy task due to the uncertainty imposed by the existence of exogenous variables such as the weather conditions. In the preceding decades, several load forecasting techniques have been studied and proposed by the researchers. These techniques can be divided into two main categories. The first category is the traditional methods such as regression techniques, kalman filtering techniques, time series and splines [1] and [3]. The second category is the artificial intelligence techniques such as neural networks, fuzzy logic systems, neuro-fuzzy systems and Radial Basis Function (RBF)
A. RBF network The application of the RBF network in STLF has been discussed in recent studies like [4-6]. An interpolation method has been used to process the input data in [4]. After processing the data, a comparison between using the grey theory and the RBF network has been introduced. In this study, the two methods have been used to forecast the peak load of 7 days and the performance index used was the RMSE. The weather factors are not included in this study. The use of the RBF network has been compared with the use of the multi-layer feed forward neural network in [5]. The weather factors have not been included in this study and the performance index was the Mean Relative Error (MRE). An improved RBF network has been proposed in [6] for the problem of STLF. This improved RBF network uses the nearest neighbor clustering algorithm to adjust the weights, the width and the centers of the RBF network. In this study, the weather conditions have been included as input data for the RBF network. This improved RBF network has been used to forecast the peak load for the next 24 hours. The performance index was the percentage error. B. Genetic programming The use of Genetic Programming (GP) in solving the STLF problem has been studied in [7-9]. In [7], the use of GP has been compared with the use of time series in solving the STLF problem. According to this study, the GP outperforms the time series method in terms of the forecasting accuracy. In this study, the load forecasting has been performed for the next 24
hour and the performance index used was the average relative error. In [8], the short term load forecasting based on meteorological factors has been studied using the GP. These metrological factors are the humidity, the temperature, the wind speed and the human comfort index. The GP has been used in this study in order to model the relation between these factors and the load size. However, in this study, the use of GP has not been compared with any other forecasting method in order to clarify the accuracy of using the genetic programming. The genetic programming has been utilized with the rough sets to develop a STLF model in [9]. The use of the GP with the rough sets has been compared with the use of the back propagation artificial neural network with the rough sets. According to this study, the use of the GP with the rough sets is more accurate than the use of the back propagation artificial neural network with the rough sets. The performance index used in this study is the absolute percentage error (APE). In this paper, a new variant of the GP; namely the multigene genetic programming (MGGP) has been introduced and compared with the standard GP and the RBF network. The MGGP has been applied to various fields of studies. It has been investigated in the modeling of complex structural engineering problems [10], the modeling of geotechnical and earthquake engineering systems [11] and the modeling of chemical process systems [12]. According to [12] and [13], the MGGP is more accurate than the standard GP approach for symbolic regression. Although the use of the standard GP has been applied previously in the area of STLF [7-9], the new variant of GP, MGGP, has not been applied or compared with the other methods in this area yet. C. Motivation As can be seen from the literature review presented in this paper, the use of genetic programming and the RBF network has been discussed and presented in several recent studies, these recent studies are trying and competing to discover and optimize the best STLF technique in order to achieve the best possible accuracy in the STLF problem. The aim of this study is to present and demonstrate the capability of the MGGP to provide an accurate short-term load forecasting. The MGGP has been studied and compared with the standard GP and the RBF network techniques based on the same input data, the same forecasting week and the same performance index. III. STLF TECHNIQUES UNDER THE STUDY
x is the input vector

f i (.) is the activation function for node . is the Euclidean norm wi is the weight between node and the output node.
ci is the RBF center for node

0 is the bias.
f1
f2
x1
x2
w1
w2
wnl
f ( x)
xn
f nl
Fig. 1 Schematic diagram of the RBF network The RBF network uses nonlinear transformation in the hidden layer and a linear transformation in the output layer. The weights between the input and the hidden layer are all equal to 1. Usually, non-linear transformation in the hidden layer is done by using the Gaussian kernel function which is given by
x ci 2 f i ( x ) = exp 2 i2
Where
(2)
i is the width parameter for node i

In order to train a RBF, a hybrid learning algorithm is usually used. This algorithm is used to determine the weights and the activation function parameters namely, the center and the width. This hybrid algorithm comprises of the use of unsupervised training to get the adaption of the activation function parameters and the use of supervised training to adapt the weights between the hidden layer and the output layer [14].
A. RBF network The RBF is a special category of the feedforward neural networks [14]. The basic idea of the RBF network is based on the function approximation theory. The radial basis function network consists of one hidden layer and one output layer. A basic RBF network with n inputs, nl hidden nodes and one output is shown in figure 1. The output from the RBF network can be given by [15].
B. Genetic programming B.1 The standard Single-gene genetic programming The genetic programming can be considered as a specialization of the genetic algorithms. In the genetic programming, the potential solutions are represented as sparse trees. The genetic programming is usually used to perform structural optimization like optimizing a computer program or a mathematical model. Figure 2 provides an example of a sparse tree representing a simple mathematical equation. In figure 2, the squares represent function nodes while the circles represent terminal nodes. The genetic programming algorithm starts by initializing the population. After this initialization, the
f (x ) = 0 + w i f i ( x c i )
i =1
nl
(1)
Where
mutation operator or the crossover operator are applied to this population based on the crossover probability and the mutation probability. There are two main types of mutation in genetic programming. The first type is to choose a certain function or terminal node randomly and replace it by a randomly generated one. The second type of mutation is to select a sub-tree randomly and then replace it by another randomly generated sub-tree. Figure 3 shows the first type of mutation, and figure 4 shows the second type of mutation.
In the crossover task, two parents are chosen probabilistically based on their fitness value, then a sub-tree is chosen randomly from the first parent and another is chosen randomly from the second parent and then these sub-trees are exchanged between the two parents to create two offsprings. The crossover task in genetic programming is shown in figure 5.
B.2 Multi-gene genetic programming In Multi-Gene Genetic Programming (MGGP), each individual of the population is a weighted linear combination of sparse trees. Each tree in this combination can be considered as a gene. An example of a MGGP representation of a potential solution is shown in figure 6, where x1, x2 and x3 are input variables and d0, d1 and d2 are the weighted coefficients. These coefficients can be determined based on the training data using ordinary least square techniques [13].
+
3
+ +
3
exp
+ *
sin
+ *
sin
X3
X2 X1
+
exp
X3
X2
y = 3 + e x2 + x1 sin x2 x3
Fig. 2 a sparse tree representing a simple mathematical equation.
X1
Parent 1
X2
X1
Parent 2
+
3
+ +
3
exp
+ *
sin
+ *
X1
exp
X3
X2 X1
+
sin
X3
X2
X1
X2
offspring 1
offspring 2
Fig. 3 function and node mutation in GP.
Fig. 5 crossover task in GP.
y = d 0 + d1 ( sqrt ( x1 x2 ) + 0.2 x3 ) + d 2 (cosh ( x2 ) + 1.4 x3 )

Fig. 4 sub-tree mutation in GP. Fig. 6 an example of MGGP representation of a potential solution.
Load (MW)
Two types of crossover are used in MGGP [13]. The first type is called low level crossover. This low level crossover is the standard GP sub-tree crossover. The second type of crossover is called the two-point high-level crossover. In this type of crossover, the genes are exchanged between the individuals. For example, if the first individual contains the following genes (G1 G2 G3 G4) and the second contains the genes (G5 G6 G7 G8), then the two point crossover can be done as follow: First, two groups of sub-genes are specified. (G1 G2 <G3 G4>) (G5 G6 <G7 G8>) Then, these genes are exchanged between the individuals resulting of the following individuals (G1 G2 G7 G8) (G5 G6 G3 G4) In MGGP, many different mutation operators can be used. These operators are as follow [13]: Random sub-tree mutation. Switch an input terminal to another selected input terminal by a random way. Gaussian perturbation of constants. Change the value of a randomly selected constant to zero. Replace a randomly selected constant with a randomly generated constant. Change the value of a randomly selected constant to 1.
maximum temperature and the day code with the peak load based on the training set of data.
2 1.9 1.8 1.7 MAPE=1.6656% 1.6 1.5 1 x 10
4
Actual load Forecasted load
7 Day reference
Fig. 7 the forecasted load using RBF network Due to the randomness included in the genetic programming algorithm, a different model has been provided for each execution of the GP algorithm. Ten different models have been obtained by running the GP algorithm for ten times based on the same set of training data. Then, each of these models has been used to forecast the peak load of week number 39. The worst MAPE value was 2.1125%. The best value was 1.4513%. The average MAPE value of the ten models was 1.8005%. Figure 8 displays the forecasted power against the actual power for one of these models.
2 1.9 Load (MW) 1.8 1.7 1.6 1.5 1 MAPE=1.7608% x 10
4
IV.
A real data set of the Egyptian electrical network has been used in this study. This data set includes the daily maximum temperature, the daily minimum temperature and the corresponding actual peak load. A day code has been assigned to each day to distinguish the weekends from the working days. The data set consists of 39 weeks. The first 38 weeks have been used for training, while the last week, week number 39, has been used to test the forecasting accuracy. Week number 39 has been given day references from 1 to 7 as shown in figures 7 to 9.
7 Day reference
Fig. 8 the forecasted load using the GP.
A. RBF In this case of study, the RBF has been simulated in the MATLAB. The number of neurons in the hidden layer has been set to 30. The orthogonal least square method has been used to determine the network parameters. Figure 7 shows the output from the radial basis function compared with the actual peak load. B. The standard Single-gene genetic programming In order to study the use of GP in STLF, the MATLAB GP toolbox provided in [16] has been used for studying both the standard genetic programming and the multi-gene genetic programming. The genetic programming has been used to develop a model for relating the minimum temperature,
C. Multi-gene genetic programming In this case of study, the use of MGGP in STLF has been studied. Due to the randomness included in the MGGP algorithm, ten different models have been recorded by running the MGGP algorithm for ten times based on the same set of training data. Each of these models has been tested by forecasting the peak load of week number 39. The worst MAPE reading was 1.8059%, and the best one was 1.2522 %. The average MAPE value of the 10 models was 1.5716%. Figure 9 displays the forecasted power against the actual power for one of these models. It can be noted that the MAPE values obtained from the MGGP are better than those obtained by the standard genetic programming. Table I summarizes the average MAPE values of the forecasting techniques under this
study. From this table, it can be noted that the best MAPE value has been obtained by the MGGP followed by the RBF network and then the Standard GP.
2 1.9 Load (MW) 1.8 1.7 x 10
4
[2]
[3]
[4]
[5]
1.6 1.5 1 MAPE=1.4895% 2 3 4 5 6 7 Day reference
[6]
Fig. 9 the forecasted load using the MGGP.

TABLE I. THE AVERAGE MAPE VALUES OF THE FORECASTING TECHNIQUES UNDER THE STUDY. [7]
Proposed system MGGP RBF network Standard GP V.
Average MAPE (%) 1.5716 1.6656 1.8005
[8]
CONCLUSION
[9]
In this paper, a new variant of genetic programming namely multi-gene genetic programming has been proposed for the problem of short-term load forecasting. To demonstrate the forecasting accuracy of the MGGP, it has been compared with the standard genetic programming and the RBF network in terms of the forecasting accuracy. The results obtained showed that the forecasting accuracy of the MGGP is superior to that of the standard GP. The forecasting accuracy of the MGGP can also compete with the accuracy of AI techniques like the RBF network. Further studies can be done in the future by utilizing the MGGP instead of the standard singlegene genetic programming, like studying the use of MGGP with the rough sets. ACKNOWLEDGMENT The authors would like to gratefully thank A. T. Ghareeb for providing the real data set on which the work developed in this paper is based. REFERENCES
[1] A. Khosravi, S. Nahavandi, D. Creighton and D. Srinivasan, Interval Type-2 Fuzzy Logic Systems for Load Forecasting: A Comparative
[10]
[11]
[12]
[13]
[14] [15]
[16]
Study, IEEE Transactions on Power Systems, Aug. 2012, Vol. 27, No. 3, pp. 1274 1282. M. Hanmandlu and B.K. Chauhan, Load Forecasting Using Hybrid Models, IEEE Transactions on Power Systems, Feb. 2011, Vol. 26, No. 1, pp. 20 - 29. S.-J. Huang and K.-R. Shih, Short-term load forecasting via ARMA model identification including non-Gaussian process considerations, IEEE Transactions on Power Systems, May 2003, vol. 18, no. 2, pp. 673679. Li Xiao-cong, Wang Le, Li Qiu-wen and Wang Ke, The Short-Term Load Forecasting Based on Grey Theory and RBF Neural Network, 2011 Asia-Pacific Power and Energy Engineering Conference (APPEEC), 25-28 March 2011, Nanning, China, pp. 1 - 4. M. Ghomi, M. Goodarzi and M. Goodarzi, Peak Load Forecasting of Electric Utilities for West Province of IRAN by Using Neural Network without Weather Information, 12th International Conference on Computer Modelling and Simulation (UKSim), 24-26 March 2010, Cambridge, UK, pp. 28 32. N. Dongxiao, J. Ling and T. Jie, Improved RBF network applied to short-term load forecasting, 2011 IEEE 2nd International Conference on Software Engineering and Service Science (ICSESS), 15-17 July 2011, Beijing, China, pp. 864 867. L. Huo, X. Fan, Y. Xie and J. Yin, Short-Term Load Forecasting Based on the Method of Genetic Programming, International Conference on Mechatronics and Automation (ICMA), 5-8 Aug. 2007, Harbin, China, pp. 839 843. Z. H.-rong, L. Y.-min and M. I.-mei, Based on meteorological factors and short-term load forecasting genetic programming, International Conference on Computer Design and Applications (ICCDA), 25-27 June 2010, Qinhuangdao, China, pp. V3-465 - V3-467. W.-C. Wang, C.-T. Cheng and L. Qiu, Genetic Programming with Rough Sets Theory for Modeling Short-term Load Forecasting, Fourth International Conference on Natural Computation (ICNC), 18-20 Oct. 2008, Jinan, China, pp. 306 310. A. H. Gandomi and A. H. Alavi, A new multi-gene genetic programming approach to nonlinear system modeling. Part I: materials and structural engineering problems, Neural Computing & Applications, 2012, Vol. 21, No. 1, pp. 171-187. A. H. Gandomi and A. H. Alavi, A new multi-gene genetic programming approach to non-linear system modeling. Part II: geotechnical and earthquake engineering problems, Neural Computing & Applications, 2012, Vol. 21, No. 1, pp. 189-201. Hinchliffe MP, Willis MJ, Hiden H, Tham MT, McKay B & Barton, GW, Modelling chemical process systems using a multi-gene genetic programming algorithm Proceedings of the First Annual Conference in Genetic Programming, 1996, USA, pp. 56-65. D. P. Searson, D. E. Leahy and M. J. Willis, GPTIPS: An Open Source Genetic Programming Toolbox for Multigene Symbolic Regression, International MultiConference of Engineers and Computer Scientists 2010 (IMECS), 7-19 March 2010, Kowloon, Hong Kong, pp. 77-80. F. Karray, C. De Silva, Soft Computing and Intelligent Systems Design: Theory, Tools and Applications. Addison Wesley, 2004. S. Chen, C.F.N. Cowan and P.M. Grant, Orthogonal least squares learning algorithm for radial basis function networks, IEEE Transactions on Neural Networks, March 1991, Vol. 2, No. 2, pp. 302 309. D. Searson, GPTIPS: Genetic Programming & Symbolic Regression for MATLAB, http://gptips.sourceforge.net, 2009.
Servo Motor Driver Design for High Performance Applications

Yusuf Yasa, Ergin Sahin, Cilem Acar, Aybuke Gozutok, Ecem Firat, Erkan Mese
Yildiz Technical University, Electrical Engineering Department
yasa@yildiz.edu.tr, erginshn@gmail.com, cilemacar@gmail.com, r.aybukegozutok@gmail.com, firatecem@gmail.com, emese@yildiz.edu.tr AbstractThe motion control applications have become quite common in recent years. This makes the design and control of the motor very popular. Especially in high performance applications, permanent magnet synchronous motors, which are often called servo motor, are used nowadays. In this paper, the process of high performance motor driver design will be presented. These stages are; main power circuit, supply circuit, fault protection circuit, ADC (Anolog Digital Converter) signal conditioning circuit, CAN (Controller Area Network) communication circuit, Mosfet drive circuit and microcontroller. To make maximum use of real estate in the driver, components and PCB card placements are optimized. The driver is supplied by battery as DC source and its energy management is important as an enabler of longer battery life. A solution is presented for a reasonable battery energy management. Some thermal analysis and subsequent design procedure is given for power switch heat sink. Simulations are performed at both circuit and system levels before the experimental prototype have been built. Validation of the driver prototype has been performed with a PMSM motor and the results are quite promising before transition starts into the production stage. Keywords; control of permanent magnet synchronous motor, electric motor driver design, servo applications.
In this paper, the stages of the designing a servo driver for PMSM will be explained. Study also considers thermal aspect of a driver design. The cost breakdown of the design is also given to shed light into production stage of the driver. II. PERMANENT MAGNET SYNCHRONOUS MOTOR
Permanent magnet synchronous motors (PMSMs) consist of two parts; stator and rotor. Rotor has permanent magnets to create constant magnetic field in electrical machine. Other time and space varying magnetic field is created with the stator windings. The windings excitation frequency should match with electrical rotation speed of the rotor. This signifies that the rotor positions must be known via either measurement or prediction. Resolver or encoder is one of the measurement techniques [1]. Mathematical model of PMSM is usually given in rotating reference frame. Quadrature (q) and Direct (d) axis PMSM equations are given below. Fig 1 shows equivalent circuit of PMSM in dq frame [2].
vqs Rsiqs
vds Rsids
d qs
I.
INTRODUCTION
In industrial applications, using of electrical equipment is rapidly increasing every year. As industrial developments occur, the requirements of high efficiency and high performance products increase at similar pace. Automobile production can be given as an example for that which has lots of robot applications. The motion capability of robot depends on the performance of motor drivers in robot arms. If each robot arm has good motion capability, the produced vehicle will be in good quality and the vehicle also will have high rate of penetration into the market. Making of high-performance robot arms would also contribute to an increase in number of daily production. In industry, precise position, speed or torque control is referred to be servo application. Permanent magnet synchronous motors (PMSMs) frequently are used in servo applications nowadays because this motor family has some attractive features such as low inertia, high efficiency, direct controllability of motor`s torque and high power density. PMSM, if it is aimed for servo applications, requires deep expertise in two subjects; first one is the design of the motor itself and second one is the design of the driver itself. Furthermore driver expertise demands both hardware and software knowledge.
qs dt ds ( Lld Lmd )ids mag Ldsids mag
dt d ds
ds
(1) (2) (3) (4)
qs ( Llq Lmq )iqs Lqsiqs

Te 3P mag iqs 22
Rs
Vsd
(5)
e sq sd
LId md Lmd If
isd
Figure 1. Direct axis equivalent circuit
Rs
Vsq
LIq sq mq Lmq
isq
Figure 2. Quadrature axis equivalent circuit
III.
DRIVER DESIGN
Servo motor driver circuit consists of three parts which are power stage, control stage and microcontroller. These parts can be seen in Fig 3.
DSP (Digital Signal Processing)
The power circuit which has 50 A maximum current in the dc bus have a power level of 1.05 kW. MOSFET family would be more appropriate in terms of price/performance for this power and voltage level. Because the output current of each phase does not exceed 25 A, 25 A current sensor would be suitable. The power circuit schematic is shown in Fig 4. Bill of material for the power circuit is given Table I.
TABLE I.
Element
POWER SUPPLY CIRCUIT
ERROR DETECTION CIRCUIT
CAN COMMUNICATION CURCUIT
ELEMENTS USED IN POWER CIRCUIT

Piece 6 4 3 1 Code IXTP98N075T KE2299B200V3550K LA25-P LV20 Price 6$ (6x1$) 13,2$ 45$ 30$
ADC OFFSET CIRCUIT
MOSFET DRIVER CIRCUIT
Mosfet Capacitor (2200F) Current Sensor
vDC
ia
ib ic
Gate Signals
POWER CIRCUIT
PMSM
Voltage Sensor
Figure 3. Driver Block Diagram
The ripple current in batteries causes the temperature rise, according to (6).
I: Current ripple R: Internal resistance of the battery A: Battery area H: Battery heat transfer coefficient
The PMSM drivers are built with three phase, two-level inverter topology in small power applications. Semiconductor switches in power circuit are driven by control signals for current and/or speed regulation. By using current, voltage and speed feedback signals, the full controllability of PMSM in wide speed range along with precise position control would become possible. Control part contains power supply circuit, error detection and protection circuit, ADC (Anolog Digital Converter) signal conditioning circuit, Mosfet driving circuit and CAN (Controller Area Network ) communication circuit. For high speed control of inverter, driver is controlled with floating point digital signal processor (DSP). By connecting the system to a Personal Computer (PC), all variables, definitions and conditions can be monitored easily. IV. POWER CIRCUIT
Due to the internal resistance of the battery, temperature increases with the current ripple [3].The rising temperature causes the reduction in battery life. In order to extend it, the ripple of the battery must be reduced. The problem can be solved by adding parallel capacitor to the battery. The whole system is simulated in MATLAB-Simulink. The results are given in Fig 5.
A. Power Circuit Design Three phase power circuit of inverter consists of 6 semiconductor switches, voltage sensor, current sensors and dc bus capacitors. The driver and PMSM are planned to use in a vehicle which has 24V dc bus created by a battery. Minimum battery voltage is assumed to be 21 V. So the PMSM motor drive system should be able to run at this lowest voltage level. Current sensors and voltage sensor are required for closed loop operation of the PMSM.
Figure 5. Battery current without(a) and with(b) dc bus capacitor
Figure 4. The power circuit of inverter
As seen from Fig 5, 8800 F capacitor would be enough to reduce current ripple to be less than 1% on the battery side [4].Table II shows the cost analysis of 8800 F capacitor. The price of 4 parallel connected 2200 F capacitors is more affordable compared to other options. Consequently, 4 units of 2200 F capacitor will be used in the driver.
TABLE II.
Value
CAPACITOR PRICE ANALYSIS

Piece 1 2 4 28$ 25$x2=50$ 3,3x4=13, 2$ Price
9000-10000F 4700 F 2200 F
that a reduction in the temperature corresponds to an exponential increase in the reliability and life expectancy of the device. Therefore, durable and reliable performance of a component may be achieved by effectively controlling the device operating temperature within the limits set by the device design engineers. Heatsink analytical design and selection is well documented in the literature [5-6]. Using these analytical techniques, a heatsink was selected based on device losses, thermal resistances, and heat sink thermal resistance values. Device and heat sink thermal resistances are matched so that excessive temperature rise is prevented in the devices junction. According to analytical calculations, case temperature (Tc) is estimated to be 47.49 oC and junction temperature (Tj) is estimated to be 52.96 oC. Validation of the analytical design of heatsink was done by using finite element analysis software Comsol Multiphysics. Temperature distribution in the heatsink cross section, outside of the heatsink and mosfet are given in Fig 6., Fig 7. and Fig 8, respectively. V. CONTROL CIRCUIT
B. Heat Sink Design Thermal management of the electrical and electronic devices in power converters becomes more important element in the overall electronic product design. Both reliability and life expectancy of an electronic equipment are inversely related to the local and global temperature rise in the equipment.
Figure 6. Heatsink Cross Section
A. Mosfet Driver Circuit Gate signals produced by the microcontroller must be amplified and isolated by the driver circuit before injected to the power devices [6]. With isolation, the electrical connection between two circuits is broken. Besides that new connection type such as an optical or magnetic connection can be established. Transformer creates magnetic link whereas optic element creates optic link between circuits. Isolation is provided with the opto-transistor in driver circuit.
ACPL-333J 100nF 5V 2.1k 0.1F 1 Vs 2 Vcc1 3 Fault 4 Vs VE 16 VLed 15 Desat Vcc2 14 13 100 D +15V 0.1F 100pF Drain 0V
Figure 7. Temperature of Heatsink

DSP Sinyali 1k
5 Cathode 6 Anode 7 Anode 8 Cathode
VEE 12 Vout 11 VClamp 10 VEE 9 10k
0.1F NPN
10k PNP 47k 10 Emiter 100nF -15 V 0V
Gate
Figure 9. Mosfet Driver Circuit
Figure 8. Mosfet Thermal Analysis
The relationship between the reliability and the operating temperature of a typical silicon semi-conductor device show
As can seen from Fig 9, amplifying process is generally implemented by using one npn and one pnp type transistor. In addition isolation is provided with isolated DCDC converter. Semi-conductor power switches must be operated under saturation, On the other hand they must not be excessively saturated because it reduces the switching speed of the device
and it increases the switching loss. Diode shown as D used to inhibit the excessive saturation operation. Elements used in mosfet driver circuit are shown in Table III.
TABLE III.
Element Driver IC Transistor Diode Element Resistor Capacitor -------
Designed signal conditioning circuit as shown in Fig 10 is used to provide offset voltage for AC signals which have negative value like current and voltage. Elements used in ADC circuit are listed in Table IV. C. Error Detection Circuit Processing speed of analog circuits are much faster than the microcontrollers. High response is required for the protection of circuit. Error detection circuit has a special significance in power electronics applications. Developed circuit can be easily integrated into any application of power electronics. Current signal information is compared with a threshold value which can be set by means of potentiometer. In case the current value exceeds threshold, circuit activates the error status and all gate signals are disconnected from the Mosfets.
10k +15V Current Signal 10k +15V
ELEMENTS USED IN MOSFET DRIVER CIRCUIT

Value Code ACPL-333J ZXTC2063E6 BAV21WS Value (Piece) 6 6 6 Piece
10 (6), 100 (6), 1 k (6), 2.1 k (6), 10 k (6), 47 k (6) 100 pF (6), 100 nF (6)
B. ADC Signal Conditioning Circuit To transfer information from an analog signal to digital for using in DSP, analog digital converter is generally used. Analog signal transfer function is given in (7). DSP has 12 bit ADC channel to read analog signals between 0 to 3V,
-15V +5V DL4002
-15V
DL4002
-5V +15V Possitive Output LM339 +15V Negative Output
LM339
(a) Comparator circuit
Where
Vin: Analog Input Voltage (0-3V), Vref+: 3V, Vref- :0V, n :12 D:Numerical Results (12 bit)
Possitive Negative Output Output
_ S
Led
SN74LVC541 OCTAL BUFFER
330
+5V +5V Output of sensor OP07 -5V +5
10k 10k 10k +10 +5
10k +15V
3.3V 100ohm 7k 3k MCP6024S

To ADC channel of DSP
_ R LATCH
LM224DR
Microcontroller Input
2.5uf
(b) Latch circuit
+3 +1.5 0
Figure 11. Error Detection Circuit

+3 +1.5 0
-5
Figure 10. ADC Signal Conditioning Circuit TABLE IV.

Element Opamp Opamp Opamp Element Resistor Capacitor ------Value OP07 LM224DR MCP6024S Value (Piece) 100 (4), 140 (3), 350 (1), 3 k (3), 7 k (3), 10 k (12) 2.5 F (4)
ELEMENTS USED IN ADC CIRCUIT

Code 1 1 1 Piece
The error detection circuit is shown in Fig 11. Even if faults temporarily occur, the error status will be still activated until DSP send reset signal to the circuit. Consequently, complete protection of the circuit is provided with the analog circuit. Elements used in ADC circuit are shown in Table V.
TABLE V. ELEMENTS USED IN ERROR DETECTION CIRCUIT
Element Opamp Opamp Latch Diyode Resistor Trimpot Value --------10 k 50 k Code ADTL084 LM339 MC14044B DL4002 ----Piece 2 2 1 2 5 2
D. Power Supply Circuit Power supply circuit is used to generate different levels of voltage for supplying different circuits and ICs. The Input voltage of the control circuit is fixed at 24 volt for controller stage. Different voltage levels such as +15V, -15V, +5V, +3,3V are required by driver. Voltage levels are implemented by using a regulator if isolation is not required. DC-DC converters are used to provide isolation and different voltage level. Power supply circuit is shown in Fig 12.
DCW03B-15
inductances and torque constant. PMSM under consideration has 16 pole pairs. Some specifications are given in Table VII. Simulation results can be seen between Fig 14 to Fig 17.
TABLE VII. PMSM MOTOR SPECIFICATIONS
Value 1.178 mH 1.81 mH 1.67 Nm/A-peak 0.069 V.s 0.0077 kg.m2 16
ia 15 ib ic
Paremeter Ld Lq Torque constant Flux linkage Inertia Pole pairs

20
24V 0,33F
+Vin
+Vout COM
+15V 2,2F 0V 2,2F -15V

10
-Vin
-Vout
(a) isolated 1 3
Currents (A)
24V 0,33F
7805 2
5V 2,2 F
5 0 -5 -10
(b) non - isolated
-15 -20 0 0.05 0.1 Time (s) 0.15 0.2 0.25
Figure 12. Isolated and non-isolated power supply circuits.
Figure 14. Simulation results of PMSM stator currents
Elements used in power supply circuit are shown in Table VI.

Load and Electromagnetic Torque (Nm)
35 30 25 20 15 10 5 0 0 Electromagnetic Torque Load
TABLE VI.
Element
ELEMENTS USED IN FEEDING CIRCUIT

Value 3w 8w 2,5 w Code DCW03B-15 DCW08B-15 LM78M05 Value (Piece) 0,22 F (15), 0,33 F (10) Piece 4 1 5
DC/DC converter DC/DC converter Regulator Element Capacitor
0.05
0.1 Time sec)
0.15
0.2
0.25
Figure 15. The variation of PMSM electromagnetic torque and load
VI.
MICROCONTROLLER
40 35 iq id iqref idref
Currents (A)
In this application, TMDSDOCK28335 series DSP used as a microcontroller which has floating point capability. This DSP has a processor speed of 150 MHz and it has special modules like PWM, eQEP and CAN communication for motor control.
30 25 20 15 10 5 0 -5 0 0.05 0.1 Time (s) 0.15 0.2
0.25
Figure 16. The variation of PMSM q and d axis currents and their references
40 35
Mechanical Speed (rpm)
Figure 13. TMDSDOCK28335 Series Microcontroller
30 nref 25 20 15 10 5 0 0 0.05 0.1 Time (s) 0.15 0.2 0.25 nmeas
VII. SIMULATION STUDY The permanent magnet synchronous motor (PMSM) is simulated under nominal operating conditions. The control of PMSM is based on vector control algorithm. Speed and current closed loop control are done by using MatlabSimulink. PMSM motor model parameters are taken from finite element analysis. These parameters include q and d axis
Figure 17. The variation of mechanical speed and reference speed of PMSM
In simulation study, 32 rpm speed is given as reference to the controller. To reach reference speed, q axis reference current is applied to the motor. D axis reference current is set to be zero to reach maximum moment point for a surface mounted PMSM according to the (8) [2]. Te is electromagnetic torque, is flux linkage and P is pole pairs. Related three-phase current and q-d axis current can be seen from Fig 14 and Fig 16, respectively. The load is changed suddenly at 0.1 sec. The dynamic performance of motor is quite satisfactory as seen in Fig 15 and Fig 17.
] (8)
Figure 21. nverter Circuit After Mounting
VIII. EXPERIMENTAL STUDY The inverter circuit is designed and put into practice by Proteus (isis & ares). The 3D designed circuit is shown in Fig 18 and Fig 19. It is composed of two parts as control board and power board. These are interlocked with each other by pins. In order to make maximum use of real estate and to create a compact design, special attention has been paid during PCB design. The inverter circuit is shown in Fig 20 and Fig 21.
Figure 22. Output currents of motor driver in experimental study
IX.
CONCLUSION
Figure 18. Control Circuit Design
Figure 19.Power Circuit Design
Fig 22 shows the output currents of motor driver. The currents are controlled and regulated with using vector control algorithm by exciting 5 kHz switching frequency.
In this paper, stages of driver design are described for permanent magnet synchronous motor which is used for high performance servo applications. Simulation of designed circuit is done by PC tools and then related circuit components have been selected by comparing the price/performance rates. In order to extend the life of the battery which is used for DC source, parallel capacitor will be used by connecting to the battery. Analysis results show that in order to achieve 1% current ripple in battery side, 8800 uF capacitor will be sufficient. Furthermore analytical design of heat sink is implemented then validation is done by using finite element analysis. Finally, some simulation and experimental results are given to demonstrate that the performance of motor and inverter is sufficient for servo applications. REFERENCES
[1] [2] [3] J. R. Hendershot and T.J.E Miller, Design of Brushless Permanent Magnet Motor Design, 2010, Magna Physics Publishing. Chee-Mun Ong, Dynamic Simulations of Electric Machiner: Using Matlab/Simulink, 2009, Prentice Hall PTR. A Technical Note from the experts in business-critical continuity,''Effects of AC Ripple Current on VRLA Battery Life'', Y. 2010 Nor, J.K., Art of Charging Electric Vehicle Batteries, WESCON/93. Conf. Rec., San Francisco, CA, pp. 521-525, 1993. Lythall,H.(2012),''Heatsinks'', http://www.sm0vpo.com/begin/heat0.htm#rate>, access:16 05 2013 Bodur, H.,''G Elektronii'', Y. 0029, 975-511-546-7, Birsen Yaynevi, stanbul, 2010
[4] [5] [6] Figure 20. nverter Circuit After Mounting
Optimal design of geometrical and physical parameters of permanent magnet machines in the purpose to reduce its vibratory behavior
N. Ferkha, M. R. Mekideche
Laboratoire LAMEL, Universit de Jijel BP 98, Ouled Aissa, Jijel, Algrie nassira.ferkha@gmail.com
A. Miraoui , A. Djerdir, A. O. n-diaye

Laboratoire Systme et Transport SET, Universit de Technologie de Belfort-Montbliard UTBM, 90010 Belfort Cedex, France
F. Peyraut
Laboratoire M3M, Universit de Technologie de Belfort-Montbliard UTBM, 90010 Belfort Cedex, France
Abstract Permanent magnet synchronous machines (PMSM) have high efficiency and torque density, and have already been employed in hybrid electric vehicles. However, one of their disadvantages is the inherent cogging torque, which is a kind of torque ripple and it would be better to minimize. This torque, sometimes, can be an important source of noise and vibrations. In this paper, the effect of the geometric characteristics of the stator on the vibratory behavior of electrical machines is illustrated. The optimum geometry for obtaining a minimum vibration level has been reached. For this purpose, an approach by using the Artificial Intelligent (AI) and the Finite Element Method (FEM) is proposed to solve the magneto-mechanical problem of geometrical parameters identification in the optimization process. The obtained results by Genetic Algorithm (GA) method have been presented. Keywords-vibration reduction; optimal design; permanent magnet synchronous machines; genetic algorithm; neural network; finite element method; magneto-mechanical coupling
(number of teeth, yoke thickness, width and height of the slots, height and width of permanent magnets ...) is generally dictated by the speed range, the design constraints of the converter, the height of the end winding, the iron losses and the copper losses. This study proposes a new structure which produces less vibration with a good compromise between torque and copper losses. With this regard, a genetic algorithm (GA) method based on inverse algorithm for geometric parameters identification of the machine, has been employed. In the proposed algorithm, a multilayer perceptron neural network (MLPNN) is used as forward model for resolving the magneto-mechanical problem, while the genetic algorithm is used to solve the optimization process in the inverse problem. II. RESOLUTION OF THE MAGNETO-MECHANICAL PROBLEM
I.
INTRODUCTION
Although permanent magnet synchronous machines (PMSM) seem to present numerous qualities, the vibrations and the emitted acoustic noise can limit their applications. The main causes of the vibration in a PMSM are due to the current, the slot harmonics and the permanent magnet producing radial magnetic force harmonics. These forces can induce some resonances with circumferential mode shapes of the stator [1]. To reduce the vibrations, we can act in two different ways, either on the driver, or on the geometric structure of the machine. The second approach has been considered in this study to reduce the vibrations. Furthermore, not all component parts of the electrical machines contribute in the same way to the vibratory behavior: it is known that the stator produces the majority of the vibrations. Our work should make possible to understand the effect of the stator geometry, on the reduction of vibrations. The choice of geometric characteristics in a permanent magnet machine
Vibrations in electrical machines are essentially due to the magnetic forces generated on their ferromagnetic parts. Thus, in order to model vibration phenomena, we should follow these steps: Determination of local magnetic force distribution, Calculation of the deformation of the considered parts of the machine.
A. Determination of Magnetic Field and Magnetic Forces Distributions The magnetic field is solved by a formulation in terms of magnetic vector potential A. The variational formulation is obtained by solving the following equation:
1 rot rot A J
(1)
where and J are respectively magnetic permeability and excitation current density. The magnetic force computation method is based on a local application of virtual works principle [2]. These nodal
forces have been calculated on each node, using the following equation:
F
and are the Lames coefficients

x 0 0 D y 0 z 0 y 0 x z 0 0 0 z 0 y x
[ P] [ A] dA s
T
(2)
where [P] and s represent respectively the magnetic stiffness matrix and the virtual displacement. B. Determination of the Dynamic Response The vibration computation is based on the use of the mode summation method where the dynamic equation (3) is solved in the space of eigenvectors [1][3].
] [C ] [d ] [ K ] [d ] [ F ] [ M ] [d
(3)
[M], [C] and [K] are respectively the mass, the damping and the stiffness matrix of the structure, [d] is the displacement vector, and [F] is the magnetic forces vector. In the case of the studied machine, the structure is considered as a free structure and the damping is considered to be negligible, therefore the equation of movement becomes:
In this case, the calculation of the resonance frequencies and their proper associated modes is going to be done from the equation (10).
([K ] i2 [ M ]) [ X i ] 0
(10)
where f i i / 2 are the natural (resonance) frequencies and [Xi] is the vector of proper modes. The vibratory response of the structure (displacement) is equal to a linear combination of associated mode shapes as follows [1][4]:
] [ K ] [d ] [ F ] [ M ] [d
(4)
To evaluate the vector of the nodal displacements [ d] in the structure (frame + stator), it is necessary to determine its corresponding mass and rigidity matrix. After that, we can compute the resonance frequencies of the mechanical structure and their associated proper modes.
[M ] [K ]
[d ]
i 1
NT
Xi
(11)
[ ]
e e
T e
[ M ]e [ ]e [ K ]e [ ]e
T e
(5) (6) (7) (8)
i are the modal coordinates associated to each mode X i.
[ ]
e
T e
[F ]
[ ]
[ F ]e
[ M ]e
[ K ]e
[N ]
Ve
T
[ N ] dVe
The finite element model described above is very expensive in terms of computing time, especially if an optimization stage is required. Moreover, a space memory problem may appear. In order to avoid these problems we have chosen to use a MLPNN for the forward modeling in the optimization process. So, in this case, the magneto-mechanical problem is replaced by the MLPNN at each iteration of the optimization problem. C. Neural Network use The using of a neural network (NN) needs three stages: learning, validation and test [5] [6] [7]. For this purpose, magneto-mechanical problem for several combinations of geometrical (yoke thickness ec and tooth shape and ) and physical parameters (Young modulo and mass density) have to be solved. For each combination we determine the vibration response at each node of the structure (frame + stator). These results will be divided on three bases: learning base, validation base and the test base of the MLP neural network, considering the vector of the geometrical and/or the physical parameters as the input vector of this network and the displacements vector as its output vector. Once it has the best performance, the NN will be used to represent the direct problem during the optimization phase based on GAs method.
[N ]
Ve
[ D] T [ H ] [ D] [ N ] dVe
(9)
[M]e and [K]e are respectively mass and stiffness matrix of an element, [F]e is the magnetic forces vector applied on an element, []e is the matrix of localization corresponding to each element, [N] groups the shape functions and is the mass density.
2 H 0 0 0 2 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
III.
GENETIC ALGORITHM AND OPTIMIZATION
The formulation of the IP has been based on the developed MLPNN model which states a relationship between the lists of geometrical and/or physical parameters and also a list of radial displacements [d]. So, we have to determine the best values of this parameters minimizing the vector [d]. For this need, we choose to minimize the sum of the elementary displacements. So for a first application we try to determine the optimal geometric parameters of the stator giving a minimum level of vibration. Thus, the treated IP-based GA technique returns to solve the following OP:
15 Min g d i i 1
determined below (problem (12)). The normalized representation is used to represent the parameters, uniform crossover and Gaussian mutation are used and an initial population of size 60 is built randomly. At each generation we choose the parameters having the six best objective functions as the new parents for the next iteration. IV. RESULTS AND DISCUSSIONS
As mentioned above, the developed model is applied for the study of a permanent magnet synchronous motor. The main parameters of this machine are summarized in table I. The computations were performed on a PC SONY with an Intel Dual Core P8600 2.40 GHz using the Matlab 7.5 language. The data are generated by means of a 2D finite element model. The mesh corresponding to the studied motor is shown in figure 2.
TABLE I. MAIN PARAMETERS OF THE STUDIED MACHINE Number of phases Number of pairs of poles Outer diameter of the stator [mm] Air gap diameter [mm] Core length [mm] Number of stator slots Number of permanent magnets
0.04 0.03 0.02 0.01
1.9 C 3.5
(12)
Pj 50
0.0278 ec 0.0317 ; 1.9977 8; 5.2871 6.7189
3 3 81.60 44.7 160 18 6
Where C and Pj represent respectively the torque and the copper losses constrains. GAs comprise a powerful stochastic optimization technique based on the processes of evolution theory [8] [9]. This method is reported to be suitable for a good approximate of global maximum or minimum value. Therefore, their use for the resolution of the problem (12) will allow us to avoid many difficulties that can be occurring using the other deterministic methods. The block diagram of GA is shown in Figure 1.
y (m)
Generation of initial parameters
0 -0.01
MLPNN
-0.02 -0.03
Evaluation Fitness
-0.04 -0.05 -0.04 -0.03 -0.02 -0.01 0 0.01 0.02 0.03 0.04 x (m) 0.05
Figure 2. Finite elements mesh
Best people: Optimal parameters
Yes
Error 0 No Selection
Generating a New Population
Crossing Mutation
Preliminary study on the mode shapes is realized to show the sensibility to parameters changes of natural frequencies. A complete investigation of the first eigenmodes shows that the important parameters, which can improve the modal characteristics of the stator, are: the yoke thickness ec and the tooth shape and as shown in figure 3. The choice of these parameters is also justified in order to keep simple stator geometry. ec
Figure 3. Geometric parameters
Figure 1. Chart of the proposed method
For GA, some parameters have to be determined. For instance, population size, range of parameters, selection and mutation operators have to be selected correctly. The ranges of variation of geometrical and physical parameters will be
A parametric study has been performed, either by a simple variation of one parameter. This analysis shows that yoke thickness is set apart from the other parameters because of its important influence on modal characteristics. The frequency of the second mode can be multiplied by 1.4, while the frequency of the third mode can be multiplied by 1.3 for the extreme values of the yoke thickness as it is shown in figure 4. This study specifies that the shape tooth parameter does not influence the machine resonance frequencies. For every mode, the stresses in the stator tooth are very small. However, the tooth shapes parameter acts particularly on mode 4 and 6. According to design constraints, the volume of the machine is held constant and the internal radius is not changed. Thus, when the yoke thickness increases, only the stator tooth length decreases.
5.5 5 4.5 x 10
4
with 56 neurons. So, the proposed MLPNN had 3 input variables (geometrical parameters) and 15 output nodes (radial displacements d).
Hidden layer Output layer Input layer
d1
ec
. . . . . . .
Figure 5. Architecture of the proposed MLPNN
-6
d2
. . .
d15
Mode 2 Mode 3 Mode 4 Mode 5 Mode 6 Mode 7
4.5
x 10
Proper frequencies (Hz)
Frquences propres (Hz)
4 3.5 3
Valeurs de test Test values Simulation RN RN simulation
3.5
Displacement (m)
29.5 30 30.5 31 Epaisseur de la culasse (mm) 31.5 32
2.5 2 1.5 1
Dplacement (m)
2.5
1.5
0.5 29
0.5
Yoke thickness (mm)
Figure 4. Natural frequencies for the first modes against yoke thickness ec
5 Points de la carcasse
10
15
Points of the frame
An examination of the electromagnetic values, like copper losses and torque, may be taken into account to confirm the validity of our approach. To compute these losses, the following simple formula has been used:
Pj cu j 2 k l N Rh s
Figure 6. Test of the proposed MLPNN

10
-7
10
-8
(13)
Fitness function
Fonction objectif
10
-9
10
-10
Where cu is the copper resistivity j is the current density (we assume that the current density is uniform), k is the winding factor, l is the mean length of turns, Ns is the stator tooth number, R is the air-gap radius and h is the stator tooth height. According to the above analysis of the influence of ( ec, , ) we have used the developed model, described above, to determine the optimal values of this parameters giving a minimum level of vibrations. The magneto-mechanical problem is replaced, in the optimization process, by the MLPNN presented in figure 5. For the training of this NN, a set of 504 parameters is simulated. This finite element simulation is carried a number of times equal to the number of sets of geometrical parameters. Several network configurations have been tried, and better results have been obtained by a network constituted by one hidden layers
10
-11
10
-12
10
-13
10
-14
10
-15
8 Gnrations
10
12
14
Figure 7. Evolution of the fitness function
Figures 6 and 7 show respectively, the test of the proposed MLPNN and the evolution of the fitness function. It can be remarked that a minimum displacement has been reached after 14 iterations for the optimal mechanical parameter presented in table II.
We have noticed that each iteration requires a smaller computation time. This justifies the interest of the implementation of this technique (NN-GA coupling) which is new in the area of magneto-mechanical coupled problems.
TABLE II. Parameters ec (mm) (Deg) (Deg)
0.04 0.03 0.02 0.01
y (m)
The second application concerns a simultaneous determination of geometrical and physical parameters giving a minimum level of vibrations. For the training of the NN, a set of 600 parameters is simulated. The proposed MLPNN architecture had 4 input variables (width and height h of the magnet, and Young modulo Es and mass density s of the stator), one hidden layer with 34 neurons and 15 output nodes (radial displacements d). The treated IP-based GA technique returns to solve the following OP:
15 Min g d i i 1
OPTIMAL PARAMETERS
Initial Optimal Computing Magnetic Copper values values time (min) torque (N.m) losses (W) 29.25 29.3 4.2284 2.6795 26 2.116 40.2567 6.7189 5.5844
Initial machine Machine initiale Optimal machine Machine optimale
1.9 C 3.8
137 10 9 E s 220 10 9 ;6900 s 8000
(14)
27.5994 57.5994; 3 10 3 ha im 6.5 10 3 ;
0 -0.01 -0.02 -0.03 -0.04 -0.05 -0.04 -0.03 -0.02 -0.01 0 0.01 0.02 0.03 0.04 x (m) 0.05
[ ] h aim . MLP d ; [ haim ] MLP C d c Es [ E s ] s s
(15)
Figure 8. Superposition of the initial and optimal machines TABLE III. Modes 2 3 4 COMPARISON OF NATURAL FREQUENCIES OF THE STATOR (HZ) Initial geometry 9938.2 14457 19316 Optimal geometry 13937 15161 23557
, haim, Es and s : optimization parameters. The range of variation of the parameter was considered based on the work
presented in [10], C: inequality constraints, which must be respected on the torque, MLPd and MLPc: neural networks, which replace the direct problem in the optimization process to define, respectively, the displacements di at each node of the structure and the torque.
Hidden layer
From table III we can show that the optimal structure of the stator has permit to increase its resonance frequencies. So, the probability of coincidences between proper frequencies of the structure and those of the magnetic forces is reduced. In this case, a minimum level of vibrations has been obtained as it is shown in the figure below.
18 16 14
e
Input layer
Output layer dd 1
Initial machine Gomtrie initiale Optimal machine Gomtrie optimale
haim s
Acceleration (m/s2) (m/s2) d'acclration Module
12 10 8 6 4 2 0 30
Es
. . . . . . .
d2
. . d. d
5 2
d1 15 1 d
Figure 10. Architecture of the proposed MLPNN (MLPd)
40
50 60 70 Angle mcanique (Deg) Mechanical angle (Deg)
80
90
The test of the proposed MLPNN and the evolution of the fitness function are represented respectively in figures 11 and 12. A minimal displacement (acceleration) has been reached after 11 iterations. The obtained optimal parameters have been regrouped in table IV.
Figure 9. Obtained displacement
1.4 x 10
-6
Valeurs de test Test values Simulation RN RN simulation 1.2
TABLE V.
COMPARISON OF NATURAL FREQUENCIES OF THE STATOR (HZ) Initial machine 9938.2 14457 19316 Optimal Optimal machine I machine II 13937 15585 15161 17038 23557 26022
Proper modes (Hz)

1
Displacement (m)
Dplacement (m)
0.8
2 3 4
0.6
0.4
V.
CONCLUSION
0.2
6 8 10 Points de la carcasse
12
14
Points of the frame
In this paper a new investigation of using MLPNN and GA for the optimal design of permanent magnetic synchronous machines has been presented. The geometrical and/or physical parameters have been changed, to obtain a minimum level of vibrations. The combination finite element-neural network allowed a significant reducing of computing time to resolve the magneto-mechanical problem. Using this technique we can consider a very important number of parameters that is not possible using classical techniques. This justifies the importance of the use of AI in this kind of problems. The optimization method proposed in this paper allows showing clearly the significance of physical parameters on the resonance frequencies of the machines. Finally, the obtained results emphasis the good contribution of the GA in IP resolution on vibratory problems of electrical machines. These results are satisfactory and open interesting prospects for the vibratory optimization in this kind of machines.
Figure 11. Test of the proposed MLPNN

7.5 x 10
-6
6.5
Fitness function
Fonction objectif
5.5
4.5
4 0 2 4 6 8 Gnrations 10 12 14
Generations
REFERENCES
S. Yu and R. Tang, Electromagnetic and mechanical characterizations of noise and vibrations in permanent magnet synchronous machines, IEEE Trans. Magnetics, vol. 42, no. 4, pp. 1335-1338, 2006. [2] A. Belahcen, Methods of calculating the magnetic forces for vibration and noise analysis in electrical machines, ACTA Polytechnica Scandinavia, Electrical engineering series, n 103, Finish Academies of Technology. Espoo, Finland, 2000. [3] M. Besbes and M. Gabsi, influence du nombre de dents sur le comportement vibratoire des MRV, Futer electrotechnique proceding, pp.404-408, 1999. [4] J. Pedro, A. Bastos, N. Sadowski Electromagnetic modeling by finite element methods, ISBN: 0-8247-4269-9, New York, 2003. [5] N.P. De Alcantara, J. Alexandre, M. De Carvalho, Computational investigation on the use of FEM and ANN in the non-destructive analysis of metallic tubes, Proc. 10th Biennial Conf. Electromagnetic Field Computation, Italy, 2002. [6] S. Haykin, Neural networks: A comprehensive foundation, Englewood Cliffs, NJ: Prentice-Hall, New York, 1999. [7] A. Fanni and A. Montisci, A neural inverse problem approach for optimal design, IEEE Trans. Magnetics, vol. 39, no. 3, pp. 1305-1308, 2003. [8] M. Srinivas, L.M. Patnaik, Genetic algorithms: a survey, Computer, vol. 27, no. 6, pp.17-26, 1994. [9] A. Nouicer, Applications des techniques rseaux de neurones et ondelettes aux dispositifs lectromagntiques, Thse de doctorat, Universit de Constantine, 2005. [10] D. GERLING, Analytical calculation of the PM machine magnetic field depending on magnet width and skewing, 13th European Conference on Power Electronics and Applications EPE '09, Barcelona, Spain, September 2009. [1]
Figure 12. Evolution of the fitness function TABLE IV. Parameters Initial values 43.05 4 190 7650 OPTIMAL PARAMETERS Optimal Computing Magnetic values time (min) torque (N.m) 42.21 3 13.97 1.97 219.8 6966.4
Machine initiale Initial machine Optimal machine Machine optimale I Optimal Machinemachine optimaleII II
(Deg) haim (mm) Es (x1011 N/m2) s (x10 3 kg/m3)

18 16 14
(m/s2) d'acclration Module (m/s2) Acceleration
12 10 8 6 4 2 0 30
40
50 60 70 Angle mcanique Mechanical angle(Deg) (Deg)
80
90
Figure 13. Obtained displacement
Voltage Drifting Method as an Islanding Detector for Inverter Based DG

Ebadollah Kamyab1,2
1
Javad Sadeh1
Electrical Engineering Department, Faculty of Engineering, Ferdowsi University of Mashhad, Mashhad, Iran
2
Department of Operation, Khorasan Regional of Electric Company, Mashhad, Iran settings. Simplicity and inexpensiveness are the main advantages of passive methods. Unfortunately, passive methods have an important drawback that has a relatively large non detection zone (NDZ) and they may not be reliable in all loading conditions [8]. Active islanding detection techniques are utilized for increasing the reliability and reducing the NDZ. Islanding can be detected by the active methods even if the DG generation matches the local load. Active methods introduce a small perturbation into the system. When the grid is disconnected, this small perturbation will make a significant change in system parameters, whereas the change will be negligible when the DG is connected to the distribution system. Both the active and the passive islanding detection techniques are used by hybrid methods. In these methods, the active technique is applied when islanding is not detectable by the passive techniques. In [9], a voltage positive feedback is designed in the synchronous d-q frame. It is applied to inverter based DG and voltage drifting method is used. The PCC voltage will be unstable due to positive feedback in islanding condition. In [10] for islanding detection the DC-link considering the PCC voltages are changed. Besides, there have been some various active methods like active frequency drift (AFD), active frequency drift with positive feedback (AFDPF) [11], slip mode frequency shift (SMS) [12], and automatic phase shift (APS) [13]. In [8], an active islanding detection method for inverter based DG is presented. The PCC voltage is drifted to detect the islanding phenomenon. Based on the presented results in this paper (Table I of [8]), the authors pointed out that the NDZ of their presented method, is small but not zero. Also, in islanding condition the voltage is always forced to increase. In this paper, the DG active power reference is determined as piecewise linear functions of voltage. This will be applied to DG as Pref after sensing the voltage variation. Using the proposed method, islanding is detected in the worst case in which the generation of DG matches the local load. Therefore, if the generation of DG and the local load are mismatched, detection of islanding will be guaranteed. The technique does not introduce any perturbations into the system; it is simple, fast and easy to implement. II. POWER SYSTEM, DG AND LOAD MODELING
Abstract In this paper an islanding detection method is presented for inverter based distributed generation (DG) using under/over voltage relay. The method is based on the design of DG active power reference (Pref) in inverter control interface. This active power reference has determined as piecewise linear function of voltage that applies after the voltage variation is sensed. Using the proposed method islanding can be detected in the worst case in which the output power of DG is the same as the active power of local load. Thus, the detection of islanding will be done in the other conditions in which the local load does not equal to the DG generation and a fraction of load is supplied from the grid system. The presented technique has no negative effect on the distribution system especially on the power system quality. The proposed method is evaluated under the IEEE 1547 and UL 1741 standards, using PSCAD/EMTDC software. Simulation results show that the proposed method operates correctly in the islanding and non-islanding conditions. Index Terms Inverter based distributed generation, Islanding detection, Non detection zone, Under/Over voltage relay.
I.
INTRODUCTION
Based on IEEE Std. 1547-2003 [1], distributed generation is defined as electric generation facilities connected to the power system through a point of common coupling (PCC) in the load site. Islanding is one of the inevitable occurrences which can take place. The islanding phenomenon occurs when a DG continues to feed power into the local load in condition that power flow from the grid system has been interrupted. Islanding phenomenon can be divided into intentional and unintentional groups. If the DG feeds the load independently in absence of distribution system, it can be aggravated the reliability and power quality, and also injured the safety of operational personnel who are repairing the distribution line. So, rapid detection of islanding phenomenon has a great importance [2]. The two main categories of islanding detection are remote and local methods. Local methods, in turn are divided into passive, active and hybrid methods. Remote techniques are based on communication between the utility and the DGs. Supervisory control and data acquisition (SCADA) [3] and power line signaling scheme [4]-[6] are used in these techniques. Although these techniques are more reliable than the local ones, they are not economic especially for small DGs [7]. Local techniques depend on the information and data at the DG site. In passive methods, islanding can be detected by setting upper and lower thresholds on a parameter of the power system in DG site. In order to detect the islanding, the parameter deviation should be greater than the threshold
Fig. 1 shows the system under study, consisting of a threephase network with a grid connected DG system and a local load. The DG is an inverter based one that is connected to a 380V system through a 100kVA transformer. The grid system is modeled by a 20kV line to line voltage source with 125
MVA short circuit capacity. The grid is connected to 0.380kV system using a 20kV/0.380kV, 100kVA transformer and a 380V distribution line. In this paper, distributed generation is modeled as a constant power with an interface control as proposed in [14] The DG operates at unity power factor (Qref =0). The DG interface control variables are controlled in the d-q axis synchronous reference frame.
T2 0.380/0.380 kV
VP CB
Line 380V
T1 0.380/20 kV
Rf
Lf
where, VP is pre-islanding and V'P is post-islanding of PCC voltage. Also Pl0=PlBI, PlAI, PDG, P and R are pre-islanding load active power, post-islanding load active power, DG active power, active power exchange between DG and the grid and local load resistance, respectively. Using (3) and (4), the load active power can be calculated as the following equation: '2 '2 '2 VP VP VP (5) PlAI = 2 PlBI = 2 P0 or P DG = 2 Pl 0 VP VP V0 Also, from (3) and (4), the active power exchange between the DG and the grid can be calculated as follows:
2 1 VP )= P DG (1 2 ) (6) '2 VP r iv where, riv=V'P/VP is called the islanding voltage ratio. The NDZ can be deduced by substituting upper and lower bounds of voltage (riv=1.1 and 0.88) into (6) when VP=V0, which makes the following inequality: P %29.132 < 100 < %17.355 (7) P DG The negative and positive sign means that the active power exchange is from the grid to the local load and from the DG to the grid, respectively. If active power exchange between the DG and the grid has been in the limit of (7), the islanding phenomenon can not be detected. For islanding detection, in such conditions, the postislanding voltage should be deviated out of the settings of the upper and lower of U/O voltage relay. It is worth noting that the worst case of islanding detection happens when the power mismatch between the load active power and DG generation is almost zero. In such condition, there is no power exchange between the DG and the grid in nominal voltage. In order to clarify the subject, suppose, Pl =P0=PDG (P=0) in nominal voltage V0. Besides, suppose that the system voltage changes to V1 due to non-islanding phenomena. Then, Pl =P0(V1/V0)2 and . In such condition, if islanding is occurred, the PCC voltage will be changed from V1 to V0 after islanding. On the other hand, by the conventional islanding detection method, islanding detection is not possible. But because the pre- and the post-islanding PCC voltage are not equal, it is possible that the islanding phenomenon is occurred. Using the proposed method, the islanding will be detected even in such cases. In other words, the PCC voltage variation is sensed by the voltage measurement and Pref switches from Pref0 to new Pref and as a result, the proposed method drifts the voltage deviation to settle out of the above mentioned limits. It in turns the islanding is detected. Since the proposed method can detect islanding in this case, the islanding detection can be guaranteed in other cases i.e. when there is a mismatch between the local load active power and DG generation. In order to present the islanding detection technique without NDZ, (6) can be rewritten as follows: 2 2 P r iv = P DG ( r iv 1) = P DG ( r iv + 1)( r iv 1) (8) In (8), if the value of riv in the first bracket takes 1.1 or 0.88, then (9) and (10) can be written. 2 P r iv 2.1 P DG ( r iv 1) (9) 2 P r iv 1.88 P DG ( r iv 1) (10)
PDG+jQDG VT
Inverter
Pload +jQload P+jQ
P = P DG (1
Rg Lg G
DG
Local Load
20 kV grid
Fig. 1. A grid connected DG system with local load.
Variables of d-q axis and DG interface control are used to calculate the modulation index and phase angle. The load active and reactive powers as function of voltage and frequency are expressed as (1) and (2) [15].
V Pl = P0 V 0
(1+ k P f )
NQ
NP
(1)
V Ql = Q0 (2) V (1+ kQ f ) 0 is the where V0 represents the system nominal voltage, voltage index of load active power, is the voltage index of load reactive power, and are active and reactive powers voltage dependency factors, is frequency deviation in per unit and P0 and Q0 represent the load active and reactive power corresponding to the nominal voltage. The indices vary between 0 and 2. Note, and can be changes from 0 to 3 and 0.2 to 0, respectively [15].
III. PROPOSED ISLANDING DETECTION METHOD If islanding phenomenon occurs, for islanding detection by U/O voltage relay, post-islanding PCC voltage should be passed the lower/upper thresholds which are 0.88/1.1 of nominal voltage [16]. If the active power exchange between the DG and the grid before islanding (P in Fig. 1) is small, the PCC voltage remains in the above mentioned limits and islanding detection is not possible. As mentioned in the authentic references [16], [17] loads with (parallel RLC) are the worst case for islanding detection. Therefore, the method is formulated for parallel RLC load. The following equations can be used for determining the NDZ considering before and after islanding PCC voltages. 2 2 V V Pl 0 = PlBI = P = P2 P0 = PDG P (3) R V0
PlAI =
'2 '2 V P VP = 2 P0 = PDG R V0
(4)
To detect the islanding, the reference of DG active power (Pref) in DG interface control is set as (11) and (12) after voltage variation for riv01 or riv0<1, respectively. riv01 P ref = P ref 0 + 2.1 P ref 0 (r iv 0 1) (11) riv0<1 P ref = P ref 0 + 1.88 P ref 0 ( r iv 0 1) (12) where, r =V V and Pref0 is the DG active power setting when the system operates at normal condition. In Fig. 2 the curve of load active power, i.e. equation (1) for P0=Pref0 and curves of (11) and (12) are drawn (curve C1 and lines L2 and L3, respectively). The intersection points of L2 and C1 are at riv0 equal to 1 (point A) and 1.1 (point B). Also the intersection points of L3 and C1 are at riv0 equal to 1 (point A) and 0.88 (point D). In this figure, point B is a stable operating point while point D is not. In order to show that the point B in Fig. 2 is stable, suppose this point be the operating point and a small deviation takes place at voltage. For instance, if the deviation is positive, DG active power reference will be smaller than the load active power (Pref <Pload). In such condition, Pload is decreasing to reduce the difference between Pref and Pload. Based on (1), in order to decrease the load active power, the PCC voltage should be decreased. It means that for positive voltage deviation, operating point will come back to point B. The same procedure will be take place if the deviation of voltage is negative. Hence, it can be concluded that point B is a stable operating point. To show that point D is unstable; suppose a small positive deviation of the PCC voltage take place at D (Pref >Pload). In this condition, Pload is increasing to reduce the difference between Pref and Pload. According to (1), in order to increase the Pload, the PCC voltage should be increased. It means that the operation point will have more deviation from D. Thus, point D is unstable operating point. In addition, if the islanding phenomenon occurs and riv01, it is possible that for some local loads there is a stable operating point between points with riv0=1 and riv0=1.1 (between points A and B) as well as between points with riv0=1 and riv0=0.88 (between points A and D). Therefore, in islanding condition, operating point can reach to one of these points and settle. Besides, point A is an unstable operating point in condition that (11) has been active power reference of DG, but for Pref as (12) point A is stable. Therefore, for riv0<1, the operating point move away from point A, but it comes back again. On the other word, the islanding detection cannot be guaranteed or the NDZ is not zero. Modification of these curves to reach to zero NDZ is the main goal of this paper. Suppose, Pref is set as Pref0 in normal condition (riv=1), while, Pref is determined as two lines that are functions of PCC voltage in abnormal PCC voltage (riv1). Section CAB shows (11) and point C along line AB is determined in such a way that the first point of intersection of line CD with load active power curve passes through point D (riv0=0.88). Point C on line AB is not unique and numerous points can satisfy the above condition. In other words, the above condition can be satisfied if point C is on line AB and 0.88<riv0<0.95. For instance, if C is selected on line AB where riv0=0.9, then line CD can be written as (13) that is drawn as L4: (13) P = P + 0.78 P (r 1) 0.132 P where D is the intersection point of C1 and L4. In this condition, if post-islanding PCC voltage is lower than its nominal voltage, then, the section AC and the line CD
iv 0 ' P 0
are the DG active power reference. A positive deviation in voltage at the intersection point of C1 and L4 (D) means Pref<Pload, then based on (5), the PCC voltage will be decreased to reduce the Pload. This means that the PCC voltage will be attracted to the intersection point of C1 and L4 (D). The same procedure will take place if the deviation of PCC voltage is negative. Thus, in this condition the intersection point of C1 and L4 (D) in Fig. 2 is a stable operating point. In Fig. 2 the stability condition of point A is the same as the stability condition of the intersection point of C1 and L2 (point D), or A is not a stable operating point. Therefore, if the inequality riv01 is satisfied i.e. the active power flow is from DG to the grid in nominal voltage, then, in islanding condition, the postislanding PCC voltage will be increased until inequality riv01.1 is satisfied. This means that islanding phenomenon is detected. In conditions that islanding phenomenon occurs and inequality riv0<1 is satisfied, the Pref is consist of lines AC and CD. Because point D (the intersection point of C1 and L4) is a stable operating point and point A is not stable, the PCC voltage will decrease till inequality riv00.88 is satisfied. Thus, in this situation, islanding is detectable, too. If the proposed method is used and the PCC voltage changes due to non-islanding conditions, the DG active power generation will be changed according to (11) and (13). In such cases, if the PCC voltage stays between the thresholds and stable for more than 500msec, the DG Pref can change to initial setting (Pref0). The property of the zero NDZ of the proposed method can be revealed referring to Fig. 3. In this figure, the curves C1-C5 are the load active power with the following equation and different values for P0.
1.4
Pref0
C1: Pl=Pref0riv02 L2: Pref=Pref0+2.1Pref0(riv0-1) L3: Pref=Pref0+1.88Pref0(riv0-1) L4: Pref=Pref0+0.78Pref0(riv0-1)-0.132Pref0
1.3
C1
B
1.2
P(MW)
L2
1.1
A L3 L3
0.9
0.8
L2
C
riv0=V'P/V0 Fig. 2. Curves of load active power and DG active power reference versus riv0
0.7 0.86
L4
0.9 0.94 0.98 1.02 1.06 1.1 1.14
Pl = P0(V'P/V0)2 (14) where, P0 is the load active power when V'P/V0=1 and is shown in Fig. 3 as curve C1. If the grid has been connected, then in nominal voltage the operating point of the system is at riv0=1 (point A). Suppose, that the islanding phenomenon occurs and
ref
ref 0
ref 0
iv 0
ref 0
Pref0 Load Equation: Pl=P0(V'P/V0)2

C1: P0=Pref0
1.8
C2: 0.975309Pref0<P0=0.985Pref0<Pref0 C3: 1.00227Pref0>P0=1.002Pref0>Pref0 C4: P0=1.05Pref0>1.00227Pref0
1.6
C5: P0=0.95Pref0<0.975309Pref0
C3
C1
1.4
P(MW)
C5 C4
1.21Pref0 E"
1.2
B C2
L1: Pref =2.1Pref0(riv0-1)+Pref0
L2: Prefi =0.78Pref0 (riv0-1)-0.132Pref0+Pref0

E' Pref0
1
F'
0.8
D" F D'
C
0.85
0.9
0.95
1.05
riv0= (V'P/V0)
1.1
1.15
1.2
1.25
1.3
Fig. 3. Power versus riv0 for different loads and DG active power reference characteristics.
inequality riv01 is satisfied, then, the active power reference of DG is defined as (11) as mentioned earlier. It intersects the load curve at a point that riv01.1, which is a stable operating point. Because point A is not a stable point in islanding condition, the PCC voltage is increased to satisfy the inequality riv01.1. It means that the islanding is detected. If inequality riv0<1 is satisfied in islanding phenomenon, active power reference of DG is defined as (11) for riv00.9 and is defined as (13) for riv0<0.9. It intersects the load curve at a point that riv00.88 which is a stable operating point. Therefore, the PCC voltage will be decreased to inequality riv00.88 is satisfied. It means that the islanding is detected. It can be seen from Fig. 3, for loads that P0 is less than Pref0 (curves C2 and C5), the inequality riv0 > 1 is satisfied. The load curve intersects line AB at a point that its riv0 is greater than 1.1 (it is not shown in Fig. 3). Also, for a load like curve C2, it intersects line L2 at point D' and line L1 in section CA at point D". It is evident that the stability condition of the points D' and D" are the same as the stability of the intersection points of C1 and L4 (D and A) in Fig. 2, respectively. Therefore, the former is stable while the latter is not. Thus, the method can be in risky condition. It can be shown that the intersection points D' and D" exist if the inequality of 0.975309Pref0<Pl0<Pref0 is satisfied. Since in this condition P0 is smaller than Pref0, the riv0 ratio is greater than 1. In contrast, the active power reference of DG does not consist of section CA of line L1 and line L2. It
means, points D' and D" cannot be considered as operating points. Because point A is an unstable operating point in islanding condition, the voltage is increasing to reach to a point that its riv0 is greater than 1.1, which in this case is a stable operating point. Therefore, islanding phenomenon is detected. On the other hand, if P0>Pref0 the inequality riv01 is satisfied. The load curves C3 and C4 intersect line L2 at point F and F', respectively. Also, for some loads like curve C3 it may be intersect the section AB of line L1 at points E' and E". The stability condition of the points E' and E" are the same as the point B and the intersection point of C1 and L4 (point A) in Fig. 2, respectively. So, E' is stable and E" is unstable. It can be shown that the intersection points E' and E" exist if the inequality of Pref0<P0<1.002273Pref0 is satisfied. Since P0 is greater than Pref0, so, riv0 is smaller than 1. Therefore, section AB is not the active power reference of DG. It means that points E' and E" cannot be considered as operating points. Because the intersection point of C1 and L4 (A) is an unstable operating point in islanding condition, the voltage is decreasing to reach to a point that its riv0 is less than 0.88, which in this case is a stable operating point (for example points F and F' in Fig. 3). Therefore, islanding phenomenon is detected again. So, the islanding is detected correctly in all cases by U/O voltage relay. On the other hand, the NDZ is only one point in which P0 is exactly equal to Pref0 that is an unstable operating point. Therefore, the NDZ of the proposed method is reduced to zero. Based on (1), at nominal system frequency and
voltage, the active power load is P0. When the system frequency is changed, the load active power equation will be changed and in nominal voltage is equal to P (1+ k f ) . As mentioned earlier, if kP=0, the method has been formulated for the worst case i.e. P0=Pref0. In condition that kP0, the worst case is taken place when P (1+ k f ) = P . Because the coefficients of the Pref formula (equations (11) and (13)) is Pref0, the method formulation is valid so. On the other hand if the changing of voltage is sensed, the active power reference will be switched from Pref0 to piecewise linear function of voltage and islanding phenomenon is detected if it takes place. It is generally accepted that the parallel RLC load (NP=2) which has a high quality factor and operate at resonance condition (unity power factor) is the worst case for islanding detection [16], [17]. Therefore, in the simulation results, the DG is designed to operate at unity power factor and the local load is modeled as a constant RLC load. If in nominal voltage the mismatch between the local load active power and DG generation is high enough, the voltage deviation in post-islanding is less/more than the U/O voltage relay settings. Therefore, application of the proposed method is highlighted when the active power imbalance is low. So, the simulation results which are presented in the next section, some cases have been considered that the power exchange between the DG and the grid satisfy inequality (7).
0 P
0 P ref 0
1.6 PCC Voltage(pu) 1.4 1.2 1.1 1 0.8 0.88 2.5
1- Pload = 50 kW 2-Pload =49.5 kW 3-Pload =50.5 kW 4-Pload =64.5 kW 5- Pload =41 kW
5 3 1 4
3 Time(sec) a) 3.5 4
200 DG active power(kW) 150
5
100 50
3 4
0 2.5
3 Time(sec) b)
3.5
Fig. 4. a. PCC voltage and b. DG active power variation for different loading.
1.4 PCC Voltage(pu) 1.2 1 0.8 0.6 0.4 2.5 3
1.1
NP=2
0.88
NP=1 NP=0
Time(sec)
3.5 4
IV. PERFORMANCE OF THE PROPOSED METHOD A. Islanding conditions The proposed method is implemented on the study system shown in Fig. 1. The islanding phenomenon occurs at t=3sec. The DG generation is considered to be 50 kW while the local load assigns different values like 50 kW, 49.5 kW, 50.5 kW, 64.5kW (lower bound of U/O voltage relay setting or P= 0.2913PDG) and 41kW (upper bound of U/O voltage relay setting or P=0.17355PDG). The DG active power reference is set as Pref0 in normal condition. Dependent on the value of riv0, 50msec after the voltage variation is sensed; the Pref is set as (11) or (13). The variations of the PCC voltage and DG active power with utilization of the proposed method are depicted in Fig.4. If the proposed method is not used and the difference between the local load active power and DG active power generation is zero or small. In these cases the islanding phenomenon cannot be detected by U/O voltage relay, while using the proposed method the voltage variations are enough for islanding detection. The DG active power is decreased or increased until the PCC voltage passes the U/O voltage relay thresholds. Note that all of the above simulation results show that the islanding phenomenon is detected by the presented technique at the permitted time. Fig. 5 shows the PCC voltage drifting for different active power load voltage indices (NP) and Pload=Pref0=50kW with Qf =2.5. It is shown that the islanding detection for NP =0 and 1 is much simpler and faster than for NP =2. In Fig. 6 the performance of the proposed method is shown for non-zero kP (kP=3) and it compares with kP=0. The resonance frequency of the local load (fr) is considered 60.4 and simulations are done for P0=50kW and 49kW (P0=49kW is the worst case of islanding detection for kP=3 and fr=60.4Hz).
Fig. 5. PCC voltage for NP= 0, 1 and 2 and P0=50kW.

1.3 PCC Voltage(pu) 1.2 1.1 1 0.9
1 2 3 4
P0=50kW and kp =3 P0=49kW and kp =3 P0=50kW and kP=0 P0=49kW and kp =0
1 0.88
3
0.8 2.5
3.5 Time(sec)
Fig. 6. PCC voltage for k P =0 and 3 and P0=50, 49kW
B. Non-islanding conditions In this section the performance of the proposed method has been tested on the system under study in non-islanding phenomena. The load is a 50kW RLC load. The phenomena like system voltage variation, capacitor bank switching and variation of local load have been considered. Results of these studies are presented in Figs. 8. Fig. 7-a shows the PCC voltage variation when the system voltage decreases or increases 10% at t=3sec as a step function, whereas the proposed method is used or not. Fig. 7-b shows the curve of the PCC voltage variation for capacitor switching; at t=2sec a 200F capacitor bank (10.89kVar) had been paralleled to the local load and at t=3sec it is switched off. The simulation is done in condition that the proposed method is used or it is not used. Also, Fig. 7-c illustrates the results when another 50 kW load with Qf =1.8 is switched on in parallel with the existing load at t=2sec and switched off at t=3 sec.
1.1 PCC Voltage(pu) 1.05 1 0.95 0.9 0.85 2.5
with using the proposed method without using the proposed method
zero. However, if the PCC voltage changes due to nonislanding phenomena, the effect of the proposed method on voltage variation is very small. The effectiveness of the proposed algorithm is demonstrated by simulation results. In addition, it is shown that for all cases, islanding phenomenon is detected at the permitted time.
4
3 Time(sec) a)
3.5
REFERENCES
[1] [2] IEEE Standard for interconnecting distributed resources with electric power systems, IEEE Standard. 1547-2003, Jul. 2003. S.K. Salman, D.J. Kin and G. Weller, New loss of main detection algorithm for embedded generation using rate of change of voltage and changes in power factors, Developments in Power System Protection, Conference Publication No.479, IEE 2001. M.A. Refern, O. Usta and G. Fielding, Protection against loss of utility grid supply for a dispersed storage and generation unit, IEEE Trans. on Power Del., vol. 8, no. 3, pp. 948-954, Jul. 1993. M. Ropp, K. Aaker, J. Haigh and N. Sabhah, Using power line carrier communications to prevent islanding, Proc. 28th IEEE Photovoltaic Specialist Conference, pp. 1675-1678, 2000. W. Xu, G. Zhang, C. Li, W. Wang, G. Wang and J. Kliber, A power line signaling based technique for anti-islanding protection of distributed generatorspart I: scheme and analysis, IEEE Trans. on Power Del., vol. 22, no. 3, pp. 1758-1766, Jul. 2007. G. Wang, J. Kliber, G. Zhang, W. Xu, B. Howell and T. Palladino, A power line signaling based technique for anti-islanding protection of distributed generators part II: field test results, IEEE Trans. on Power Del., vol. 22, no. 3, pp. 1767-1772, Jul. 2007. P. Mahat, Z.Chen and B. Bak-Jensen, A hybrid islanding detection technique using average rate of voltage change and real power shift, IEEE Trans. on Power Del., vol. 24, no. 2, pp. 764-771, April 2009 . H.H. Zeineldin and J.L. Kirtley, A simple technique for islanding detection with negligible non-detection zones, IEEE Trans. on Power Del., vol. 24, no. 2, pp. 779-786, April 2009. T.T. Ma, Novel voltage stability constrained positive feedback antiislanding algorithm for the inverter-based distributed generator systems, IET Renew on Power Gener., vol. 4, no. 2, pp. 176-185, March 2010. H.Vahedi, R.Noroozian and A.Jalilvand, Gharehpetian G.B., A new method for islanding detection of inverter based distributed generation using DC-link voltage control, IEEE Trans. on Power Del., vol. 26, no. 2, pp. 1176-1186, April 2011. M.E. Ropp, M. Begovic and A. Rohatgi, Analysis and performance assessment of the active frequency drift method of islanding prevention, IEEE Trans. on Energy Convers., vol. 14, no. 3, pp. 810816, Sept. 1999. G.A. Smith, P.A. Onions and D.G. Infield, Predicting islanding operation of grid connected PV inverters, Proc. IEE Electric Power Applications, vol. 147, pp. 1-6, Jan. 2000. G. Hung, C. Chang and C. Chen, Automatic phase shift method for islanding detection of grid connected photovoltaic inverter, IEEE Trans. on Energy Convers., vol. 18, no. 1, pp. 169-173, Mar. 2003. X. Wang, W. Freitas, W. Xu, and V. Dinavahi, Impact of DG interface controls on the sandia frequency shift anti-islanding method, IEEE Trans. on Energy Convers., vol. 22, no. 3, pp. 792794, Sept. 2007. P. Kundur, Power System Stability and Control, McGraw Hill Inc. 1994. IEEE Recommended practice for utility interface of photovoltaic systems, IEEE Standard 929-2000. UL 1741 Standard for Converters and controllers for use in independent power systems, UL 1741Standard, 2001 UL 1741 Standard for Static inverter and charge controllers for use in photovoltaic systems, UL 1741Standard, 2001.
1.08 PCC Voltage(pu) 1.06 1.04 1.02 1 0.98 1.5 2 2.5 Time(sec) b)
1.05 PCC Voltage(pu) 1 0.95 0.9 0.85 1.5 2 2.5 Time(sec) c) 3 3.5 4
3.5
[3]
[4]
[5]
Fig. 7. PCC voltage for different cases. a. system voltage change b. capacitor switching. c. load switching
[6]
C. Effects of load quality factor on the proposed method As a fact, the islanding detection is more difficult for loads with high quality factor [18]. Therefore, this subject has been mentioned in different references [17], [18].
1.1 PCC Voltage(pu) 1.05 1 0.95 0.9 0.88 0.85 2.5 3 Time(sec)
[7]
[8]
1-Qf =0.5 2-Qf =1.5 3-Qf =2.5 4-Qf =4
[9]
3,4 2
3.5 4
[10]
Fig. 8. Effect of quality factor on the performance of the proposed method
In this paper, in order to investigate the performance of the proposed method, different load quality factors have been considered. Fig. 8 shows the simulation results for different load quality factors. It can be seen that for each quality factor, islanding phenomenon is detected at the permitted time. V. CONCLUSION In this paper a new islanding detection technique is presented. The method is based on the change in the DG active power reference. The DG active power reference is set as a constant value Pref0. If the PCC voltage changes the Pref is determined as piecewise linear functions of PCC voltage. If islanding phenomenon occurs the proposed method makes the PCC voltage to pass to the upper or lower thresholds of the U/O voltage relay. The proposed method is reliable in all cases, that it means the method detects the islanding phenomenon regardless of amount of mismatch in active power between the local load and DG generation. If the voltage variation is sensed due to islanding phenomenon, the Pref will be changed and as a result, islanding is detected. On other words, the NDZ of the proposed method is only one point in which the power exchange between DG and grid is
[11]
[12]
[13]
[14]
[15] [16] [17] [18]
Design of a Mechanical Load Simulator Having an Excitation Current Controlled Eddy Current Brake
Mehmet Onur GULBAHCE
Department of Electrical Engineering Istanbul Technical University Istanbul, Turkey ogulbahce@itu.edu.tr
Handan NAK
Department of Control Engineering Istanbul Technical University Istanbul, Turkey nak@itu.edu.tr
Derya Ahmet KOCABAS

Department of Electrical Engineering Istanbul Technical University Istanbul, Turkey kocabasde@itu.edu.tr
Abstract Behaviour of an industrial load can be described with the total inertia of rotating parts and load torque. It is important to know both the load torque and power in order to determine the rated values of the electrical motor since the needed power is supplied by the motor during continuous operation. The mechanical behaviour of the electrical motor is so important, since they dont operate at constant torque or power in practice. Therefore electrical motors must have the capability to provide the desired driving torque or power for a load having variable demand. Additionally, the method for choosing the right motor depends on the calculated effective torque or power related to the operational characteristics including the change of load versus time. In this study, a current controlled eddy current brake for loading a motor with the desired load scenario as a load simulator is used for determining the operating capability under different load conditions which is similar to the industrial applications. A closed loop control system including a conventional PI controller is designed for an eddy current brake to obtain the needed torque for different load characteristics. The coefficients for the PI controller is optimized by genetic algorithm. The designed load simulator can be used to simulate any desired load scenario for any motor. Keywords-component; eddy current brakes; load simulator; excitation current control; conventional PI controller; genetic algorithm.
change. The analysis depends on both the properties of the electrical motor and characteristics of the industrial load. There are several mechanical load types in industrial applications and an electrical motor can be exposed to any of them. Generally, load torque can be related to some variables depending on the application type such as shaft speed, rotation angle, path and time where the most important one is angular speed. The mathematical relation between the load torque and the angular speed changes depending on the load type. The load torque (TL) can be expressed in terms of the power of the angular speed () which can be between -1 and 2 [1]. Load torque can be inversely proportional to angular speed (TL-1) Load torque can be independent from angular speed. (TL0) Load torque and angular speed can be directly proportional (TL) Load torque can change quadratically with angular speed (TL2)
The operational demand is also important in choosing the right motor. Modes of operation such as S1, S2, S8 are classified in international standards [2]. In order to analyse the operational response of a motor which can be electrical or else, the system can be simulated before setting up. To achieve this goal, a load simulator is needed to be designed. The simulator can be coupled with the mathematical model of the motor for simulation which is possibly electrical. Also it can be used to load a motor with the desired scenario after implementing the necessary equipment. An eddy current brake can be used for creating the load demand as a mechanical load simulator. The eddy current brake is an electromechanical conversion system which converts energy into thermal energy [3]. It consists of a conductive disk which rotates between two static electromagnets. Ferromagnetic or nonmagnetic material can be used for conductive disk which is connected to the shaft of a motor to be tested. When the electromagnet is supplied by direct current, conductive disk falls into a time-varying
I.
INTRODUCTION
Accordance with the principles of conservations of energy, power and torque demand of an industrial load are supplied mostly by electrical motors. Change of load torque versus time is needed to be known while choosing the motor to drive the load. For determining the name plate values of the electrical motor, effective load torque is needed to be calculated related to the operation mode. Every type of motor is firstly required to accelerate the industrial load up to the operation speed and then to supply the power need at this speed in addition to resist temporary overloading. The operational response of the motor for a demand of an industrial load and also the transient behaviour of an electrical motor is needed to be known. The transient analysis is necessary for starting, braking and possible load torque
magnetic flux and eddy currents are induced around the flux lines [4]. These currents have a circular route inside the material [5]. In addition to that, induced eddy currents consumes the mechanical energy at the rate which is proportional to square of the effective value of the eddy current and the resistance of the conductive material. In other words, mechanical input energy of the rotating system is transformed into heat inside the rotating disk [3,4,5]. In this study, a previously design eddy current brake model is used to design the mechanical load simulator and the excitation current is controlled to simulate the load torque. A conventional PI controller is designed by means of genetic algorithm. The simulation model includes a frequently met load scenario which is suitable for any (electrical) motor to drive the load. The model can also be used for analysing any transient behaviour. II. LOAD SIMULATOR MODEL The aimed load simulator is designed by using a previously designed eddy current brake model fed by a DC chopper driven by a PI controller. The purposed model is given in Fig. 1. The input and output of the simulator are excitation current (ie) and braking torque to load the motor (Te), respectively. Te,ref represents the desired torque to be obtained by adjusting the dc excitation current of the eddy current brake where d is the duty cycle of the DC chopper.
III.
MODELLING OF EDDY CURRENT BRAKE
Eddy current brake given in Fig 2 can be used to convert the mechanical energy into thermal energy whilst loading a rotating machine. The rotational design must contain a magnetic field producing source which can be either a permanent magnet or an electromagnet. Previous studies [6,7] attempted to obtain a mathematical formula for breaking force or torque induced in Eddy current brake. In [7], the breaking torque is mentioned to change linearly related to angular speed, conductivity and air gap length and also quadratically according to magnetic flux. The complexity of the system and the equation makes it impossible to solve the complex mathematical formulation. Several methods can be used to obtain the time response of the whole system. Instead of solving the eddy current brake dynamic equations, finite element analysis or practical study can be used to associate the input and output quantities of the brake. In this study, finite element analysis results of a previously designed eddy current brake [8] is used to model the load simulator. The finite element analysis provides the variation of the ECB torque or power to rotor speed, and excitation current as a lookup table. This case leaves no need for an analytical formulation of torque or power. Nevertheless, there are some polynomial based methods to model eddy current brakes in literature [9]. In this paper, the ECB in [8] is modelled by using look up table. The inputs of the eddy current brake model are excitation current and speed, and the output is the induced braking torque. The numerical data obtained previously consist of information about excitation current between 0-10 A, shaft speed between 0-7500 rpm and calculated induced braking torque values for the given input values. The used eddy current brake which has 8 electromagnets on each side of the conductive disk has 3D characteristics between braking torque, speed and excitation current as in Fig. 3 and total power dissipation, mechanical speed and excitation current as in Fig. 4.
Te , ref
ie
Te
Figure 1. Block scheme of the eddy current brake torque control system.
Braking Torque [Nm]
Conductive disk
0 -5 -10 -15 -20 -25 8000
Pole and coil
6000
2000 Shaft Speed [rpm]
4000
10
Excitation Current [A]
Figure 3. 3D braking torque, mechanical speed and excitation current characteristics.
Figure 2. geometry of the eddy current brake.
Braking Torque [Nm] 10

-1 5
-2 5
0 -2
5 -1
0 -1
-5
5 -1 0 -1
8
-5
brake changes with the excitation current and shaft speed as can be seen in Fig. 3. The main task of controller is tuning the duty cycle (d) of the dc chopper for controlling the excitation current of the brake. A conventional PI controller which defined in (1) is utilized for this purpose where u(t) is control signal, e(t) is error, Kp is the proportional coefficient and Ki is the integral coefficient of the controller..
7
-5
-1 0
5
-5
Figure 4. 2D braking torque, mechanical speed and excitation current characteristics.
0
-10
u ( t ) = K p e ( t ) + Ki e ( ) d
0
(1)
0 0
-5
1500
3000 4500 Shaft Speed [rpm]
6000
7500
Total Power Dissipation [W]
For optimising the PI controller parameters (Kp, Ki), genetic algorithm (GA) is used. GA is a global optimisation method with heuristic characteristic imitating natural selection and evolution process [9]. General flow chart of the genetic algorithm is given in Fig. 5. In the algorithm, a proportion of the population is selected after initializing the population, fitness function is calculated and stopping criteria are checked. It is possible to stop the algorithm by the values of fitness function, iteration number or stall generation. The next step is generating a new population by using genetic operators (reproduction, crossover, mutation) and then re-checking the stopping criteria after calculating fitness function until obtaining the desired stopping criteria [10, 11, 12]. The most critical step is to determine the fitness function. Authors preferred to use the integral of time multiply squared error (ITSE) as the performance index. ITSE is defined as in (2) where t is time and e(t) is the error signal in time domain:
-2000
-4000
-6000 10 8 6 4 Excitation Current [A] 2 0 2000 4000 Shaft Speed [rpm] 6000 8000
J = t e 2 ( t )dt
0
(2)
Here, the error is equal to the difference between desired and actual ECB torque given in Fig.1.
Figure 5. 3D total power dissipation, mechanical speed, excitation current characteristics.

Total Power Dissipation [W] 10
-4 0 00
e ( t ) = Te , ref ( t ) Te ( t )
(3)
-6000
-5 000
-5000
-4 000
0 0 0
-2 000
8
-1 000
-4000
0 00 -3
-3 000
-3000
-2 0 00
-2 000
-2000
-1 0 00
1500
-1000 3000 4500 Shaft Speed [rpm]
-1 000
6000
7500
Figure 6. 2D total power dissipation, mechanical speed, excitation current characteristics.
IV.
CONTROLLER DESIGN
Figure 7. Flow chart of the genetic algorithm.
To control the torque of eddy current brake, excitation current is controlled, since the produced torque of eddy current
In this paper, Matlab is used for optimisation of controller parameters to minimise the performance index and simulation diagram in Fig. 6 is built in Simulink to execute the optimization. V. SIMULATION RESULTS
To obtain the results for variable speed-constant torque operation dramatic speed change input for a constant torque demand is used for measuring the response of the simulator as can be seen in Fig. 9. Fig. 10 shows that ECB can provide the desired load torque while speed changes abruptly. Fig. 11 illustrates the torque error for this duty.
5500 Shaft Speed [rpm] 5000 4500 4000 3500 3000 2500 0 10 20 30 Time [s] 40 50 60
The simulation diagram in Fig. 6 is constructed in Matlab/Simulink to examine the system behaviour with different load scenarios. Controller parameters are calculated by GA as Kp=8.209 and Ki=15.11 by using the performance index ITSE given in Fig. 6. Three different duties are taken into consideration for simulation which are variable torque constant speed operation, constant torque variable speed operation and intermittent duty. To measure the performance of the controller clearly, a load scenario including dramatic load torque change for 2500 rpm is used. As seen in Fig. 7, induced torque of the ECB tracks the reference torque signal while keeping the shaft speed at 2500 rpm. Fig. 8 illustrates the torque error of the simulator when the industrial load needs to be driven at constant speed while the torque demand varies.
speed Ref erence e u
Figure 11. Shaft speed versus time.

13 12 Torque [Nm] 11 10 9
Reference ECB Torque Actual ECB Torque 0 10 20 30 Time [s] 40 50 60
Eddy current brake Controller
Teddy
Figure 12. Reference and actual torque of eddy current brake.

1 0.8 0.6 0.4 0.2 0 -0.2 0 10 20 30 Time [s] 40 50 60
error
1 s Clock
itse
Figure 8. Simulation diagram of the mechanical load simulator system with eddy current brake.
14 13 Torque [Nm] 12 11 10 9 0 10 20 30 Time [s]
Torque Error [Nm]
Figure 13. Torque error of the system.

12 Reference ECB Torque Actual ECB Torque Torque [Nm]
Reference ECB Torque Actual ECB Torque 40 50 60
11
10
Figure 9. Reference and actual torque of eddy current brake at 2500 rpm shaft speed.
0.8 Torque Error [Nm] 0.6 0.4 0.2 0 -0.2 0 10 20 30 Time [s] 40 50 60
10
20
30
40 Time [s]
50
60
70
80
Figure 14. Reference and actual torque of eddy current brake at 2800 rpm shaft speed.
Figure 10. Torque error of the system at 2500 rpm shaft speed.
[8] M. O. Gulbahce, Orta Guclu Bir Girdap Akimi Freni Tasarimi ve Sonlu Elemanlar Yntemi ile Analizi (Design and Finite Element Analysis of a Medium PowerEddy Current Brake), Msc. Thesis, Istanbul Technical University, Graduate School of Science Engineering and Technology, 2012. [9] S. Anwar, A parametric model of eddy current electric machine for automative braking applications, IEEE Trans. Control Syst. Technol., vol. 12, no. 3, pp. 422-427, May 2004. [10] Zhang, H., Cai, Y., Chen, Y. Parameter Optimization of PID Controllers Based on Genetic Algorithm, 2010 International Conference on EHealth Networking, Digital Ecosystems and Technologies, IEEE Conference, 17 18 April 2010, Shenzhen, China. [11] Tan, M. K., Chin, Y. K., Tham, H. J., Teo, K. T. K., Genetic Algorithm Based PID Optimization in Batch Process Control, ICCAIE 2011, IEEE Conference, 4 7 December 2011, Penang, Malaysia. [12] Jie, L., Ding, F., Bo, G., Parameters Optimization of Aeroengine PID Controller Based on Genetic Algorithms, IFCSTA 09, 25 27 December 2009, Chongqing, China.
13 Torque Error [Nm] 12 11 10 9
10
20
30
40 Time [s]
50
60
70
80
Figure 15. Torque error of the system at 2800 rpm shaft speed.
To simulate an intermittent and aperiodic duty, a possible load torque change to be met in practice as shown in Fig. 12 is used and it was seen that reference and actual torque values are substantially equal while the shaft speed is kept constant at 2800 rpm. Fig. 13 shows the torque error for this duty. As seen in all simulation result figures given above, the breaking torque of the eddy current brake is capable to track the reference signal properly in different duties. For all three duties the system responses have low overshoot and settling time. Steady state error for torque is also zero as seen clearly in Fig. 12. VI. CONCLUSION
In this paper, a mechanical load simulator is designed and the operational data of a previously designed eddy current brake is used in the load simulator while the excitation current is controlled to simulate the desired load torque. Genetic algorithm is used to design a conventional PI controller to regulate the dc excitation current of the eddy current brake in order to obtain the desired torque. Three major load scenarios are simulated to measure the performance of the controller. Simulation results prove load simulator is successful in creating the desired load scenarios with the designed built-in PI controller. REFERENCES
[1] [2] [3] W. Schuisky, Ilhami Cetin, Elektrik Motrleri (Electrical Motors), 1st ed. (in Turkish), Istanbul: Fatih Yaynevi, 1990. International Standards of Rotating Electrical Machines IEC 60034-1 Gulbahce, M. O., Kocabas, D. A., Habir I. Finite Element Analysis of a Small Power Eddy Current Brake, Mechatronica, 2012, IEEE Conference, 5-7 December, 2012. Gulbahce, M. O., Kocabas, D. A, Atalay A. K., A Study to Determine the Act of Excitation Current on Braking Torque for a Low Power Eddy Current Brake, IEMDC-13, IEEE Conference, 12-15 May 2013 Chicago-Illinois. Gulbahce, M. O., Kocabas, D. A, Atalay A. K., Determination of the Effect of Conductive Disk Thickness on Braking Torque for a Low Power Eddy Current Brake, POWERENG-13, IEEE Conference, 13 17 May 2013, Istanbul, Turkey. Schieber, D., Braking Torque on Rotating Sheet in Stationary Magnetic Field, Proceedings of the Institution of Electrical Engineers, vol. 121, no. 2, pp. 117122.,1974 Wiederick, H. D., Gauthier, N., Campbell, D. A., and Rochon, P., Magnetic Braking: Simple Theory and Experiment, American Journal of Physics, vol. 55, no. 6, 500503., 1987
[4]
[5]
[6]
[7]
Simulation and Dynamic Control of a Hybrid System for Controlling the Voltage and Frequency of an Island Network
Behrooz Ghahremani
Dapartment of Electrical Engineering Najaf Abad Branch, Islamic Azad University Isfahan, Iran ghahremani@sel.iaun.ac.ir
Saed Abazari
Dapartment of Electrical Engineering Shahrekord University Shahrekord, Iran saedabazari@yahoo.com
Ghazanfar shahgholian
Dapartment of Electrical Engineering Najaf Abad Branch, Islamic Azad University Isfahan, Iran shahgholian@iaun.ac.ir
AbstractIn this paper, Dynamic effect of a hybrid system consisting of fuel cell and battery in controlling voltage and frequency of an island network has been investigated. In this hybrid system, Using a suitable control system, battery has been used as a momentarily active power generation source and fuel cell as a delayed power generation source with a suitable time constant. Since fuel cells have a relatively high time constant for active power generation due to the nature of their performance, using this method, the fuel cell will have the sufficient time in the active power generation process; and this increases lifetime of fuel cell besides keeping DC terminal voltage of fuel cell constant about the nominal voltage. Keywords-Dynamic Control; Fuel Cell; Battery; Simulation; Island Network
I.
INTRODUCTION
Power generation in the fuel cells is a chemical process in which amount of generated active power is controlled by regulating amount of input hydrogen gas and air [1]. This causes a delay in power generation process, and hence, a hybrid power generation system coupled with fuel cell is mostly employed in the dynamic studies aimed at improving dynamic performance of fuel cell and compensating the delayed performance [1, 2]. A battery and inverter together with fuel cell have been used in the present paper to improve performance of fuel cell. Battery and inverter will generate instantaneous power (through controlling the frequency and voltage) in the initial moments of power variation in the system, and then, the generated active power of fuel cell will increase using an appropriate control system and with a suitable time constant while the active power generation of battery and inverter will proportionally decrease. The whole demanded active power will be supplied by the fuel cell after a couple of seconds. This controlling method leads to proper power generation performance of fuel cell followed by enhancement of fuel cell life and maintenance of its terminal voltage in the range of nominal voltage. The respective hybrid system also has a very quick dynamic response which can improve the dynamic stability in the networks lacking suitable dynamic performance. Generated reactive power of fuel cell
varies with a very low time constant but does not bring about a limitation for the fuel cell because the generated reactive power never causes abrupt changes in the amount of generated power in the fuel cell. This leads to improvement of generation capacity of reactive power compared to the former methods [2] and will promote voltage stability in voltage drop moments or in short-circuits which cause voltage reduction. As such, fuel cell system and its inverter will have the ability to independently generate instantaneous reactive power; which is one of the advantages of the proposed method in comparison with former models which deployed singleinverter system. Additionally, the generated active power of fuel cell is completely governed by controlling its inverter and its variation value is perfectly controlled at any time. In reference [2], a hybrid system consisting of fuel cell, capacitor, battery, and an inverter has been used; power generation is controlled by the inverter and no control is imposed on the active power generated by the fuel cell. II. DYNAMIC MODEL OF FUEL CELL IN DIGSILENT SOFTWARE
Numerous dynamic models have been extracted and applied to simulate fuel cell in the power system [1, 3, 4, and 5]. In the current paper, a model for fuel cell has been simulated in DIGSILENT software extracted from the simulated models of the papers [1]. The fuel cell (FC) systems have several other deficiencies such as cold starting and output voltage fluctuation. Due to the low temperature of FC at the beginning of a FC starting, they need more time to produce the desired power. If the FC system is forced to deliver the power to a heavy load during this period or during step load, it could be damaged. To solve these problems, a secondary energy source such as an electrochemical battery or Super Capacitor system needs to be connected to the FC system to produce power during transient states [1]. A number of literatures have studied the modeling, control, and performance of FC systems. Wang and Nehrir discussed
the modeling, control and fault handling of PEMFC 1 system with active and reactive power demand [7]. Lee and Wang investigated small signal stability analysis of an autonomous hybrid renewable energy. In their paper a simple model of FC system was used [6]. S. Obara proposed a new model of FC system based on experimental research as first order lead lag system [8]. A. K. Saha implemented a method to control solid oxide fuel cell (SOFC) system in distributed power generation application [9]. Padulles et al.introduced a simulation model of an SOFC power plant. In their model electrochemical and thermal processes were simulated as first order lead lag transfer function [10]. Li et al. controlled SOFC power plant and investigated the SOFC dynamic behaviors under a gridconnected condition [11]. Moreover, different previous works have already investigated the active and reactive power flow control of FC systems. Unfortunately, in the control strategy design, the authors did not consider the effect and behavior of FC system utilization factor under disturbance. However in this paper the dynamic model of FC system has been modified by considering the effect of utilization factor to operate in optimal value to enhance FC system performance and lifetime. A. Control AND POWER SYSTEM MODELING OF THE FUEL CELL IN DIGSILENT SOFTWARE With further current extracted from the fuel cell, a voltage drop occurs on the fuel cell terminal due to presence of internal resistances. This voltage drop is compensated by controlling partial pressure of the fuel and causes a dynamic behavior in the fuel cell under load change conditions. This dynamic behavior is modeled with control system of fuel cell in Fig. 1 whose input and output respectively are intensity of the current extracted from the fuel cell terminal and the signal representing value of additional generated voltage in the dynamic conditions. This signal can be converted into an effective voltage on the output terminal of fuel cell by applying it to a dependent voltage source connected in series. General scheme of the fuel cell control system can be observed in Fig. 1. In the power system of fuel cell, standard ideal driving force (E0) was modeled using a DC voltage source and additional generated voltage in dynamic conditions was represented by another voltage source connected in series with the former voltage supply; and internal resistance of sources was modeled by an internal resistance inside the connected voltage source. General scheme of power system of the fuel cell can be observed in Fig. 2.
FIG. 2. General scheme of power system of the fuel cell
The values of fuel cell control system parameters used in the current paper are shown in Table 1 that is extracted from the simulated models of the papers [1].
TABLE 1. VALUES of FUEL CELL CONTROL SYSTEM PARAMETERS Model K1 K2 K3 K4 K5 K6 K7 K8 K9 K10 K11 K12 K13 K14 K15 Value 1000 2.74e-6 0.2 2.33e-6 1186.24 0.0383 3558.71 0.0128 0.8734 1.16e-6 396.8254 0.3436 0.0548 450 0.001
proton exchange membrane fuel cell
FIG. 1 General scheme of control system of the fuel cell
B. Block-Frame Diagram of Fuel Cell in DSL2 Block-frame diagram related to fuel cell for connection between current measurement device and control and power systems of fuel cells is demonstrated in Fig. 3.
FIG. 3 Block-frame diagram of fuel cell
C. The Proposed Control Method Battery and inverter ensemble act as the instantaneous power supplier, and in the event of variation in the demanded power, immediately regulate their delivered power proportional to these variations. They keep supplying the power charge if no other supply source is available in the system. Yet, immediately following injection of power supply from another source, this ensemble adjusts its delivered power commensurate with amount of the injected power. Fuel cell and inverter ensemble is used as power generation system which connects to the circuit with a delay. This is possible by
2
applying two inverters, and according to papers [3], the absence of control on the output power of fuel cell does not guarantee a safe productivity of the fuel cell. This delayed performance also contributes to enhancement of fuel cell life, and also, maintenance of DC Bus voltage in the nominal value. The system is capable of abruptly generating reactive power with a short delay parallel to battery and inverter ensemble. The delay is due to the absence of identical working point to supply the reactive power with the battery-inverter ensemble in the transient state; hence, a working point is created for the network in transient states. In single-inverter systems, amount of generated reactive power is less than that of the proposed hybrid system and fuel cell is not used as independent generation source of abrupt reactive power [2]. Additionally, single-inverter systems, due to having less generated reactive power sources than the two-inverter systems, exhibit lower capability to supply reactive power of network in short circuit cases when the voltage drops severely. III. MODEL OF HYBRID SYSTEM
General scheme of the network under study is illustrated in Fig. 4. The whole setting of fuel cell, battery as hybrid system, and gas turbine feed the load in island network.
Digsilent Simulation Language
INVERTER
Fuel Cell
Battery
Hybrid System
INVERTER FIG. 4 GENERAL SCHEME OF THE NETWORK UNDER STUDY
IV.
CONTROL METHOD OF FUEL CELL INVERTER
As mentioned earlier, the demanded active and reactive powers are instantaneously supplied by battery and inverter ensemble in the network under study. In Fig. 5 the amount of consumed active power load (PL), the reactive power delivered by battery to the load (Qb), active power (PFC), and reactive power (QFC) delivered to inverter-fuel cell are all sent to the controller which determines values of d-axis and q-axis currents. Since the variations load might occur in steps, intensity of reference d-axis current (Idr) in the output of controller undergoes sudden changes. If this current is directly applied to the inverter, an abrupt DC current will be extracted from the fuel cell which might be unable to support and the terminal might suffer from severe voltage decline disturbing performance of inverter as a consequence. Accordingly, these abrupt changes can be completely eliminated by passing signal of reference d-axis current through a low-pass filter and the current extracted from fuel cell will always decrease or increase with a gentle slope. Besides enhancement of fuel cell
life, this causes the DC Bus voltage of fuel cell to permanently remain close to the nominal voltage and inverter will always have a safe performance in this state. Amount of reactive power delivered by battery is also compared with reactive power of inverter-fuel cell and the resulting error signals are transmitted to a conventional PI controller and value of Reference d-axis current (Idr) is determined. This current is given to the inverter as the reference current of q-axis after passing through a low-pass filter with a very small time constant. Selection of reactive power generated by battery as the reference reactive power is due to the fact that increase in reactive power generated by fuel cell reduces the reactive power supplied by the battery, and also, when these two values are equal and reactive power of consumption load are supplied equally, both sources will have greater capacity of reactive power generation compared to the state where one of them supplies the reactive power of consumption load individually. Network frequency and the zero-passing voltage point are given by PLL in the form of two signals (Sin r, and Cosr) to the inverter.
FIG. 5 The control system used in the system
A. Control System of Fuel cell Inverter Fig. 6 represents this control system which consumption load power is compared with fuel cell inverter delivered power and an error signal is generated and controller determines values of d-axis and q-axis currents by means of conventional PI controllers proportional to this error signal. After passing through a low-pass filter, this currents are delivered to the inverter as reference d-axis and q-axis currents (Idr , Iqr).
P-ref=PL
variations of load, that the active and reactive power in different times were applied. Figs. 12, 13 show the curves of voltage and frequency ac terminal to which load is connected. Simulation results for duration of 200 seconds are presented in Figures 7 to 13.
Needed active power of load
Kp+Ki/S
1/1+ST
Id-ref
Changes of active power
Generated active power of fuel cell
PFC
PI Controller Kp+Ki/S
Low Pass Filter

1/1+ST
Q-ref=Qb QFC
Iq-ref
Needed active power of load ____ Generated active power of Fuel cell -----FIG. 7
Time(s)
FIG. 6 Control system of fuel cell inverter
The respective system is simulated by DIGSILENT software on which RMS simulation studies are applied. The simulation results are subsequently presented. To check efficiency of the proposed control system, abrupt variations of load active and reactive power in different times were applied according to Table 2. Generated powers of fuel cell and battery were evaluated during these variations in Figures 7 to 9. Results indicate that battery and inverter immediately supply the power changes in the initial instants of variations in the system, and after a couple of seconds, generation power of fuel cell increases slowly and generation power of battery decreases as the former goes up. It must be noted that reactive power delivered by fuel cell is able to vary instantaneously because DC current of fuel cell will not vary for changing the reactive power of fuel cell inverter.
TABLE 2 ABRUPT CHANGES of ACTIVE AND REACTIVE POWERS of LOAD Changes of reactive power 50% Changes of active power 75% Time )Second( 40
Changes of active power
V.
SIMULATION RESULTS
Time(s)
FIG. 8 Active power of battery Needed reactive power of load

Changes of reactive power
Generated reactive power of fuel cell & battery Case

Time(s)
-25% 50% -80%
-50% 25% -50%
80 120 160
2 3 4
Needed reactive power of load - - - Generated reactive power of fuel cell ....... Generated reactive power of battery ____ FIG. 9
Using this method, variations of fuel cell terminal dc voltage and the dc current extracted from the fuel cell remain close to the nominal voltage and current (Figs. 10, 11). In order to check the effect of the proposed hybrid system on the network, this hybrid system was connected to an island network consisting of a load, a gas turbine and abrupt
Magnitude of voltage(pu)
Time(s)
FIG. 10 Dc voltage of fuel cell terminal
Also by placing this proposed hybrid system in the network, an improvement in the controlling voltage and frequency of island network was achieved in comparision with the state. The absence of that system showed high performance of hybrid system, eliminated steady state fault of governer of gas turbine and minimize deviation of nominal frequency. On the other hand, hybrid system had an influence on the performance voltage regulation of gas turbine and the value of voltage also stabilized in the nominal value.
Time(s)
Magnitude of current(pu)
REFERENCES
[1] M. Nayeripour, M.Hoseintabar, T.Niknam, A new method for dynamic performance improvement of a hybrid power system by coordination of converters controller, Journal of Power Sources, pp. 40334043, Elsevier 2011. [2] K. Sedghisigarchi, A. Feliachi, Dynamic and transient analysis of power distribution systems with fuel cellspart II: control and stability enhancement, IEEE Transactions On Energy Conversion, Vol. 19, No. 2, June 2004 .[3] K. Sedghisigarchi, A. Feliachi, Dynamic and transient analysis of power distribution systems with fuel cellspart I: fuel-cell dynamic model, IEEE Transactions On Energy Conversion, Vol. 19, No. 2, June 2004. [4] C. Wang, M. HashemNehrir, A physically based dynamic model for solid oxide fuel cells, IEEE Transactions On Energy Conversion, Vol. 22, No. 4, December 2007. [5] A. Gebregergis, P. Pillay, D. Bhattacharyya, R. Rengaswemy, Solid oxide fuel cell modeling, IEEE Transactions On Industrial Electronics, JANUARY 2009. [6] L. Dong-Jing, W. Li, Small signal stability analysis of an autonomous hybrid renewableenergy power generation/energy storage system part 1: timedomain simulations, IEEE Trans. Energy Convers., 23 (1) (2008), pp. 311 320. [7] C. Wang, M.H. Nehrir, Control of PEM fuel cell distributed generation systems, IEEE Trans. Energy Convers., 21 (2) (2006), pp. 586595. [8] S. Obara, Analysis of a fuel cell micro-grid with a small-scale wind turbine generator, J. Hydrogen Energy, 32 (2007), pp. 323326.
FIG. 11 Dc current of Fuel cell
Magnitude of voltage(pu)
Time(s)
FIG. 12 Voltage curve of AC termianl connected to the load With hybrid system
_____
Without hybrid system
------
50
Frequency(hz)
Time(s)
FIG. 13 Frequency curve of AC terminal connected to the load With hybrid system
[9] A.K. Saha, S. Chowdhury, S.P. Chowdhury, Application of solid-oxide fuel cell in distributed power generation, IET Renew. Power Gener., 1 (4) (2007), pp. 193202 [10] J. Padulles et al. An integrated SOFC plant dynamic model for power systems simulation, J. Power Sources, 86 (1) (2000), pp. 495500. [11] Y.H. Li, S. Rajakaruna, S.S. Choi, Control of solid oxide fuel cell power plant in a grid connected system, IEEE Trans. Energy Convers., 22 (2) (2007), pp. 405413. [12] J. Padulles et al. An integrated SOFC plant dynamic model for power system simulation, J. Power Sources, 61 (1996), pp. 495500. [13] Y. Zhu, K. Tomsovic, Development of models for analyzing the loadfollowing performance of micro turbine and fuel cells, J. Electric Power Syst. Res. 62 (2002) 111.
_____
Without hybrid system
------
VI.
CONCLUSIONS
Simulation results implied that in the proposed control method the respective hybrid power generation system is capable to immediately supply the demanded power and provides a highly favorable dynamic response in the case of sudden load variation. Furthermore, generation power of fuel cell did not increase suddenly which created favorable performance conditions for the fuel cell so that it had sufficient time for supplying the needed power and as well DC voltage of its terminal remained around the nominal value. Consequently, an appropriate performance was achieved for fuel cell inverter.
IRFOC vs DTC Performance Comparison Analysis

S.E. Rezgui, A. Mehdi, S. Legrioui, H. Meddouce, A.M. Boulahia, and H. Benalla
Electrotechnic Department Faculty of Technology Sciences, Constantine 1 University Constantine, Algeria salaheddine.rezgui@lec-umc.org; adel_hdm@yahoo.com; said.legrioui.1989@gmail.com; houssam@yahoo.fr; boulahiamalek@yahoo.fr; benalladz@yahoo.fr
AbstractThis paper presents an experimental studies of two well known strategies of induction machine (IM) drives; field oriented control (FOC) and Direct Torque control (DTC). The two Control methods are used in speed regulation using proportional-Integral (PI) and Fuzzy Logic controllers (FLC). A common hardware platform is realized, and a series of tests are conducted to emphasize the relative advantages and disadvantages of each method. The Experimental test bench is carried out under Matlab/Simulink with real time interface (RTI) and the dSpace (DS1104) board. electronic document is a live template. The various components of your paper [title, text, heads, etc.] are already defined on the style sheet, as illustrated by the portions given in this document. (Abstract) Keywords-IRFOC; DTC; dSpace;PI and Fuzzy control.
I.
INTRODUCTION
Since they have been invented in the late of 1960s and middle of 1980s respectively, the FOC [1] and DTC [2,3], are until nowadays the most privileged techniques when high performances control are required. Compared with the Rotor field oriented control, the DTC has many advantages such as less machine parameter dependence, intrinsically speed sensorless, simpler implementation and quicker dynamic torque response, with no need of currents controllers neither a coordinate transforms. But it suffers of current's oscillations, torque ripples and variable switching frequency. In counterpart the FOC presents good dynamic, more stability at steady state, low switching frequency and less currents and torque ripples, but it major drawbacks are; more computation time (needed in Parks transformation, and decoupling), more sensitive to parametric variations, several controllers loops and requires a speed sensor [4]. Taking into mind the industrial interest of these two major techniques, many researchers have taken the task to compare them by simulation and experiment, but each of authors had its own visions and aims. The most popular papers are in [5-10]. Authors had accomplished comparison between DFOC and DTC, based on the fact that they have the same particularity; flux estimation, and doesnt requiring a speed sensor. In [5] a summarized comparison results presented a slight advance of DTC scheme compared to FOC scheme regarding the dynamic flux control performance and the implementation complexity. In [6], the paper compared the discrete space vector modulation DTC (DSVM) scheme and DFOC, the behavior of DFOC scheme at
steady state operation was characterized by lower values of the three-phase currents ripple with respect to the DTC scheme, but in transients behavior the results showed that in DTC a better torque response can be achieved in terms of settling time and maximum overshoot. In [7], the FOC shows slower dynamics but better steady-state tracking compared to DTC. Authors in [8], concluded that DFOC and DTC were good in dynamic response, the parameter sensitivities were low and medium, and the torque ripples were high and low in DTC and DFOC respectively. The results in [9] have shown that both techniques achieve a similar level of transient torque performance, however the DTC scheme was disadvantaged due to the possibility of loss of flux control at lower speeds/loads, higher torque/current ripple, however the FOC scheme is disadvantaged by greater parameter sensitivity and complication with setting up the current control loops. In [10], a new DTC scheme was presented that uses the stator flux vector control, but the comparison with the RFOC doesnt appear explicitly in term of performance criteria analysis. A recent works consecrates comparisons between these schemes, as in [11] where the paper dealt with the comparative analysis of three control schemes for electric vehicles efficiency improvement; FOC, DTC, and DTC-SVM. DTC techniques exhibit improved transient torque response compared to standard RFOC. The disadvantage of the compared schemes is increased current distortion. However DTC-SVM exhibit low current distortion and also fast torque response and low torque ripple. Therefore, the authors indicated that the DTC-SVM scheme was the best candidate for an EV drive optimized control. In [12] the current IFOC was used for comparison vs. an optimum voltage vector selection table DTC in permanent magnet synchronous motor (PMSM) drive. Regarding at starting response and speed reversal the proposed DTC presented a best behavior. As both FOC and DTC have drawbacks, a combination of DTC and vector control (VC) seems to be interesting, the works that are presented in [13] for induction motor (IM), and [14] for PMSM drives, use a current VC scheme together with a DTC type switching table. The performance comparison between the proposed control with respect to those obtained under classical DTC and VC presents a faster dynamics than that of VC and less system variable pulsations than those under DTC and VC. Until nowadays, a lot of papers deal with improving both schemes; one can cite but non-exhaustively [15-17].
Research supported by Constantine Electrotechnic Laboratory (LEC).
Several drive systems employ a conventional controller such as a PI type controller, which as well-known works perfectly but only under unchanged system parameters or load conditions. But if it occurs, the performance of the closed loop system will be deteriorated, resulting in an unstable system. Thus, it need other type of controllers which can account nonlinearity, or somewhat adaptable to varying conditions in real time, without needing an operator reprogramming. Many controllers are now being employed such as fuzzy in order to achieve a desired performance level. The fuzzy logic controllers have been proven to be successfully used for a number of complex and nonlinear processes, with more robustness and less sensitive to parametric variations than conventional PI controllers. A lot of researchers in the domain of IM drive pay attention to the design and implementation of FLC, as in [18-20]. Finally, both DTC and FOC schemes have their advantages and disadvantages, by this work we aimed to make a contribution in the literature by adding an experimental comparison investigation. In order to achieve a regular comparison we selected the current IFOC which has a topology close enough to the DTC, hysteresis controller current regulators are considered because of the fast response that can provide, like in DTC. The selected criteria to evaluate their performances are studied in both steady-state and transient operating conditions when using a PI speed and FLC controllers. II. DTC AND IRFOC DESCRIPTION
(3)
Where J is the moment of inertia, TL is the load torque, fv is the viscous friction coefficient, and =/p is the mechanical rotor speed, and is the derivative. B. Description of IRFOC The principle of indirect rotor field-oriented control is when the d axe of the reference frame is locked to the rotor flux vector, this results in a decoupling of the variables so that flux and torque can be separately controlled by stator directaxis current isd, and quadrature-axis current isq, respectively, like in the separately excited DC machine. To carry out the alignment on a reference frame revolving with the rotor flux requires information on the modulus and position of the rotor flux. The electromagnetic torque, the synchronous angular speed, and the position are expressed as: = = = Where: = . . . / . (4) (5) (6)
The Fig.1 shows the global scheme of IFOC.
A. IFOC theory The dynamic model of three-phase, Y-connected induction motor can be expressed in the d-q synchronously rotating frame as: = = (1)
Where is=[isd, isq]T, r=[rd, rd]T, and vs=[vsd, vsa]T are respectively the vectors of stator currents, rotor flux linkages, and stator voltages. s is the synchronously rotating angular speed, is the electrical angular speed of the rotor, and sl = s is the slip frequency. And; = 1/ , = / , = / , 0 1 . = / , = / , = , = 1 0 Where Rs and Rr are the stator and rotor resistances, Ls and Lr are the stator and rotor inductances, Lm is the mutual inductances between the stator and the rotor winding, is the total leakage factor. The = 1 ( L2 m / Ls Lr ) electromagnetic torque Te can be expressed in terms of stator currents and rotor flux linkages as: = . (2)
Figure 1. Schematic of the IFOC wiht PI or FLC controllers..
The IM is fed by a current-controlled PWM inverter, the speed control loop uses a PI or FLC controller to produce the motor torque reference Te* which gives the quadrature-axis current reference isq*. The block dq/abc is used to convert isd* and isq* into three currents references isa*, isb*, and isc* which will be compared with the measured currents, thus, the errors are fed into the hysteresis current regulator. C. DTC theory The term direct control of torque and flux is based on the fact that from the errors between the reference values of the torque (and the flux) and those estimated, it is possible if its know the flux angle (thus the sectors) to directly control the states of the voltage source inverter (VSI) to reduce errors within hysteresis band controllers as shown on Fig. 2.
Where p is pole pair number. The motional equation of IM is described as:
The DTC development is carried out on the stationary reference fame (,), the electrical equations of IM are then: = 0= (7)
Error e(k) and change in speed error signals de(k) are used as the inputs of the speed controller, and one output is taken to provide the torque command. They are multiplied by a respective scale factors to adjust the fuzzy domain [-1.2, 1.2]. For each value of error, error change and torque command, the degree of membership () is evaluated for all the membership functions as illustrated at Fig. 3, and associated with the fuzzy variable which consists of five fuzzy sets; positive large (PL) and negative large (NL) which are represented by trapezoidal membership function (mf), and triangular (mf); positive small (PS), zero (ZE), and negative Small (NS), and positive large (PL).
NL NS ZE PS PL 1
Where is=[is, is], s=[s, s], r=[r, r], vs=[vs,vs] are respectively the vectors of; stator currents, stator/rotor flux linkages, and stator voltages.
3~
Te*
Te
V3 V4 V5
V2
0.8 D e g re eo fm e m b e rsh ip
V1 V6
0.6
s*
s
0.4
0.2
-1
-0.8
-0.6
-0.4
-0.2
0.2
0.4
0.6
0.8
Figure 3. Membership distribution of inputs and output. Figure 2. Shematic of the DTC
The operation of fuzzy logic is based on a set of control rules of IF ... THEN as represented in Table I.
TABLE I. LINGUISTIC RULE TABLE
Stator- flux components are estimated by: = . (8)

de(k) e(k) NL NS ZE PS PL
Modulus and angle of flux can be obtained as follows: | |= = tan Thus, the electromagnetic torque can be estimated via the stator flux and currents: = . (10) (9)
NL
NS
ZE
PS
PL
NL NL NL NS ZE
NL NL NS ZE PS
NL NS ZE PS PL
NS ZE PS PL PL
ZE PL PL PL PL
III.
EPERIMENT IMPLEMENTATION
As afore mentioned, the DTC is based on using two hysteresis controllers, they evaluate the difference between requested and estimated values, and thereby determine if the flux and torque vectors should be increased, decreased, or constant. It is proven that the band width of the flux hysteresis comparator has influence on current distortion (harmonics), while torque ripples and switching frequency of the VSI device are affected by torque hysteresis band [4]. A logic signals are produced from the switching table and used to trigger the switches of the three-phase VSI. There are six possible active combinations of these logic signals with the corresponding active input voltage vectors of the inverter depending on sector number, as shown in Fig. 2. D. Fuzzy Logic Speed Controller (FLC) In this work the FLC was designed using the Fuzzy Control Toolbox provided within Matlab, with the Mamdanis minmax decision inference engine.
A three phase Y connected; 1kW, 2880rpm, 230/400V, 4/2.8A, squirrel cage IM is tested to achieve the comparison. The IM parameters were identified through the standard tests and are given on appendix. The IM is loaded via 1kW DC generator, and coupled to a 1024pt incremental encoder for speed measurement. The control system is implemented on dSpace DS1104 board with the help of Matlab/Simulink (RTI) blocks. Fig. 4 shows the hardware snapshot of the experimental setup. Through the ControlDesk software a control panel has been developed (Fig. 5); it provides an easy handling interface between dSPACE and Simulink blocks, and useful tool for online variation of the operating parameters of the system. A. Performance Analysis of PI based DTC and IRFOC The sampling time Ts=100s, and the carrier frequency is set at fc=10 kHz. The values of the regulators are; Kp=0.13, Ki=0.001, for both IRFOC and DTC. We should note that these values are different from those calculated by the well known methods (because the drives did not worked); we were
Rotor speed(rpm)
2000
Rotor speed(rpm)
constrained to adjust them online from the ControlDesk control panel. This is certainly due to the measure uncertainties and simplifying assumptions at the parameters identification.
DTC (PI)
3000
IRFOC (PI)
3000
2500
2500
2000
1500
1500
1000
1000
500
500
0
0 0 2 4 6 8 10 12 14 16
-500
10
12
14
16
18
time(s)
time(s)
6 4 5
3.5
2.5
Torque(N.m)
Torque(N.m)
0 2 4 6 8 10 12 14 16
1.5
0.5 0 0 -1 -0.5
-2
-1
10
12
14
16
18
time(s)
time(s)
Figure 4. Snapshot of the hardware setup
0.8
12
0.6
10
0.4 8 0.2
flus,b(Wb)
isq,isq*(A)
-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8
-0.2 2 -0.4
-0.6
-0.8 -0.8
-2
10
12
14
16
18
flux,a (Wb)
time(s)
10 8 6 4 2 0 -2 -4 -6 -10 -8 -10 0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16 18
10
is,abc(A)
-5
time(s)
is,abc(A)
time(s)
Figure 5. Control pannel via ControlDesk software.
Figure 6. High speed PI performance with multiple torque command.

DTC (PI)
550 500 450 400 500 600
Figs. 6 and 7, show the IFOC and DTC when the IM is operating at high and low speed. Starting with 50% of load disturbance, and following with sharp changes to 0% and up to 100% of nominal load torque respectively. On Fig. 6(a), it is seen that DTC presents significant overshoot and slow disturbance rejection with steady state error, On Fig. 6(a) and Fig. 6(c) the IFOC presents good dynamics and disturbance rejection with no overshoot, the current isq follows its reference isq* instantly, and the flux follows a circular trajectory (Fig. 6(c)). At steady state the same response time is observed, in IFOC there is lesser torque ripple (around 0.2 N.m) and the currents have more uniform waveform.
IFOC (PI)
Rotor speed(rpm)
350 300 250 200 150 100 50 0 0 2 4 6 8 10 12 14 16
Rotor speed(rpm)
400
300
200
100
0 0 2 4 6 8 10 12 14 16 18
time(s)
time(s)
1.5
1.5 1
Torque(N.m)
Torque(N.m)
0 2 4 6 8 10 12 14 16
It is clear that the speed tracking is much better in IFOC scheme and the use of hysteresis control in DTC causes some performance degradation with torque overshoot in the torque transient and more ripples at steady state regime. Due to the unlimited current in DTC during start-up the machine exceeds more than IFOC the rated current as shown in Fig.6 (d,d).
0.5
0.5
0 0
-0.5 -0.5 0 2 4 6 8 10 12 14 16 18
time(s)
time(s)
Figure 7. Low speed PI performance with multiple torque command.
Rotor speed(rpm)
Rotor speed(rpm)
At low speed the motor is following a command 0 rpm and 500 rpm, both schemes present good response time (Fig.7), but we note the well known low speed problem in DTC which presents more speed and torque ripples approximately 0.5N.m, which caused acoustic noises and vibrations. B. Performance Analysis of FLC based DTC and IRFOC The performances of each method are examined when intelegent controller (i.e fuzzy controller) are introduced in speed control loop. It is noted that the sampling period was increased up to Ts = 600s, because the algorithm require more calculating time and we have got on an "over-run" during compelation from simulink to dSpase, unfortunatly this is the main limitation of the DS1104. Torque transients is observed when the machine was subjected to a series of torque changes, starting with 10% and up to 50% of nominal torque. Figs. 8(b,b) present the corresponding torque response of the two schemes. It can be compared that, in the steady state, the torque ripple in the DTC is less than in IFOC. Howerver, both schemes achieve high dynamic performance in response to changes in torque demand. The flux trajetory is rather hexagonal with larger band, this is due to the large sample time. Additionally, we can see on Fig. 8(a) that the speed tracking performance of DTC has become more efficient with respect to the DTC with PI control. The Fig. 9 presents low speed behavior when step reference change form 0 rpm and 600 rpm, we can see a very good dynamic response in trensients or at steady state regime. A slight superiority of IFOC upon the DTC is observed in setling time and disturbance rejection, although the IFOC presents more ripples, but this mainly due to sampling time. C. More Performance Analysis Finally, we added a test of robustness to sudden changes in speed reference with reversal when IM is operating at half load capacitance. From Fig. 10(a,a), which give the speed tracking evolution; we can see that the IRFOC present fast response and less oscillations but with little overshoot. The current isd follow the reference value isd* (Fig. 10(b)) with some undesirable peaks when changing the step commands which are visible on the is,abc currents depicted on Fig. 10(d). During acceleration mode the motor delivers full torque until the speed reaches its requested reference, the torque then drops to the same value as the load to keep the speed constant. In DTC the electromagnetic torque presents more ripples than IRFOC as it is seen on Fig. 10(c,c). Now we resume tests with FLC implementation. At Fig.11(a) the DTC prsents less oscillations than in PI speed control, but it takes approximately the same responding time, thus the IRFOC still have faster speed dynamic. On Fig. 11(c,c), the torque tracking performances are the same , DTC presents more ripples than IRFOC but not as with PI controller.
DTC (FLC)
3000 3000
IFOC (FLC)
2500
2500
2000
2000
1500
1500
1000
1000
500
500
0 0 2 4 6 8 10 12 14 16 18 20
0 0 2 4 6 8 10 12 14 16 18 20
time(s)
time(s)
4 4 3 3 2 2 1
Torque(N.m)
Torque(N.m)
0 2 4 6 8 10 12 14 16 18 20
-1
-1
-2
-2
-3
-3
-4
-4
10
12
14
16
18
20
time(s)
time(s)
10
0.8
0.6
0.4
isq, isq* (A))

-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
flus,b(Wb)
0.2
-0.2
-2
-0.4
-4
-0.6
-6
-0.8
-8
-1 -1
-10
10
12
14
16
18
20
flux,a (Wb)
temps(s)
10
4 5 2
is,abc(A)
is,abc(A)
0 2 4 6 8 10 12 14 16 18 20
-2 -5 -4
-10
-6
-8
10
12
14
16
18
20
time(s)
time(s)
Figure 8. High speed -FLC- performance with multiple torque command.

DTC (FLC)
600 600
IFOC (FLC)
500
500
Rotor speed(rpm)
400
Rotor speed(rpm)
0 2 4 6 8 10 12 14 16 18 20
400
300
300
200
200
100
100
0 0 2 4 6 8 10 12 14 16 18 20
time(s)
time(s)
2.5
1.5 1
Torque(N.m)
Torque(N.m)
0 2 4 6 8 10 12 14 16 18 20
0.5
-1
-0.5
-1
-2
-1.5
-2 -3 -2.5 0 2 4 6 8 10 12 14 16 18 20
time(s)
time(s)
Figure 9. Low speed FLC- performance with multiple torque command.
DTC (PI)
la vitesse de rotation
3000
3000
IFOC (PI)
Lm= 0.7209H, J = 0.00207 kg.m2, fv=0.000173 kg.m/s. REFERENCES

[1] [2] [3] Bose B.K., Modern Power Electronics and AC Drives, Prentice Hall, 2002, ch 8. I.Takahashi, T.Noguchi, A new quick response and high efficiency control strategy of an induction motor, IEEE Trans. Industrial Application, vol. IA-22, pp. 820 827, Sept./Oct. 1986. M. Depenbroak, Direct Self-Control (DSC) of Inverter-Fed Induction Machine IEEE Trans. Power Electronics, Vol. 3, No. 4, pp. 420 429, Oct. 1988. Peter Vas: Sensorless Vector and Direct Torque Control, Oxford University Press, London, 1998, ch.1. H. Le-Huy, Comparision of field-oriented control and direct torque control for induction motor drives, in Conf. Rec. IEEE Industry Applications Soc. Annual Meeting, 1999, pp. 1245 1252. D. Casadei, F. Profumo, G. Serra, and A. Tani, FOC and DTC : Two Variable Schemes for Induction Motors Torque Control, IEEE Power Electronics, vol. 17, no. 5, pp. 779 787, Sept. 2002. T. A. Wolbank, A. Moucka, and J. L. Machl, A comparative study of field-oriented and direct-torque control of induction motors reference to shaft-sensorless control at low and zero-speed, in Proc. IEEE Int'l. Symp. Intelligent Control, pp. 391 396, 2002. D. Telford, M.W. Dunnigan, B.W. Williams, A Comparison of Vector Control and Direct Torque Control of an Induction Machine, IEEE Power Electronics Specialists Conference. vol. 1, pp. 421 426, 2000. Cruz, M. A. Gallegos, R. Alvarez, and F. Pazos, Comparison of several nonlinear controllers for induction motors, IEEE Int'l. Power Electronics Congress (CIEP), 2004, pp. 134 139. Tripathi, R. S. Anbarasu, and R. Somakumar, "Control of ac motor drives: performance evaluation of industrial state of art and new techinique," in IEEE Int'l. Conf. Industrial Tech. (ICIT), 2006, pp.3049 3054. A. Haddoun, M.E.H. Benbouzid, D. Diallo, R. Abdessemed, J.Ghouili, K. Srairi, Comparative Analysis of Control Techniques for Efficiency Improvement in Electric Vehicles, IEEE Vehicle Power and Propulsion Conference, Sept 2007, pp. 629 634. S. Dwivedi, B. Singh, Vector Control Vs Direct Torque Control Comparative Evaluation for PMSM Drive, Proc. IEEE Power Electronics, Drives and Energy Systems, 2010, pp. 18. S. Vaez-Zadeh, E. Jalali, Combined vector control and direct torque control method for high performance induction motor drives, in J. Energy Conversion and Management, Vol. 48, Iss. 12, Dec. 2007, pp. 30953101. Z. Jiang, S. Hu, W. Cao, A New Fuzzy Logic Torque Control Scheme Based on Vector Control and Direct Torque Control for Induction Machine, in 2008 Proc. IEEE Innovative Computing Information and Control. A. Lokriti, I. Salhi, S. Doubabi, Y. Zidani, Induction motor speed drive improvement using fuzzy IP-self-tuning controller. A real time ISA Transactions Vol.52, Iss. 3, May 2013, pp. 406-417. Y. Zhang, J. Zhu, Z. Zhao, W. Xue, D. G. Dorrell, An Improved Direct Torque Control for Three-Level Inverter-Fed Induction Motor Sensorless Drive, IEEE Trans. Power Electronics, vol. 27, no. 3, March 2012, pp. 1502 1513. S.A. Zaid O.A. Mahgoub K.A. El-Metwally, Implementation of a new fast direct torque control algorithm for induction motor drives, IET Electr. Power Appl., 2010, Vol. 4, Iss. 5, pp. 305 313. M. Nasir Uddin, M. Hafeez, FLC-Based DTC Scheme to Improve the Dynamic Performance of an IM Drive, IEEE Trans. Industry Applications, vol. 48, no. 2, pp. 823 831, March/April 2012. T. Sreenivasa Murthy,T. Vamsee Kiran, Comparative Analysis of FUZZY DTC and ANN DTC for Induction Motor, Inter. Jour. Emerging Trends in Engineering and Development, Issue 3, Vol.2 March 2013, pp. 483 493. M.A. Fnaiech, F. Betin, G.-A. Capolino, F. Fnaiech, Fuzzy Logic and Sliding-Mode Controls Applied to Six-Phase Induction Machine With Open Phases, IEEE Trans. Industrial Electronics, vol. 57, no. 1, Jan. 2010, pp. 354 364.
2000
2000
la vitesse(tours/min)
1000
1000
N (rpm)
-1000
-1000
-2000
-2000
-3000
-4000
-3000
10
12
14
16
18
20
10
12
14
16
18
20
temps(s)
le couple electromagnetique
5
8
time(s)
[4]
6
le couple electrom agnetique(N.m))
4 3
[5] [6] [7]

0 2 4 6 8 10 12 14 16 18
2 1 0 -1 -2
-4 2
Te(N .m )
0 2 4 6 8 10 12 14 16 18 20
-2
-3 -4 -5
-6
-8
temps(s)
time(s)
Figure 10. PI based DTC and IFOC behavior at sharp speed command.
DTC (FLC)
la vitesse de rotation
3000
3000
[8] [9] [10]
IFOC (FLC)
2000
2000
la vitesse(tours/min)
1000
1000
n(rpm)
-1000
-1000
[11]
0 2 4 6 8 10 12 14 16 18 20
-2000
-2000
-3000
-3000
10
12
14
16
18
20
temps(s)
le couple electromagnetique
20
8
time(s)
[12]
6
le couple electrom agnetique(N.m ))
15 10 5 0 -5 -10 -15 -20
[13]
Te(N .m )
-2
[14]
-4
-6
10
12
14
16
18
20
-8
10
12
14
16
18
20
temps(s)
time(s)
[15] [16]
Figure 11. FLC based DTC and IFOC behavior at sharp speed command.
IV.
CONCLUSION
This result was expected according to literature, the IFOC has a lower torque ripple and fast speed response, but not sufficiently to say definitively that the IFOC is superior to DTC. This could be caused by the choice of hysteresis control for inverter switching, instead of more sophisticated methods. Introduction of a fuzzy approach in both IRFOC and DTC has demonstrated; fast speed response, rapid load disturbance rejection, but an optimization procedure should be considered to decrease the sampling time which has a direct influence on ripples. APPENDIX p=2, f= 50Hz, Rs=6.58, Rr=5.81, Ls=0.749H, Lr= 0.749H,
[17] [18] [19]
[20]
Experimental Verification of Torque Ripple Calculation for Induction and Synchronous Motor
Tetsuya Kojima1, Haruyuki Kometani
Mitsubishi Electric Co. 8-1-1, Tsukaguchi-Honmachi, Amagasaki, Hyogo 661-8661, Japan 1 Kojima.Tetsuya@eb.MitsubishiElectric.co.jp
Mitsuhiro Kawamura, Yuichi Tsuboi, Ritaka Nakamura, Hiroyuki Masuda, Yoshihiro Ohgashi
Toshiba Mitsubishi-Electric Industrial Systems Co. 2-4, Suehiro-cho, Tsurumi-ku, Yokohama 230-0045, Japan
AbstractIn large capacity motor drives, torque ripple evaluation is important in order to prevent problems such as unexpected lifetime reduction caused by mechanical resonance with the load. However, enormous cost and time are necessary to measure torque ripple in large capacity motor drives. This paper discusses torque ripple calculation methods using electrical signals of a motor and then proposes an accurate calculation method. The experimental results with 400W induction motor and 2.2kW synchronous motor verified that the proposed method can accurately calculate torque ripple caused by harmonics such as inverter voltage distortion and motor spatial harmonics and is useful for torque ripple evaluation. Keywords-component; synchronous motor torque ripple; induction motor;
II.
TORQUE CALCULATION
A. Theoretical Consideration Motor torque: is given as follows[4],[5].
Pm I
V R I dt
where Pm is the pole pair, is the flux linkage, V is the stator voltage , I is the stator current, R is the stator resistance. On the other hand, motor torque is given by
I.
INTRODUCTION
Pm Po / r
The use of large capacity drives above 10MW has been increasing because, for example, motor drives are substituted for turbine drives for the purpose of higher efficiency[1],[2]. In large capacity drives, torque ripple can be caused by inverter voltage distortion, motor spatial harmonics, control period, and inverter DC link voltage. In frequency range, especially lower than fundamental frequency of a motor, the harmonic current and the torque ripple are large because the impedance is small. Moreover, the torque ripple can be amplified by the resonance with the load. To prevent the unexpected lifetime reduction, torque ripple should be evaluated in detail. To evaluate torque ripple, measuring torque requires enormous cost and time especially in large motor drives, because it needs a torque transducer installed between a motor and a load. On the other hand, torque ripple calculation methods using electrical signals of a motor have been studied[3],[4]. However these calculation methods are primarily intended to control torque, further study is necessary for validity of torque ripple calculation. This paper firstly discusses torque ripple calculation methods using electrical signals of a motor at constant motor speed, and then proposes an accurate calculation method. The experimental results with 400W induction motor and 2.2kW synchronous motor verify that the proposed method accurately calculates the torque ripple and is useful for torque ripple evaluation.
where Po is the output power, r is the rotor electrical frequency. It will be shown first that the both of Eq. (1), (2) and Eq. (3) give a same result, then that torque is calculated with stator voltage, stator current, and stator resistance without rotor electrical frequency. In case of an induction motor, Combining Eq. (1) and (2) gives the following equation.
Pm V I R I 2 / s
where s is the stator electrical frequency. Rotor input power: Pr is given as follows, however iron loss is neglected because it is small in low frequency range where torque ripple becomes problematic.
Pr p I
p V R I
Rotor input power: Pr, rotor copper loss: Pcr, and output power: Po have the following relations.
Pr : Pcr : Po 1 : s : 1 s s s r / s
Substituting Eq. (7) and (8) into Eq. (3) gives torque as follows.
Pm Pr s
with the mechanical transfer function from the motor to the torque transducer. The transfer function shown in Fig. 2 is adjusted by hammer impact test based on the design data. This system can accurately measure torque below nearly 40Hz. Operating voltage and frequency of the motor are 42V and 5Hz respectively in consideration of the control voltage source capacity and the torque measurement range. Configuration of harmonics verified is shown in Tab. II. In case of integer order symmetric 25Hz, harmonic voltages are injected to 3 phases as follows.
Substituting further Eq. (5) and (6) into Eq. (9) gives the same result as Eq. (4). Therefore, the torque of Eq. (1) and (2) equals to that of Eq. (3). In addition, this torque proves to be torque of motor output power after stator and rotor copper losses are subtracted. In case of a synchronous motor, stator and rotor electrical frequencies are the same and rotor copper loss doesnt generate. Therefore, Eq. (4) given by combining Eq. (1) and (2) apparently equals to Eq. (3). However, in this paper, constant motor speed is assumed in both the motors: an induction motor and a synchronous motor. B. Calculation Method This paper proposes the torque ripple calculation method with stator voltage, current, and resistance based on Eq. (1) and (2) as follows. (1) Measuring the stator resistance of a motor: R (2) Rotating the motor at constant speed and record the stator voltage: V and the stator current: I for a certain period. (3) Calculating p = V RI with the data recorded. (4) Executing Fast Fourier Transform analysis (FFT) about pcalculated. (5) Calculating the flux linkage: by integrating the FFT result of pwith respect to the each frequency components. (Eq. (2)) (6) Calculating the torque by Pm (I). (Eq. (1)) However simple integration based on Eq. (2) can calculate the flux linkage, it is sensitive to measurement errors. Practical data recorded have a lot of measurement errors because they are acquired by sampling waveforms that include various disturbances such as inverter switching frequency. Integration process enlarges an error at a time point further to all the time points afterward. III. EXPERIMENTAL VERIFICATION
vuh m Ve cos2 f h t vvh m Ve cos2 f h t 2 / 3 vwh m Ve cos2 f h t 2 / 3
where Ve is the amplitude of fundamental voltage, m is the harmonic voltage amplitude ratio to Ve, fh is the harmonic voltage frequency. In case of non-integer order asymmetric 17 Hz, harmonic voltages are injected to only W phase as follows.
vwh m Ve cos2 f h t
Figure 1. Torque ripple calculation test system for IM. TABLE I. CONFIGURATION OF RECORDER Sampling frequency Recording period Low pass filter
15 10 5 0 -5 -10 -15 -20 -25 0 10 20 30 40
20 kHz 5 sec 5 kHz
gain [dB]
50
60
70
80
90 100
frequency [Hz]
A. Induction Motor The proposed calculation method was verified with an induction motor (IM). Fig. 1 shows the experimental system for verification. Control Voltage Source I supplies voltage including harmonics to the induction motor of 400W 400V 60Hz. A recorder records the torque measured with a torque transducer. The configuration of the recorder is shown in Tab. I. In this verification, measured torque is acquired by compensating the torque measured with a torque transducer
Figure 2. Transfer function from IM torque to torque transducer. TABLE II. CONFIGURATION OF HARMONIC VOLTAGE FOR IM No. 1 2 3 4 Type and frequency Integer oder, Symmetric, 25 Hz Non-integer order, Asymmetric, 17 Hz Equation (10)-(12) 0.03 (13) Amplitude m(1) Load [%] 0 100 0 100 Result figure Fig. 3 Fig. 4 Fig. 5 Fig. 6
(1) m : Ratio to fundamental voltage.
Torque ripple calculation results, in case of integer order symmetric harmonics, are shown in Fig. 3 and 4. The proposed method can calculate torque ripple of 30Hz caused by harmonics of negative sequence 25Hz. Results, in case of noninteger order asymmetric harmonics, are shown in Fig. 5 and 6. The proposed method can calculate torque ripple of 12 and 22 Hz caused by harmonics of 17Hz that include both of positive and negative sequences. In addition, the proposed method can calculate torque ripple of 36Hz in Fig. 4 and 6 that will be attributed to rotor slot harmonics.
torque amplitude [pu]
0.14 0.12 0.10 0.08 0.06 0.04 0.02 0.00 0 10 20 30 40
measured calculated
B. Synchronous Motor The proposed calculation method was verified with a synchronous motor (SM) similarly. Fig. 7 shows the experimental system for the verification. The main configuration of the system including the recorder is the same as that of the induction motor. The specification of the synchronous motor is 2.2kW 200V 60Hz. The mechanical transfer function from the motor to the torque transducer is shown in Fig. 8, with which torque measured with the torque transducer is compensated. This system can accurately measure torque below nearly 120Hz. Operating voltage and frequency of the motor are 48V and 10Hz in consideration of the control voltage source capacity and the torque measurement range. Configuration of harmonics verified is shown in Tab. III. In case of integer order 50Hz and non-integer order 2Hz, harmonic voltages are injected to 3 phases as described in Eq. (10)-(12). In case of over excitation, excitation current of the motor is increased from 1pu to 2.5pu to generate 7th order spatial harmonics. Torque ripple calculation results, in case of integer order harmonics, are shown in Fig. 9 and 10. The proposed method can calculate torque ripple of 60Hz caused by harmonics of negative sequence 50Hz. Results, in case of non-integer order harmonics, are shown in Fig. 11 and 12. The proposed method can calculate torque ripple of 12Hz caused by harmonics of negative sequence 2Hz. Results, in case of over excitation, are shown in Fig. 13 and 14. The proposed method can calculate torque ripple of 60Hz caused by 7th order spatial harmonics of positive sequence 70Hz.
frequency [Hz]
Figure 3. Torque Ripple Calculation for IM in No.1(Tab.). 0.14
0.12 0.10 0.08 0.06 0.04 0.02 0.00 0 10 20 30 40
measured calculated
frequency [Hz]

0.04 0.03 0.02
gain [abs]
measured calculated
Figure 7. Torque ripple calculation test system for SM.
10 8 6 4
c
0.01 0.00 0 10 20 30 40
frequency [Hz]
2 0 0 50 100 frequency [Hz] 150 200

0.04 0.03 0.02 0.01 0.00 0 10 20 30 40
measured calculated
Figure 8. Transfer function from SM torque to torque transducer. TABLE III. CONFIGURATION OF HARMONIC VOLTAGE FOR SM No. 11 12 13 14 15 16 Type and frequency Integer oder, 50 Hz Non-integer order, 2 Hz Over excitation Equation (10)-(12) 0.02 Amplitude m(1) 0.05 Load [%] 0 100 0 100 0 50 Result figure Fig. 9 Fig. 10 Fig. 11 Fig. 12 Fig. 13 Fig. 14
frequency [Hz]
Figure 6. Torque Ripple Calculation for IM in No.4(Tab.).
(1) m : Ratio to fundamental voltage.
0.2 torque ampliltude [pu] 0.15 0.1 0.05 30 40 50 60 70 80 frequency [Hz] Figure. 9 Torque Ripple Calculation for SM in No.11(Tab.). 0 0 10 20
0.2 torque ampliltude [pu]

measured calculated
0.15 0.1 0.05
measured calculated
30 40 50 60 70 80 frequency [Hz] Figure. 12 Torque Ripple Calculation for SM in No.14(Tab.). 0.06 torque ampliltude [pu] 0.05 0.04 0.03 0.02 0.01 30 40 50 60 70 80 frequency [Hz] Figure. 13 Torque Ripple Calculation for SM in No.15(Tab.). 0.06 torque ampliltude [pu] 0.05 0.04 0.03 0.02 0.01 30 40 50 60 70 80 frequency [Hz] Figure. 14 Torque Ripple Calculation for SM in No.16(Tab.) 0 0 10 20 measured calculated 0 0 10 20 measured calculated
10
20
0.2 torque ampliltude [pu] 0.15 0.1 0.05 30 40 50 60 70 80 frequency [Hz] Figure. 10 Torque Ripple Calculation for SM in No.12(Tab.). 0 0 10 20 measured calculated
0.2 torque ampliltude [pu] 0.15 0.1 0.05 30 40 50 60 70 80 frequency [Hz] Figure. 11 Torque Ripple Calculation for SM in No.13(Tab.). 0 0 10 20 measured calculated
IV.
CONCLUSION [1]
REFERENCES
M. Tsukakoshi, M. A. Mamun, K. Hashimura, H. Hosoda and T. Kojima, Performance Evaluation of a Large Capacity VSD System for Oil and Gas Industry, in Proc. IEEE Energy Conversion Congress and Exposition, Sep. 20-24, 2009, pp. 3485-3492. M. A. Mamun, M. Tsukakoshi, K. Hashimura, Innovation of a Large Capacity 5-level IGBT Inverter for Oil and Gas Industry, in Proc. IEEE Energy Conversion Congress and Exposition, Sep. 17-22, 2011, pp. 1964-1971. G. Buja, D. Casadei, G. Serra, Direct Torque Control of Induction Motor Drives, in Proc. IEEE International Symposium on Industrial Electronics, July 7-11, 1997, vol. 1, pp. TU2-TU8. L. Youb, A. Craciunescu, A comparison of various strategies for direct torque control of induction motors, in Proc. Internatioinal Aegean Conference on Electrical Machines and Power Electronics, Sep. 10-13, 2007, pp. 403-408. A. Veltman, D. W. J. Pulle, R. W. De Doncker, Fundamentals of Electrical Drives, Springer, 2007, pp. 235-237.
In large capacity motor drives, torque ripple can cause problems such as unexpected lifetime reduction caused by mechanical resonance with the load. They should be evaluated in detail, however, measuring torque requires enormous cost and time especially in large motor drives. Therefore, this paper discussed torque ripple calculation methods using electrical signals of a motor and then proposed an accurate calculation method. The proposed method was verified with 400W induction motor and 2.2kW synchronous motor, and proved to be able to accurately calculate the torque ripple caused by harmonics such as inverter voltage distortion and motor spatial harmonics and be useful for torque ripple evaluation.
[2]
[3]
[4]
[5]
Analysis of Waveform Approximation for the AC Current of a Line-Commutated Converter

Mehdi Daryabak Department of Electrical and Computer Engineering University of Manitoba Winnipeg, Canada umdaryab@cc.umanitoba.ca Shaahin Filizadeh Department of Electrical and Computer Engineering University of Manitoba Winnipeg, Canada shaahin_filizadeh@cc.umanitoba.ca
AbstractThis paper analyzes mathematical approximations of the ac current of a line-commutated converter (LCC), which is the building block of conventional HVDC systems. The paper derives analytical expressions for the ac current of an LCC during the commutation and non-commutation periods and continues with calculating the percentage error of both the fundamental frequency component of the ac current and the power factor of the converter when the ac current is approximated with either a square or a trapezoidal waveform. The results quantitatively show the benefit of the trapezoidal waveform approximation in terms of its percentage error and simplicity. Keywords- Error analysis; fundamental frequency component; Line commutated converter (LCC); power factor
I.
INTRODUCTION
High-voltage direct current (HVDC) transmission systems are becoming increasingly prevalent in modern power networks. Technical advantages of them over ac transmission systems include (i) linking unsynchronized ac networks, (ii) transmitting large amounts of power over long distances with lower capital cost and lower losses, and (iii) transmission via underground/submarine cables with lower losses [1], to name a few. Although voltage-source converter (VSC) based HVDC is being pursued as an option for special applications [2], the majority of the installed HVDC transmission systems are presently based on line-commutated converters (LCC). The LCC-based HVDC offers the benefits of a more mature technology and is available in much larger ratings than the currently available VSC-based HVDC systems. Due to switching phenomena, the ac-side current of a linecommutated converter is not purely sinusoidal and its waveform depends on such factors as the magnitude of the applied ac voltage, ac line inductance, dc-side current, firing angle (), commutation period (), and frequency of the system [3]. When LCC-HVDC systems are simulated using an electromagnetic transient (EMT) simulator, the ac line current waveform is represented with adequate accuracy provided that a small enough simulation time-step is selected. In other studies, such as high-level controller design or in the study of slow dynamics of the converter and the ac system,
however, simplified representation of the ac current often suffice. The accuracy of the results will therefore depend on how faithful these simplified representations are in capturing the important properties of the actual ac line current. One area of particular interest that significantly benefits from a simple yet reasonably accurate representation of the ac current is when average models of the converter are to be developed. Most averaging techniques depend on an approximation of the ac and dc side quantities of the converter [4], [5], [6], [7]. It is therefore important to quantify the error introduced by approximate waveforms in the development of average models. Commonly used approximations of the ac line current of an LCC include the square waveform [8], [9] and the trapezoidal waveform [10]. This paper develops analytical expressions for the ac line current and its square-wave and trapezoidalwave approximations and presents an analysis of the error of these two common waveforms in representing the fundamental frequency component of the ac current and the power factor of the converter. II. LINE-COMMUTATED CONVERTER
A. Principles of Operation Fig. 1 shows a schematic diagram of a line-commutated converter. This converter block consists of six thyristor valves, which are fired consecutively (as numbered in Fig. 1) with a delay of known as firing angle with respect to a particular instant, most commonly selected to be the positive zero crossing of the line voltage. The dc side of the converter is represented using a constant current source of id. In practice large smoothing reactors are deployed on the dc side of LCCHVDC systems to ensure that the dc line current is essentially constant and with minimum ripple. On the ac side, the converter is connected to a set of balanced three-phase voltages via an inductive element. The inductive element could for example be the leakage reactance of the converter transformer or a line inductor in series. For the analysis that follows a lossless converter is assumed, i.e. the switches do not incur losses during turn-on and turn-off periods and also have no losses during current conduction.
Fig. 2 shows the dc voltage and ac current waveforms of the LCC. It also includes the square and trapezoidal waveform approximations of the ac current. It is noted that the ac line current (Fig. 2(b)) contains a finite duration of time over which the current moves from one constant value to another, e.g. from 0 to Id or from Id to 0. This period of time starts at the switching instant and has an angular duration of , and is widely known as the commutation period. The commutation period results from the ac-side inductance (Lc). The switches that are involved in the process of commutation will be ON during this period; one switch will have diminishing current flow while the other one will gain increasing current. During the overlap period the ac line current has a nonlinear transition between its two boundary constant values as seen in Fig. 2(b). The equivalent circuit of the LCC during commutation period between switches 1 and 5 is showed in Fig. 3. The commutation period between various switches is the same as long as operation conditions remain intact.
Lc + + vb _ _ + vc _ Lc Lc ia ib ic 4 6 2 _ 1 3 5 id=Id +
Lc + + vb _ _ + vc _ Lc Lc
ia ib ic
id=Id +
vd
va
Fig. 3. Equivalent circuit of the LCC during commutation (valves 1 and 5).
vd
va
Fig. 1. Line commutated converter circuit diagram (LCC). vab Voltage vac vbc (a) vd 561 61 612 12 123 23 234 34 Id Current (b) t A (c) t B 345 45 456 56 t vba vca vcb
B. Waveform Analysis The two approximations of the ac line current are also shown in Fig. 2. The square wave approximation (Fig. 2(c)) ignores the overlap period and shows sudden jumps at the instant of switching. The trapezoidal approximation however includes a period of time equal in length to the actual overlap period, during which the current is assumed to vary linearly between the two boundary values. In the following analytical expressions will be developed for the actual ac line current and its approximations. It will then be followed by an analysis of the percentage error of approximation. Assume that the phase voltages of the ac system are as follows. va = Vm sin(t + / 6) (1) va = Vm sin(t / 2) v = V sin(t + 5 / 6) m a To find the phase-a current (ia) during a given overlap period, e.g. between valves 1 and 5 when the dc voltage transfers from vcb to vab as shown in Fig. 3, the sets of equations in (2) must be solved. dia dic va Ls dt = vc Ls dt i = I i (2) d a c i (t = ) = 0 a ia (t = + ) = I d
-Id Id Current
Solution of these equations which results in the following expression for ia. ia (t ) = I s (cos cos(t )) (3) where
3Vm (4) 2 Ls and Vm is the peak value of ac phase-voltage, Ls is the sourceside inductance, and is the nominal angular frequency of the system. With this expression for the phase-a current during the given overlap interval, the following expression for ia for the entire period can be obtained. Is =
-Id Id Current (d) t 0
-Id
Fig. 2. Line commutated converter waveforms. (a) three-phase line voltages and the dc voltage; (b) ac line current. (c) square-wave approximation of the ac line current. (d) trapezoidal approximation of the ac line current.
t + I s (cos cos(t )) 2 Id + t + 3 2 2 2 I d I s (cos cos(t )) + t + + 3 3 3 2 0 + t + + 3 ia (t ) = + t + + I s (cos + cos(t )) 5 + + t + I d 3 2 5 5 )) I d + t + + I s (cos + cos(t 3 3 3 5 + t + 2 + 0 3
III.
RESULTS AND DISCUSSIONS
The fundamental components of the three current waveforms constitute the foundation of the presented studies. Derivation of the fundamental components of the square and trapezoidal waveforms is straightforward as will be shown later. Although it is possible to derive a closed form expression for the fundamental component of the actual ac current [11], it provides little practical value in modeling and control design; as such the study presented in this paper has opted for a highaccuracy numerical Fourier analysis. A. Fundamental Frequency Component of the AC Current The peak value of fundamental component of the squarewave approximation (Fig. 2(c)) is obtained as follows. I (1) = 2 3
Fig. 4. Fundamental frequency component of the ac current over dc current.
Id
(6)
Similarly the following expression is obtained for the peak value of the fundamental frequency component of the trapezoidal approximation. sin( ) 2 I (1) = Id (7) ( ) 2 Fig. 4 shows the variation of the fundamental component of the ac current versus the overlap angle (). For convenience of comparison, the y-axis is per-unitized by the dc current Id. A family of curves is shown for = 0 and 60 (rectifier mode of operation) and for = 120 and 150 2 3
The following expression is used to calculate the percentage error between the actual and approximate values of the quantities shown. approximate - exact percentage error = 100% (8) exact Fig. 5 shows the calculated percentage error for the normalized fundamental frequency component of the ac current. Table I compares the maximum percentage errors for both approximate waveforms. As shown the largest absolute percentage error in estimating the fundamental component of the ac current is 4.54% for the square wave and 1.49% for the trapezoidal approximation, and both occur at =60. In [3] the maximum percentage error of the square wave was reported as 4.3%, which is due to the division of the error by the approximate value and not by the exact value as is done in this paper in (8).
AC current magnitude/DC current [I(1)/Id]
(5) In the following section, the fundamental component of both the square wave and the trapezoidal approximations are compared with that of the above expression. The power factor of the converter system resulting from the actual current waveform and its two approximations will also be studied. The aim of the paper is to investigate the percentage of error that is introduced in the fundamental component of the line current and the converter power factor as a result of squarewave and trapezoidal approximations.
(inverter mode of operation). Although the overlap angle is a function of the firing angle, it is shown as an independent variable as its value can be adjusted by other parameters, e.g. the ac side inductance, dc current and ac voltage. The overlap angle cannot exceed 60, as can be deduced from Fig. 2(b) by observing the distance of /3 between the two points A and B. This period encompasses the overlap angle and as such the largest attainable value for the is 60, as indicated on the x-axis in Fig. 4. Also note that + cannot exceed 180; therefore for every value of the firing angle the corresponding curve continues only until =180. As shown in Fig. 4, the square waveform approximation depends neither on the firing angle nor on the overlap angle and therefore has a constant value. The trapezoidal waveform, on the other hand, only depends on the overlap angle; the actual current waveform depends on both the firing angle and the overlap angle. As seen the approximate waveforms present varying degrees of error depending on the firing and overall angle combinations.
5 4 3 2 1 0 -1 -2 0 10 20 30 40 Overlap angle ( ) 50 60 Trapezoidal approximation Square approximation
IV.
CONCLUSION
Fig. 5. Percentage error of fundamental frequency component of ac current over dc current.
In practice the overlap angle is normally much smaller and hence the percentage error of the two approximations will be less. For example for overlaps angles less than 30 the square wave and the trapezoidal approximations will have percentage errors of less than 1.14% and 0.38%, respectively.
B. Converter Power Factor For calculating the power factor of the converter it is assumed the converter is lossless, i.e. the real power at its ac terminals equals the real power on its dc side. Therefore, the following expression can be stated. V I (1) (9) 3( m )( ) PF = V d I d 2 2 where Vd is the average dc voltage and is given in as (10) [3], and PF is the power factor. Vd =
This paper presented a detailed analysis of the ac current of a line-commutated converter and its approximation with square and trapezoidal waveforms. These approximations affect the calculation of both the fundamental frequency component of the ac current and the power factor of the converter. The paper developed closed form expressions for the fundamental ac current component for the two approximating waveforms and analyzed their percentage error relative to the exact solution. Although both approximations have small errors, the trapezoidal waveform produces much higher accuracy and has a conveniently simple form. The percentage error of the trapezoidal method is well below 0.5% for operating conditions when the overlap angle is below 30, and does not exceed approximately 1.5% for other operating conditions. It is therefore concluded that the combination of the simplicity of its form and the accuracy of its estimation makes the trapezoidal approximation a reliable alternative to the exact form of the ac line current in studies such as averaging and dynamic phasor based modeling of the LCC.
3 3
Vm cos
Ls I d
(10)
From (9) the power factor of the converter is simply derived as follows. 2V I 1 (11) PF = d d (1) 3Vm I As seen in (11) the approximation used for fundamental component of the ac current affects the calculation of the converter power factor. Using a formula similar to (8), the percentage error in estimating the converter power factor is calculated for the two approximating waveforms. The results are shown in Fig. 7 and the largest percentage errors are reported in Table I. It is again seen that the trapezoidal approximation provides a much more accurate representation of the converter power factor.
Fig. 6. Percentage error of the power factor of the converter. Table I Percentage error of both types of approximation Maximum percentage error Square Trapezoidal of approximation approximation AC current magnitude 4.54 -1.49 AC current magnitude for normal operating condition 1.14 -0.38 (30) Converters power factor -4.34 1.52 Converters power factor for normal operating condition -1.13 0.38 (30)
Percentage error (%)
References
[1] [2] J. Arrillaga, High voltage direct current transmission, IEE, 1998. N. Flourentzou, V. G. Agelidis, and G. D. Demetriades, "VSC-based HVDC power transmission systems: An overview," IEEE Trans. on Power Electronics, vol. 24, no. 3, pp. 592-602, March 2009. [3] E. W. Kimbark, Direct Current Transmission. New York: Wiley, 1971, vol. I. [4] S. Sudhoff and O. Wasynczuk, "Analysis and average-value modeling of line-commutated converter-synchronous machine systems," IEEE Trans. on Energy Conversion, vol. 8, no. 1, pp. 92-99, March 1993. [5] S. Sudhoff, "Analysis and average-value modeling of dual linecommutated converter-6-phase synchronous machine systems," IEEE Trans. on Energy Conversion, vol. 8, no. 3, pp. 411-417, September 1993. [6] S. Chiniforoosh, J. Jatskevich, V. Dinavahi, R. Iravani, J. Martinez, and A. Ramirez, "Dynamic average modeling of line-commutated converters for power systems applications," in proc. PES general meeting, pp. 1-8, 2009. [7] S. Chiniforoosh, J. Jatskevich, A. Yazdani, V. Sood, V. Dinavahi, J. Martinez, and A. Ramirez, "Definitions and applications of dynamic average models for analysis of power systems," IEEE Trans. on Power Delivery, vol. 25, no. 4, pp. 2655-2669, October 2010. [8] K. Padiyar, HVDC power transmission systems: technology and system interactions, John Wiley & Sons, 1990. [9] W. Hammer, "Dynamic modeling of line and capacitor commutated converters for HVDC power transmission," Ph.D. dissertation, Dept. Electrical Eng., Swiss Federal Institute of Technology, Zurich, 2003. [10] Q. Qi, L. Jiao, Z. Yan, Y. Ni, S. Chen, and F. Wu, "Modeling and simulation of HVDC with dynamic phasors," in proc. of the CSEE, vol. 12, pp. 32-38, 2003. [11] J. Arrillaga, Y.H. Liu, and N.R. Watson, Flexible Power Transmission: The HVDC Options, John Wiley & Sons, Ltd, 2007.
A new approach for optimal reactive power flow solution of multi-terminal hvdc systems using genetic algorithm
Ula Kl
Assistant Professor, Celal Bayar University Faculty of Hasan Ferdi Turgutlu Technology Manisa, Turkey ulas.kilic@cbu.edu.tr
Abstract Optimal reactive power flow (ORPF) is one of the known problems of the power systems. Many numerical and heuristic methods were used to solve this problem so far. As seen from these studies in literature, heuristic methods are more effective and faster than numerical methods. This case is to make more attractive and mandatory the using of heuristic methods in optimal power flow solution of High Voltage Direct Current (HVDC) systems. In this study, ORPF solution of multi-terminal HVDC systems is accomplished by using the genetic algorithm (GA) that is one of the heuristic methods. A new approach is used in opposition to the current-balancing method used mostly in literature for the first time. The proposed approach is tested on the modified IEEE 14-bus test system. The obtained results are compared to that reported in the literature to show validity and effectiveness of the new approach. Keywords-Optimal reactive power flow; HVDC system; multiterminal; genetic algorithm
Krat Ayan
Assistant Professor, Sakarya University Faculty of Computer and Information Sciences Sakarya, Turkey kayan@sakarya.edu.tr GA is essentially search algorithm that is based on natural genetics. It evaluates the optimal solution with randomized, structured exchanges of information among exist solutions. The constraints on solution space are not implemented during the process in GA approach. Therefore GA approach is a robust method. GA approach was only applied to ORPF of purely AC power systems until today [11-13]. The purpose of this paper is to solve the ORPF in multi-terminal HVDC systems using GA approach. GA is essentially search algorithm that is based on natural genetics. It evaluates the optimal solution with randomized, structured exchanges of information among exist solutions. The constraints on solution space are not implemented during the process in GA approach. Therefore GA approach is a robust method. The power of GA comes from its ability to exploit historical information from previous solution guesses to increase performance of future solutions [14]. In this study, ORPF solution of multi-terminal HVDC systems is accomplished by using GA that is one of the heuristic methods. A new approach is used in opposition to the current-balancing method used mostly in literature for the first time. The proposed approach is tested on the modified IEEE 14-bus test system [15]. The obtained results are compared to that reported in the literature to show validity and effectiveness of the new approach. II. THE MODELING OF DC TRANSMISSION LINK
I.
INTRODUCTION
Many studies were performed for power flow solutions of multi-terminal HVDC systems so far [1-3]. Most of them use the numerical methods. There are two basic approaches for solving the power flow equations of AC-DC power systems in literature. The first is the sequential approach [4-5]. In this method, the alternating current (AC) and direct current (DC) equations are solved separately by successive iterations. Although the implementation of the sequential method is easy, it has convergence problems associated with certain situations and are the state vector doesnt contain explicitly the dc variables. The second approach is known as the unified approach [6]. The scientists have used many different methods for solving ORPF problem of purely AC power systems so far [710]. These methods are numerical and heuristic methods. According to the results reported in literature, it can be seen that heuristic methods are superior from the numerical methods [9-10]. The important one of the advantages of heuristic methods is that they convergence to the optimum solution in more short time than others and convergence fewer local minimum.
Before analyzing DC transmission system, it is necessary to model DC transmission link and the converters. The modeling is based on the accepted assumptions in the literature [6]. Furthermore, a representation of an AC bus having the generators, AC lines, shunt compensators and converters can be represented by Fig. 1 [4]. The active and reactive power equalities at such a general ac bus are given by Eqs. (1) and (2).
Pgk = Plk + Pdck + Pk

Q gk + Q sk = Qlk + Qdck + Q k
(1) (2)
Vk k
Pk + jQk
I dck
Pgk + jQ gk
Qck
Pdck + jQdck
k
Vk k
Pdck
tdck
V dck
Qdck
I k k
Plk , Q lk
Figure 1. General AC bus representation [4]
Figure 2. Equivalent Representation of a Converter Terminal [4]
Xc
I dck
In Fig. 1 P , Q , V and represent active power outputs, reactive powers, bus voltage magnitudes and bus voltage angles, respectively. The subscripts g , l , s and dc also represent generator, load, shunt reactive compensator and DC link, respectively.
3 2
Vk tdck cos( k )
Vdck
Pk = Vk Qk = V k
V (G
j j =1 N
kj
cos kj + Bkj sin kj sin kj B kj cos kj
)
)
(3)
Figure 3. Converter Circuit Model[4]
V (G
j j =1
kj
(4)
where V j and Vk are the voltage magnitudes of j th and k th buses; G kj and B kj are the conductance and susceptance between buses k and j of the bus admittance matrix ( Ybus ); kj is the angular difference between buses k and j , N is a number of all buses of the power system.
A. DC system equations Direct voltage of m-pulse rectifier in a monopolar HVDC system can be expressed as follows [16]: Vdck = nmtdck
The expression of V dcok = 3 2V k t dck / in the first term of the Eq. (6) is defined as the open circuit direct voltage. Thus, for the actual quantities of HVDC system, the direct voltage, the phase shifted, the active power, and the reactive power expressions can be given by equalities, respectively.
Vdck = Vdcok cos k
Vdck = Vdcok cos k
X ck I dck
X ck I dck
(7) (8)
k = cos 1
V dck V dcok
(9) (10) (11)
2Vk
nmX ck sin cos k I dck 2 m
(5)
Pdck = V dck I dck
where n is the converter number in series, m is peak number on load voltage per period, t dck is the tap ratio of the transformer at dc side, Vk is the secondary effective voltage of the transformer, k is delay angle for the rectifier operation (or extinction advance angle for inverter operation), X ck is the commutation reactance, I dck is dc current. The equivalent representation of a converter terminal and converter circuit model is given in Figs. 2 and 3, respectively. Rearranging the equation (5) for m = 6 and n = 1 ,
V dck = 3 2
Qdck = Pdck tan k

where
represents phase shifted.
B. Per unit transformation After defining the base active power PACbase and the base voltage V ACbase of ac side, the base current I ACbase and base impedance Z ACbase for ac side can be determined as follows: PACbase = PDCbase I ACbase = PACbase 3V ACbase (12) (13)
V k t dck cos k
X ck I dck
(6)
Z ACbase =
V ACbase 3 I ACbase
(14)
q dck = p dck tan k
(29)
After defining the base active power PDCbase , the expressions of V DCbase , I DCbase , and Z DCbase for dc side can be
determined by defining C b = 2 nm sin / as follows: m
V DCbase = C bV ACbase
I DCbase = 3 I ACbase Cb
(15) (16) (17)
III. ILLUSTRATION OF GA GA is heuristic algorithm based on natural selection. GAs were firstly utilized by Holland in 1975 for solving optimization problems [17]. The base logic of the algorithm is that gens of powerful individuals are based to be carried over next generation and others are based to be detached in next generation. In natural selection, a human born, grows, and dies. These stages of human life correspond to the different operators in the algorithm. The simple flow scheme of GA is shown in Fig. 4 [18]. The detailed information related to GA operators is in Ref. [17-20].
2 Z DCbase = C b Z ACbase
where V DCbase , I DCbase and Z DCbase show the base active power, the base direct voltage, the base direct current, and the base impedance of dc side, respectively. Per unit expressions by dividing the actual ac and dc expressions to ac and dc base expressions are given by the following equalities:
vk = Vk V ACbase
(18) (19) (20)
x ck =
rck =
X ck Z ACbase
nmX ck / 2 nmX ck / 2 x ck = 2 = Z DCbase C b Z ACbase 4nm sin 2 ( / m )
vdck = idck =
Vdck VDCbase I dck I DCbase
(21) (22)
Figure 4. The simple flow scheme of GA [18]
where v k , x ck , rck , v dck , and idck show ac voltage, commutation reactance, commutation resistance, dc voltage and dc current in per unit, respectively. According to these, the others are determined as follows:
V dck V DCbase nmt dck = 2V k
sin cos k m
V DCbase
nmX ck I dck 2 V DCbase
IV. THE CALCULATION OF FITNESS VALUE It is significant the determination of the state and the control variables to solve OPF problem of two-terminal HVDC system. Furthermore, these control variables should be the same as those of the problem to be optimized. The control variables of this optimization problem are selected as follows: u = [u AC , u DC ] u AC = p g 2 ,..., p gNg , v g1 ,...,v gNg , t1 ,...,t NT , qc1 ,..., q Nc
(23)
Vdck VDCbase
nmtdck =
nmX ck sin cos k I dck m 2 Z DCbase I DCbase 2 nm sin VACbase / m 2Vk
(24)
u DC = [ p dc1 , q dc1 , q dc 2 , q dc3 , i dc12 , i dc13 ]
(30) (31) (32)
where p gi except for the slack bus is the generator active power outputs, v gi is the generator voltage magnitudes, t is the transformer tap setting, and q c is the shunt compensation. N g , N T and N c is the number of generator buses, transformers, and shunt VAR compensations, respectively. The state variables of this optimization problem are selected as follows:
v dck = v k t dck cos k rck i dck p dck = v dck i dck v dcok = v k t dck
(25) (26) (27) (28)
k = cos 1 (v dck / v dcok )
x = [x AC , x DC ] x AC = p gslack , q g1 ,..., q gNg , v L1 ,..., v LNl
(33)
x DC = v dc1 , v dc 2 , v dc3 , t dc1 , t dc 2 , t dc3 , 1 , 2 , 3
i dc 2 = i dc 23 i dc12
Step 8: idc 3 is calculated by the Eq. (42) as follows:
(41)
(34) (35)
i dc3 = i dc 23 i dc13
Step 9: p dc 2 is calculated by the Eq. (43) as follows:
(42)
where p gslack is the slack bus active power output, q gi is the reactive power outputs, v Li is the load bus voltage magnitudes, N l is the number of load buses, respectively. The representation of multi-terminal HVDC system is shown in Fig. 5.
v1 1
p dc 2 = v dc 2 i dc 2
Step 10: p dc3 is calculated by the Eq. (44) as follows:
(43)
p dc3 = v dc3 i dc3
(44)
t dc1
qdc1
rc 1
Step 11: The active and reactive powers of DC links are considered as load. Thus, the active and reactive powers of the buses connected to DC links are updated by Eq. (45).
update p lj = p lj + p dcj update q lj = q lj + q dcj
pdc1
j = 1,2,3
(45)
idc 1
vdc1
Step 12: Load flow analysis is realized by Newton-Raphson method. All the state variables of AC side are certain and the unknown state variables of DC side are obtained step by step. Step 13: j is calculated by the Eq. (46) as follows:
idc3
rc 3
idc12
v2 2
t dc 2
idc13
rdc13
pdc 2
qdc2
rc 2
idc2
rdc12
rdc23
pdc3
qdc3
t dc3
v33
j = tan 1
vdc2
idc23
vdc3
qdcj pdcj
j = 1,2,3
(46)
Figure 5. The representation of multi-terminal HVDC system
Step 14: t dcj is calculated by the Eq. (47) as follows:
The power flow calculation for each individual in heuristic methods is performed and a fitness value Fi for each individual to evaluate its quality is calculated as follows: Step 1: The initial conditions of the control variables are determined by heuristic method. Step 2: i dc1 is calculated by the Eq. (36) as follows:
t dcj =
v dcj v j cos j
j = 1,2,3
(47)
Step 15: v doj is calculated by the Eq. (48) as follows:
v doj = v dcj + i dcj rcj
j = 1,2,3
(48)
Step 16: j or j is calculated by the Eq. (49) as follows:

(36)
i dc1 = i dc12 + i dc13

Step 3: v dc1 is calculated by the Eq. (37) as follows:
j , j = cos 1
v dcj v doj t dcj

Ng
j = 1,2,3
(49)
v dc1 = p dc1 i dc1

Step 4: v dc 2 is calculated by the Eq. (38) as follows:
(37)
Step 17: Power loss is calculated by the Eq. (50) as follows:
v dc 2 = v dc1 i dc12 rdc12

Step 5: v dc3 is calculated by the Eq. (39) as follows:
(38)
p loss =
i =1
p gi
p
j =1
lj
(50)
v dc3 = v dc1 i dc13 rdc13

Step 6: i dc 23 is calculated by the Eq. (40) as follows:
(39)
Step 18: The fitness value for each individual is obtained by the Eq. (51) as follows:
lim Fi = ploss + pgslack pgslack +
) (q
2 i =1
Ng
lim + gi qgi 3
) (v v )
2 li i =1
Nl
lim 2 li
i dc 23 = (v dc 2 v dc3 ) rdc 23
Step 7: i dc 2 is calculated by the Eq. (41) as follows:
(40)
(
j=1
lim 2 vdcj vdcj +
) (
j=1
lim 2 tdcj tdcj +
) (
j=1
2 j lim j
(51)
4
lim lim lim lim lim where p gslack , q gi , v Li , v dc t dc and lim show the limits
2.5 2 F itn e s s v a lu e 1.5 1
x 10
V. SIMULATION RESULTS AC transmission lines among the buses 5, 4 and 2 in the original IEEE 14-bus test system is replaced to multi-terminal HVDC system and the modified IEEE 14-bus test system is obtained as shown in Fig. 6 [15].
9 8 7 10 20 30 40 50 60 Iteration number 70 80 90 6 100
0.5 0
Figure 7. The variation of fitness value and power loss versus iteration number
Table 1. AC system results

Bus no. 1 2 3 4 5 6 7 8 9 10 11 12 13 14
pg
0.80542 0.89868 0.95773 ------------
qg
-0.01685 0.13818 0.40552 ------------
vg
1.0643 1.0582 1.0593 1.0168 1.0369 1.0432 1.0212 1.0972 1.0076 1.0078 1.0281 1.0420 1.0328 0.9993
0.0000 -1.0377 -1.9637 -4.6589 -5.2260 -8.2141 -6.0349 -6.0349 -6.7225 -6.1544 -4.1753 -3.1347 -3.4644 -6.3509
Figure 6. The modified IEEE 14-bus ac-dc test system [15]
A lot of trial is made for the different iteration numbers and population sizes. Great iteration number (>>1000) and population size (>>200) is not preferred because the great computational time is necessary for reaching an optimum. Therefore, in selection of the iteration number several experiments are performed for different population sizes. As a result of the performed experiments, the best individual number is determined as 50, the best child number is determined as 25, and the best individual number undergo mutation is determined as 5 for the GA. The running time of software for test system is obtained as 104.5 s. The variation of fitness value and power loss versus iteration number is shown in Fig. 7. As seen from Fig. 7 that the algorithm converges an optimum within 25 iterations in terms of the fitness value. It is clear that there is no increase in power loss after 25 iterations. AC and DC system results obtained by the new approach are given in Tables 1 and 2, respectively. Furthermore, t47 , t49 , t53 , qc 6 , and qc8 are obtained as 1.0621 , 0.9486 , 1.0147 , 0.5508 , and 0.4733 in p.u respectively. In Table 3, the results obtained by the proposed new approach are compared to that reported in the literature. As seen from Table 3 that the power loss obtained by the proposed approach is lesser than the others.
Table 2. DC system results

Conver. no. 1 2 3 Active power (p.u.) 0.6185 -0.9447 0.3374 Reactive power (p.u.) 0.1532 0.3999 0.0994 DC current (p.u.) 0.6355 -0.9804 0.3449 Control angles (degree) 9.2723 18.6737 15.3280 Effective transformer ratio 0.9670 1.0291 0.9639
Table 3. Comparative results Power loss (MW) GA 6.0545 Ref. [15] 6.640 Ref. [16] 8.532
VI. CONCLUSION In this study, a new approach is used to solve ORPF of multi-terminal HVDC system using GA for the first time. ORPF of multi-terminal HVDC system can be solved without using the current-balancing method by this approach. As seen that, the results obtained by this approach are lesser than those reported in Ref. [15-16] by 8.81% and 29.03%, respectively. This approach together with different heuristic methods can be also used for OPF and ORPF solutions of HVDC systems.
P o w e r lo s s (M W )
of the related variables, respectively.
Fitness value
Power loss (MW)
11 10
REFERENCES
1. U. Arifolu, The power flow algorithm for balanced and unbalanced bipolar multiterminal ac/dc systems, Electric Power Systems Research, vol. 64, no. 3, pp. 239-246, 2003. U. Arifolu, Load flow based on newtons method using norton equivalent circuit for an AC-DC multiterminal system, European Transactions on Electrical Power, vol. 9, no. 3, pp. 167-174, 1999. M.M. El-Marsafawy, and R.M. Mathur, A New, Fast Technique for Load-Flow Solution of Integrated Multi-Terminal DC/AC Systems, IEEE Transactions on Power Apparatus and Systems, vol.99, no.1, pp.246-255, 1980. C.N. Lu, S.S. Chen, and C.M. Ong, The incorporation of HVDC equations in optimal power flow methods using sequential quadratic programming techniques, IEEE Trans. Power Syst, vol. 3, no. 3, pp. 1005-1011, 1988. J. Reeve, G. Fahmy, and B. Stott, Versatile load flow method for multi-terminal HVDC system, IEEE Transactions on Power Apparatus and Systems, vol. 96, no. 3, pp. 925-933, 1977. T. Smed, G. Andersson, G.B. Sheble, and L.L. Gigsby A new approach to AC-DC power flow, IEEE Transactions on Power Systems,vol. 6, no. 3, pp. 1238-1244, 1991. N. Grudinin, Reactive power optimization using successive quadratic programming method, IEEE Trans Power Syst, vol. 13, no. 4, pp. 1219-1225, 1998. M.K. Mangoli, K.Y. Lee, and Y.M. Park, Optimal real and reactive power control using linear programming, Electr Power Syst Res, vol. 26, no. 1, pp. 1-10, 1993. K. Ayan, and U. Kl, Artificial bee colony algorithm solution for optimal reactive power flow, Applied Soft Computing, vol. 12, no. 5, pp. 1477-1482, 2012. S. obanl, A. ztrk, U. Gven, and S. Tosun, Active power loss minimization in electric power systems through artificial bee colony algorithm, Int Rev Electr Eng, vol. 5, no. 3, pp. 22172223, 2010. K. Iba, Reactive power optimization by genetic algorithm, IEEE Trans Power Syst, vol. 9, no. 2, pp. 685-692, 1994. Q.H. Wu, Y.J. Cao, and J.Y. Wen, Optimal reactive power dispatch using an adaptive genetic algorithm, Int J Electr Power Energy Syst, vol. 20, no. 8, pp. 563-569, 1998. Y. Liu, L. Ma, and J. Zhang, Reactive power optimization by GA/SA/TS combined algorithms, Int J Electr Power Energy Syst; vol. 24, no. 9, pp. 765-769, 2002. D. Walters, and G. Sheble, Genetic Algorithm Solution for Economic Dispatch with Valve Point Loading, IEEE Transactions on Power Systems, vol. 8, no. 3, pp. 1325-1332, 1993. F. Yaln, and U. Arifolu, A new approach based on genetic algorithm for optimal reactive power flow solution in multiterminal AC-DC systems, Przeglad Elektrotechniczny, vol.89, no. 3a, pp. 231-235, 2013. U. Arifolu, Optimal Power Flow Using Sequential Power Flow Approach For an AC-DC Power System, Ph.D. Thesis, Istanbul Teknik University, 1993. J. H. Holland, Adaptation in Natural and Artificial Systems, Ann Arbor: University of Michigan Press 1975. A.. ahin, B. Kl, and U. Kl, Optimization of heat pump using fuzzy logic and genetic algorithm, Heat and Mass Transfer, vol. 47, no. 12, pp. 1553-1560, 2011. MATLAB Optimization Toolbox 5 Users Guide 2012, The Math Works, Inc http://en.wikipedia.org/wiki/Genetic_algorithm
2.
3.
4.
5.
6.
7.
8.
9.
10.
11. 12.
13.
14.
15.
16.
17. 18.
19. 20.
Lyapunov Function-Based Control of Active Power Filters to Reduce the Adverse Effects of Harmonics on Induction Motors
Ibrahim Aliskan
Dept. of Electrical and Electronics Eng. Bulent Ecevit University Zonguldak, Turkey ialiskan@beun.edu.tr
Kayhan Gulez
Dept. of Control and Automation Eng. Yildiz Technical University Istanbul, Turkey gulez@yildiz.edu.tr
Gurkan Tuna
Dept. of Computer Programming Trakya University Edirne, Turkey gurkantuna@trakya.edu.tr
M. zgr Kizilkaya
Aeronautics and Space Technologies Institute Turkish Air Force Academy Istanbul, Turkey mkizilkaya@hho.edu.tr
AbstractIn this study, a pulse width modulator controller inverter type active power filter is used for reducing the adverse effects of harmonics. Induction motor control system used in this study comprises a proportional-integral controller, a direct torque control algorithm, and an inverter module. Nonlinear controller algorithm proposed in this study enables setting different parameters in the controller system. Our experimental results show that more planar electromagnetic torque can be obtained by using the proposed controller-based filter than a classical controlled-based filter. Keywordsactive power filters; harmonics; Lyapunov function
redistribution of power to keep the system balanced and electromagnetic interference reduction [2]. Various filters are used in motor control systems to neutralize the effects of these harmonics on motor torque [3], [4]. In these systems, direct torque control is used to minimize torque ripples [3], [5], [6]. In this study, a nonlinear controller designed with Lyapunov function for active power filter system is proposed. The results of the experimental studies show that adverse effects on motor torque can be neutralized by using the proposed nonlinear controller-based filter system. II. PROPOSED SYSTEM
I.
INTRODUCTION
Harmonics, one of the most common power quality problems, have adverse effects on almost all systems in the industry [1]. To deal with adverse effects, passive inductancecapacitance filters and/or active power filters are used. An efficient solution to deal with harmonics is the utilization of active filters. The advantages of active filters are the elimination of unwanted harmonics, power factor correction,
The system consists of an induction motor, a vector controller based inverter, and a proportional integral (PI) speed controller. Fig. 1 shows the schematic diagram of the model and Fig. 2 shows the active filter structure. The shape of supply voltage provided to the system is given in Fig. 3.
Figure1. The schematic diagram used in the experimental studies.
Figure 2. The model of the active power filter.
Figure 3. The shape of supply voltage.
For a better explanation of the overall system in detail, firstly, induction machine model needs to be written in the general reference frame.
s = Ls .i s + Lm .i r r = Lr .i + Lm .i
where
g g r
(3) (4)
g s
V s = Rs .i s +
g r
g d s + j.wg . s dt g
(1)
V s is stator supply voltage, i s and i r are stator and rotor currents, wg is general reference speed, wr is rotor
speed.
g d r 0 = Rr .i + + j ( wg wr ) r dt
(2)
Rs , Rr are stator and rotor resistances, s and r
are stator and rotor flux linkages. The superscript g refers to the reference frame [5]. Mechanical dynamics of the induction motor are given in Eq. (5) and Eq. (6).
g g 3 p Te = . .Im s .i s (5) 2 2 dw 2 dw J . m = J . . r = Te Tload (6) dt p dt where Te is the motor torque, J is the moment of the inertia
Gc ( s ) = K p (1 +
1 ) i .s
(13) (14)
K p = K u / 2.2 = 20 / 2.2 & i = Pu /1.2 = 1/1.2
of the system, and Tload is the torque of system load [5]. Direct torque controller (DTC) algorithm is the hysteresis control of the magnitudes of stator flux and electromagnetic torque. The controller selects one of the six non-zero and two zero voltage vectors of the inverter [6].
Required angular velocity is the reference signal for the system. As commonly used at design phases, unit step function is used as the reference input. 0.2N/(m/s) is the friction coefficient of the system and 0.407kg.m2 is the inertial load structure. Time delay is 0.12s and time constant is 2.5s. When these values are used, the results are that the proportion is 20.85 and time constant of the classical controller is 0.0192s. In 2 the system, Lyapunov function is V = k .e / 2 . V = 2.k .V is the required dynamic structure for exponential stability. Nonlinear controller function is
f (e , e ) = e + (e active filter output ) .0, 00031 dt .
( (
Te = Teref Te
(7)
g
Here, error is vector variable.
e = harmonics active filter output

Dynamics of the filter
sg = sref s
where
(8)
output are d active filter ouput / dt = k .e d harmonics / dt . k is
Te and sg are torque and flux errors,

g s
taken from the classical controller design. III. EXPERIMENTAL STUDIES
respectively. The switching vector selection depends on the fact that the errors must be taken into the hysteresis band. The vectors of V are determined using Eq. (9) [6].
2. 4. j. j. g 2 V s = .Vdc . Sa + e 3 .Sb + e 3 .Sc 3
(9)
The quarter electrical vehicle system built for experimental studies is shown in Fig. 4. The experimental results with DC power supply are shown in Fig. 5, 6, 7 and 8. It is powered by a direct current power supply. In the experimental studies, {0.5 0.6 0.8 1.1 1.3 1.5 1.65 1.8 1.9 2.1 3 3.8 6 7.2 9 10.2 11.3 12.1}s switching times are used for nonlinear load to simulate the real system.
If right voltage vector of inverter is selected, stator flux is rotated to desired frequency (ws), which is inside a specified band [8]. When stator ohmic drops are neglected, stator flux dynamic is calculated using Eq. (10). (10) Required stator flux and electromagnetic torque are obtained by the selection of correct voltage vector by using Eq. (11) and Eq. (12) [7], [9]. Here, is the angle between stator and rotor fluxes.
s = V s .t
s = V s i s .Rs dt
3 s s s s Te = .P. ( ds .iqs qs .ids ) 2
where
s s s
(11) (12)
V s and i s are the measured stator voltage and stator

Figure 4. The quarter electrical vehicle system for experimental studies.
current signals. s is the calculated stator flux vector. In the system, the aim of the speed controller is to provide speed error value of zero to the system by the use of the PI controller. For tuning, Ziegler-Nichols tuning rule is used. Eq. (13) gives the parameters of the controller. The controller tuning function is selected as 1sin2t. The parametric values are calculated using Eq. (14).
Stator flux positions of the motor are shown in Fig. 5, mechanical angular velocity of the rotor is shown in Fig. 6, mechanical speed of the rotor when only DC power supply is used is shown in Fig. 7, and finally electromagnetic torque of the stator is shown in Fig. 8. As it can be seen in these figures, electrical and mechanical signals of the induction motor system are disturbed by harmonic signals. Total harmonic distortion is 36.25%.
The effectiveness of the system using AC voltage with harmonic distortion as in Fig.3 with and without proposed controller is shown in the following figures. Fig. 9 and Fig. 10 show stator flux positions of the motor for the classical controller-based filter and for the nonlinear controller-based filter, respectively. Flux positions finally settle at 0.3Wb reference value. Fig. 11 and Fig. 12 show mechanical angular velocities of the rotor for the classical controller-based filter and for the nonlinear controller-based filter, respectively.
Figure 5. Stator flux positions of the motor.
Figure 9. Stator flux positions of the motor for the classical controller-based filter.
Figure 6. Mechanical angular velocity of the rotor.
Figure 10. Stator flux positions of the motor for the nonlinear controller-based filter.
Figure 7. Mechanical speed of the rotor when only DC power supply is used.
Figure 11. Mechanical angular velocity of the rotor for the classical controllerbased filter. Figure 8. Electromagnetic torque of the stator.
IV.
CONCLUSION
In this study, a nonlinear controller designed with Lyapunov function for active power filter system is proposed. The advantage of this controller over classical controllers is that dynamic effects of error signal are added to the controller function. As a result, more planar electromagnetic signal can be obtained. The experimental studies conducted in this study proved the effectiveness of the proposed controller. When time sharing algorithm was applied to the control operation of the motor supplier inverter module, total harmonic distortion value of the nonlinear controller based system was 9.72%, whereas the distortion of the classical controller based one was 33.08%. REFERENCES
[1] K. Srinivasan and R. Jutras, Conforming and Nonconfirming Current for Attributing Steady State Power Quality Problems, IEEE Transactions on Power Delivery, 1998, vol.13, no.1, pp. 212-217. K.Gulez, I. Aliskan, T. V. Mumcu, and G. Cansever, Neural Network Based Control of AC-AC Converter for Voltage Sags, Harmonics and EMI Reduction, Lecture Notes in Computer Science, 2007, vol. 4681, pp. 534-544. K. Gulez, A. A. Adam, and H. Pastaci, Passive Filter Topology to Minimize Torque Ripples and Harmonic Noises in IPMSM Derived with HDTC, International Journal of Electronics, 2007, vol. 94, no. 1, pp. 23-33. K. Gulez and A. A. Adam, Compound Passive Filter to Minimize Torque Ripples and EMI Noises in PMSM Drives, Proceedings of SICE Annual Conference, 2007, pp. 77-82. N. R. N. Idris and A. H. M Yatim, Direct Torque Control of Induction Machines with Constant Switching Frequency and Reduced Torque Ripple, IEEE Transactions on Industrial Electronics, 2004, vol. 51, no. 4, pp. 758-767. H. F. A. Wahab and H. Sanusi, Simulink Model of Direct Torque Control of Induction Machine, American Journal of Applied Sciences, 2008, vol. 5, pp. 1083-1090. C. M. Ong, Dynamic Simulation of Electric Machinery Using MATLAB/Simulink, New Jersey: Prentice Hall, 1998, pp. 415-438. M. C. Di Piazza, A. Ragusa, and G. Vitale, Effects of Common-Mode Active Filtering in Induction Motor Drives for Electric Vehicles, IEEE Transactions on Vehicular Technology, 2010, vol. 59, no. 6, pp. 26642673. M. C. Di Piazza, A. Ragusa, and G. Vitale, An Optimized Feedback Common Mode Active Filter for Vehicular Induction Motor Drives, IEEE Transactions on Power Electronics, 2011, vol. 26, no. 11, pp. 3153-3162.
Figure 12. Mechanical angular velocity of the rotor for the nonlinear controller-based filter.
Fig. 13 and Fig. 14 show stator electromagnetic torques of the motor for the classical controller-based filter and for the nonlinear controller-based filter, respectively. When the figures are compared, the advantages of the nonlinear controller-based filter can easily be seen. For example, more planar results are obtained in 4.4s - 5.5s time period when the nonlinear controller-based filter is used.
[2]
[3]
[4]
[5]
[6]
[7] [8] Figure 13. Electromagnetic torque of the stator for the classical controllerbased filter.
[9]
Figure 14. Electromagnetic torque of the stator for the nonlinear controllerbased filter.
Congestion Effects on Regional & System Emission and Consumers Allocated Cost
M. Pourakbari-Kasmaei, M. J. Rider, and J. R. S. Mantovani
Department of Electrical and Electronic Engineering UNESP Ilha Solteira, Brazil mahdi.pourakbari@ieee.org
Abstract This paper considers the congestion effects on emission and consumers allocated cost. In order to consider some environmental and operational effects of congestion, an environmental constrained active-reactive optimal power flow (AROPF) considering capability curve is presented. On outage conditions, the total cost of the system will increase. On the other hand in power systems, the operating cost and system emission have conflicted objectives, then it may be concluded that the outage in the system may lead to a total emission decrease. In this paper the famous Aumann-Shapley method is used as a pricing methodology. Two case studies such as 14-bus and 118-bus IEEE test systems are conducted. Results demonstrate that, although the line outage in power systems leads to increase the total cost, the amount of emission depending on the place where the outage occurs can be more than, less than or equal to the normal conditions emission. Also results show that although from power sellers standpoint the well-known Aumann-Shapley method is a precise pricing method to cover the incurred cost with an acceptable error that can show the real effect of congestion on consumers cost, from consumers standpoint it is not a good method for cost allocation, because some consumers will face with an increase in cost and the others will face with a decrease on their cost. Keywords- Aumann-Shapley pricing; active-reactive OPF; regional emission limit; system emission limit; congestion.
with active power generation, where the generation and transfer of reactive power yields to active power loss and hence consumes energy. There are lots of methodologies to find a solution for OPF where some of them are analytical, and the others are heuristic search methods such as linear programming (LP) [3], NewtonRaphson (NR) [4], nonlinear programming (NLP) [5], quadratic programming (QP) [6], interior point [7], genetic algorithm (GA) [8], miscellaneous artificial intelligent (MAI) [9], evolutionary programming (EP) [10], ant colony optimization (ACO) [11], particle swarm optimization (PSO) [12], fuzzy logic (FL) [13], etc. Recently, electricity power pricing has become a crucial issue in restructured power system [14]. Spot or real-time pricing of electricity has provided the economic structure for many of new service options. In the literature most of the studies have focused on reactive power pricing builds on marginal cost theory, which has been applied in the spot price for real power [15], and this method of electric power pricing does not consider some concepts such as block rate, demand charges, backup charges, and so on, and in this regard the role of energy marketplace is important. The spot price depends on supply and demand conditions at that spot [16], [17]. Real-time pricing of reactive power in most researches are based on active power pricing. Also in these days there is a growing concern on harmful environmental impacts of generating electricity focusing on pollution [18], [19]. Since 1990 the clean air act amendments (CAAA), the utilities are to modify their design or operational strategies to deplete the pollution and atmospheric emissions [20] [21]. The limitation on producing emission can be considered as system emission limit (EMS) and regional emission limit (EMA) as well. Power system congestion will effect on pricing and also on amounts of regional and system emission. Transmission congestion occurs when there is not sufficient transmission capacity to meet the transmission service constraints among busses or within a region. In other word, a power system will face with a congestion condition when a transmission line flow does not meet the reliability limits. One of the methods that can help to increase the line flow and consequently may alleviate the congestion limitations is producing more reactive power to
I.
INTRODUCTION
According to fast growing power demand associated with the fuel cost increase, economic-oriented tools such as economic load dispatch (ED) and optimal power flow (OPF) have become crucial issues in power system operation. As the ED and OPF are used in real-time energy management, they have been considered as the kernel of a power system [1], [2]. In order to have a proper power system considering operational and environmental constraints, the role of a robust and precise tool is completely undeniable. The primal objective of an OPF is to minimize the total cost of active and/or reactive generation considering active and reactive power balance, power flow limits, and active and reactive power generation limits. On the other hand, the profound effect of reactive power on power system security is the inseparable part of a proper power system, as it affects the voltage profile of the system and also it has a close relation
This work was supported by FAPESP under Grant 2011/13995-5, CNPq, and FEPISA from Ilha Solteira- Sao Paulo.
increase the transmission of active power in lines, and then the role of reactive power and its optimal dispatch is very important. Moreover, the electricity cannot be stored economically, and transmission congestion may prevent a free exchange among control areas, then in power systems considering the optimal active and reactive power flow and allocating the congestion cost among consumers are considered as two important problems. The well-known Aumann-Shapley (A-S) pricing method which is based on a game-theoretic framework is a good approach for price allocation [22]. In this paper in order to make a good price allocation, some practical and environmental constraints via a modified OPF is used, where the capability curve, EMS, and EMA are taken into consideration. Effect of congestion on regional and system emission is taken into account, and also via the AmannShapley pricing method its real effect on consumers allocated cost are considered and the drawback of this method to allocate cost among consumers is considered. In this regard two IEEE test system such as 14-Bus and 118-Bus systems as case studies are conducted. The present paper is organized as follows: Section II formulates the environmental constrained active-reactive OPF and cost allocation approach; in section III, case study and result are presented, and section IV presents the concluding remarks. II. ENVIRONMENTAL CONSTRAINED ACTIVE-REACTIVE OPF AND COST ALLOCATION APPROACH
Pg imin Pg i Pg iM (Qg i ), Qg t
min ik min i
i = 1, ..., ng i = 1, ..., ng i , k = 1, ..., nl
(7) (8) (9) (10) (11)
Qg i Qg
max ik
max i
Em i EMA
i A ing
t ik t
Em i EMS
where Qg i is reactive power generation, Qd i , and Pd i are active and reactive demand, Qi and Pi are active and reactive injection, Vi is voltage magnitude, t lk is transfer tap, Pg iM (Qg i ) is the maximum active power limit based on the reactive power generation, Emi is the generation's emission calculate by (12), EMA and EMS are regional and system emission limits. The amount of emission is calculated by a quadratic function as (13).
Emi ( Pg i ) = i ( Pg i ) 2 + i Pg i + i
(12)
where, i, i, and i are the emission coefficients. The Pg iM (Qg i ) limit is imposed by the capability curve and turbine, where it has three different portions as it has shown in Figure 1.
In this section to consider the effects of congestion on consumers allocated cost and also on regional and system emission, at first a formulation for environmental AROPF under normal and congestion condition is presented and then a method for cost allocation via environmental AROPF is taken into consideration. A. Environmental Constrained Active-Reactive OPF The objective function of an OPF in normal operation condition is as follows:
Min f N = C i ( Pg i )
i =1
ng
(1)
Figure 1. Capability curve of synchronous generator
where, ng is the number of generator, and
Ci ( Pg i ) is the cost
of active power and approximated by a quadratic function such as (2).
C i ( Pg i ) = ai ( Pg i ) 2 + bi Pg i + ci
where ai, bi, and ci are cost coefficients and
(2)
Pgi is the active
power generation. System operating constraints are as follows:

Qg i Qd i Qi (V , , t ) = 0, i = 1, ..., ng Pg i Pd i Pi (V , , t ) = 0, i = 1, ..., ng flik (V , , t ) fl Vi
min max ik
Pg i (Qg Qg min ), Qg min Qg Qg i i i i i Qg Pg min i i Pg iM (Qg i ) = Sg i2 Qg i2 , Qg i Qg i Qg i (13) 2 max 2 2 2 Vi ( Ei ) (Qg i Xs i + Vi ) , Qg i Qg i Qg imax Xs i
(3) (4) (5) (6)
Eimax as the maximum excitation voltage is calculated by (14). Ei

max
, i, k = 1, ..., nl i = 1, ..., ng
Vi Vi
max
( Xsi Pg i) 2 + ( Xsi Pg i + Vi ) 2 2 Vi
(14)
When the congestion occurs, the optimal solution will change based on the power system rebalancing. In this paper in order to consider the transmission line outage, the following formulation is used.
Min f C = C i ( Pg iC )
i =1
ng
Pd i K Qd i Qd i = K Pd i =
(29) (30)
(15)
where Ci ( Pg iC ) is the cost of active power as (16).
The procedure of sequentially solving the objective functions of fN and fC for K times, where the amount of demands are defined by (31) and (32).
Pd i (k ) = k Pd i , Qd i (k ) = k Qd i , k = 1,..., K k = 1,..., K
C i ( Pg iC ) = ai ( Pg iC ) 2 + bi Pg iC + ci
C i
(16)
(31) (32)
where Pg is the active power generation at congestion event. System operating constraints are as follows:
Qg iC Qd i QiC (V C , C , t C ) = 0, Pg Pd i Pi (V , , t ) = 0,
C i C C C C
i = 1, ..., ng i = 1, ..., ng
(17) (18) (19) (20) (21) (22) (23) (24) (25)
As the destination is to find the reactive power pricing then the differences between marginal costs of two aforementioned steps is taken into account as it shown by Eq. (28) and Eq. (29).
Pd =
i
fl (V , , t ) fl
C ik C C C
max ik
,
C i
i, k = 1, ..., nl i = 1, ..., ng i, k = 1, ..., nl
Vi min Vi C Vi max , Pg t
min i C i M i
i = 1, ..., ng i = 1, ..., ng
Pg Pg (Qg ),
C ik max ik
Qd
1 K i ( Pd i (k )) Pdi ( Pd i (k ))] [Pd K k =1 1 K i (Qd i (k )) Qdi (Qd i (k ))] = [Qd K k =1
(33) (34)
Qg imin Qg iC Qg imax ,
min ik
t t
C i
where, the () () is the marginal cost of active/reactive at demand level k defined in Eq. (26) and Eq. (27). III. CASE STUDIES AND RESULTS Two case studies are considered in order to show the effects of congestion on consumers allocated cost, regional emission, and system emission. The first case is the IEEE 14-Bus test system with two regional emission limits and the second one is the IEEE 118-Bus test system with 2 regional emission limits. In both cases the power factor is held as 0.9. The presented approach finds the optimal solution using a modeling language for mathematical programming (AMPL) [23]. A. IEEE 14-Bus system This system contains 11 load busses, 5 generators with 20 transmission lines [24]. The environmental active-reactive OPF is applied to this system and the incurred cost by congestion is allocated between consumers. For this system, in normal condition (NC) and congestion condition (CC), three different cases such as: system without emission limits (NL), system with regional emission limit (EMA), and system with regional and system emission limits (EMA&EMS) are taken into consideration. For this case study, EMS is supposed to be 150 kg. The transmission line outage is taken place for the line between busses 1 and 2. The additional data of this system are given in Table I.
Bus 1 2 3 6 8 TABLE I. ADDITIONAL D ATA, IEEE 14-B US SYSTEM EMA i i S [p.u] Xs i Region Limit(Kg) 0.007 -0.52 25.8 3.324 0.825 1 115 0.007 -0.54 26.9 1.4 0.925 0.004 -0.49 30.1 1.0 1.098 0.004 -0.53 25.3 1.0 1.098 2 190 0.008 -0.40 23.9 1.0 1.098
Em iA
ing
EMA
EmiC EMS
B. Cost allocation Approach Sometimes congestion in power system occurs after a sudden increase in demand, sometimes because of a transmission line outage or generally it occurs when there is not sufficient transmission capacity to meet the transmission service constraints. Real-time congestion pricing can show the effect of congestion on consumers. The allocated congestion price is calculated by (26).
Pd i C Q = [ Pd i Qd i ] Qd i
(26)
In order to find the pricing factors, a well-known, robust and powerful game theoretic method of Aumann-Shapley is used. The pricing factors are defined using marginal costs, then at first the short-run active and reactive marginal cost of power consumers located at node i are computed by (27) and (28) [10].
Pd =
i
( f N ) , Pd i
Qd =
i
( f N ) Qd i
(27) (28)
( f C ) i = , Pd Pd i
( fC ) i = Qd Qd i
To compute the prices to calculate consumers costs, the active and reactive demands at each bus is divided to small quantities and usually divided by a large number, K.
TABLE II. GENERATION AND ALLOCATED COSTS UNDER NORMAL AND CONGESTION CONDITION FOR NL, EMA, AND EMA&EMS CASES, IEEE 14-BUS
SYSTEM
Bus 1 2 3 4 5 6 7 8 9 10 11 12 13 14 Total
NL 5511.917 1071.5526 1158.4625 0 0 0 0 339.2537 0 0 0 0 0 0 8081.1858
Cost of NC ($) EMA 3824.8406 1132.2144 1735.4751 0 0 415.1848 0 1105.4673 0 0 0 0 0 0 8213.1822
EMA&EMS 3513.0006 1094.7796 1744.3555 0 0 1134.8879 0 790.0664 0 0 0 0 0 0 8277.09
NL 4110.4806 1215.4209 2342.0499 0 0 0 0 817.1306 0 0 0 0 0 0 8485.082
Cost of CC ($) EMA 3824.274 1221.5089 2448.0479 0 0 0 0 997.1544 0 0 0 0 0 0 8490.9852
EMA&EMS 3451.2309 1203.5589 2311.4580 0 0 665.9238 0 888.9455 0 0 0 0 0 0 8521.1171
NL 0 47.2174 163.8474 66.6554 9.0726 13.7327 0 0 41.6857 12.5418 4.5872 7.5157 17.0482 20.2710 404.1751
Allocated Cost ($) EMA EMA&EMS 0 0 35.0936 31.3149 122.4735 106.6760 42.7693 37.8867 5.0922 4.4850 7.8737 7.5750 0 0 0 0 26.9688 22.6970 8.0005 6.8448 2.7872 2.5000 4.2902 4.0753 9.9269 9.2830 12.5458 11.0007 277.8217 244.3384
Table II contains the fluctuations of generation and allocated costs of 14-bus system under normal and congestion conditions for NL, EMA, and EMA&EMS cases. Results demonstrate that for both NC and CC, when emission limits are taken into account, the total generation cost is increased while on the other hand the allocated costs have decreased; exactly when the emission limits are taken into account the gap between the generation costs under NC and CC conditions for the limited cases of EMA and EMA&EMS are decreased, then this is why the allocated costs have faced with a decrease.
TABLE III. PRODUCED EMISSION UNDER NORMAL AND CONGESTION CONDITION FOR NL, EMA, AND EMA&EMS CASES, 14-BUS SYSTEM Emission of NC (Kg) Emission of CC (Kg) Bus NL EMA EMA&EMS NL EMA EMA&EMS 1 189.12 98.51 84.43 112.22 98.49 81.75 2 16.51 16.49 16.50 16.51 16.51 16.50 3 19.32 16.44 16.40 15.14 15.10 15.17 6 25.30 20.24 13.54 25.3 25.3 17.61 8 21.09 18.95 19.13 19.08 18.90 18.97 Total 271.34 170.63 150.00 188.25 174.30 150.00
Figure 2. Allocated cost for NL, EMA, and EMA&EMS ($), IEEE 14-Bus system
Figure 2 compares the incurred cost and the allocated cost of NL, EMA, and EMA&EMS. Results show that the allocated cost can cover the incurred cost while the allocation errors for NL, EMA, and EMA&EMS cases are 0.069%, 0.0067%, and 0.1276% respectively, which are the acceptable errors.
Table III shows the produced emission under different conditions for NL, EMA, and EMA&EMS states. Results show that as it was expected, based on the conflicting objectives of cost and emission and by considering the increasing order of total cost for NL, EMA, and EMA&EMS states, the amounts of emission for these states are sorted in decreasing order. Although considering NL and CC for EMA&EMS state shows that the CCs cost has increased (Table II), in both NC and CC the amounts of emission has kept constant and hit the upper limits. The bold face numbers in this table reveal that although in Table II the cost of EM state after congestion has faced with an increase about 278 $ (8213.18 $ for NC, 8490.98 $ for CC), the corresponding emission has faced with an increase about 4 Kg (170.63 Kg for NC, 174.30 Kg for CC). Using Table III and the aforementioned results, it reveal that based on the place where the outage has occurred it even has led to emission increase on busses 2, 6, and 8 which are highlighted part of the table. This issue is discussed in detail via the IEEE 118-bus test system.
Figure 3. Regional emission for NL, EMA, and EMA&EMS (Kg) , IEEE 14Bus system
B. IEEE 118-Bus system This system has 99 Load busses, 186 transmission lines, 54 generators, and 2 critical emission regions. The first region contains generators of busses 70, 73, 74, 76, 77, and 80 and
the second region contains generators of busses 89, 90, 91, and 92. In this case study we focus on the places where the congestion occurs; two different study in congestion is taken into account, where in each study three outages of transmission lines are considered. The regional and system emission limitations are 1250 kg and 7500 kg respectively. 1) Normal lines outage: in this case the outage of three normal lines are considered; these lines are called normal lines because they do not transmit high amount of active and reactive power. The outaged lines and the active and reactive transmitted powers of these lines under normal condition and for both of NL and EMA&EMS conditions are in Table IV. In this case, for normal condition (NC) and contingency condition (CC) two studies on NL and EMA&EMS conditions are considered. Costs and the amounts of emission in region 1 (R1), region 2 (R2), and the total system cost and emission system are in Table V.
TABLE IV TRANSMITTED POWERS OF CANDIDATE LINES FOR OUTAGE, IEEE 118-BUS SYSTEM Candidate Lines from 69 to 77 from 80 to 81 from 94 to 96 Case P Q P Q P Q (MW) (MVAr) (MW) (MVAr) (MW) (MVAr) NL 72.48 -11.91 7.67 -60.20 14.99 -2.88 EMA&EMS 73.78 -13.02 -4.26 -59.48 15.95 -3.13
with a decrease in cost, whereas the consumers with positive congestion allocated cost will face with an increase on their cost. This happens as in outage condition, some areas will face with more limitation to receive or transmit the power to other area, then the optimal solution will change and consequently the additional allocated cost among consumers of some areas are negative, and on the other hand some consumers will face with positive additional allocated cost. Although the Aumann-Shapley pricing method is an acceptable method to cover the incurred cost, Figure 4 shows that it is not fair for all consumers as it increases some consumers cost and decrease some other consumers cost.
Figure 4. Additional allocated cost for normal outage case under NL condition, IEEE 118-Bus System
From Table V, it is clear that in NL condition and for R1, R2, and total consideration, amounts of emissions have faced with an increase in CC comparing with NC, also the corresponding costs have faced with an increase. For EMA&EMS case, amounts of emission for CC is more than or equal to NCs emission. The difference part of Table V confirms that the outage of aforementioned lines yield to increase in cost and emission. Also the considerable impact of this outage is related with the cost of first region while for NL and EMA&EMS cases its effect on this region are respectively 1228.57 % and 978.82 % more than its effect on the total costs.
TABLE V RESULTS OF NORMAL LINES OUTAGE Emission (Kg) Case Condition R1 R2 Total R1 NC 1434.4 1360.4 8209.0 14107.8 NL CC 1525.4 1399.5 8255.8 15130.8 Difference 91.0 39.1 46.8 1023.0 NC 1250.0 1244.5 7500 14042.6 EMA& CC 1250.0 1250.0 7500 15122.5 EMS Difference 0.0 5.5 0 1079.9
2) Critical Line outage: in this case two lines are under normal operation (69-77 and 80-81) but the other one (8-9) is under critical operation. The candidate lines and the active and reactive transmitted powers of these lines under normal condition and for both of NL and EMA&EMS conditions are in Table VI. Output active power of line 8-9 shows its critical role in this system.
TABLE VI TRANSMITTED POWERS OF CANDIDATE LINES FOR OUTAGE, , IEEE 118-BUS SYSTEM Candidate Lines from 69 to 77 from 80 to 81 from 8 to 9 Case P Q P Q P Q (MW) (MVAr) (MW) (MVAr) (MW) (MVAr) NL 72.48 -11.91 7.67 -60.20 -394.58 -74.80 EMA&EMS 73.78 -13.02 -4.26 -59.48 -379.72 -79.01 TABLE VII RESULTS OF CRITICAL LINE OUTAGE, , IEEE 118-BUS SYSTEM Emission (Kg) Cost ($) Case Condition R1 R2 Total R1 R2 Total NC 1434.4 1360.4 8209.0 14107.8 14185.9 129660.7 NL CC 1464.0 1374.3 7759.9 14830.3 14277.7 133897.9 Difference 29.6 13.9 -449.1 722.5 91.8 4237.2 NC 1250.0 1244.5 7500.0 13965.9 13406.5 129756.6 EMA& CC 1250.0 1250.0 7479.9 13901.8 13444.2 133941.9 EMS Difference 0.0 5.5 -20.1 -64.1 37.7 4185.3
Cost ($) R2 14185.9 14443.5 257.6 13406.5 13441.1 34.6
Total 129660.7 129737.7 77 129756.6 129856.7 100.1
The allocated costs of normal outage case under NL and EMA&EMS conditions are respectively 78.03 $ and 101.2 $ corresponding to 77 $ and 100.1 $ of incurred costs. In this case the allocation errors of NL and EMA&EMS are %1.34 and %0.92 respectively. Figure 4 shows the congestions allocated cost between consumers under NL condition. Expounding this figure from consumers standpoint reveals that, congestion can have a positive or negative effect on their cost. In this figure, the consumers with negative congestion allocated cost will face
Table VII shows the regional and total costs and emissions for NL and EMA&EMS cases under normal and contingency conditions. Considering differences show that unlike Table V in this critical outage, the considerable impact is on total cost and emission where in both NL and EMA&EMS cases, total
costs have faced with an increase corresponding with a total decrease in emission. On the other hand, unlike the NL case which in both regional cost and regional emission has faced with an increment, in EMA&EMS case, the first regional cost has faced with a decrement corresponding with no change on environmental emission (because it has hit its limit). In this case the allocated cost for no limit condition is 4242.04 $ and for limited condition is 4188.02 corresponding with 4237.2 $ and 4185.3 $ of incurred cost respectively. The allocating errors are %1.14 and %0.64 respectively. IV. CONCLUSIONS In order to find the effects of congestion on regional and system emission and also on consumers allocated cost, two case studies has conducted. Results show that the effects of congestion have a close relation with the place, critical role, and the region of the transmission line outage such that in some cases increasing in cost yields to decrease in emission and in some cases increasing in cost yields to increasing in emission. Also results show that, although the Aumann-Shapley pricing method is a good, precise and robust method to cover the incurred costs, which is really important from sellers standpoint, this method is not good to allocate the incurred cost among consumers as it works based on lagrangian multipliers and the power flow in congestion condition will find another solution than normal condition for demands which yields to increase or decrease in some consumers cost. It is recommended to consider a method to allocate the incurred cost such that it satisfies the seller and consumers desired objective. REFERENCES
[1] [2] H. W. Dommel and W. F. Tinney, Optimal Power Flow Solutions, Power Apparatus and Systems, IEEE Transactions on, vol. PAS-87, no. 10, pp. 18661876, 1968. M. Pourakbari-Kasmaei and M. Rashidi-Nejad, An effortless hybrid method to solve economic load dispatch problem in power systems, Energy Conversion and Management, vol. 52, no. 89, pp. 28542860, 2011. F. G. M. Lima, F. D. Galiana, I. Kockar, and J. Munoz, Phase shifter placement in large-scale systems via mixed integer linear programming, Power Systems, IEEE Transactions on, vol. 18, no. 3, pp. 10291034, 2003. K. L. Lo and Z. J. Meng, Newton-like method for line outage simulation, Generation, Transmission and Distribution, IEE Proceedings-, vol. 151, no. 2, pp. 225231, Mar. 2004. D. Pudjianto, S. Ahmed, and G. Strbac, Allocation of VAr support using LP and NLP based optimal power flows, Generation, Transmission and Distribution, IEE Proceedings-, vol. 149, no. 4, pp. 377383, Jul. 2002. A. Berizzi, M. Delfanti, P. Marannino, M. S. Pasquadibisceglie, and A. Silvestri, Enhanced Security-Constrained OPF With FACTS Devices, Power Systems, IEEE Transactions on, vol. 20, no. 3, pp. 15971605, 2005. W. Yan, J. Yu, D. C. Yu, and K. Bhattarai, A new optimal reactive power flow model in rectangular form and its solution by predictor corrector primal dual interior point method, Power Systems, IEEE Transactions on, vol. 21, no. 1, pp. 6167, 2006. L. J. Cai, I. Erlich, and G. Stamtsis, Optimal choice and allocation of FACTS devices in deregulated electricity market using genetic algorithms, in Power Systems Conference and Exposition, 2004. IEEE PES, 2004, pp. 201 207 vol.1. H. Mori and Y. Goto, A parallel tabu search based method for determining optimal allocation of FACTS in power systems, in Power
[10]
[11]
[12]
[13]
[14]
[15] [16] [17]
[18]
[19]
[20] [21]
[3]
[4] [5]
[22]
[23] [24]
[6]
System Technology, 2000. Proceedings. PowerCon 2000. International Conference on, 2000, vol. 2, pp. 1077 1082 vol.2. W. Ongsakul and P. Jirapong, Optimal allocation of FACTS devices to enhance total transfer capability using evolutionary programming, in Circuits and Systems, 2005. ISCAS 2005. IEEE International Symposium on, 2005, pp. 4175 4178 Vol. 5. L. Shi, J. Hao, J. Zhou, and G. Xu, Ant colony optimization algorithm with random perturbation behavior to the problem of optimal unit commitment with probabilistic spinning reserve determination, Electric Power Systems Research, vol. 69, no. 23, pp. 295303, 2004. J. G. Vlachogiannis and K. Y. Lee, Determining generator contributions to transmission system using parallel vector evaluated particle swarm optimization, Power Systems, IEEE Transactions on, vol. 20, no. 4, pp. 17651774, 2005. N. P. Padhy, Congestion management under deregulated fuzzy environment, in Electric Utility Deregulation, Restructuring and Power Technologies, 2004. (DRPT 2004). Proceedings of the 2004 IEEE International Conference on, 2004, vol. 1, pp. 133 139 Vol.1. M. Rashidinejad, H. Khorasani, and A. Rashidinejad, Transmission Expansion Planning in Restructured Electricity Industry Using a Hybrid Heuristic Technique, International Review of Modelling and Simulations, vol. 3, no. 3, pp. 283289, Jun. 2010. M. C. Caramanis, R. E. Bohn, and F. C. Schweppe, Optimal Spot Pricing: Practice and Theory, Power Apparatus and Systems, IEEE Transactions on, vol. PAS-101, no. 9, pp. 32343245, 1982. M. L. Baughman and S. N. Siddiqi, Real-time pricing of reactive power: theory and case study results, Power Systems, IEEE Transactions on, vol. 6, no. 1, pp. 2329, Feb. 1991. X. Ma, A. A. El-Keib, and T. A. Haskew, Marginal cost-based pricing of wheeling transactions and independent power producers considering security constraints, Electric Power Systems Research, vol. 48, no. 2, pp. 7378, 1998. M. Pourakbari-Kasmaei and M. Rashidi-Nejad, An effortless hybrid method to solve economic load dispatch problem in power systems, Energy Conversion and Management, vol. 59, no. 89, pp. 28542860, 2011. M. Basu, P. K. Chattopadhyay, R. N. Chakrabarti, and T. K. Ghoshal, Economic Emission Load Dispatch with Non-Smooth Fuel Cost and Emission Level Functions Through an Interactive Fuzzy Satisfying Method and Evolutionary Programming Technique, Journal of the Institution of Engineers (india). Electrical Engineering Division, vol. 87, pp. 4146, 2006. J. W. Lamont and E. V Obessis, Emission dispatch models and algorithms for the 1990s, Power Systems, IEEE Transactions on, vol. 10, no. 2, pp. 941947, May 1995. M. P. Kasmaei, M. J. Rider, and J. R. S. Mantovani, A Novel Straightforward Compromising Method for Dynamic Economic and Emission Dispatch Considering Valve-Point Effect, in Environment and Electrical Engineering (EEEIC), 2013 12th International Conference on, 2013. M. Junqueira, L. C. da Costa, L. A. Barroso, G. C. Oliveira, L. M. Thome, and M. V Pereira, An Aumann-Shapley Approach to Allocate Transmission Service Cost Among Network Users in Electricity Markets, Power Systems, IEEE Transactions on, vol. 22, no. 4, pp. 15321546, 2007. R. Fourer, D. M. Gay, and B. W. Kernighan, AMPL: A Modeling Language for Mathematical Programming. Duxbury Press, 2002. Online, http://www.ee.washington.edu/research/pstca/pf14/pg_tca14bus.htm.
[7]
[8]
[9]
Comparative Study of Adding PV/Wind Energy Systems to Autonomus Micro Grid

Abstract the micro grid system changes the traditional distribution system with single power source and radial network into a new distribution system with several power sources. Depending on the interaction level between micro grid and main grid, a micro grid can be classified as isolated or grid connected. In isolated micro grid a connection with the main grid is not existing requiring the micro grid to be self-sustaining. In this type of micro grid it acts as a small power system with its all component (generation, transmission, and distribution). The main driver for isolated micro grid is the availability of local energy resources. This paper introduces a comparative study of adding Photo Voltaic (PV) or wind energy systems to the micro grid in order to enhance the isolated operation of the micro grid. The study conduct a comparison based on adding the same capacity of PV or wind energy to the autonomous micro grid. The amount of energy not supplied to the load and for how long some loads are shredded are the two main factors through the comparison. A typical case study of the micro grid system located at CairoEgypt will be analyzed in details. A simulation model that calculates the electrical power output based on actual data at the site with respect to changes in the environmental parameters for both PV and wind energy systems will be introduced. The model was implemented using Matlab m-file to accommodate any location in the world. The integration of the micro grid system with renewable energy resources will be an outcome of a development of the existing electricity networks towards a sustainable distribution network system. At the end the results is presented showing weather PV or Wind energy affecting the isolated operation of the micro grid. Keywords-micro grid; Photovoltaic ; Wind energy; CHP power; Distribution system Matlab
I.
INTRODUCTION
The concept of micro grid is in the research domain for many years. Micro sources of electric energy in micro grid can include wind energy, photo voltaic, fuel cells, combined heat power system, and micro turbines. In this context new local infrastructure become a challenge these infrastructures are called smart grids, which are intelligent bi-directional electricity networks. These distribution networks may use information technology (IT) to manage energy production and schedule the demands according to each consumer. This paper focuses on comparing benefits associated with the PV power or wind energy system connected to the micro grid in the isolated mode. Micro grids have the potential to provide many benefits and published literature provides complementary set of details.
The reference [1] investigated the impact of micro grid on the reliability of the distribution system. This reference evaluates the reliability of the distribution system with micro grid considering it as a load point to external. The work in [1] concludes that micro grid could enhance the reliability of the overall distribution network to large extent. A strategic micro grid for the improved reliability is addressed in [2] which investigate that the constraints to integrate large volumes of DG to power system (PS) can be limited by strategically integrated micro grids in future power system. The work in [2] was proposed an algorithm that can improve the reliability of future power system by increasing the integration of the DG. The algorithms that have been presented in [2] calculate the distribution system reliability indices at four different scenarios to prove that the strategic micro grid can enhance the reliability of the system. Sustainable micro grids for isolated systems have discussed in [3] a central point of this work is to make clear that sustainable energy integrates the economic ,social ,and environmental dimensions. The work presented in [3] finds that the multi objective algorithm (MOO) is the most suitable optimization approach for the micro grids accountable to make their preferences. Reference [4] shows that the network structure characteristic is the basic issue in reliability assessment. Two patterns to construct micro grid based on distribution network in china have been presented in [4] either for medium voltage or low voltage levels. The work concludes that the proper micro grid structure helps improving the reliability level of micro grid throughout the outage rate of DG units. The concepts of micro grid, the integration of renewable energy resources, and the DG technologies have been presented and discussed in [5]. The work discuss that the concept of micro grid with the integration of DG (wind energy, photo voltaic, combined heat and power, and micro turbines) together with the new DG technologies can meet the Energy, Economy , and Ecology ( 3E requirements )[5]. A methodology for economic design and optimal operation of micro grids with renewable energy resources was presented in [6]. A mathematical model for economical operation of specific micro grid was presented and applied. The model shows that the micro grid can foster the services of smaller loads with cleaner more efficient more reliable technologies [6].a methodology to find optimal capacity and placement of Distributed energy resources (DER) in respect to security six bus meshed distribution micro grid is addressed in [7]. The application of the approach founds in [7] shows that the systematic and rational placement of DER improves system reliability and security by improving the voltage profile of buses and the results indicates that the
inclusion of the DER in the micro grid does not always guarantee the minimum of line loss. The presented the three E Energy, ecology, economy development requirements have opened the opportunities for onsite power generation application in the renewable energy conversion technology [8]. It is clearly seen that DER can enhance the reliability of the micro grid from different views. The effect of adding DER to micro grid on its reliability will vary depending on the type of added DER and its capacity. The most popular and more sustainable renewable resources are wind energy, and photovoltaic [5]. The main target of this paper is to conduct a comparison between wind energy and photo voltaic in terms of their reliability effect on the micro grid system. This work will study the reliability issues under two main configuration of the micro grid. The proposed approach will be applied to micro grid located as a part of the distribution system at Hurgada Egypt. Throughout this study an answer to the main question of which affects the reliability of the isolated operation of the micro grid system more wind energy or photo voltaic? will be found and analyzed. II.
PROPOSED ALGORITHIM
are; the short-circuit current (ISC), the open circuit voltage (VOC), the voltage at point of maximum power (Vmpp), the current at point of maximum power (Impp), the number of cells in series in a module (n), the reference cell temperature (TR) and the temperature coefficient for short-circuit current (Ki) in A/K.
Figure1. Schematic diagram of the micro grid system
The main target of this work is to decide which affects the operation of the micro grid more the wind energy or photo Voltaic (PV). A micro grid with total capacity of 50 MW located at Hurgada. A Hurgada has recorded higher solar insolation as well as higher wind speed among Egyptian cities [9]. The adequacy assessment of the micro grid system used for this study is carried out for two different cases of operation. Case 1 supplies its load from 25MW from conventional CHP generators and 25MW from wind turbine generators. Case2 is held out with half of its generating capacity from CHP units and the other half from PV panels. The micro grid is composed of 36 units of Combined Heat and Power (CHP) generators of 0.7 MW each in addition to either 28 units Wind Turbine Generator (WTG) of 0.9 MW each or a photo voltaic array to produce 25MW. The total customers load of the micro grid is recorded load composed of residential and commercial customers. The power from conventional generators is 25 MW it presents 50% of the total generated power and the other half of generated power are either from the PV or from the wind generators. Figure1. Shows a schematic diagram of the micro grid system. The given micro grid is simulated using MATLAB. The detailed simulation steps is given in the rest of this section. A.Photovoltaic Model A single diode PV model is used that is completely based on actual values [10]. The actual PV parameters from data sheet are used for the calculations of the module output power. Where no assumptions are made or left for try and error estimation, The main parameters taken from data sheet
The model is shown in Figure 2, the current output of a PV module (IPV) is calculated using the light generated photocurrent (IPh), the diode current (ID) and the leakage current (ISh) passing through the cell shunt resistance (Rp) [11]. Ipv = IPh -ID - ISh IPh =I SCG(1+ki(TC-TR)) TC=Ta+0.2G (1) (2) (3)
Where the ambient temperature (Ta) in Kelvin and solar radiation (G) in kW/m2, are input values according to the location of implementation for the PV module. Using Shockley's Diode Equation, The diode current (ID) in ampere [11]:
(4)
(5) (6)
(7)
Where, K is Boltzmann's constant (1.380658*10-23 J/K), q is the electron charge (1.6*10-19 C), Eg is the band gap energy for semiconductor (1.12eV for the silicon), IOR is the reverse saturation current of diode in ampere, Vt is the thermal voltage in volt and Vpv is output voltage from the PV module. Newton Raphson's method is used to solve non-linear equation, in order to obtain numerical values for Ipv and Vpv to calculate the output power of the PV module (PPV) in W. P_PV= I_PV V_PV (8)
2. The model is based on assuming that the ever changing WTG power obtained from Eq. (11) is divided into four-states. 3. The Mean Time to Fail (MTTF) and Mean Time To Repair (MTTR) is calculated for each state. 4. A Mont Carlo simulation technique is used for the convergence of MTTF's and MTTR estimated. 5. The state duration approach [8] is employed to calculate the simulated WTG power using the estimated MTTF's and MTTR Step1: Modelling PV system A 25MW PV system is modeled and simulated on Matlab using datasheet parameters of Pmax of 245 W, ISC =37.5A VOC of 8.73v, Impp of 8.04A, Vmpp of 30.5V, Ki at 4.6269 mA/K, and n of 60 cells for each panel. A typical hourly recorded temperature data at hurgada is used to model the PV output power. The solar radiation is calculated according to the algorithm described previously in this section based on the geographic location. The calculated solar radiation together with the given data sheet parameters and the hourly recorded temperature at the site will be the input of the PV modeling algorithm [10]. The model is used to produce 25MW output power. Step2: Modeling WTG A WTG is modeled and simulated based on the described model at part B in this section. The recorded wind speed data at Hurgada is used together with rated power of each WTG to calculate the output power of each WTG. The cut in, rated, and cutout wind speeds at Hurgada. A 25 MW installed capacity of wind energy is integrated to the micro grid through different connections. The installed capacity from wind energy is composed of 28 WTG each of 0.9MW. The hourly recorded wind speed at the location is used with the data of the wind generator to simulate the hourly wind power. Step3: Modelling CHP generating units The micro grid under the study has 25MW total capacity of CHP generator. This generating capacity comes from 36 unit of 0.7MW each. All of which are simulated using state duration approach. This approach is the most common approach that used for the simulation of different types of conventional generating units for adequacy studies. The state duration approach is based on that each conventional generating unit is presented as a two state model. The given data for each CHP unit are the repair rate () and failure rate ( ) and rated power of each unit. The parameters Mean Time to Fail MTTF and Mean Time to Repair (MTTR) are calculated directly from faliur and repair rate of each unit, (12) (13) Step4: Load Model
Figure 2. One-diode PV Circuit
The PPVA (photovoltaic array power) can be calculated using the number of PV modules connected in series (NS) and parallel (NP) by [12]: (9) Solar Radiation Calculation The solar radiation (G) in kW/m2 can be calculated at the instant of any hour during the year based on the geographic location, day and the time [13]. (10) Where, Gn is the beam radiation measured in the direction of the rays on the (n day) of the year in W/m2 and n is the day number of the year (varies between 1 and 365). () is the latitude of the location in degrees, () is the declination of the sun in degrees and () is the solar hour angle in degrees B.Wind Turbine Generator model The wind turbine generator model will be simulated according to the model suggested in [14]. The algorithm procedure of this model will result in a WTG model for reliability analysis. The model simulates the WTG out power as a four states. The WTG will be presented at this work by the model described in [14]. 1. The hourly output power from WTG (Pc) is calculated from simulated hourly wind speed using Eq. (11). The constants A, B, and C are presented in details [14].
(11)
An hourly recorded load at Hurgada for the micro grid is used during the simulation steps. The loads connected to the micro grid are composed of residential and commercial consumers. Step5: evaluation steps for the micro grid system. The total generated power of the micro grid in both cases can be simulated by combining the total hourly CHP generated power with the total hourly renewable energy resource power. Either from the wind energy or from the PV resources. Pg(i)=Pc(i)+Pr(i) (14)
III.
RESULTS
A. Generation Modeling 1. Generated power from PV A 25MW installed capacity of PV power is simulated using the algorithm described in previous section. Figure 3 shows the P-V characteristic of the proposed PV model. The simulation model results in the hourly power from PV arrays according to the temperature and solar insulation at Hurgada. An inverter with 77% efficiency is used to convert the power from PV to AC power in order to supply AC loads.
Where: Pg: hourly total generated power in the ith hour Pc: hourly generated power from conventional units in the ith hour Pr: hourly power from renewable energy resource in the ith hour Renewable energy resource could be PV or wind. The hourly total generated power is produced in both cases and compared with the load in order to held adequacy analysis. The adequacy studies is tested for two different energy resources to determine which is more adequate to the micro grid system. Where the total micro grid generation is examined to determine its adequacy to meet the total system load requirement. The adequacy assessment for this micro grid is done with three different indices the LOEE (Loss of Load Expectation) or in some other words the Expected Energy not Supplied (EENS) in MWh/yr, LOLE (loss of load expectation) in hr/yr, and LOLF (loss of load frequency) in occurrence/yr. the calculation of this indices is done by the use of eq.15, 16, and 17.Monte Carlo simulation procedure is used for the convergence of the results. The simulation time will be as longer as the convergence of the results occurs.
Figure3. Photovoltaic Power-Voltage Curve
2. Generated power from WTG A 25MW installed capacity of wind power will be simulated using 28 WTG unit of 0.9 MW each rated power. The WTG units will locate at Hurgada Egypt. The cut in, rated cutout wind speed are 7, 9.3, and 10.2 m/h respectively. 3.Generated power from CHP generators A 25MW installed capacity of CHP generators are simulated using state duration approach for reliability studies. The system is composed of 36 generators of 0.7 MW each. The generators are connected to the micro grid in different places. Each CHP unit has scheduled outage of 120 hrs/yr. Figure4. Shows the simulated power of total CHP power over one year
(15) (16) (17) Where: j : number of load failure, N :is the total number of failures per year, tj: is the duration of the j-th failure, Ej: is the unsupplied energy at the j-th failure and Fj: is the number of failures The above indices is calculated for the micro grid using different cases of generation. A comparison between the results will be held from different views and discussed to determine which affect the reliability of the micro grid more the wind energy or the PV energy.
Figure4.Total CHP units Power over 1 year
B. System analysis As described in the previous section that the micro grid analysis will be checked through three main indices LOEE or EENS, LOLE, and LOLF. These indices will be calculated for both cases of system generation resources. Case1. 25MW CHP generating power in addition to 25MW power comes from wind energy system. The generation scheme of case1 is checked with total demand in order to study the adequacy factors for each year. The simulation program is run typically as case1 of generation over 250 iteration. Each iteration forms the hourly presentation over one year. A MonteCarlo Simulation procedure is applied for the results conversion. Figure5. Shows the LOLE index for the generation condition of case.1. The EENS, LOLE and LOLF are calculated during the simulation steps and converged with Monte-Carlo algorithm at 4.02MW/hr/yr, 382.5 hrs/yr, 350.4 Occurrence/yr. Case2. In this case the micro grid with half of its generation (25MW) from conventional generating units in addition to 25MW comes from the PV panels. The study of case 2 is carried out with the same simulation program parameters of case 1as the simulation time, sampling time, and convergence criterion. A Monte-Carlo Simulation procedure is applied for the results conversion.
from both resources each at a time. The study are done based on using the same simulation time, evaluating same indices using the same convergence criterion. Figure 6 illustrate the LOEE in both cases.
Figure6. Convergence of EENS for both cases Figure 6 illustrates the energy not supplied to the system load in both cases of study presented in the LOEE or EENS in the Y-axis. The figure shows that the convergence of the EENS is done at 4.02MW with the wind energy case of integration, while it converges at 5.48MW with the use of PV power. This concludes that the integration of wind energy with the system will improve the unsupplied energy to the load by 2.92% than the integration of PV panels. Comparing the LOLE index for the two cases of generation the LOLE converges at 389.99 hrs/yr when the micro grid uses the wind energy and converges at 384.2 hour per year when the system is integrated PV. These results shows that the integration of PV energy with the system will improve the interruption time presented in LOLE of the system than the integration of wind energy do by 1.57% with the same presentation. Finally, the frequency of interruptions presented in Loss of Load Frequency (LOLF) has been calculated for both micro grid generation cases. This LOLF shows how many times any one of the micro grid customers lose totally or partially its demand. LOLF converges at 350 occurrence when the micro grid is using wind energy as 50% of its generation power. And converges at 478 occurrence for case 2 of generation as half the generation capacity comes from PV panels.
V. CONCLUSION
Figure5. Convergence of LOLE for case1 The EENS, LOLE and LOLF are calculated during the simulation and converge at 5.48MW, 246.5hrs/yr, and 478 Occurrence/yr respectively for case2 of generation. IV: RESULTS DISCUSSION Comparing the results obtained from both cases will allow to discriminate between the two renewable energy resources from the adequacy study point of view. The comparison is conducted between the wind energy and PV energy. The comparison is based on integrating same generating capacity
A micro grid system is modeled using Matlab software package. The PV is modeled based on actual values of PV cell and all environmental changes of temperature and solar radiation is considered. Solar radiation equations were presented to accommodate any location in the world. The PV model was extended to simulate a 25MW connected to the micro grid. The wind energy is modeled using different units of WTG based on the recorded hourly wind data at Hurgada. The cutout, cut in and rated wind speed is taken form the site. The WTG model is suitable or any type of WTG. The WTG model
is utilized to simulate a 25MW presenting half generating capacity of the micro grid under the study. Judging whether the integration of the micro grid with PV or with wind energy resources will be more adequate for the micro grid. Comparing the results obtained from both cases will allow to discriminate between the two renewable energy resources from the adequacy study point of view. The comparison is conducted between the wind energy and PV energy. The comparison is based on integrating same generating capacity from both resources each at a time. The study are done based on using the same simulation time, evaluating same indices using the same convergence criterion. The results show that the integration of wind power with the system will help the micro grid to improve its energy not supplied and the number of interruptions by 2.92%, 1.572% respectively each year. On the other hand the integration of Photo Voltaic power will only improve the interruption duration of the system by1.46%than the integration of wind energy. The integration of the micro grid system with renewable energy resources will be an outcome of a development of the existing electricity networks towards a sustainable energy system. REFERENCES
Chengming Wu , Jie Liu and Yabin Wu," the Impacts of micro-grid on the reliability of distribution system," 2nd international conference on consumer electronics, communications and networks (CECNet), Apr. 2012. [2] H. Zavaleta-Vidal , D. Jayaweera," Strategic micro-grids for the improved reliability of future power systems," conference on innovative smart grid technologies Asia (ISGT), IEEE, PES, PP.1-6 , Nov. 2011. [3] R. Martnez-Cid and E. ONeill-Carrillo," Sustainable micro-grids for isolated systems," transmission and distribution conference and exposition ,IEEE PES , PP.1-7, Apr.2010. [4] Y. Luo, L. Wang, G. Zhu and G. Wang," Network Analysis and Algorithm of Micro-grid Reliability Assessment," Power and Energy Engineering Conference (APPEEC), Asia-Pacific ,PP.1-4, Mar. 2010. [5] X. Liu and B. Su," Micro-grids - an Integration of renewable energy technologies ," China international conference on electricity distribution, CICED 2008, PP.1-7, Dec. 2008. [6] H. Asanol and S. Bandol," Economic analysis of micro-grids," Power conversion conference - Nagoya, PCC '07, pp. 654 - 658 , Apr. 2007. [7] A.K. Basu, A. Bhattacarya and S.P. Chowdhury ," Relilibility study of a micro-grid system with optimal sizing and placement of DER ," CIRED Seminar on Smart-grids for distribution, IET-CIRED, pp.1-4, Jun. 2008. [8] B. N. Alajmi, K. H. Ahmed, S. J. Finney, and B. W. Williams," Fuzzylogic-control approach of a modified hill-climbing method for maximum power point in micro-grid standalone photovoltaic system," IEEE transactions on power electronics ,Vol. 26 ,No.4 ,PP. 1022 - 1030 ,2011 [9] Annual report New and renewable Energy association of Egypt,2010. [10] Ibtihal Zahran,Eman Beshr,Magdy Eiteba Reliability Peformance of a Grid-Connected Photovoltaic System in Egypt,IEEE conference, APPEC,March 2012. [11] I. Houssamo, M. Sechilariu, F. Locment, G. Friedrich, "Identification of Photovoltaic Array Model Paramters. Modelling and Experimental Verification," International Conference on Renwable Energies and Power Quality, 23rd to 25th March, 2010, Granada, Spain.J. Clerk Maxwell, A [1]
Treatise on Electricity and Magnetism, 3rd ed., vol. 2. Oxford: Clarendon, 1892, pp.6873. [12] A. ElShahat, "PV Cell Module Modelling & ANN Simulation for Smart Grid Applications," Journal of Theoretical and Applied Information Technology, 2005-2010. [13] K. Leban, E. Ritchie, Selecting the Accurate Solar Panel Simulation Model, Nordic Workshop on Power and Industrial Electronics, June 911, 2008. [14] Eman BeshrY.Galal,,Y.Hegazy,M.AL.BadrNovel approach for modeling wind turbine generators for reliability analysis IEEEconference,PECON 2012
Investigating Converter Options for Automotive Grade Permanent Magnet Sychronous Generators
Erkan MESE*, Yusuf YASA*, Hakan AKCA*, Mustafa G. AYDENIZ*, Murat AYAZ**, Murat TEZCAN* (*)Yildiz Technical University, Electrical Engineering Department (**)Kocaeli University, Technical Education Faculty
emese@yildiz.edu.tr, yasa@yildiz.edu.tr, hakca@yildiz.edu.tr, aydeniz@yildiz.edu.tr, murat.ayaz@kocaeli.edu.tr, mutezcan@yildiz.edu.tr
Abstract In this paper, converter options for an automotive class permanent magnet synchronous generator (PMSG) are investigated. There are two options to rectify and regulate voltage from AC to DC. These are uncontrolled rectifier/dc-dc converter and controlled rectifier. Each converter topologies are explained and their comparison is performed. The comparison considers generator apparent power, efficiency, power flow ad cost. As far as converter effects on the generator apparent power rating and system efficiency are concerned, uncontrolled rectifier/dc-dc converter seems to be superior. On the other hand, controlled rectifier option has cost advantage over uncontrolled rectifier/dc-dc converter topology. A tradeoff exists between converter cost and PMSG utilization while making decision between two topologies. Keywords- Permanent magnet synchronous generator, hybrid electric vehicle, uncontrolled rectifier, dc-dc converter, controled rectifier.
with PMSG. Automotive is not the only area where PMSG is considered. Other prospective areas have been seen in the horizon such as aviation, defense, marine, renewable energy and etc. This paper discusses PMSG operation for automotive applications where generator operation occurs over much wider speed range because of the speed range of the vehicles crankshaft. In this study generator speed is bounded between 1500 rpm and 6000 rpm. This suggests that AC to DC conversion should take place in the face of very wide range of AC input voltage. II. GENERATOR SIDE CONVERTER OPTIONS
I.
INTRODUCTION
Two AC to DC conversion options are investigated as shown in Fig 1. In the first option, an uncontrolled rectifier and a buck type DC to DC converter is studied. In the second option, a controlled rectifier is studied. For both options, simulation and experimental results will be given in the following sections. Impacts on PMSG as well as cost breakdown of two converters will be discussed.
Electrical accessories in a conventional vehicle are supplied by a Lundell alternator and battery combination. In hybrid electric vehicles (HEV), DC/DC converter is used for the same purpose [1]. Another cost effective solution for HEVs so called electric accessory drive system (EADS) has single electric motor and this motor drives all accessory loads through a belt [2]. A separate Lundell alternator is driven by EADS motor. Furthermore, another proposed solution is permanent magnet dual-winding electric machine as EADS prime mover which consists of motor and generator parts. Windings of the machine are concentrated. Motor and generator windings are electrically and magnetically isolated [3]. When the traction motors of the vehicle shut down, the motor part of the dual-winding electric machine runs and drives mechanical accessory loads. Generator part in the machine runs and provides electric power to electrical accessories. With this technique, both motoring and generating operations can be implemented in a single housing of electric machine. The system offers packaging advantages for HEVs. Above discussion indicates that, both conventional vehicles and some HEVs are using an alternator for low-voltage electric power generation. Lundell alternators in these applications are suffering from problems such as power and packaging limitations [4]. One solution is to replace Lundell alternator
Figure 1. Voltage regulation options of PMSG.
A. Uncontrolled Rectifier/ DC-DC Converter Output of the generator is connected to the uncontrolled rectifier as shown in Fig 2. Average DC output of an uncontrolled rectifier is given as in (1).
Figure 3. (a) Generator phase voltage wave forms Figure 2. Schematic diagram of uncontrolled rectifier and dcdc converter.
Vd = 1 3 3 /6 /6 Vmax cos(t )dt = Vmax sin(t ) /6 = Vmax / 3
/6
(1)
Where, Vl-l is line-to-line voltage of the generator. Vd is the mean output voltage value of the uncontrolled rectifier. Three phase uncontrolled rectifier is connected to a bucktype dc-dc converter. Buck converter equations are; Ts = 1 / f s (2) Ts = Ton + Toff (3)
Ton = Ts
Figure 3. (b) Generator current wave forms B. Controlled Rectifier PMSG is connected to a controlled rectifier. The dc output voltage is controlled with adjusting SCRs trigger angle. The schematic can be seen in Fig 4. Mean value of the output voltage of the controlled rectifier is given in (8). Vlmax is peak line-to-line PMSG voltage. is trigger angle of SCRs.
Vd = Vd = 1 3 /6+ /6 Vl max cos(t )dt = Vl max sin(t ) /6+ / 3 3 2
/6
Toff = (1 )Ts
(4) (5)
Where fs and Ts are the switching frequency and period of the converter, respectively. On and off durations of switches are represented with Ton and Toff. is duty cycle. Following expressions show the output voltage (Vout) and current (iout) of a buck converter. Vout = Vd (6)
I out = 1
(8)
Id
(7)
Vl cos =
3 6
V ph cos 2.34V ph cos
Simulation and Experimental Results Coupled simulation has been performed for uncontrolled rectifier/DC-DC converter options. Coupled simulation is a new simulation technique which uses finite element (FE) analysis that provides more reliable results compare to others. Uncontrolled rectifier and dc-dc converter models are built in Ansoft-Simplorer and then coupled to PMSG which runs as Ansoft-Maxwell FE model. In simulation, PMSG is rotated at 2500 rpm. The simulation and experimental results of generator voltages and currents are given in Fig 3. Figure 4. The circuit scheme of controlled rectifier. Simulation and Experimental Results Controlled rectifier simulation is performed with coupled simulation technique. Simulation results for generator voltage and current can be seen in Fig 5. Experimental results are given in Fig 6.
Generator Voltages 25.00
ANSOFT
40 30
Vas, Vbs, Vcs [V]
12.50
Voltage (V)
Curve Info
20 10 0 -10 -20
0.00
-12.50
-25.00 60.00
InducedVoltage(PhaseGA) InducedVoltage(PhaseGB) InducedVoltage(PhaseGC)
-30 -40 2
62.50
65.00
67.50 Time [ms]
70.00
72.50
75.00
6 7 Time (s)
10 x 10
-3
(a)
Generator Currents 250.00
100
ANSOFT
Figure 6 (a) Generator voltages

150
Ias, Ibs, Ics [A]
125.00
50 Current (A)
Curve Info
0.00
-50
-125.00
Current(PhaseGA) Current(PhaseGB) Current(PhaseGC) 62.00 64.00 66.00 Time [ms] 68.00 70.00
-100
-250.00 60.00
-150 0.106
0.107
0.108
(b)
Vdc 16.00 14.00 12.00 10.00 Vdc [V] 8.00 6.00 4.00 2.00 0.00 -2.00 0.00 20.00 40.00 60.00 Time [ms] 80.00 100.00
ANSOFT
0.109 0.11 Time (s)
0.111
0.112
0.113
Figure 6 (b) Generator currents III. COMPARATIVE STUDY
So far, some simulation and experimental results related to the controlled rectifier and dc-dc converter are given. In this section, comparative analysis of two converter options is performed. These two options are compared with each other cost and electrical performance. (c)
Idc
ANSOFT
250.00 200.00 Idc [A] 150.00 100.00 50.00 0.00 0.00 20.00 40.00 60.00 Time [ms] 80.00 100.00
A. Cost Analysis In cost study, it is assumed that the circuits will be produced in quantities of 1000. Cost study results are given in Table-I and Table-II for dc-dc converter and controlled rectifier, respectively. The power capacity of each converter is assumed to be 3 kW which corresponds to generator output power capacity.
TABLE I. DC-DC CONVERTER COST ANALYSIS
DC-DC Converter No Item Number Product Name Qty Price ($) Item Subtotal ($)
(d) Figure 5. (a) Generator voltages, (b) generator currents, (c) dc output voltage and (d) dc output current
1 2 3 4 5 6 7
IXFN140N30P VUO110-16N07 TLP250 DCW08B-15 2200uF/250V 1.5uf/1000V 100 V 200 A
Power MOSFET Three Phase Rectifier Bridge TOSHIBA Photocoupler DC/DC Converters 8W Input and output capacitors Snubber capacitors Output inductor
3 1 1 1 2 1 1
23,8 51,9 0,87 13 21 6 35

Total ($)
71,4 51,9 0,87 13 42 6 35

220,17
DC-DC converter production cost is around 220 $/per converter and controlled rectifier is around 163 $/per converter.
TABLE II. CONTROLLED RECTIFIER COST ANALYSIS
Controlled Rectifier No Item Number Product Name Qty Price ($) Item Subtotal ($)
1 2 3 4 5
MCC72-12IO1B TLP250 DCW08B-15 DCW03B-15 2200uF/250V
Thyristor/Diode Modules TOSHIBA Photocoupler DC/DC Converters 8W DC/DC Converters 2W Output capacitors
3 6 1 3 1
36,28 0,87 13 5 21
Total ($)
108,8 5,22 13 15 21
163,02
Figure 7. Generator efficiency comparison for two converters
Input capacitance and output inductance of dc-dc converter are the major causes of this difference. Controlled rectifier does not require input inductance because it uses generator inductance. So it gives cost advantage to the controlled rectifier. B. Effects on Generator Performance In this section, effects of the converters on generator performance are analyzed. In analysis; the effects of using each option on generator, converter and system efficiency are done. In addition to that, torque ripple and voltage/current harmonic analysis are also performed. The effects of each converter on generator, converter and system efficiency for different power levels can be seen in Table III and Table IV. Fig 6 shows that uncontrolled rectifier/dc-dc converter topology and controlled rectifier topologies have around same converter efficiency.
Figure 8. System efficiency comparison for two converters The variation of generator efficiency by using different topologies is given in Fig 7. Fig 8 shows the system efficiency comparison of these two topologies. In low power levels, generator efficiency levels are comparable for both converters. However with increasing power, efficiency of generator using dc/dc converter topology becomes better than the one with controlled rectifier. The reason is that, effective value of the current in controlled rectifier is much more than dc/dc converter topology because of high reactive current. Finally, the system efficiency, which combines generator and converter efficiency, can be mentioned: PMSG with uncontrolled rectifier/dc-dc converter has much higher efficiency than PMSG with controlled rectifier has. Harmonic analysis of generator phase voltage and current is performed with Fast Fourier transform (FFT) technique to see the effects of each converter on generator. Fig 9 and Fig 10 show waveforms for dc-dc converter option and controlled rectifier option, respectively.
Figure 6. Experimental efficiency comparison between uncontrolled rectifier/dc-dc converter and controlled rectifier
Figure 9. Experimental generator phase current (left) and voltage (right) when uncontrolled rectifier/dc-dc converter is used.
harmonics seem to be contributing to both active and reactive power flow. Whereas in the controlled rectifier case, harmonics are mainly contributing to reactive power flow. Current harmonic data shows that RMS current in the controlled rectifier case is higher than its counterpart. These observations imply that PMSG with controlled rectifier has to supply higher apparent power than PMSG with uncontrolled rectifier and DC-DC converter option. IV. CONCLUSION
Figure 10. Experimental generator phase current (left) and voltage (right) when controlled rectifier is used.
TABLE III. EXPERIMENTAL FFT RESULTS OF UNCONTROLLED RECTIFIER/DC-DC CONVERTER
5th 11.52% 15.38% 66 0 7th 7.25% 13.49% 59.6 0 11th 0.99% 4.1% 29.1 0 13th 1.36% 5.35% 47.8 0 cos 0.92
In this paper, converter options of a PMSG have been investigated. There are two options to convert and regulate voltage from ac to dc which are uncontrolled rectifier/dc-dc converter and controlled rectifier. Each converter topologies are explained and their comparison is performed. Cost and performance are two metrics for comparison. As far as the cost is concerned, controlled rectifier has advantage on uncontrolled rectifier/dc-dc converter topology. However, for electrical performance such as efficiency and apparent power rating, uncontrolled rectifier/dc-dc converter is superior. ACKNOWLEDGMENT The work has been supported under the grant number 110E111 by the Scientific and Technological Research Council of TURKEY (TUBITAK)
Harmonic Magnitude Current Voltage Phase (o) Current Voltage
TABLE IV.
EXPERIMENTAL FFT RESULTS OF CONTROLLED RECTIFIER

5th 4.17% 39.5% 152.6 50.8 7th 3.39% 45.5% 135.6 31.6 11th 1.57% 30.55% 136.2 28.4 13th 1.57% 37.84% 114.2 5.8 cos 0.73
Harmonic Magnitude Current Voltage Phase(o) Current Voltage
REFERENCES
[1] [2] [3] N.R. Trevett, X-by-Wire_New Technologies for 42V Bus Automobile of Future, Msc. Thesis, The South Carolina Honors College, 2002. R.K. Serrels, Accesory Drive System, US Patent Application, Patent Number 20080020875, 2008. E. Mese, Y. Yasa, H. Akca, M.G. Aydeniz, M. Garip, A New Electric Accessory Drive System for Hybrid Electric Vehicles, ECCE 2012 IEEE Energy Conversion Congress&Exposition, pp. 1909-1916, September 15-20, 2012, Raleigh, NC, USA. D.J. Perreault and V. Caliskan, Automotive Power Generation and Control IEEE Transactions on Power Electronics, vol. 19, no. 3, pp. 618630, May 2004.
FFT results demonstrate that both converters have common harmonics on voltages and currents which are 5th, 7th, 11th and 13th harmonics. This suggests that extra power flow would occur due to harmonics. Whether this is active or reactive power depends on the phase shift between associated voltage and current harmonic components. From the phase shift data, in the uncontrolled rectifier/ DC-DC converter case
[4]
New Hybrid System of Fuel Cell Power Plant and Wind Turbine for Household Consumption
Mohammad Ebrahim Talebian Assistant professor Managing Director (MD) of Neka Power Plant, Mazandaran, Iran metalebian@gmail.com www.neka power plant.ir www.drtalebian.ir Sobhan Sobhani M.Sc student Department of electrical Engineering Science and Research Branch Islamic Azad University, Mazandaran, Iran sobhan.sobhani2011@yahoo.com Arjang Borzooi M.Sc student Department of electrical Engineering Science and Research Branch Islamic Azad University, Mazandaran, Iran arjangpower@yahoo.com
AbstractThis paper has been suggested a hybrid system of fuel cell power plant and wind turbine. The proposed system connected to grid for home power consumption. The operational cost of the system in wind energy has different effects. In fuel cell power plant (FCP), Genetic algorithm is used to find optimum output power. FCP can produce both electrical energy and thermal energy at the same time, Therefore it tried to supply thermal load and electrical load. Combination of wind turbine and FCP with local grid results in lower operational cost than that of individual units. For analyze the operational cost a Cost function consist of cost of energy, electrical power from wind energy, power trade with the local grid, thermal recovery from the FCP, Setup Fee of FCP and maintenance cost are used. The tariff for electricity is variable and loads have changed in every time. A wind energy unit is used to commute in Mashhad, Iran. The operation of the fuel cell power plant system is planned to optimize cost according to thermal load and electrical load demand, available wind power and different tariffs for buy and selling electricity of local grid. The obtained results confirm the capability of the proposed system. Keywords- fuel cell power plant; wind turbine; cost function
I.
INTRODUCTION
Conventional energy sources are no longer considered as the solely way of supplying energy then societies try to use distributed generator systems (DG) beside conventional ones [1].The term DG means any small-scale generation which is located near the consumers load instead of being in the center or remote locations. DGs advantages, over other systems, such as less waste of energy over long transmission or distribution lines [2]and being quite flexible in a sense that there is always the ability to add smaller hardware during peak times make renewed interest in the DGs operating in parallel with the distribution network and make hybrid systems. The term hybrid energy system is commonly used to describe a power system with more than one type of supplier or generator, usually a generator powered by a gas or diesel engine, and a renewable energy source such as a wind, photovoltaic (PV), or hydroelectric power generator. Nowadays, the use of hybrid renewable energy systems not only due to the above mentioned
disadvantages of conventional systems but also for supplying less costly the power demand of various regions has attracted some researchers attention. For example in [3] electrical demand of the biggest island of Turkey was examined to realized how it could be possible to supply that with renewable energy sources. In [4] the viability of adding wind turbines to an existing diesel plant of a remote aria in Saudi Arabia was studied. Another feasibility study is described in [5], where hybrid systems supplied by hydrogen are evaluated for applications in Newfoundland, Canada. Therein most of these studies, and also [6], hybrid electricity generation systems are often considered less costly and more reliable than systems that rely on an individual source of energy. wind energy sources, due to the fluctuating behavior, cant employ as the solely source of power supply, they should be accompanied with other energy sources such as FCP to make less costly and more reliable source of energy[6].Recently, the combined use of renewable energy sources, especially FCP is becoming increasingly fascinating[7].Proton Exchange Membrane (PEM) fuel cells for having a lot of advantages such as: high efficiency (35% - 60%), low to zero emissions, quiet operation, high reliability due to the limited number of moving parts, modularity, scalability, quick installation, gives good opportunities for cogeneration operations and the ability to be placed at any site in a distribution system without geographic limitations [8-11] show great promise for use as DGs. All of these advantages lead to a deep study of this type of fuel cell in order to supplying residential load. So, this study began to investigate the feasibility of adding a FCP to the then already present local grid of a residential house in the Mashhad. On a second stage, we further investigate the viability of adding wind energy to the current utility system and FCP in order to achieve a more economical and environmentally friendly power supply to support both the electricity and thermal demand of residential house in Mashhad. The combination of FCP and wind energy systems in the form of CHP system can be considered as a potential choice to satisfy electrical and thermal demand of residential house. In such kind of system, the thermal and electrical energy generation should
be managed by a robust management strategy in a way to minimize the total cost with regard to satisfy constraints. Economical aspect of fuel cell has been presented in references [8-10, 12-15]. In order to find the optimal output power from FCP at the presence of constraints an economic model has been introduced in [8, 9]. The model of these two articles just considers the feasibility of trading energy with grid, and the usage of thermal recovery from FCP. In [10] the amount of the stored hydrogen is also included to the model. In this paper the model of above mentioned papers has been extended to integrate wind energy. Not only does the tariffs for selling and buying electricity from the local grid are not constant but also they vary according to the time of using electricity and volume of demand. Different strategies for managing the output power from FCP and wind energy units can be defined. Since wind energy systems have low to zero operating cost so in this paper wind energy unit always is operated at its full capacity. The remaining part of the paper is organized as follows: Section 2 gives a complete structure of the system besides the type of FC and wind energy unit which is used in this study. Formulation of economic model is presented in section 3. The GA algorithm and parameter adjustments are explained in section 4 introduces the solution methodology. Test results and conclusions are discussed in Sections 5 and 6, respectively. II. SYSTEM CONFIGURATION
A. Network Structure The diagram of wind energy PEM Fuel Cell hybrid energy system connected to grid is indicated in Fig. 1. The system is constituted of wind energy turbine, local grid, heat pump, PEM Fuel Cell stack and load unit
Heat energy
Fule cell
Electrical power
Natural GAS
abandoned so, thermal energy is just recovered from the reformer where the temperature goes up to about 360 C. Hot water and space heating considers as thermal load in this paper and adds to electric loads of PEM FCP. The thermal load is fulfilled by using the recovered thermal energy from the FCP and, if necessary, by use of natural gas. On the basis of the difference between the thermal load and the recovered thermal energy and supplying electrical load by FCP or wind energy, six different strategies can be defined. In the first three strategies the only source of electrical energy supplier is PEM FCP connected to grid, while in the other strategies wind energy unit is added to the source of electrical energy supplier and operates at its full capacity allows rapid start-up [17], inexpensive materials than high temperature fuel cells [18]. PEMs can be made extremely thin and the thinner the polymer electrolyte the higher the conductance and lower the resistance losses. Therefore this type of fuel cell presently receives the most attention among all kind of fuel cells. Hence, in this paper 6.3 KW PEM fuel cell power plant due to its advantages is used. When FCP works at full load it can produces thermal energy as much as electrical energy [19]. In order to manage excess thermal or electrical energy, it is vital to have a robust management strategy. In PEM FCP, due to the lower operating temperature, thermal recovery from the stack is abandoned so, thermal energy is just recovered from the reformer where the temperature goes up to about 360 C. Hot water and space heating considers as thermal load in this paper and adds to electric loads of PEM FCP. The thermal load is fulfilled by using the recovered thermal energy from the FCP and, if necessary, by use of natural gas. On the basis of the difference between the thermal load and the recovered thermal energy and supplying electrical load by FCP or wind energy, six different strategies can be defined. In the first three strategies the only source of electrical energy supplier is PEM FCP connected to grid, while in the other strategies wind energy unit is added to the source of electrical energy supplier and operates at its full capacity. C. Wind energy unit 2/6kw wind energy unit is used to commute wind energy to electric power. In order to protect the turbine from damage a turbine's cut-out and cut-in speed are determined by the manufacturer. The cut-in speed is the point where the turbine starts to generate electricity from turning. The cutout point shows the boundary of speed that denotes how fast the turbine can spin before reach to danger zone. Hence, some sort of brake mechanism is used to prevent the turbine from reaching to this danger zone. Technical information of the utilized turbine is available on the internet [20] and is given in table I.
TABLE I. TECHNICAL INFORMATION OF WIND TURBINE
Heat pump
Heat energy
home
Electrical power
Electric line
Power grid
Wind turbine
Figure 1. Network structure
B. Fuel cell PEM stands for polymer electrolyte membrane or proton exchange membrane. Advantages of PEM fuel cells can be mentioned as: their high efficiency compared with other energy conversion devices [16], low operation temperature cause to reach the operation point rapidly and to have a robust management strategy. In PEM FCP, due to the lower operating temperature, thermal recovery from the stack is
Power 1 2 3 4
Cut in wind speed Rated power Rated wind speed Cut out wind speed
Unit
4.5 m/s 6 Kw 12.5 m/s 16 m/s
III.
ELECTRICAL EQUATIONS
A. Wind turbine The electric power output of the wind turbine P WE.T at interval i with respect to wind speed can be declared as below [21].
rtf,t=1.0785PLRt4-1.9739PLRt3+1.5005PLRt20.2817PLRt+0.6838 IV. SYSTEM COST COMPONENT
(12)
A. Cost Of fuel Cost of fuel for producing electrical energy by the FCP. (13)
(1) { The constants, a, b and c can be determined by the equation (2) which are presented in [22]. B. Purchased electrical energy cost Electrical energy purchased from Grid when FCP is the only supplier of demand energy. ( ) (14) Electrical energy purchased from Grid when FCP is the only supplier of demand energy. ( ) Electrical energy Purchased from local grid for storing surplus thermal energy when FCP supply demand energy. (
{
Electrical energy Purchased from local grid for storing surplus thermal energy when wind energy unit is added to supply electricity power.
( )
B. Formulation of economic model
(3)
Term of 0.2max (Pth.t-Lth.t) illustrates requested electrical energy for storing surplus thermal energy.
(4) (5) (6) (7) (8) (9)
C. Gas cost for purchasing thermal energy Gas cost can be added to the cost function if thermal load is more than recovered thermal energy and it is calculated as follow. D. Startup and maintenance cost ( )
Part Load Ratio (PLR) is used to determine efficiency and thermal to electrical ratio [19]. These are calculated in two categories by considering PLR as follow. For Plrt< 0.05 t-0.2716 , rTE,t 0.6801 (10) For PLRt 0.05 t 0.9033PLRt5-2.9996PLRt4+3.6503PLRt3-2.0704 2 PLRt +0.4623PLRt + 0.3747 (11)
E. Cost of wind unit Although wind O&M cost trends had been decreasing for the expansion in overall wind farm size, it cant be negligible. High O&M costs associated with generators, gearboxes and drive trains. In fact, O&M costs are not fix and estimated to increase, on average, 25% over the 20-year life of the various wind machines [23]. Average values of O&M costs are $0.027/kWh. Therefore, formula for O&M cost of the paper is (20). 0&M cost=(0.027)*(PWE,t*6min /60min) (20)
V.
SYSTEM INCOMES COMPONENTS
A. Selling surplus electrical energy Surplus electrical energy sold by FCP is calculated as follows: IELst =celstTmax(pFCL-Lell,0) (21) At the presence of wind energy unit surplus electrical energy income will be calculated as follows: IELst =celstTmax(pFCL+pWEt -Lell,0) (22) VI. AN OVERVIEW OF GENETIC ALGORITHMS
a) b) c) d) e)
A simple genetic algorithm is like the following steps: Generate randomly a population of initial population within the feasible ranges of the decision variables Calculate the fitness for each string in the population. Create offspring strings through reproduction, crossover, and mutation operation. Evaluate the new strings and calculate the fitness for each string (chromosome). If the search goal is achieved, or an allowable generation is attained, return the best chromosome as the solution; otherwise go to step Flowchart of extended GA based solution methodology is displayed in figure 2.
Initial population using random generator
XCO2 wind turbine is added to the complex and make integrated hybrid system. In each case (except base case) GA defines optimal electricity output power of the FCP with respect to consider electricity trading tariffs, thermal and electrical load, FCP constraints and electric output power of wind turbine. Electricity trading tariffs are shown in Table 3. As table 3 show selling price in all the time is cheaper than purchasing price then it is encourage grid to buy electricity from FCP. Electrical and thermal residential load are depicted, in figure 3. When FCP runs, thermal energy is produced as a byproduct besides electrical energy. After recovering this energy we must be sure it will be used by neighborhoods. Hence in order to encourage them to use this energy, its price should be lower than other ways of supplying thermal energy. So, as it is obvious from Table 2, thermal energy selling price with FCP is considered lower than fuel price for residential load.
TABLE II. FCP AND GENETIC ALGORITHM PARAMETERS
New generation
T=0
Set population of individuals
individuals
Evaluate fitness
Parameters Maximum limit of generating power, Pmax (kW) Minimum limit of generating power, Pmin (kW) Hot start -up cost, a($) Cold start- up cost, ($) The fuel cell cooling time constant, (h) Minimum up-time, MUT (number of intervals) Minimum down-time, MDT (number of intervals) Lower limit of the ramp rate, PD (kW) Upper limit of the ramp rate, Pu (kW) Length of time interval, T (h) Maximum number of starts-stops, Nmax Maximum number of evolutionary generation Number of individuals Fuel price for residential load,Cg ($/kWh) Price of natural gas for FCPP, Cf ($/kWh) Thermal energy selling price Cth,s ($/kWh) Thermal storage efficiency, st,th (%)
Time (hour) Purchasi ng Tariff Selling Tariff 0-6
0.05 0.03
value
6.3 0.0 0.05 0.15 0.75 2 2 0.5 0.4 0.1 5 5000 450 0.6 0.4 0.4 90
21
0.1 5 0.1
TABLE III. TARIFF OF TRADING ELECTRICAL ENERGY WITH LOCAL GRID

T=T+1 Parent selection
6-8
0.07 0.05
9
0.09 0.07
1011
0.1 0.07
1216
0.11 0.08
17
0.1 3 0.0 9
1819
0.14 0.1
20
0.1 7 0.1 4
22
0.1 0.0 7
23
0.0 7 0.0 5
Crossover &mutation
T=max generation Y stop
Figure.2 Genetic algorithm
VII. CASE STUDY In order to supply residential load three different cases are tested. The base case the residential load demand is supplied just by local grid. In the second case a 6.3kw PEM FCP is applied beside local grid to supply residential load. Data for PEM FCP with GA parameters and thermal energy trading tariffs are given in Table 2.in the last case a 6kw
Figure 3. Electrical and thermal load
According to preliminary studies[24] by the Iranian Renewable Energy Organization (SUNA), Iran's mountainous terrain are characterized by unique wind corridors and are able to produce at least 6500 MW. Iranian government has committed, on its fifth Five-Year Economic Development Plan (20102015), to generate 1650 MW of
wind energy by March 2014. So, Iranian residents can rely on renewable energy sources of Iran, especially wind, to supply their demands. In this article to prove this assertion wind speed data in Mshhad of Khorasane-Razavi province have been used. Wind speed data and the electric output power of turbine, after placing data to (1), are shown in figure 4and 5 respectively.
Figure 7. Electrical Power Trade with Local Grid
Figure 4.Wind speed changes in 24 h of a day at 10 m elevation Figure 8. Thermal Load and Generation
In this case daily cost of energy supplying is equal to 14.5385$ it means a 2.5308 $ reduction in each day that result in 923.7420$ saving per year. C. Case2 By adding another source of energy to the case 1, case 2 is formed. In this case, in addition to the FCP unit the model is tested in using a total capacity of 6 Kw wind energy units which is operated at its full capacity all the time. Wind speed data, figure 4, and wind energy unit date, table 1, are used to calculate power output of wind energy unit. Wind energy unit can affect the amount of power generated by FCP and in turn changes the answers. Electrical load and wind energy power generation plus FCP power generation are depicted in figure 9. Electricity trading with local grid is shown in figure10. Figure 11 shows thermal recovery from FCP. Total cost of this case, which is incomparable with case 2, is 9.3626 $. Therefore from the result and [6] it is obvious that hybrid systems can cause lower cost than a single-supply system.
Figure 5. Electric output power of XCo2 turbine
A. Test and results Base case: In this case both electrical and thermal load are supplied through the local grid and natural gas, respectively. Base case shows the cost of supplying residential load without considering the FCP and WE unit. In this test case daily cost of supplying both thermal and electrical energy will lead to 17.0693$. B. Case 1 In this case a combination of FCP and local grid is used to supply both electrical and thermal load. If the recovered thermal energy is less than thermal load the lack can be compensate by using natural gas otherwise surplus recovered thermal energy is being sold to other neighborhoods. Electrical load and power generation is shown in figure 6while electrical energy trade with local grid is shown in figure 7. Thermal load and recovered thermal energy is depicted in figure 8.
Figure 9. FCP and wind energy power generation with electrical load
VIII. CONCLUSION The paper suggests practical concepts regarding operational cost modeling of the system. In the first case the
Figure 6. Electrical Load and Generation
fuel cell power plant supplies electrical and thermal power as a DG system but in the last case wind energy unit is added to the complex and based on the available power from wind energy unit, the fuel cell power plant supplies both electrical and thermal power to a residential demand. The integration of wind energy and FCP system in an economic model which consist of power trade with the local grid with different tariffs and thermal recovery from FCP is introduced in this paper. Based on wind speed, low cost wind turbine energy production and wind energy unit runs at full capacity most of the time. The main factor that affects the operation of the FCP is thermal load. For instance some times, based on the system economics, FCP inclines to generate electrical energy more than the electric load during high thermal consumption periods and produces low electrical energy during low thermal periods. Test results on a 6.3 kW fuel cell power plants and 6 kW wind energy indicate the feasibility of the suggested approach and its potential to find the optimal power output from the FCP subject to the connected constraints and wind energy unit power. REFERENCES
[1] S. Tichi, M. Ardehali, M. Nazari, Examination of energy price policies in Iran for optimal configuration of CHP and CCHP systems based on particle swarm optimization algorithm, Energy Policy, 38 (10) (2010) 6240-6250. P. Nema, R. Nema, S. Rangnekar, A current and future state of art development of hybrid energy system using wind and PV-solar: A review, Renewable and Sustainable Energy Reviews, 13 (8) (2009) 2096-2103T. Zhou, B. Franois, Energy management and power control of a hybrid active wind generator for distributed power J. Clerk generation and grid integration, Industrial Electronics, IEEE Transactions on, 58 (1) (2011) 95-104. R. Ramakumar, P. Chiradeja, Distributed generation and renewable energy systems, in: Energy Conversion Engineering onference, 2002. IECEC'02. 2002 37th Intersociety, IEEE, 2004, pp. 716-724.
[14] [15]
[16] [17] [18] [19]
[20] [21] [22]
[23] [24]
genetic algorithms, in: Evolutionary Computation, 1995., IEEE International Conference on, IEEE, 1996, pp. 479. G. Erdmann, Future economics of the fuel cell housing market, International Journal of Hydrogen Energy, 28 (7) (2003) 685-694. N. Srinivas, K. Deb, Muiltiobjective optimization using nondominated sorting in genetic algorithms, Evolutionary computation, 2 (3) (1994) 221-248. L. Carrette, K. Friedrich, U. Stimming, Fuel cellsfundamentals and applications, Fuel cells, 1 (1) (2001) 5-39. A.J. Appleby, Fuel cell handbook, 1988. D. Feroldi, M. Basualdo, Description of PEM Fuel Cells System, PEM Fuel Cells with Bio-Ethanol Processor Systems, (2012) 49-72. M.B. Gunes, Investigation of a fuel cell based total energy system for residential applications, Virginia Polytechnic Institute and State University, 2001. www.renewableenergyworld.com, in, 2013. Catalogue of European Urban Wind Turbine Manufacturers, in. R. Billinton, G. Bai, Adequacy evaluation of generation systems including wind energy, in: Electrical and Computer Engineering, 2002. IEEE CCECE 2002. Canadian Conference on, IEEE, 2002, pp. 24-29. http://www.suna.org.ir P. Giorsetto, K.F. Utsurogi, Development of a new procedure for reliability modeling of wind turbine generators, power apparatus and systems, ieee transactions on, (1) (1983) 134-143.
[2]
[3]
[4]
[5]
A. Demiroren, U.Yilmaz, Analysis of change in electric energy cost with using renewable energy sources in Gkceada, Turkey: An island example, Renewable and Sustainable Energy Reviews, 14 (1) (2010) 323-333.
S. Rehman, I. El-Amin, F. Ahmad, S. Shaahid, A. Al-Shehri, J. Bakhashwain, A. Shash, Feasibility study of hybrid retrofits to an isolated off-grid diesel power plant, Renewable and Sustainable Energy Reviews, 11 (4) (2007) 635-653. M. Khan, M. Iqbal, Pre-feasibility study of stand-alone hybrid energy systems for applications in Newfoundland, Renewable energy, 30 (6) (2005) 835-854. M. El-Sharkh, A. Rahman, M. Alam, Evolutionary programmingbased methodology for economical output power from PEM fuel cell for micro-grid application, Journal of power sources, 139 (1) (2005) 165-169. A.M. Azmy, I. Erlich, Online optimal management of PEMFuel cells using neural networks, Power Delivery, IEEE Transactions on, 20 (2) (2005) 1051-1058. M. El-Sharkh, M. Tanrioven, A. Rahman, M. Alam, Impact of hydrogen production on optimal economic operation of a grid-parallel PEM fuel cell power plant, Journal of power sources, 153 (1) (2006) 136-144. M. El-Sharkh, M. Tanrioven, A. Rahman, M. Alam, Cost related sensitivity analysis for optimal operation of a gridparallel PEM fuel cell power plant, Journal of power sources, 161 (2) (2006) 11981207. F. Barbir, T. Gomez, Efficiency and economics of proton exchange membrane (PEM) fuel cells, International Journal of Hydrogen Energy, 22 (10) (1997) 1027-1037. Y. Zoka, H. Sasaki, J. Kubokawa, R. Yokoyama, H. Tanaka, An optimal deployment of fuel cells in distribution systems by using
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
A New Fault Tolerant Scheme for Cascaded HBridge Multilevel Converter

Saeed Ouni, Negar Noroozi, Mahmoud Shahbazi, MohammadReza Zolghadri, Member, IEEE, Hashem Oraee, Senior Member, IEEE Department of Electrical Engineering, Sharif University of Technology saeedouni@ee.sharif.edu, nenoroozi@gmail.com, mahmoudshahbazi@outlook.com, zolghadr@sharif.edu, oraee@sharif.edu
Abstract- Multilevel converters are mainly used in high power applications (usually more than a few hundred kW). It is therefore important that they can continue working under faulty condition. In this paper, a new method is proposed to improve performance of CHB inverter under faulty condition. The proposed method will be compared with existing methods in term of voltage availability. The converter is simulated in Simulink and the results are provided. These results confirm the effectiveness of the proposed scheme, and are in accordance with the theoretical results.
Keywords: Cascaded Availability. H-Bridge Inverter, Fault Tolerant, Voltage Figure 1. 11-Levels CHB-MLI
I.
INTRODUCTION
Recently the multilevel converters have been widely used in the power industry. The primary three-level inverter, neutral clamped diode (NPC) inverter has been proposed by Nabae in 1981[1]. Then in 1988, cascade H-Bridge inverter (CHBMLI) structure was presented [2]. The flying capacitor multilevel inverter (FC-MLI) is another proposed structure for multi-level inverter [3, 4]. It can be mentioned as a drawback for DC-MLI and FC-MLI converters that, the number of the clamping diodes and the flying capacitors increases dramatically with the voltage level. Because of the CHB modular structure and other features, this inverter with seven to eleven voltage levels with low device switching frequencies (below 500 Hz) has been increasingly used in high-power, medium-voltage (MV) drives. For example, Siemens Perfect Harmony is a CHB-MLI inverter used for power and voltage range of 0.2-132 MVA and 2.313.8 kV [5]. The main drawback of this inverter is the need of isolated dc supply for each H-bridge module. Fault tolerant ability is necessary for power electronic converters to make the continuity of service as much as possible. For example, in [6, 7] three fault tolerant structures for AC/DC/AC converter are presented. Also, in [8, 9] a fault tolerant structure for matrix converter is proposed. Since the multilevel converters are mainly used in high power application, it is important that they can continue working under faulty condition. Different methods are proposed for MLIs to achieve this importance. For example, in [10] and [11] a fault-tolerant solution for an NPC converter and a five level diode clamped inverter, respectively, under short-circuit fault is proposed. In [12] by using three pair switches the faulty leg can be connected to the NP of the converter under
any fault condition, regardless of whether it is in open or short circuit. In these methods, the inverter is able to continue working but it may be necessary to reduce the working modulation index. In [13] a redundant leg is used to replace with defective leg under faulty condition, and then the inverter structure doesnt change after applied fault tolerance strategy. Several methods are proposed for fault tolerant operation of CHB-MLIs in the literature as bellow: A) Use of a Redundant Cell The cascaded inverter uses its modularity advantageously to introduce the idea of redundancy of cells instead of using redundancy of components. [14, 15] B) Bypass Operative Cells When a fault occurs, the number of operative cells per phase is no longer the same; thus, unbalanced output voltages are applied to the load. A simple way to recover the balanced operation is to bypass as many cells as necessary, in order to
Figure 2. Angle between phases and available voltage amplitude by using NS method for an 11-levels CHB-MLI with 2 faulty cells [16].
Figure 3. Typical examples of faulty space vectors caused c by failed power cells [20].
evel CHB-MLI Figure 4. 11-Le
operate with the same number of cells on each e phase [16, 17, and 18]. C) Neutral Shift method: [16, 18, and 19] In this method, only the faulty cells are bypassed, and to provide balance line voltages, the phase reference voltages must be modified. This approach will be called c neutral shift (NS) and is equivalent to introducing a zero-sequence component into the reference voltages. In Fig. 1 an 11-level CHB-MLI is shown. This method is used for this inverter with 2 faulty cells. The angle between ph hases and available voltage amplitude are shown in Fig. 2. It is clear that in various fault conditions the reference values for phase voltages may be different in amplitude and angle values. D) Use of Redundant States [20-23] A well-known characteristic of ML conv verters is that they have many redundant states. These states can be easily used trates some typical to overcome fault operation. Fig. 3 illust examples of the space vector diagram when w one or more power cells in a 7-level inverter are damage ed. The dots with a circle around represent the space vectors affected by faulty cells, and therefore cannot be used. In this paper, a new fault tolerant scheme e is presented for a CHB converter. First, the proposed schem me is explained and then its voltage availability is compared with the existing
TABLE I
methods. Finally, the convert ter is simulated in Simulink environment, and the simulatio on results are compared with theory. II. PROPO OSED METHOD In this scheme, a different approach a is used in order to guaranty the symmetrical opera ation of the converter after the fault occurrence. In order to do o so, after a fault event in one of the phases, same fault is deliberately applied to the other phases. For example consideri ing Fig. 4 suppose there is a short circuit fault in one of the upper u switches of the cell 1 of phase A. in this condition, the bottom switch cannot become on and will stay always off. Sam me condition must be provided for the b and c phases. In other r words, in one of the cells the upper switch is always on an nd its corresponding bottom switch is always off. It should be noted that this change can be applied in the cells in a ro otational order. By using this approach, output phase voltages s of the inverter will have a dc value. However, since this dc va alue is the same for all phases, it is not present in the line-line voltages. v In the following, difference of this method with other methods is studied. This comp parison is carried out for two conventional switching techn niques used for CHB-MLIs; Space Vector Modulation (SVM M) and multi carrier PWM. For a better comparison, first two fa ault types are defined. F1 fault shows a fault in the cell from which the cell can no longer produce the +Vdc voltage. It is s visible that this fault can be the result of a short circuit fau ult in S1 switch (Fig. 4), and because of that the lower switch h of this leg S2 must be always open to avoid short circuiting the t source. This fault can also be a result of an open circuit fau ult in S2. F2 fault indicates the fault that blocks the possibility y of producing Vdc voltage, that can similarly occur in two cases. c
THE RECTIFIER PARAMETER RS

Available Voltage V Fault Type F1 in phase a F1 in phase a,b F1 in phase a,b,c F1 in phase a , F2 in phase b F1 in phase a , F2 in phase b,c NS method 92.90 86.03 80.02 86.03 80.02 P Proposed method 89.95 89.95 89.95 80.02 80.02
Figure 5. Space vector diagram for: a) 5-Levels, 5 b) 5-Levels under fault condition, c) 7-Levels, d) 7-Levels s under fault condition,
A) As mentioned in the introduction, on ne way of creating fault tolerance is by using redundant vector rs. We know that it is possible to use redundant vectors in vector control of the CHB-MLI. In Fig. 5 produced vector of the e proposed method and faulty cell bypassing method are show wn for 5- level and 7-level converters. Although number of ve ectors is less in the proposed method, but the maximum availab ble vector to create a sinusoidal voltage is the same in both methods. m Therefore one can verify that in case of using vector method for control of the converter, the available voltage is the e same. B) In Table I, the available voltage for the t two mentioned modes is shown for different faults in an 11-level converter. From this table one can see that the availa able voltage in the proposed method for an F1 fault in phase a in 3% lower than that of the NS method. However, it is nota able that in the NS method one cell is completely removed and d in other words, if this cell has been connected to some sola ar panels, all these panels are out of the circuit. In the simulation results this will be clarif fied more. Another important observation is that in case of another a F1 fault in one of other phases, the available voltag ge in the proposed method will be 4% more than NS method. m Also it is important to note that in this case, too many m solar cells are
omitted in NS method. This difference will reach a value as high as 10 % for another fault in i the third phase. Also in this table, these two methods are com mpared for two more cases. In one case, an F1 fault is occurr red in the A phase and an F2 fault is occurred in the B phase. In this case, the available voltage of the proposed method is 6 % less than the NS method, and three cells are by ypassed here, versus the two bypassed cells in the NS metho od. However, the advantage of the proposed method is the sym mmetrical use of all cells. In the last case, an F1 fault is applied in phase A, and another one is occurred the b,c phases. The e available voltage in both methods are similar, while the e above mentioned advantage still holds true for the proposed method. III. SIMULA ATION RESULTS A typical 11-level CHB-MLI has h been simulated in Simulink with parameters shown in Table e II.
TABL LE II
THE CHB-MLI PARAMETERS P

Number of Level 11 Output voltage 2.2 kV L-L rms DC Link k Voltage e 450 V Pout 400 kW fs 500 Hz
2500 2000 1500 Inverter Output Voltage 1000 500 0 -500
applied to the phases b and c. In this condition the inverter output voltages are shown in Fig. 6.a. It is clear that these three voltages are symmetric with the same dc component it is shown in Fig 6.b. The line-line voltages rms and dc values are shown in Fig 7. It is visible that no dc value is presented in these voltages. Also, the rms values are reduced to 90% of pre-fault values. In Table III simulation results are presented and compared with the results of the NS method. For this fault, it can be seen that the available voltage in the proposed method is slightly less than NS method. However, here all faulty cells (three cells) have equally provided power whereas in the NS method, faulty cell is completely out of operation. Also other cells are forced to provide unequal power with higher values. It can be said that this advantage can make this method interesting for a photo voltaic inverter. In this application by using this method under such fault condition, all photo voltaic panels can be used. Table III shows also the simulation results for another fault condition, when an F1 fault is occurred in phases a and b. Not only the proposed method offers a higher available voltage, but again all the cells participate in power conversion. Based on theoretical and simulation results, it can be concluded that the voltage availability of the proposed method might be slightly less or more than the NS method. On the other hand, in the proposed method, all cells are engaged in power conversion process, in contrary with the NS methods. IV. CONCLUSION In this paper, a new fault tolerant method for the CHB converter has been proposed. In this scheme, same fault is applied to the two healthy phases, in order to eliminate the DC voltage in line-line voltages, which is produced by the asymmetry caused by the fault. Although in some cases the available voltage is smaller than that of the NS method, but this method is interesting due to its symmetrical operation and not disconnecting too many DC sources, that can be solar panels. Therefore, although this method has better voltage availability in some cases, in other cases it can be interesting as well, due to the above-mentioned reasons. It is possible to
-1000 -1500 -2000 -2500 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
Time(s)
a
300
250 DC Value of Inverter Output Voltage
200
150
100
50
-50 0.08
0.09
0.1
0.11 Time(s)
0.12
0.13
0.14
b Figure 6. a) Inverter output voltages, b) DC component of Inverter voltage

R M S V alue of Line-Line V oltages D C value of Line-Line V oltages
100 50
X: 0.1241 Y: 0.6123
0 -50 -100 0.08
0.09
0.1
0.11 Time(s)
0.12
0.13
0.14
2250 2200 2150 2100 2050 2000 1950 0.08 0.09 0.1 0.11 Time(s) 0.12
X: 0.1255 Y: 1981
0.13
0.14
Figure 7. Line-line voltages, a) rms value, b) dc values
Let us first suppose that an F1 fault in the phase a is occurred in t=0.1 sec. In the proposed method, same fault is
TABLE III Simulation Result and comparison between the proposed method and NS method. Available Voltage Proposed method 90% 90% Power of first cell of each phase Proposed Method a F1 in phase a F1 in Phase a,b 92.9% 86.1% 44% 44% b 44% 44% c 44% 44% Neutral Shift Method a 0% 0% b 98% 0% c 88% 86.5% Power of other cells of each phase Proposed Method a 90% 90% b 90% 90% c 90% 90% Neutral Shift Method a 94% 81% b 98% 90% c 88% 86%
Type of Fault
NS Method
Note: The rated value of output line-line voltage is 2.2 kV and the rated power of each cell is 26kW.
use a combination of these two methods. REFERENCES

[1] Nabae A, Takashi I, Akagi H. A new neutral-point clamped PWM inverter. IEEE Trans Ind Appl 1981;17:51823. [2] Marchesoni, M.; Mazzucchelli, M.; Tenconi, S., A nonconventional power converter for plasma stabilization, Power Electronics, IEEE Transactions on, vol.5, no.2, pp.212-219, Apr 1990 [3] 2. M. F. Escalante, J. C. Vannier, et al., Flying Capacitor Multilevel Inverters and DTC Motor Drive Applications, IEEE Transactions on Industrial Applications, Vol. 49, No. 4, pp. 809815, 2002. [4] L. Xu and V. G. Agelidis, Flying Capacitor Multilevel PWM Converter Based UPFC, IEE Proceedings on Electric Power Applications, Vol. 149, No. 4, pp. 304310, 2002. [5] Siemens Global Website. [online]. Available: http://www.siemens.com [6] M. Shahbazi, P. Poure, S. Saadate, M. R. Zolghadri, FPGA-based Fast Detection with Reduced Sensor Count for a Fault-Tolerant Three-Phase Converter, IEEE Trans. on Industrial Informatics, 2012 (accepted, article in press, doi: 10.1109/TII.2012.2209665). [7] M. Shahbazi, E. Jamshidpour, P. Poure, S. Saadate, M. R. Zolghadri, Open and Short-Circuit Switch Fault Diagnosis for Non-Isolated DC-DC Converters Using Field Programmable Gate Array, IEEE Trans. on Industrial Electronics, vol. 60, pp. 4136-4146, Sep. 2013. [8] D.U. Campos-Delgado, D.R. Espinoza-Trejo and E. Palacios; Faulttolerant control in variable speed drives: a survey , IET Electr. Power Appl., 2008, Vol. 2, No. 2, pp. 121134 [9] Sangshin Kwak; , "Four-Leg-Based Fault-Tolerant Matrix Converter Schemes Based on Switching Function and Space Vector Methods," Industrial Electronics, IEEE Transactions on , vol.59, no.1, pp.235-243, Jan. 2012 [10] S. Li and L. Xu, Fault-tolerant operation of a 150 kW 3-level neutral point clamped PWM inverter in a flywheel energy storage system, in Conf. Rec. 36th IEEE IAS Annu. Meeting, Sep./Oct. 2001, vol. 1, pp. 585588. [11] G. Sinha, C. Hochgraf, R. H. Lasseter, D. M. Divan, and T. A. Lipo, Fault protection in a multilevel inverter implementation of a static condenser, in Conf. Rec. 30th IEEE IAS Annu. Meeting, Oct. 1995, vol. 3, pp. 25572564. [12] S. Li and L. Xu, Strategies of fault tolerant operation for three-level PWM inverters, IEEE Trans. Power Electron., vol. 21, no. 4, pp. 933 940, Jul. 2006. [13] E. R. da Silva,W. S. Lima, A. S. de Oliveira, C. B. Jacobina, and H. Razik, Detection and compensation of switch faults in a three level inverter, in Proc. 37th IEEE Annu. Power Electron. Spec. Conf., Jun. 2006, pp. 17. [14] W. Song and A. Q. Huang, Control strategy for fault-tolerant cascaded multilevel converter based STATCOM, in Proc. 22nd IEEE Applied Power Electron. Conf., Feb. 2007, pp. 10731076. [15] H. Iman-Eini, S. Farhangi, J.L. Schanen, and M. Khakbazan-Fard, "A Fault-Tolerant Control Strategy for Cascaded H-Bridge Multilevel Rectifiers," Journal of Power Electronics, vol. 10, 2010, p. 34-42. [16] P. W. Hammond, Enhancing the reliability of modular medium-voltage drives, IEEE Trans. Ind. Electron., vol. 49, no. 5, pp. 948954, Oct. 2002. [17] W. Song and A. Q. Huang, Control strategy for fault-tolerant cascaded multilevel converter based STATCOM, in Proc. 22nd IEEE Applied Power Electron. Conf., Feb. 2007, pp. 10731076. [18] J. Rodrguez, P. Hammond, J. Pontt, R. Musalem, P. Lezana, and M. J. Escobar, Operation of a medium-voltage drive under faulty conditions, IEEE Trans. Ind. Electron., vol. 52, no. 4, pp. 10801085, Aug. 2005. [19] P. W. Hammond, Multiphase power supply with series connected power cells with failed cell bypass, U.S. Patent 6 222 284, Apr. 24, 2001.
[20] S. Wei, B. Wu, F. Li, and X. Sun, Control method for cascaded Hbridge multilevel inverter with faulty power cells, in Proc. 18th IEEE Applied Power Electron. Conf., Feb. 2003, vol. 1, pp. 261267. [21] P. Correa, M. Pacas, and J. Rodrguez, Modulation strategies for fault tolerant operation of H-bridge multilevel inverters, in Proc. IEEE Int. Symp. Ind. Electron., Jul. 2006, vol. 2, pp. 15891594. [22] P. Correa and J. Rodrguez, Control strategy reconfiguration for a multilevel inverter operating with bypassed cells, in Proc. IEEE ISIE, Jun. 2007, pp. 31623167. [23] Y. Zang, X. Wang, B. Xu, and J. Liu, Control method for cascaded Hbridge multilevel inverter failures, in Proc. 6th WCICA, Jun. 2006, vol. 2, pp. 84628466. [24] 9G. Brando, A. Dannier, A. Del Pizzo, and R. Rizzo, Quick identification technique of fault conditions in cascaded H-bridge multilevel converters, in Proc. ACEMP, Sep. 2007, pp. 491497. [25] Lezana, P.; Pou, J.; Meynard, T.A.; Rodriguez, J.; Ceballos, S.; Richardeau, F.; , "Survey on Fault Operation on Multilevel Inverters," Industrial Electronics, IEEE Transactions on , vol.57, no.7, pp.2207-2218, July 2010.
Placement and Performance Analysis of STATCOM and SVC for Damping Oscillation
Gokhan Cakir
Electrical and Computer Engineering Department Tennessee Technological University Cookeville, TN 38505 USA gcakir42@students.tntech.edu
Abstract Power system oscillation is one of the major problems in power system operation. If not damped, these oscillations can grow and decrease transmission capacity of the lines which may cause interruption in energy supply. Several methods for damping of these oscillations are reported in literature. Traditionally, these oscillations have been damped by power system stabilizers. Recently, FACTS devices such as static synchronous compensator (STATCOM), static var compensator (SVC) equipped with a power oscillation damper (POD) have been also efficiently used for damping oscillation. It should be noted that, the main objective of FACTS-controllers is to enhance voltage stability and increase power flow transfer capability the transmission network. This paper proposes application of the residue factor method to obtain the best location for STATCOM and SVC for damping oscillations. The residue factor used is related to the critical oscillatory modes. Also presented in this paper is a comparison of SVC and STATCOM for the enhancement of oscillations damping. IEEE 14-Bus test system was used to demonstrate the effectiveness of the proposed method. From the simulation results it is seen that STATCOM has more ability in damping oscillation when compared to SVC. Index TermsPower System Oscillation, Static Synchronous Compensator, Static Var Compensator, FACTS
Ghadir Radman
Electrical and Computer Engineering Department Tennessee Technological University Cookeville, TN 38505 USA gradman@tntech.edu severe disturbances. They may even lead to leading power factor operation, and losing stability [5]. In addition to providing reactive power, and voltage control, flexible AC transmission system (FACTS) controllers equipped with supplementary controllers effectively damp out power system oscillations. In this work, Power Oscillation Damper (POD) was applied to STATCOM and SVC as a supplementary controller. FACTS controllers sometimes are found to provide much better damping for inter-area mode of oscillations than the PSSs [3]; Shunt FACTS controllers, such as Static Var Compensator and STATCOM, are capable of damping power swing mode effectively [6]. This paper deals with two aspects. One aspect is to determine the best location of STATCOM and SVC in order to damp out oscillations. The other aspect is to make a comparison between STATCOM and SVC for their effectiveness in enhancing damping oscillation. The first aspect is studied with the help of residue method presented in [7], where the method is applied to STATCOM and SVC. The second aspect is evaluated using simulation program called Power System Analysis Toolbox (PSAT). This paper is structured as follows: section II presents STATCOM unit; section III presents SVC unit; section IV presents residue method; section V presents FACTS controller POD design; section VI presents simulations and results, and section VII concludes the paper. II. STATCOM UNIT STATCOM is a shunt-connected reactive power compensation device. It is a device used to provide voltage support to the system by injecting or absorbing reactive power to/from the system. Fig. 1 shows the main three components of a STATCOM: voltage source converter (VSC) with a capacitor in the DC side, coupling transformer, and the control system. The relation between the AC system voltage and the voltage at the STATCOM AC side terminals provides the control of reactive power flow. If the voltage at the STATCOM terminals is higher than the system voltage, reactive power will be injected from STATCOM to the system and STATCOM will work as a capacitor. When the voltage at the STATCOM is less than the AC voltage, STATCOM will work as an inductor, and reactive power flow will be reversed [9].
I. INTRODUCTION Power system stability problem is usually associated with insufficient damping of oscillations. Power system oscillations are usually in the range between 0.1 and 2 Hz depending on the number of the generators in a system, and can be classified as local and inter-area oscillations. In local mode of oscillations, with a frequency between 1.0 and 2.0 Hz, one generator oscillates against the rest of the system while interarea oscillations, with a frequency 1.0 Hz and less, are related to the phenomenon where synchronous generators in one area oscillate with the ones in another area. The traditional approach to damp out inter-area and local oscillations is to install PSS that provides supplementary control action through the generator excitation system. PSSs are widely used to damp out local and inter-area modes of oscillations [1, 2]. Power utilities worldwide have been using PSSs as an effective excitation controller in order to enhance the system stability [4]. However, there have been problems experienced with PSSs over the years of operation. Some of these problems were owing to the limited capability of PSS as it is capable of damping only local modes of electromechanical oscillations. Furthermore, PSS's may also cause great variations in power system voltage profile under
STATCOM control can be designed to keep the voltage of the bus to which it is connected close to a set reference value by controlling the AC side voltage of the VSC through a PIcontrol. Under normal operating condition, the phase shift between the internal and terminal voltages will be small causing the STATCOM to absorb small amount active power to support the internal losses [9].
AC SystemBus
Vref
PI AC Voltage Regulator
Converter Dynamics
Vac
K ACP
uc
K ACI s
1 1 sTc
V1
sTw 1 sT1 1 sT3 1 sTw 1 sT2 1 sT4
Fig. 2. STATCOM with Power Oscillation Damper (POD)

Phase Locked Loop
III. SVC UNIT
jX t
PI
Qref
I dc
VSC
R
C
Vdc
Fig. 1. STATCOM Model
The STATCOM equations in d-q reference frame are summarized as follows [8]:
Static VAR Compensator (SVC) is a shunt-connected static VAR generator or absorber whose output is adjusted to exchange capacitive or inductive current so as to maintain or control specific parameters of the electrical power system such as bus voltages. SVC is an important FACTS device that has been used for a number of years to improve transmission line economics by resolving dynamic voltage problems. Accuracy, availability and fast response enable SVCs to provide high performance steady state and transient voltage control compared with classical shunt compensation. SVCs can perform the duty of providing rapidly controlled Vars appropriately and thus maintaining the voltage, and inherently improving transient stability. Although the main objective of SVC is to regulate voltage at its terminal, they are also used for damping oscillation in power systems. Fig. 3.a and Fig. 3.b show typical configuration of a SVC with voltage control and its V-I characteristic respectively. The Fixed Capacitor FC provides reactive power and is designed to act as a harmonic filter. The other two branches with thyristor valves, Thyristor Controlled Reactor (TCR) and Thyristor Switched Capacitor (TSC), are controlled to regulate the required reactive power by the system. Not every SVC needs all of the above elements.
where , , , , and are the d-axis, and q-axis STATCOM current and voltage components, , are the resistance and leakage reactance of the coupling transformer, is the capacitor voltage, represent the leakage resistance of the electronic component, and is the angular frequency [9]. The reactive output power of the compensator is varied to control the voltage at connection point in order to keep the voltage within the permissible limits. STATCOM can provide reactive power almost instantly via controlling the modulation index m and hence improving system transient stability [9]. The STATCOM damping controller structure is shown in Fig. 2.
Fig. 3. (a) SVC Configuration (b) V-I Characteristic
The SVC damping controller structure is shown in Fig. 4 [15]. The susceptance of the SVC, , could be expressed as:
(4) where is the SVC reference susceptance that is tuned for steady state operation.
B
ref BSVC min SVC
can be expanded using partial fractions in terms of c and b vectors, and the right and left eigenvectors are as follows [10]: (12) where N is the total number of eigenvalues.
uSVC
max uSVC
Ks 1 sTs
BSVC
Each term of the summation in the numerator is a scalar called residue, , expressed as follows: (13)
max BSVC
min SVC
sTw 1 sT1 1 sT3 1 sTw 1 sT2 1 sT4
Where and denote the right and left eigenvectors respectively associated with the ith eigenvalue [7]. This can be considered in terms of mode controllability and observability. The modal controllability measure is as follows: (14)
Fig.4. SVC with Power Oscillation Damper (POD)
The modal observability measure is as follows: (15) According to (14) and (15), (16) can be expressed as follows: (16) The residue of a particular mode i gives the measurement of that modes sensitivity to a feedback between the output y and the input u for a SISO system. The residue is the product of the modes observability and controllability [7]. A. Eigenvalue Analysis Stability issue can be analyzed by studying the eigenvalues. An operating point is stable if all of the eigenvalues are on the left-hand side of the imaginary axis of the complex plane; otherwise it is unstable [11]. Let be the ith eigenvalue of the state matrix A. The real parts of the eigenvalues give the damping, and the imaginary parts give the frequency of oscillation. If a real eigenvalue is negative, the associated mode decays over time. The larger the magnitude of the mode, the quicker it decays. On the other hand, if one of the real eigenvalues is positive, the corresponding mode is unstable [12]. Given the state matrix A is real, the complex eigenvalues always occur in conjugate pairs. Complex conjugate pair eigenvalues correspond to an oscillatory mode. A pair with a positive represents an unstable oscillatory mode because these eigenvalues yield an unstable time response of the system. In contrast, a pair with a negative represents a stable oscillatory mode. The damping ratio is calculated by: (17) (11) Generally, the oscillatory modes having damping ratio less than 3% are said to be critically or poorly damped oscillatory
IV. RESIDUE METHOD Mathematical model of the overall dynamic system is expressed using a set of non-linear differential equations as follows: (5) The overall linearized system model including STATCOM/SVC is represented by the following equation: (6) (7) where b and c are the columnvector and the row-vector, respectively. Active power flow deviation which is a local signal has been used as the feedback signal for FACTS supplementary controller (POD). Modulation index m, and susceptance , are used as an input to power system for STATCOM and SVC, respectively. Assuming , , and are the diagonal matrix of eigenvalues and matrices of right and left eigenvectors, respectively; then we have: (8) (9) (10) A mode is uncontrollable if the corresponding row of the matrix is zero. A mode is unobservable if the corresponding column of the matrix is zero. If a mode is either uncontrollable or unobservable, feedback between the input and output will have no effect on the mode [10]. The open loop transfer function of the system is as follows:
modes, and eigenvalues corresponding to these damping ratios are called dominant modes because their contribution dominates the time response of the system. However, in power systems, states are considered to be well damped if the damping ratio for all eigenvalues is greater than 5% [11]. V. FACTS POD CONTROLLER DESIGN POD is used as a feedback controller. Active power flow deviation which is local signal has been used as the feedback signal for STATCOM and SVC supplementary controller (POD). When applying the feedback control, eigenvalues of the system are changed. The change of the eigenvalues must be directed towards the left half complex plane for damping improvement [10]. The movement can be achieved with a transfer function consisting of an amplification block, a washout block, and stage of lead-lag blocks [10]. Transfer function of the POD controller is (18)
Each generator of the test system is equipped with the AVR Type II which is the simplest AVR model that can be used for rough stability evaluations [13]. The generator is described by six order non-linear mathmetical model while exciter by third order. The sixth order model of generator is obtained assuming the presence of a field circuit and an additional circuit along the d-axis and two additional circuits along q-axis. The generator state variables are while exciter has the following state variables [13]. The state variables are defined in [14]. Eigenvalue analysis was obtained in PSAT, and the results are calculated for three different cases: no controller, with SVC, and with STATCOM. Table I gives the eigenvalues and damping ratios for the critical mode.
TABLE I EIGENVALUE ANALYSIS OF THE CASE STUDY NETWORK Critical Mode States No Controller e1q_Syn_1, vf_Exc_1 e1q_Syn_1, vf_Exc_1 e1q_Syn_1, vf_Exc_1 Dominant Eigenvalue -0.058588.1392i -1.40568.2109i -1.54398.2249i Damping Ratio 0.0072 0.1687 0.1845
where K is a positive constant gain, and is the transfer function of the combined wash-out and lead-lag blocks. The washout time constant is usually in the range of 5-10 s. [10]. The lead-lag parameters are determined in [10] as follows:
With SVC With STATCOM
where e1q_Syn_1 is q-axis voltage of generator-1, and vf_Exc_1 is field voltage of generator-1. Table II shows the residue values for different cases described as follows: STATCOM is connected to each bus. Residue values are calculated, and shown in column 2. Similarly, SVC is connected to each bus. Residue values are calculated, and shown in column 3. The largest residue indicates the most effective location of STATCOM and SVC [10]. According to Table II, the best place for STATCOM is Bus number 2, and for SVC is Bus number 4 since they are the highest residue values.
TABLE II RESIDUE VALUES DUE TO CORRESPONDING STATCOM AND SVC LOCATION Bus No 1 Residue Values for STATCOM 0.0003024 0.2031 0.1150 0.0723 0.1221 0.0097 0.011 0.003 0.012 0.0069 0.0042 0.0022 0.0043 0.0035 Residue Values for SVC 0.0422 0.000211 0.0626 0.0826 0.0631 0.0114 0.0268 0.0089 0.0418 0.0353 0.0140 0.0029 0.0123 0.0304
(19)
,
where is the corresponding critical mode number, denotes phase angle of the residue , is the frequency of the critical mode to be damped in rad/sec, is the number of compensation stages (usually =2) [10]. VI. SIMULATION AND RESULTS The performance of STATCOM and SVC have been verified on IEEE 14 Bus test system [13] shown in Fig. 5, and the results are presented in this section.
2 3 4 5 6 7 8 9 10 11 12 13
Fig. 5. IEEE 14 bus test system
14
The damping ratio of the base case is 0.0072 (0.72%) which is relatively low. The eigenvalue pair of 0.058588.1392i corresponding to critical mode for this system is the poorly damped oscillatory mode. Table I demonstrates that adding STATCOM controller to the power system increases the damping ratio to 0.1845 (18.45%) while adding a SVC to the system increase the damping ratio to 0.1687 (16.87%) for critical mode. This shows that STATCOM has slightly better ability in damping oscillation when compared to SVC. A. Performance Evaluation of the System In order to test the best location, simulation of the system was performed using PSAT. A three phase fault was applied at Bus 5 in order to observe the impacts of STATCOM and SVC devices on damping oscillation. It is observed from Fig. 6 that the oscillation is damped out in about 5s with STATCOM placement, which is quite less when compared with the SVC placement. Fig. 7 shows the frequency at generator-4 for three diferent case. Similarly, Fig. 7 shows that STATCOM has better ability than SVC in damping system oscilations.
1.066
VII. CONCLUSION This paper studied STATCOM and SVC placement using residue method, and compared the performance of STATCOM and SVC using small signal analysis. The best placement of STATCOM and SVC was obtained by using residue method. The IEEE 14 Bus test system was used for the study. Eigenvalues of the test system were computed in order to find poorly damped oscillatory mode. A supplementary controller called POD was designed for STATCOM and SVC. The results show that STATCOM is slightly more effective in damping oscillation when compared with SVC. All of the simulations were performed using Power System Analysis Toolbox (PSAT) in MATLAB environment. REFERENCES
[1] P. Pourbeik, and J.G. Michael, Simultaneous coordination of power system stabilizer and FACTS device stabilizers in a multi-machine power system for enhancing dynamic performance, IEEE Trans. Power System, 1998, 13, (2), pp. 473-479 N. Martins, and T.G.L Leonardo, Determination of suitable location for power system stabilizers and static var compensators for damping electromechanical oscillations in large scale power systems, IEEE Trans. Power System, 1990, 5, (4), pp. 1455-146 H. Okamoto, A. Yokoyama, and Y. Sekine, Stabilizing control of variable impedance power systems: application to variable series capacitor system, Electr. Eng. Jpn., 1993, 113, (2) M. A. Abido, Power system stability enhancement using FACTS controllers, A Review, the Arabian Journal for Science and Engineering, Volume 34, Number 1B, pp. 153-171, April 2009. R. Eriksson, Security-Centered coordinated control in AC/DC transmission systems, Master thesis, 2008, TRITA-EE_2008_034. N. Mithulananthan, C.A. Caizares, J. Reeve, and G. J. Rogers, Comparison of PSS, SVC and STATCOM controllers for damping power system oscillations, Mitsubishi Electric, Mitsubishi Power System Stabilizer (PSS), 2001, pp. 1-7. N. Magaji, and M.W. Mustafa Optimal location of TCSC device for damping oscillations, ARPN Journal of Engineering and Applied Sciences, Vol. 4, No. 3, May 2009, pp, 28-34 K. R. Padiyar, and A.M. Kulkarni, Design of reactive current and voltage controller of static condenser, International Journal of Electrical Power & Energy Systems, vol. 19, pp. 397-410, 1997. A. F. Abdou, A. Abu-Siada, and H. R. Pota, Application of a STATCOM for damping subsynchronous oscillations and transient stability improvement, Universities Power Engineering Conference (AUPEC), pp. 1-5. R. Sadikovic, P. Korba, and G. Andersson, Application of FACTS devices for damping of power system oscillations, Power Tech, 2005 IEEE Russia, pp. 1-6 J. G. Ambafi, M. N. Nwohu, H. O. Ohize, and O. J. Tola, Performance evaluation of PSS and STATCOM on oscillation damping of a NorthCentral power network of Nigeria grid system, International Journal of Engineering and Technology, vol. 2, no. 2, pp. 209-219, Feb. 2012. A. F. Snyder, Inter-Area oscillation damping with power system stabilizers and sysnchronized phasor measurements, thesis (Msc. In Elect.) Faculty of Virginia State University, Feb. 1997. F. Milano, Power system analysis toolbox documentation for PSAT version 1.3.4, July 14, 2005, pp. 156-168, 390-392. P. W. Sauer, and M. A. Pai, Power System Dynamics and Stability, New Jersey: Prentice Hall, 1998, p. 103. M.E. Aboul-Ela, A.A. Salam, J.D. McCalley and A.A. Fouad, "Damping Controller Design for Power System Oscillations Using Global Signals", IEEE Transactions on Power Systems, Vol. 11, No. 2, May 1996, 767773.
[2]
No Controller With STATCOM With SVC
1.064
[3]
1.062
[4]
voltage (pu)
1.06
[5]
1.058
[6]
1.056
[7]
1.054 0 2 4 6 8 10 12 14
time (s)
[8]
Fig. 6. Voltage response at Bus 1 with a three phase fault of Bus 5 [9]
1.0012
1.001
1.0008
With SVC With STATCOM No Controller
[10]
1.0006
frequency (pu)
[11]
1.0004
1.0002
[12]
1
0.9998
[13] [14]
0 1 2 3 4 5 6 7 8 9 10
0.9996
0.9994
[15]
time (s)
Fig. 7. Frequency for Generator 4 with three phase fault at Bus 5
A Novel AC-AC Converter With Minimum Snubber Requirement

Subhajyoti Mukherjee, Sunit Kumar Saxena, Tarak Saha Honeywell Technology Solutions Bangalore, India Sunit.Saxena@honeywell.com
AbstractMatrix converters inherently have problems with commutation which are generally mitigated by using snubbers. Problem of commutation also places a limitation on their voltage transfer ratio. In this paper, a new high frequency linked AC-AC converter topology is proposed which ensures reliable commutation, taking care of parasitic inductances, with minimum snubber on the input side. A simple switching algorithm to achieve this is also suggested. The topology has the flexibility to improve the voltage transfer ratio of a conventional matrix converter while maintaining unity power factor on the input AC side. Simulation results support the feasibility of the scheme. Keywords - ac-ac converter; high frequency link; zero current switching.
line side converter, the secondary side inverter is switched in such a way that its zero vectors are placed during the switching of the primary side converter. However, its also admitted in [6] that snubbers on transformer primary will still be needed to take care of the primary leakage. In this paper a topology is proposed, with switching scheme, which will minimize the snubber requirement while keeping all the desirable features of an AC-AC high frequency link converter intact. II. PROPOSED TOPOLOGY For the topology shown in Fig. 1, switching in Stage1 takes place when Stage3 is undergoing a freewheeling state. This theoretically ensures that during switching of Stage1 entire load current is freewheeling through Stage3 and Stage1 experiences a zero current switching. However, practically the transformer leakage will prevent such switching in Stage1 until snubbers are placed across transformer primary. To minimize snubber requirements, while maintaining unity input power factor for high frequency linked AC-AC converters, an additional H-bridge is introduced in the transformer primary side and a topology, as shown in Fig. 2, is proposed. The input stage is now composed of two parts - one converting the input AC voltage to DC (Stage1); the other (Stage2) converting this DC voltage to high frequency AC. The switching in Stage2 will be done in such a way that switching in Stage1 always occurs at zero current, while maintaining the volt-second balance across the transformer. A switching scheme to do this is also proposed. This will eliminate requirements of snubbers on the transformer primary. The bidirectional switch of the Stage2 ensures the continuity of current during switching in Stage2 and hence also eliminates snubber requirement for Stage2. Moreover, by introducing two stages, the control algorithm becomes simpler than that of topology shown in Fig. 1. In situations where unity input power factor is not desired, 4quadrant switches of Stage1 can be replaced by 2-quadrant switches, thereby reducing the switch count. Four stages of the proposed converters are described below: Stage1: This Stage converts the input AC voltage to a DC voltage such that a desired power factor is maintained at the input. Stage2: This Stage operates at a constant duty ratio D and converts the DC voltage available from Stage1 to a high frequency AC which is stepped up by the high frequency transformer.
I.
INTRODUCTION
A lot of research is currently being carried on in the field of Power Converters for direct AC-AC converter topologies without an intermediate DC link capacitor. The matrix converter is the one of the most deserving candidate in this field. Direct Matrix and Indirect Matrix converter are discussed in literature [1] [2]. However, complicated commutation [3] and involvement of snubber often prevents the achievement of the theoretical modulation index. One way to solve the commutation problem is to use a resonant ac link to achieve soft switching. One such recent example is in [4]. However resonant link converters are complex in design and subjected to several constraints. Moreover the switches in such converter are subjected to additional current or voltage stress. The commutation problem has been to an extent solved for an Indirect Matrix Converter in [5] where the output inverter stage goes through a freewheeling state during a switching of the input converter stage. However, increase in the voltage transfer ratio still remains a concern. This imposes a limitation when high voltage transfer ratio is desired. A simple solution is to include a transformer in the input or output side of the matrix converter. However, such transformer will be designed at power line frequency and thereby resulting in a bulky transformer. The best way to solve the problem is by using a high frequency transformer and this introduces the concept of AC-AC high frequency linked converters. High frequency linked DC-AC conversions have been extensively studied in literature but very few report on high frequency linked AC-AC converters are found in literature. One such example is in [6] as shown in Fig. 1. In [6], to eliminate snubbers on the input
Figure 1. AC- AC high frequency linked converter as proposed in [6].
Figure 2. Proposed AC- AC high frequency linked converter.
Stage 3: This Stage converts the stepped up AC voltage back to DC voltage. Its switching is same as that of Stage2. Stage 4: This is a standard voltage source inverter which converts the DC from Stage 3 to the required ac output voltage. Total number of 26 switches and 20 gate drivers are needed to implement the proposed topology. III. THEORY OF OPERATION To ensure zero current switching of Stage1, with minimum snubbers on the input and no snubber on the transformer primary, Stage2 has to be switched in a way such that a freewheeling mode (S7 and S9 ON or S8 and S10 ON) of Stage2 is synchronized with the switching of Stage1. A. Control of Stage1 To obtain unity power factor at the input while maintaining maximum positive dc output voltage, Stage1 has to be switched depending on the location of the input voltage vector. At any point of time only two switches, one of the top switches (S1 S3 or S5) and one of the bottom switches (S2 S4 or S6) of Stage1 needs to conduct. The position of the vectors for Stage1 is shown in Fig.3. Assume the voltage vector to be in sector1 for Stage1. The vectors AB and AC form the boundary of Sector1 and hence the input power factor is to be controlled by switching VAB (S1 and S4) or VAC (S1 and S6). The switch S1 of phase A is permanently ON in this sector and S4 and S6 are alternately switched within each switching cycle to obtain unity power factor.
Considering the input phase voltages as:
V AN = V p Sin A = V p Sinin t
VBN = V p Sin B = V p Sin(in t
(1)
2 ) (2) 3 2 VCN = V p Sin C = V p Sin(in t + ) (3) 3 Where, V p is peak of the phase voltage and in is the input supply frequency, the switching duration of S4 and S6 are given by
dB = dC =
sin B sin A sin C sin A
(4) (5)
where, d B + d C = 1 Similar values can be computed for other sectors.

Let X = d B be the normalized ON duration of the switch S4, as computed by the control algorithm for control of the input power factor. So for Stage1, S1 and S4 are ON for time 0 < tri < X and S1 and S6 are ON for time X < tri < 1 , where tri represents the normalized triangular carrier wave at an instant. Thus, there is a switching from S4 to S6 in Stage1 at the instant corresponding to X.
If, ref1 < tri < ref 4 , S9 ON and S10 OFF If, ref 2 < tri < ref 3 , S7 ON and S8 OFF Now, Stage2 undergoes the other freewheeling mode (S8 and S10 ON) when Stage1 undergoes a switching. The switching of Stage3 can be same as that of Stage2 or Stage3 can be switched at 50 percent duty ratio, switching alternately S11, S14 and S12, S13.
Figure 3. Stage1 voltage vectors and switches to be operated to achieve unity power factor.
B. Control of Stage2 and Stage3 In a normal DC-AC high frequency linked converter Stage2 will be switched at 50 percent duty ratio such that for half the switching period S7 and S10 are ON and for the next half of the switching period S8 and S9 are ON. To ensure zero current switching of Stage1, a freewheeling mode of Stage2 is to be placed around the instant corresponding to X. This can be done by choosing the references for Stage2 as ref1 = X (0.5 D / 2) (7) ref 2 = X (0.5 + D / 2) (8) ref 3 = X + (1 X )(0.5 D / 2) ref 4 = X + (1 X )(0.5 + D / 2) (9) (10)
Fig. 4 explains the switching scheme. The switching for Stage2 is done such that If, ref1 < tri < ref 4 , S7 ON and S8 OFF If, ref 2 < tri < ref 3 , S9 ON and S10 OFF Where, each of the switches in the pair S7, S8 and S9, S10 are complementary to each other. Thus a freewheeling mode of Stage2 (S7 and S9 ON) is ensured when switching of Stage1 occurs. Again to ensure the voltsecond balance of the transformer over one switching cycle, at point B the switching logic for Stage 2 is changed as
Figure 4. Proposed switching scheme.
C. Control of Stage4 Introduction of the freewheeling modes in the switching of Stage2 will make the corresponding portions of the transformer input voltage as zero. These zero portions will be carried forward and will appear in the output of Stage3. The output of Stage3 is input to Stage4 and the presence of the zero portions will cause distortions in output current. To avoid such distortion, the zero vectors ((000) or (111)) of Stage4 have to be synchronized with the freewheeling mode of Stage2. Thus, during the switching of Stage1, load current is freewheeling through Stage4 and transformer primary current through Stage2. The freewheeling mode of Stage2, when Stage1 is switched, ensures that the primary current (magnetizing current and leakage current for this particular case) always has a path to flow. This implies that no snubbers are needed on transformer primary and also ensures that Stage1 is always switched at zero current. To avoid nonlinear distortions in the output current, the zero vector of Stage4 has to be synchronized with the freewheeling states of Stage 2 and this can be easily done following [5] [6]. Space vector modulation is used to control the Stage4 converter. The voltage vectors are shown in Fig 5. Let the output voltage vector be in sector1 where the active vectors (100), (110) and the zero vectors (000) and (111) of Stage4 is to be used to control the output voltage and frequency. Let T0, T1 and T2 be the computed time intervals where the vector (100) is to be applied for time T1, vector (110) for time T2 (000) and (111) each for time T0/2. These time intervals T0, T1 and T2 are computed by the control algorithm depending on the output voltage and frequency requirements. For a required output voltage of magnitude Vo 3Vo sin( ) T1 = 2Vd 3
3Vo sin 2Vd T0 = TS T1 T2 T2 =
Figure 5. Voltage vectors of Stage4.
become (000)-(100)-(110)-(111)-(110)-(100)-(000)-(100)(110)-(111)-(110)-(100)-(000). The zero vector (111) of Stage4 is now synchronized with the freewheeling state of Stage2 which is in turn aligned with the switching instant of Stage1. Simulation results of Fig.6 confirms this. Similarly the zero vector (000) of Stage4 can also be aligned with the freewheeling state of Stage2. Implementation of this switching algorithm will introduce the following constraints on the magnitude of the output voltage. a) The introduction of freewheeling modes of Stage2 will bring down the average dc value available for Stage4. b) The duration of the application of zero vector of Stage4 has to be longer than the duration of the application of the freewheeling mode of Stage2. This will limit the maximum achievable modulation index of Stage4. But these two drawbacks can be compensated by increasing the turn ratio of the high frequency transformer by the required amount. IV. SIMULATION RESULTS The proposed topology, suggested in Fig. 2, is simulated, in open loop to verify the proposed switching, using MATLABSIMULINK under the following conditions Input phase voltage 115 V rms Switching frequency 9.6kHz Transformer ratio 1:2 Load Resistor 7 Load Inductor 1.5mH Fig. 6 and Fig. 7 shows simulation results with input frequency 400Hz and output frequency 400Hz. Fig. 4 shows that a switching of Stage1 (from state1 to state0 or vice versa) occurs when Stage2 and Stage4 each are undergoing a freewheeling state with (S7 and S9) or (S8 and S10) conducting in Stage2 and (S15 S17 S19) or (S16 S18 S20) conducting in Stage4.
(11) (12) (13)
And Where, 0 < <
, Vd is the input dc voltage and TS is one
switching interval. Let, ref 5 = X * T0 / 2 ref 6 = X (T0 / 2 + T1 ) ref 7 = X (T0 / 2 + T1 + T2 ) ref8 = X + (1 X ) * T0 / 2 ref 9 = X + (1 X )(T0 / 2 + T2 ) ref10 = X + (1 X )(T0 / 2 + T1 + T2 ) The switching for the inverter phases is such that If ref 5 < tri < ref10 S11 ON, S12 OFF If ref 6 < tri < ref9 S13 ON, S14 OFF (14) (15) (16) (17) (18) (19)
If ref 7 < tri < ref8 S15 ON, S16 OFF The switching is explained in Fig.4. The normally used switching pattern (000)-(100)-(110)-(111)(111)-(110)-(100)-(000) in one switching cycle will now
1.5
Stage1 Switching
1 0.5 0 -0.5 1 0 2.21 2.22 2.23 2.24 2.25 2.26 2.27 2.28 2.29 -1 1.5 1 0.5 0 1.5 -0.5 1 2.21 2.22 2.23 2.24 2.25 2.26 2.27 2.28 2.29 2.3 x10
-3
Stage2 output
2.21
2.22
2.23
2.24
2.25
2.26
2.27
2.28
2.29
2.3 x10
-3
2.3 x10
-3
Fig. 8 shows simulation results with input frequency 50Hz and output frequency 400Hz. Nomenclature of traces is identical as Fig. 7. Fig. 9 shows simulation results with input frequency 400Hz and output frequency 50Hz. Nomenclature of traces is identical as Fig. 7. The input line current can be smoothened by placing an input filter to eliminate switching frequency ripple. It can be understood from the fourth traces of Fig. 7, Fig. 8 and Fig. 9 that the fundamental component of input line current is in phase with the input voltage, making the input power factor to be unity.
300 O utput of S tage1 (V ) 200 100 0 600 O utput of S tage3 (V ) 400 200 0 50 Load c urrent(A ) 0 -50 0 Input V oltage (s c aled)(V ) and Input C urrent(A ) 50 0 -50 0 0.002 0.004 0.006 0.008 0.01 0.012 time(sec) 0.014 0.016 0.018 0.02 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Swb
Swa
0.5 0 1.5 -0.5 1 2.21 2.22 2.23 2.24 2.25 2.26 2.27 2.28 2.29
2.3 x10
-3
Swc
0.5 0 -0.5 2.21 2.22 2.23 2.24 2.25

tim e (se c)
2.26
2.27
2.28
2.29
2.3 x10
-3
Figure 6. (Top to bottom) Switching of Stage1, Output of Stage2 (1 - S7 and S10 ON, 0 - S7 and S9 ON or S8 and S10 ON, -1 - S8 and S9 ON) Swa - Switching of Phase a of Stage4: 1 - top switch (S15) ON, 0 - bottom switch (S16) ON. Swb - Switching of Phase b of Stage4: 1 - top switch (S17) ON, 0 - bottom switch (S18) ON. Swc - Switching of Phase c of Stage4: 1 - top switch (S19) ON, 0 - bottom switch (S20) ON.
300 O utput o f S tage 1(V ) 200 100 0 600 O utput of S ta ge3 (V ) 400
9 x 10
10
-3
Figure 8. (Top to bottom) Output of Stage1 (V), Output of Stage3 (V), 3phase output load current (A), scaled down input phase voltage (red) and current (blue).
300 O u tp ut of S tag e 1(V ) 200 100 0 600 0 O u tp ut o f S ta ge 3 (V ) 400 200 0 50 L o ad C urren t(A ) 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
200 0 50 3 4 5 6 7 8 9 x 10 10
-3
Loa d c urre nt(A )
0 -50 3 4 5 6 7 8 9 x 10 10
-3
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
0.018
0.02
Inp ut v oltage (s c ale d)(V ) an d Input c urrent(A )
50 0 -50 3 4 5 6 time(sec) 7 8 9
10 x 10
-3
0 -50 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 50 0 -50 0 0.002 0.004 0.006 0.008 0.01 0.012 time (sec) 0.014 0.016 0.018 0.02
First trace in Fig. 7 shows the DC output voltage after Stage1 and second trace shows the DC output voltage after Stage3. Third trace shows the 3-phase sinusoidal output currents with the R-L load. Fourth trace shows the input phase current (without input filter) which is in phase with the input voltage (sinusoidal).
In pu t v o lt ag e (s c ale d)(V ) and In pu t c u rre n t(A )
V.
CONCLUSION
A minimized snubber AC-AC converter topology is proposed and a switching algorithm is also proposed to control the same. The Proposed AC-AC converter introduces an additional H-bridge in the main power pass which, without much impacting the efficiency of operation, provides a robust mechanism of current commutation in AC-AC converters. It promises a deterministic design for reliable AC-AC converter, independent of stray parameters. Simulation results show the verification of the proposed control algorithm. The proposed switching algorithm ensures no snubber requirement on transformer primary. The presence of the high frequency transformer gives flexibility over the voltage transfer ratio. Further, the essential features of matrix converter like, no DC capacitor and bidirectional power flow with unity input power factor is present in the scheme. The topology can be used for applications which require controlled output voltage/frequency operating with input supply of varying voltage/frequency nature like Variable Speed Constant Frequency (VSCF) systems or applications where constant input
voltage/frequency source is used voltage/frequency motor drive applications. REFERENCES

[1]
for
variable
[2]
[3]
[4]
[5]
[6]
J.W.Kolar, T.Friedli, J.Rodriguez, P.W Wheeler, Review of Three Phase PWM AC-AC Converter Topologies IEEE Trans on Ind Electronics, Nov 2011.vol.58, issue 11. pp 4988-5006. P.Wheeler, J.Rodriguez, J.C Clare, L.Empringham,A.Weinstein, Matrix Converters: a technology review IEEE Trans on Ind Electronics, April 2002,vol.19, issue 2. pp 276-288. B.H Kwon, B.D Min, J.H Kim, Novel commutation technique of ACAC converters IEE Proceedings on Electric Power Applications,July 1998, vol.145, issue 4, pp 295-300. M.Amirabadi, A.Balakrishnan, H. Toliyat, W. Alexander, High Frequency ac-Link PV Inverter IEEE Trans on Ind Electronics, Feb 2013.vol.PP, issue 99. pp 1. L.Wei, T.A Lipo, A novel matrix converter topology with simple commutation. Industry Applications Conference, Chicago,USA, Sept 2001.Vol3. pp- 1749-1754. H.J Cha, P.N Enjeti, A three-phase ACAC high-frequency link matrix converter for VSCF applications. Power Electronics Specialist Conference, June 2003.vol.4,pp.1971-1976.
Battery/UC Hybridization for Electric Vehicles via a Novel Double Input DC/DC Power Converter
Furkan Akar
Duzce University, Technology Faculty Electrical and Electronics Engineering Duzce, Turkey furkanakar@duzce.edu.tr
Bulent Vural
Yildiz Technical University Electrical Engineering Istanbul, Turkey bvural@yildiz.edu.tr
Abstract In this work, for electric vehicles (EVs), a novel double input DC-DC power converter, that enables utilization of a battery and ultra-capacitor (UC) in parallel while increasing the overall performance of electric vehicles and recovering regenerative breaking energy, is introduced. Since UCs have higher power density when compared to batteries, by the use of a UC as a power input, DC bus voltage regulation at transients and peak loads is achieved easily, thus the life of the battery and efficiency of the system are increased. The average value model of the proposed converter is created in MATLAB, Simulink and SimPowerSystems environment, then its dynamic performance is tested under the load determined from the ECE15 drive cycle. Keywords-DC/DC converter, electric vehicles, hybrid electric energy systems, li-ion battery, ultra-capacitor
I. INTRODUCTION The fact that the global warming which has become a serious threat for the world, and the increasing prices of fossil fuels have accelerated the development of environmental friendly and high efficient vehicles that use alternative energy sources. Pure electric vehicles are the most popular technology among the others [1]. The batteries have higher energy densities however lower power densities when compared to the capacitor based energy storage systems [2]. Due to the fact that EVs require high power especially during the acceleration, in the case of using only a battery as a single power source in an EV, the performance and comfort of the vehicle are influenced adversely; moreover, a decrease in the battery life is inevitable. So, a hybrid system with a battery and another energy storage device needs to be formed. The batteries and UCs are foremost energy storage devices in hybrid EVs [3]. Unlike the batteries, UCs are known for their high power densities while suffering from their low energy storage capabilities [4]. By utilizing batteries and UCs simultaneously, a system that gathers advantages of these two devices can be developed. The performance of a hybrid system utilizing two different energy storage devices depends on the performance of the converter used in this system. Therefore, it is critical to have a proper DC-DC converter that can enable power transfer from the input energy sources to the output along with load sharing.
In [5], a hybrid system with a unidirectional converter and a bidirectional converter is developed. It can be seen that this topology allows load sharing despite of unsatisfying DC bus voltage regulation. In [6], a multi-input converter is introduced; this converter enables the input energy sources to power the output individually or simultaneously via a multi winding transformer. However, this kind of isolated DC-DC converter topology has several drawbacks, such as, complicated circuit structure, high cost, and the leakage inductance of transformer. In order to overcome these issues, a double input bidirectional DC-DC converter that utilizes a coupled inductor is given in [7]. This converter shows a really good performance on DC bus voltage regulation and load sharing while a fuel cell and a UC are used as input energy sources. However, it can be easily stated that this converter cannot recover the regenerative breaking energy due to the diode located at its output; in this work, it is aimed to come through this disadvantage, by adding a switch parallel to the aforementioned diode. Moreover, a reverse current diode is added and the diode that protects the fuel cell is removed. After these modifications, a novel double input bidirectional DC-DC converter having the capability of recovering regenerative breaking energy by both inputs is developed. As given in Fig.1, the battery and UC can be charged and discharged according to the state of switches, the instantaneous voltages and power of the inputs and output.
Figure 1. The topology of the proposed DC/DC converter.
TABLE I OCELL 8Q1P LI-ION BATTERY PARAMETERS Parameters Nominal Capacity Nominal Voltage Figure 2. The winding diagram of coupled inductor Maximum Charge Voltage Discharge Cut-off Voltage Weight Approximate Resistance TABLE II BMOD165 UC PARAMETERS Parameters Capacitance Maximum ESR Nominal Voltage Maximum Peak Current Leakage Current Maximum Charge Voltage Values 165 F 6.3 m 48 V 51 V 1900 A 5.2 mA Values 110 Ah 100 V 116.8 V 80 V 133 kg 60 m
II.
SYSTEM DESCRIPTION AND METHODOLOGY
A. Converter topology and its operating principle The proposed converter topology is shown in Fig. 1. Here it can be seen that the input sources can feed the output simultaneously. In addition, in the case that one of these sources fails or disconnects, other can continue its operation. For the sake of control simplicity, Q3 switch is hold closed during the simulation, therefore it is assured that the regenerative breaking energy is recovered whenever possible. Furthermore, the duty cycle of Q0 is set to 0.5 because of a possible over current problem. According to the duty cycles of Q1 and Q2 switches, the proposed converter can operate in buck-boost modes in both discontinues inductance current (DIC) mode and continues inductance current (CIC) mode. DIC mode decreases reverse-recovery loss and switching loss, as CIC mode decreases conduction loss and input current ripple; however, CIC mode increases the core losses. Thus, the inductors should be coupled as demonstrated in Fig. 2 [7]. The mutual inductance and leakage inductance values of a coupled inductance can be computed according to (1)-(3),

(3) = where, k is the coupling coefficient, LM is the mutual inductance, L1 are L2 self inductances, L1` and L2` are leakage inductances. The proposed converter can be considered the combination of two boost converter and a one buck converter. Based on this observation, the output current equation can be written as is the sum of currents flowing through the (4). Here, leakage inductances IL1` and IL2`. =
(1) (2)
Finally, the average value model of proposed converter can be constructed as shown in Fig. 3. As can been seen here, in this modeling technique, the switches are replaced by controlled current and voltage sources. Despite of decreasing the practicalness of the simulation model due to the ignored switching dynamics and losses, this method increases the speed of simulation, and allows long time analysis. It is also worthy to note that in this technique, the inductances are represented by resistors because sudden changes in the currents are ignored. Furthermore, from Fig.3, one can see that the currents flowing through the both leakage inductances are bidirectional. In other words, the both input sources can transfer their energy to the output or the other sources, while they can recover the regenerative breaking energy. B. Battery and UC modeling According to the parameters given in Table I, the equivalent capacitance of the battery can be calculated. First, the battery energy in joule is computed as 39.6 MJ via (5). Then, the capacity can be found as 10kF via (6) for Vmax (116.8 V) and Vmin (80 V). In addition the serial resistance of battery is neglected in the simulation. Note that, a detailed battery model is not used here to save time. 2 (6) In (5), E denotes the energy in joule, is the nominal capacity in Ah while is the nomival voltage. In (6), C is the equivalent capacitance, Vmax and are maximum and minimum battery voltages, respectively. UC is modeled based on the parameters in Table II as shown in Fig. 4. Here, ESR is the equivalent serial resistance while Rp denotes the parallel resistance that models the UC self-discharge; it can be easily computed as 9.23 k by dividing the nominal voltage to the leakage current. = = 3600 (5)
(4)
Figure 3. The average value model of the proposed power converter
UC+
UC-
100 50
Iload (A)
Rp
ESR
0 -50
50
100 (a)
150
200
Figure 4. UC electrical equivalent circuit model

235
VUC (V)
III. SIMULATION AND RESULTS In the designed system, an ultra-capacitor bank composed of three UCs whose parameters given in Table II is created for a successful DC bus voltage regulation. Since UCs are connected in serial, the UC bank equivalent capacitance, and voltage will be 55 F, and 120 V, respectively. Additionally, the initial voltage of battery is assumed to be 100 V. The battery/UC hybrid system used in the simulation is depicted in Fig. 5. As can be seen from this figure, it is aimed that the designed hybrid system operates to meet the required amount of energy of ECE-15 drive cycle. Here, DC bus voltage reference is determined 188 V [8]. In this work, in order to show the some of the operation modes of converter, a simple control algorithm is used. According to the this algorithm, UC bank current is controlled to regulate the DC bus voltage via one of the PI controllers that controls the duty cycle of switch Q1. In addition, the duty cycle of Q2 is controlled by the other PI controller to ensure that the demanded power below 9 kW is supplied by the battery. As mentioned earlier, the duty cycles of Q3 and Q0 are kept constant at 1 and 0.5, respectively. A. Simulation Results In the Fig. 6, the variations in load current, and DC bus voltage versus the time are given when the demanded power is determined from the ECE-15 drive cycle. As can be seen from Fig. 6, the DC bus regulation is accomplished successfully after transients thanks to the high power density feature of the UC.
188 V bus (V) 141 94 47 0 0 50 100 (b) Time (s) 150 200
Figure 6. a) Load current, b)DC bus voltage
IUC (A)
50 25 0 -25 -50 0 20 40 60 80 100 (a) 120 140 160 180 200
125 120 115
20
40
60
80
100 (b)
120
140
160
180
200
0.9 dS1 0.8 0.7 0 20 40 60 80 100 (c) Time (s) 120 140 160 180 200
Figure 7. a) UC current, b) UC voltage, c) S1 duty cycle
Fig. 7 shows the variations in UC bank current, voltage and duty cycle of Q1. According to this figure, it can be easily seen that UC voltage usually tends to increase. In other words, it is typically charged because of the regenerative breaking energy. Moreover, UC current reaches about 20 A around 140th second since the demanded power is more than the maximum output power of battery in this period. At this stage, UC bank comes into play, and feeds the output for DC bus regulation.
Figure 5. The block diagram of the whole system
100 Ibattery (A) 50 0 0 20 40 60 80 100 (a) 120 140 160 180 200
100 Vbattery (V) 99.9 99.8 99.7 0 20 40 60 80 100 (b) 120 140 160 180 200
Fig. 9 presents the changes in the output power according to ECE-15 drive cycle, battery power and UC bank power. Here, one can see from Fig. 9-b that battery power is always positive, and less than 9 kW due to the control strategy, and the fact that the battery is never charged. Moreover, according to Fig. 9-c, UC bank power takes negative values when the regenerative breaking energy is available; this figure also shows that in the periods when the demanded power is higher than 9 kW, UC bank discharges. IV. CONCLUSIONS This paper presents a novel double input DC\DC power converter that forms a hybrid energy storage system which consists of a battery and an ultra-capacitor bank for EVs. Associated switches are controlled in a way that the output power less than 9 kW is supplied from the battery, and UC bank energy is transferred to the output when the battery is insufficient. In addition, UC bank recovers the regenerative breaking energy. Simulation results demonstrate that the proposed converter topology shows a good performance on both holding the DC bus voltage constant, and recovering the regenerative breaking energy. Besides, the peak power loads and transients do not affect the battery; by this way, it can be possible to increase its lifetime and efficiency. Due to the simplicity of the control, all operation modes of the converter cannot be shown in this work. For example, the regenerative breaking energy is only captured by the UC bank despite the fact that the proposed converter can enable a power transfer from the output to the battery as well; in the future work, it is targeted to improve the control strategy, and demonstrate the power transfer in all directions. REFERENCES
[1] [2] C. C. Chan, The state of the art of electric and hybrid vehicles, in Proc. 18th Annu. IEEE Conf. Computer Security Application, Washington, 2002, pp. 245-275. A. F. Burke, Batteries and ultracapacitors for electric, hybrid, and fuel cell vehicles, Proceedings of the IEEE, 2007, pp. 806-820. G. Wang, P. Yang, J. Zhang, Fuzzy Optimal Control and Simulation of Battery-Ultracapacitor Dual-Energy Source Storage System for Pure Electric Vehicle, International Conference on Intelligent Control and Information Processing, August 13-15, 2010 Khaligh A, Li Z. Battery, ultracapacitor, fuel cell, and hybrid energy storage systems for electric, hybrid electric, fuel cell, and plug-in hybrid electric vehicles. ,IEEE Trans Veh. Technol 2010;59(6):2806e14. O. Erdinc, B. Vural, M. Uzunoglu, A wavelet-fuzzy logic based energy management strategy for a fuel cell/battery/ultracapacitor hybrid vehicular power system, J Power Sources 2009;194(1):369e80. B. Vural, O. Erdinc, M. Uzunoglu, Parallel combination of FC and UC for vehicular power systems using a multi-input converter-based power interface, Energy Convers Manag 2010;51(12):2613e22. B. Vural, FC/UC hybridization for dynamic loads with a novel double input DC-DC converter topology, International Journal of Hydrogen Energy 38 (2013) 1103-1110 B. Vural, O. Erdinc, M. Uzunoglu, Parallel combination of FC and UC for vehicular power systems using a multi-input converter-based power interface, Energy Conversion and Management
1 dS2
0.5
20
40
60
80
100 120 (c) Time (s)
140
160
180
200
Figure 8. a) Battery current, b) battery voltage, c) S2 duty cycle
15000 (W) 10000 5000 0 -5000 0 20 40 60 80 100 (a) 120 140 160 180 200
P
battery
Load
15000 (W) 10000 5000 0 -5000 0 20 40 60 80 100 (b) 120 140 160 180 200
[3]
15000 (W) 10000 5000 0 -5000
[4] [5]
0 20 40 60 80 100 120 (c) Time (s) 140 160 180 200
UC
Figure 9. a) Output power, b) battery power, c) UC power
[6] [7]
The battery current, voltage and duty cycle of Q2 switch are demonstrated in Fig. 8. As one can notice from this figure that the maximum current of battery is approximately 90 A due the fact that its maximum power is set to 9 kW. Furthermore, it seems that the battery is never charged since the duty cycle of Q2 is higher than the one of Q1. However, based on a different control strategy, it can be possible to charge the battery as well.
[8]
Thyristor Controlled Rectifiers (TCR) for traction problems and solutions

Vitaly Gelman
VG Controls Vernon, NJ USA vgelman@vgcontrols.com
AbstractThe paper discusses specifications for the Thyristor Controlled Rectifiers (TCR) in traction applications. Covered are TCR voltage selection, output filter requirements, harmonics, reactive power, power circuit configuration, power recuperation and cooling. Compares TCRs and diode rectifier usage in traction. It summarizes TCR requirements specific for traction. Keywords - traction; Thyristor Controlled Rectifiers;TCR; DC filtering; AC harmonics; voltage selection; reactive power; recuperation
voltage regulation at higher current) limits Traction Power Sub-Stations (TPSS) spacing to about 1 mile. We have more leeway with selection of TCRs output voltage because TCR regulates the DC voltage to compensate for changes in incoming line voltage and load current. US transportation authorities specify TCR rated voltage from 825 to 850 VDC; this voltage stays constant with load current changing from 0% to an inflection point (selected anywhere from 100% to 300% current). The lower voltage (e.g. 825 V) would leave more margins for regenerative braking power transfer. Assuming linear diode rectifier loading curve with 6% regulation and 750 V nominal voltage and a TCR with 825 VDC nominal voltage and 150% inflection point, we arrive at Figure 1. As you can see from Figure 1, there is TCR voltage gain of 125 V for loads above 150% comparing to Diode rectifier.
I.
INTRODUCTION
There is a significant interest to the TCRs from the transportation authorities yet the majority of traction projects use diode rectifiers. The Thyristor Controlled Rectifiers (TCR) advantage compared to diode rectifier is higher train voltage, but this comes at a cost: we need a controller to regulate the voltage and an output filter to reduce the DC voltage ripples. There are very few publications comparing diodes and TCRs for traction and no standards or specifications for traction TCRs; this paper attempts to fill the gap. We will show how TCRs improve train voltage regulation, reduce system cost by increasing traction substation spacing and provide energy savings through the power recuperation. This paper presents analysis of TCR power circuit options and summarizes the requirements for traction rectifiers. II. TCR VS. DIODE RECTIFIERS, OR WHY DO WE NEED TCR
International Standard IEC1287-1 (1995) Power converters installed on board rolling stock[2] specifies the train equipment shall operate normally at 1.2 nominal voltage. US standard nominal train voltage is 750 VDC, such train shall operate normally up to at 1.2*750=900 VDC. What is the maximum nominal voltage of diode rectifier to operate 750 V trains? (1) Using standard US utility contract specifying 10% incoming AC voltage regulation, the maximum no load voltage should be 900/1.1=818 VDC. (2) Using 6% regulation (typical in traction) we get nominal voltage of 818* 0.94=769 VDC. (3) To allow for 2% ripples we get 769*0.98=754 VDC In practice the nominal voltage for Diode rectifier is specified as 750 V for 750 V trains. This voltage (together with
Figure 1. Voltage regulation of Diode rectifier and TCR
The higher TCR voltage allows us to increase TPSS spacing and reduce their number with corresponding savings, and improve train performance see [1], [4] and [5] for details. Many new transportation systems use 1500 V trains. Since all voltages are proportional to the rated train voltage, for the 1500 (=750*2) trains we can use the rated TCR voltage in the range of 1650 to 1700 V (double the 825 to 850 V). Consequently for 1500 V trains with similar assumptions as for we arrive at 250 V voltage gain for TCRs, leading to similar reduction in TPSS quantity.
Additional advantage is fault current limiting by the TCR for both close and far faults to 300 450%, depending on design. This reduces stresses on the transformer and the breakers, extending service life. The separate issue of output capacitor transient fault current is discussed below. III. BASIC TRACTION RECTIFIER REQUIREMENTS
Without output filter the TCR peak voltage exceeds 938 VDC for 12 pulse mode and 1000 VDC for 6 pulse for typical TCR parameters. Increasing incoming AC line voltage by even 5% will bring peak voltage to 985 Vpk (12 pulse) or 1050 Vpk (6 pulse). This is well above allowable train voltage 900 Vpk. Solution to the TCR peak voltage problem is installing large output filter capacitor to eliminate output ripples and satisfy IEC1287 requirements. We have the following requirements for selecting the output capacitance: The resonance frequency between the transformer inductance and filter capacity should be well below the six harmonic of the mains (e.g. 360 Hz for 60 Hz system) to avoid parallel resonance in 6 pulse operating mode and control circuits oscillations. The capacitor current rating should be below actual ripple current.
From my experience the traction applications require from all types rectifiers the following: Ability to withstand short circuits. Nominal Power is 1 to 5 MW. Typical nominal rectifier power is 3 MW. For 850 V nominal voltage this corresponds to rated current of 3.5 kA, for 1700 V nominal voltage rated current is 1.75 kA. Up to 300 to 400% overload. For 3MW rectifier 400% current translates into 14 kA DC for 850 V nominal voltage and 7 kA for 1700 V. Nominal DC train voltage of 750 or 1500 VDC.
It is possible to use both electrolytic and film capacitors, but at 850 V and above the film capacitors are more practical: Film capacitors have high voltage rating; electrolytic capacitors are limited to 450 V, so they would require series connection for 850 V. Film capacitors have lower conduction losses, thus they have higher current rating. Film capacitors have service life at least 10 times longer than electrolytic one. Electrolytic capacitors are limited to few thousand hours service life at maximum temperature, to extend service life we need to de-rate the current rating and provide good cooling. Film capacitors are available in large sizes; electrolytic capacitors are not. To satisfy voltage and current requirements with electrolytic capacitors we will need both parallel and series connection of many small units.
For diode rectifiers the nominal output voltage is 750 or 1500 V, respectively. For TCRs the nominal output voltage is 825 to 850 V for 750 V systems and 1650 to 1700 V for 1500 V systems. IV. OUTPUT RIPPLES AND AC INPUT HARMONICS
A. Do we need an output filter? The Standard IEC1287-1 (1995) Power converters installed on board rolling stock specifies allowable train overvoltages as a function of pulse duration (see Figure 2).
Figure 2. Allowable train overvoltage
For the above reasons the film capacitors are preferable for traction TCR output filters. For a typical 3 MW, 850 V TCR we will need about 20 mF to avoid problems with resonance. At this capacitance level and 1000 V voltage, the film capacitor current rating exceeds 1000 A, few times higher the actual capacitor ripple current. Also the output ripples are reduced well below the diode rectifier ripples. Conclusion: to utilize the TCR advantages of increasing TPSS spacing and improved train performance without jeopardizing trains equipment, we need to use output filter, preferable parallel film capacitor. B. Is capacitive filter safe for the circuit? Somebody might claim that using a capacitor filter is very dangerous because a short circuit will lead to an extremely large fault current and will destroy everything in its path. In reality the peak fault current I pk
The peak overvoltage should be less than 1.2 nominal voltage for pulses with duration longer than 1 ms. The higher TCR voltage (see above) allows us to increase TPSS spacing and reduce their number with corresponding savings, but this brings increased peak voltage problem: increasing TCR DC voltage requires higher rectifier transformer secondary AC voltage, so TCR peak voltage is higher comparing to diode rectifier. Also, because of the delayed firing required for voltage regulation, TCR has higher output ripples than diode rectifier.
I pk
Vd
where:
Vd is DC voltage at the time of fault
L C
is a characteristic impedance,
L and C are total loop inductance and filter capacitance, respectively If we have a fault just at the feeder breaker terminals outside the substation, the loop inductance will be still few uHy. Even assuming it to be just 2 uHy (very close positive and negative buses inside the TCR and throughout the substation) and using an output capacitor of 20 mF, we get
L C 2 20000 0.01
TCR substation (12 pulse, parallel bridges, capacitive filter, no IPT) with network short circuit power of 125, 190 and 360 MVA (denoted as x1 of 1.53, 1 and 0.63, respectively). These are the actual system parameters for one of the TCRs installations in the US.
Assuming the TCR voltage peak current
Vd is 850 V, we arrive at the

85 kA
Figure 3. Voltage distortion of 3 MW TCR with different system impedances
I pk
Vd
850 0.01
This current is actually decaying AC current with the frequency of

f 1 2 LC 795 Hz
Even if we assume the fault circuit has high Q and the fault current will not decay substantially by the time the feeder breaker contacts separate, the peak current is well within DC breaker interrupting capability (over 120 kA per ANSI C37.14 and C37.16) and it is an AC current, it is easier to interrupt than DC. So the typical feeder DC breaker will have no problem interrupting the fault current right at the feeder breaker terminals. In a more practical situation with the fault occurring at the tracks, the loop inductance will be above 50 uHy; inductance increase by a factor of 25 will decrease the capacitor current and its frequency by a factor of 5. Thus the capacitor fault current will be only 17 kA with 160 Hz frequency; it will be even easier to interrupt. Additional argument is a small amount of energy stored in the TCR output filter. At 850 V a 20 mF capacitor has energy
Ecap CV 2 20 10 3 8502 7.2 kJ 2 2
We can see that 12 pulse TCR harmonics are well below IEEE519 requirements (3% THD vs. 5% limit), even for a weak system with 125 MVA short circuit power. We can see even 6-pulse operation yields voltage THD below IEEE519 requirements of 5%. The distortion for stronger system is much lower. To verify the simulation, we measured the actual 12 pulse TCR voltage and current harmonics and compared them with simulated; the results are shown in Figure 4. We can see very good correspondence between measured and simulated values. The only difference is high value for the 5th and 7th voltage harmonics. The last two columns show system impedance calculated from the measured voltage and current (ZsysCalc) and theoretical system impedance based on system short circuit power (ZsysThr). We can see the good correspondence between ZsysCalc and ZsysThr for high order harmonics but large difference for the 5th and 7th harmonics, indicative for the resonance around 5th harmonic. Further investigation revealed a capacitor bank connected in parallel to the 13 kV line without series inductors, apparently it was creating a parallel resonance.
Traction DC breaker can dissipate over 100 times higher energy when interrupting 100 kA DC fault. Conclusion: TCR output capacitor doesnt create dangerous transient fault currents. The current is limited by loop inductance, is zero crossing and should be cleared by traction DC breaker with no problem. C. How big are TCR AC harmonics? TCRs have higher harmonics contents than diode rectifiers due to delayed firing required for voltage regulation. However the voltage distortion for 12 pulse TCR is well below IEEE519 standard requirements of voltage THD (Total Harmonic Distortion) to be below 5% with individual harmonics below 3%. Figure 3 shows simulation results for THD for 2.5 MW
Figure 4. Comparison between measured and calculated TCR harmonics
Figure 5 shows TCR voltage and current oscillograms at 67% load, we can see the voltage distortion is low.
Assuming +10% over-voltage and customary 100% of SCR voltage margin, we arrive at required SCR voltage of 4400 V. There are plenty of high current SCRs available at voltage above 4400, standard devices go above 6500 V. Conclusion: we should use the parallel connection of the bridges. C. Do we need an IPT? Inter Phase Transformer (IPT) is connected between the parallel rectifier groups, in our case between two bridges, to reduce circulating currents. While the IPT are necessary for single-way rectifiers (e.g. 3 phase double WYE circuit 45), the 12 pulse double-way systems have the circulating current limited by the rectifier transformer impedance. Loosely coupled standard Wey-Delta rectifier transformer sufficiently limits the circulating current to eliminate the need for an IPT. Successful operation TCRs without IPT (e.g. DART) proves the IPTs are not needed. Figure 7 shows the power circuit of 12-pulse reversible TCR with parallel bridges connection, output capacitive filter and no IPT. Conclusion: we do not need an IPT. VI. POWER FACTOR AND REACTIVE POWER
Figure 5. 12 pulse TCR primary voltage and current at 67% load
The TCRs had been successfully operated at numerous locations causing no problems with either harmonics or power factor (in the US it is Dallas and Phoenix). V. POWER SECTION CONSIDERATIONS
A. Should TCR use 6 or 12 pulse mode? 12 pulse rectifiers allow us to eliminate half of the harmonics and reduce total harmonic distortion (THD) by about 30 to 40%, as we showed in the previous section. Since the required DC current determines total number of SCRs, the difference between 6 and 12 pulse systems boils down to marginal cost increase of the control system and about 5% increase in the rectifier transformer cost. The cost of power filters to remove additional harmonics present for the 6-pulse system is much higher. Conclusion: we should use 12 pulse TCRs. B. Parallel or series bridge connection 12-pulse system is combined of two 6-pulse bridges that can be connected either in parallel or in series. With series connection we do not have to worry about current sharing between the bridges but the SCR losses are almost twice as big as for parallel connection. The DC current has to go through four SCRs for series bridge connection versus two SCRs for parallel connection. Parallel connection requires the SCR withstanding voltage to be twice as big and their forward voltage drop might be higher, but the difference in forward voltage drop is below 15% (one of the offsetting factor is lower current, under parallel connection each SCR is carrying half the current). So from the point losses reduction the parallel bridge connection reduces the losses by almost 50%. Additional advantage of parallel connection is redundancy: if one of the bridges is inoperable (cooling failure, multiple SCRs failure, etc.) the rectifier could continue to operate with single bridge at reduced current (and higher harmonics). In a rectifier with series bridge connection the failure of one bridge means failure of the whole rectifier. One might ask if we have to connect more that one SCR in series to achieve required voltage ratings. For a rectifier with 1700 VDC nominal output voltage we will need about 1400 VAC incoming voltage (margins for firing angle delay, etc.).
The reactive power is phenomenon related to a phase shift between AC voltage and current. The current component shifted 90 degrees from voltage transfers no power but it creates losses and it is a major source of AC voltage change. It is customary to use power factor (a cosine of angle between voltage and current) as a measure of reactive current. Though it is adequate for steady loads, it is very misleading for wide changing loads such as traction where the substation operates for 15 seconds at 250%, then 70 seconds at 100% and then draws 10% power for the rest of 300 seconds cycle. The real negative effects of the reactive current are voltage sags (proportional to the current) and additional losses (proportional to the current square). If we have poor power factor at low power level both effects are negligible. It is more practical to use the reactive current directly or its product with voltage (called reactive power) to estimate the negative effects. A. Comparing reactive power for diode rectifier and TCR Lets neglect harmonics and assume for simplicity that all losses are located on the DC side. First lets consider diode rectifier with a 6% voltage regulation (industry standard). Lets further assume the resistive component provides 2% regulation and commutation inductance provides 4% regulation. Basic rectifier equations are:
P 3 V1 I1 cos ; Pd Vd I d 1 P Pd ; Id I1 1 6 Vd Vd 0 cos ; Vd Vd 0 xI d ; Vd 0 3 2
V1
Vd 0 Vdout Vd rI d ; Pdout Vdout I d
cos
Vd
where: P 1, P d , V1 , I1 , Vd and I d are primary power, DC power before taking losses, RMS values of primary voltage and current, and DC voltage before losses and DC current. Vd 0 , , Vdout and Pdout are no load DC voltage, angle between voltage and current, output DC voltage after losses, and output DC power after losses, respectively. x and r are inductive and resistive impedances, 4% and 2%, respectively. From these equations we can calculate output voltage Vdout and cos for any current and finally calculate reactive power Q as a function of output power Pdout.
negative effects of reactive current (voltage change and additional losses) are relevant only at high current where both diode rectifiers and TCRs behave similarly. Conclusion: TCR has higher reactive current (lower power factor) than diode rectifier but the difference is insignificant in the high current region. VII. ENERGY RECUPERATION A. Why do we need energy recuperation? All modern trains have regenerating capability and can supply mechanical energy to the DC bus during braking; we might get up to 80% of mechanical energy converted back to DC power at trains terminals. If there is another train accelerating nearby it can absorb the braking energy and reduce the overall system power consumption. Theoretically we could capture almost all recuperated energy if we would always have another train accelerated at the same location and at the same time our train is braking, but it is impossible to achieve in practice. With short headway of 2.5-3 min during peak hours the trains are close to each other and there are many of them. Consequently, the braking energy is transferred efficiently; the lost energy is small, 8% of total. With long headway of 30 min during off hours the lost energy could be up to 60% of the traction power supplied by traction substation [6, Table5]. Both simulation and actual operating experience in [6] show the total energy savings of about 8 to 11%. To utilize braking energy we can add recuperating converters to traction substations and transfer the recuperated energy to the power grid. Essentially we use the power grid as a huge energy storage system [7]. Since the power grid always has loads, it can easily absorb the recuperated braking energy. Many states now require utilities to accept the power from customers and bill them on the net energy (net metering). Acceptance of reversible TPSS by power utilities depends on their acceptance of distributed power generation. The issue is a matter of policy, not so much of technology. Since the power demand is typically measured as a running average over 15 minutes interval, recuperating substation should cut the demand charges as well, even for traction substation with individual metering because the time between train braking and acceleration is typically 30 sec, well below 15 min. The power demand reduction would be even more apparent for a system with a combined metering where AC distribution lines belong to the transportation authority (e.g. BART). B. How expensive is adding energy recuperation to TCR? The RTCR is actually two thyristor controlled rectifiers connected anti-parallel. By firing either one of the bridges (forward or reverse) we can transmit the energy either from AC to DC or in the opposite direction. Since we can choose to fire only one bridge at a time we can eliminate circulating currents without adding the inductors. Comparing to the regular TCR the modifications are adding reverse bridge; there is minimal addition to the control system to add reverse bridge functionality. Assuming we can use the same transformer for
Q1 3 V1 I1 sin 3 V1 I1 1 cos 2
It is important to calculate the reactive power as a function of output power (not the current) to compare different rectifiers because TCR has higher voltage at the same current and trains require power, not current.
1.600 1.400 1.200 1.000 0.800 0.600 0.400 0.200 0.000 0.0
Q*
TCR Diode
0.5
1.0
1.5
P*
2.0
2.5
3.0
Figure 6. Relative reactive power Q* as afunctin of output power for diode rectifier and TCR
We can write similar equations for the TCR, the only difference will be the equation for Vd
Vd Vd 0 cos xI d
where: is the firing angle. The TCR changes firing angle to keep the output voltage Vdout constant while the current changes from zero to inflection point. Above the inflection point the firing angle stays at minimum value (5 to 10 deg) to assure proper SCR firing. Figure 6 shows the relative reactive power Q* as a function of relative DC power P* for diode rectifier and TCR with 150% current inflection point
Q
*
Q Pd rated
P
*
Pdout Pd rated
where Pd rated is the rated output power (after the losses). We can see on figure 6 that the reactive power at low loads (below inflection point of 150%) is lower for the diode rectifiers due to the TCR delayed firing. But once we are above 150% the difference almost disappears. It is understandable because above inflection point the TCR operates with minimum firing angle, almost like a diode rectifier. The
both forward and reverse thyristor bridges, we can estimate the transformer power increased by about 15% for the additional current due to reverse power flow. We can use many common components (bus structure, cooling fans, etc.) between forward and reverse bridges. Though the number of semiconductor switches is doubled, the overall cost increase to add energy recuperation of about 35%. We can estimate cost of 3MW TCR plus rectifier transformer with 400% maximum current to be $440k, while 3 MW reversible TCR with transformer is $600k [7]. Figure 7 shows power circuit of reversible TCR. The harmonics and power factor in the recuperating mode are the same as in the rectification mode. If we use 12 pulse bridges for both power flow directions we keep the harmonics below IEEE519 limits for forward operation even at 400%. The reverse current is 100%, so the harmonics in reverse operation are lower than in the forward direction. Running the power in reverse direction should not cause any problems because the current is much lower.
Reverse Bridge
TCRs use forced air-cooling with the fans driven by variable frequency drives to keep their speed low at lower load. This way both fan noise and their service life are improved. Conclusion: we need to use forced air cooling. B. Traction Substation Cooling Since the TCRs require forced air-cooling the air going through the TCR must be clean. In urban environment the air quality is often poor with high amount of dust and other pollutants. We can try to use air filters at the TCR cabinet and/or at the substation level but then air filter changes becomes a major maintenance problem; in Dallas (DART) they had to change filters weekly. A better approach is to use close loop system: have redundant air conditioners to cool the whole substation. This eliminates filter changing maintenance problem and has additional benefits of running all substation electronics and other equipment in a clean environment at low temperature (e.g. 30oC). Conclusion: we need to use close loop system with redundant air conditioners to cool TCR substation. IX. CONCLUSIONS
Figure 7. Power circuit of 12 pulse reversible TCR
Conclusion: adding energy recuperating to the TCR is inexpensive and can be justified on the energy savings [7]. VIII. COOLING A. TCR cooling: Forced air or convectional? Modern diode traction rectifiers customarily use convectional air-cooling. This is practical because the diodes for traction rectifiers have high operating junction temperature of 160oC through 190oC. Assuming maximum ambient temperature of 40oC and a safety margin of 15oC, we arrive at the difference between junction and ambient of 190-40-15=135oC for 190oC diode, for 160oC diode the difference is 105oC. The TCRs use SCRs for rectification, their maximum junction temperature of 125oC. With the same assumptions we arrive at the difference between junction and ambient for SCRs as only 125-40-15=70oC. The lower junction-to-ambient temperature difference means the thermal impedance junction to ambient must be 30 to 50% lower for the SCRs than for diodes (additional factor here might be 5 to 15% higher losses in the SCRs). To decrease thermal impedance 50% while operating at lower temperature we need to at least double the number of heat sinks and rectifier footprint; this would be very costly. In practice the
The TCR for traction applications should be built as 12 pulse systems with parallel connection of two 6 pulse bridges and no IPT. Output capacitive filter is necessary to keep peak train voltage within standard IEC1287-1 requirements. Though both AC harmonics and reactive power of the TCRs are higher than of a diode rectifiers, they are within the IEEE519 standard requirements and do not represent the problems. Adding energy recuperation capability to the TCR is inexpensive and can be justified on the energy savings. The traction TCRs should use forced air cooling; TCR traction substations should use redundant air conditioners. ACKNOWLEDGEMENTS The author wishes to thank Moustapha Ouattara from WMATA for helpful discussion. REFERENCES
[1] V. Gelman, Braking Energy Recuperation (Reversible ThyristorContrlled rectifiers). September 2009, IEEE Vehicular Technology Magazine [2] The Standard IEC1287-1 (1995) Power converters installed on board rolling stock [3] IEEE-519: IEEE Recommended Practices and Requirements for Harmonic Control in Electric Power Systems [4] V. Gelman, S. Sagareli Implementation Of New Technologies In Traction Power Systems. Proceedings of JRC 2004: 2004 ASME/IEEE Joint Rail Conferences [5] V. Gelman, Thyristor rectifier for traction practical at last. Proceedings of 2002 APTA conference. [6] J.M. ORTEGA et al. Kinetic energy recovery on railway systems with feedback to the grid. 9th World Congress on Railway Research, May 22-26, 2011 [7] V. Gelman, Comparison between wayside storage and reversible thyristor controlled rectifiers (rtcr) for heavy rail applications. Proceedings of JRC 2013: 2013 ASME/IEEE Joint Rail Conference
Multi-Purpose RF Antenna for Partial Discharge and Oil Quality Monitoring

Ayman H. El-Hag, Nasser Qaddoumi, Rida Mourtada, Esam Al Murawwi, Ahmad Nimer, Kamal AlMazam, Mohammed Hirzallah and AbdelRahim Huwair
Electrical Engineering Department, American University of Sharjah, Sharjah, UAE
Abstract In this paper the possibility to use an RF antenna for both partial discharge (PD) detection and transformer oil quality monitoring is investigated. First, several RF antenna designs such as the dipole, axial helix and circular loop, have been compared to obtain the best optimum design for the detection of partial discharge. It was found that the circular loop antenna has shown the best capability and accuracy of detecting the partial discharge. Then, the selected design has been compared with a commercial acoustic emission (AE) sensor. The RF antenna showed better performance than the AE sensor in terms of high signal to noise ratio and ability to detect PD signal at an early stage. Finally, the RF antenna demonstrated the ability to differentiate between old and new oil samples in terms of their water content and dielectric constant. Index TermsPartial discharge, RF antenna, AE sensor and transformer oil.
I. INTRODUCTION Partial discharge (PD) is a small electrical spark that occurs within the electrical insulation of transformers, switchgears, cables, and windings in large motors and generators. Frequent occurrences of PD can lead to the deterioration of the electrical insulation and finally failure of the power system apparatus. Therefore, it is very important to detect PD for different power system components at an early stage and take the necessary actions [1]. Among the different power apparatus power transformer is the most expensive and vital equipment for network reliability. Therefore, efficient and cost optimized condition monitoring is highly demanded. Several PD on-line detection techniques have been developed to monitor the PD activities inside power transformers. These techniques can be classified as chemical, acoustic and electrical methods. Dissolved gas analysis (DGA) is considered as the most frequent chemical test where its results are highly correlated with different types of faults including partial discharge [2]. The major drawback of using DGA to monitor PD is its limited capability to detect PD online as it is usually conducted as time based maintenance practice. Moreover, DGA cant be used to locate PD sources. Alternatively, on-line PD detection using radio frequency (RF) antennas, acoustic emission (AE) sensors and high frequency current transformers (HFCT) have been used
extensively. The sensitivity of several monopole antennas inserted in transformer tank has been tested [3]. Conical monopole antenna has shown relatively better broadband response compared to other monopole antennas. Moreover, it has been found that increasing the antenna size significantly improves the antenna sensitivity [3]. Tang et al. used an antenna array to locate PD inside a single phase transformer model [4]. Two sensors and Hyperboloid-Genetic algorithm have been used successfully to find the PD location with total error of few centimeters. In a different application, RF antenna has been used to measure the water content and permittivity of different dielectric materials. A short monopole antenna probe connected to broadband vector reflectometer has been used to measure the permittivity for air, 1-Butanol, and Teflon [5]. The complex permittivity of saline solutions in the frequency range 725-1300 MHz has been measured using a monopole antenna. The measured data are in good agreement with those obtained by other techniques [6]. A dual monopole antenna model for a two-wire dielectric probe was used to measure the water content of brain cells. The measured results for water and saline are found to be in agreement with the expected results [7]. The objective of this work is to design and utilize a multipurpose RF antenna that can detect the partial discharge in transformer's oil and monitor the oil's quality. II. MATERIALS AND METHODS In order to generate PD, high electric field is required. A 15 kVA, 220V/20kV transformer connected to a variac was used to generate a variable high voltage at its secondary side which is connected to a sharp electrode, as shown in Figure 1. The distance between the sharp electrode and the ground tank was adjusted to 1cm which will result in a very high concentration of the electric field that will lead to PD generation. For the PD detection part, two sensors were used. First, an AE sensor was attached at the outer surface of the tank using a magnet. The sensor is connected to a preamplifier that amplifies the captured signal with a gain of 46 dB (25-80 KHz). The second sensor is an RF antenna that was inserted inside the tank as shown in the Figure 1. The antenna detects the electromagnetic waves resulting from the occurrence of PD. The measured signal passes through an amplifier with a gain of 20 dB and bandwidth of 1-150 MHz. A digital
oscilloscope was used to monitor the signal of the AE sensor and the RF antenna.
Figure 2: Un-calibrated Measurement System.
Figure 1: PD system setup used to generate and measure the PD signal.
Mineral oil is usually used in distribution and power transformers. Two samples of new and old mineral oil were acquired from a local utility company to investigate the potential of utilizing the same RF antenna used to detect PD in determining the quality of oil. Furthermore, the breakdown voltage and the water content tests were conducted on the two samples and the results are shown in Table 1.
Table 1: Standard transformer oil tests laboratory results
Figure 3: Calibrated Measurement System.
III. RESULTS AND DISCUSSION

New Oil Results 44.9 12
Tests Breakdown voltage (KV) Water content (ppm)
Method IEC 60156 ASTM D 4928
Old Oil Results 27.7 30
To investigate the potential of utilizing the RF antenna to determine the oil quality, two reflectometers were assembled. Figure 2 shows the un-calibrated reflectometer, in which the signal generated by the frequency sweeper is fed to port 1 of a directional coupler. Most of the signal comes out at port 2 to the RF antenna and part of it goes to port 3. The signal at port 2 propagates through the oil and part of it gets reflected back to the RF antenna and to port 2. Now, the signal goes to port 3 where a detector diode is used to pick up the microwave signal and converts it to a DC voltage. Port 4 is connected to a matched load. The reflected signal from the oil depends on its dielectric properties. Thus, as the oil properties vary, the reflected signal changes and that fact is exploited to determine the optimal frequency to differentiate old and new oil. After knowing our operating frequency, a calibrated system as shown in Figure 3 was built to estimate the oils dielectric property. The signal comes from the frequency sweeper and it goes to port one of the coaxial slotted line then it goes out from port two to the RF antenna. Having two signals travelling in opposite directions produces standing waves and a diode detector can be used to measure the voltage standing wave ratio.
A. Comparing RF antenna designs Three different antennas (Figure 4) have been simulated, tested and their sensitivities to PD detection were investigated. MATLAB simulation was performed to compare the Electric Field Intensity of the antennas. Figure 5 shows the E-Field Intensity distribution of these antennas. The dipole and circular loop antennas have donut-shape E-Field pattern with even distribution. Thus, they have a good reliability of detecting the PD signal. However, it can also detect external signals from outside the oil tank having confusion with the PD signal. Since the antennas are operating on a uniform ground plane and according to the phenomenon of image theory of ground planes, the electric field will be reflected totally inside the tank and thus having doubled the electric field.
Figure 4: Different types of RF antennas
The helix follows a much different pattern shape where most of its electric field intensity is concentrated in one direction forming a major lobe, while having a minor lobe in the opposite side, representing the rest of the intensity. This results in an advantage if the antenna is directed towards the inside of the tank. Thus, the axial helix has a high reliability of detecting the partial discharge signal if oriented horizontally, where as it detects much less signals from outside the tank making it useful to be more focused. However, there might be
a difficulty in the detection process if the PD is not in the line of sight of the major lobe.
detectability and polarization for the three different antennas. It is evident from Table 2 that circular loop antenna is the most sensitive antenna for PD measurements and hence it was selected for the other studies.
(a)
(b)
(c) Figure III5: Electric Field Intensity, (a) Dipole Antenna, (b) Circular Loop antenna, (c) Helix Antenna. Figure III6: Typical PD signals measured using different RF antennas, (a) circular loop antenna, (b) dipole antenna, (c) helix antenna.
The voltage source connected to the sharp electrode was raised gradually to generate high intensified field and hence PD which will be measured using the antennas. Typical voltage waveform for the measured PD for the three different antennas is depicted in Figure 6. The voltage level at which the antenna detects PD was recorded and it is usually called PD inception voltage. Table 2 shows the measured PD inception voltage,
Table 2: Inception voltage for three different antenna designs
Antenna type Circular Loop Antenna Dipole Antenna Helix Antenna
Inception voltage (kV) 3.7 5.9 4.0
Detectability Most of the Tank Most of the Tank Line of Sight
Polarization Circular Linear Circular
To check the consistency of the previous observation, the PD signal was measured at different voltage levels. Around 20 readings of PD signal were recorded at each voltage level and the average peak value of the PD signal was calculated. It is evident from Figure 8 that the level of the detected PD signal is higher for RF antenna compared to AE sensor. This is because the RF antenna is less affected by the noise since its positioned inside the steel tank and closer to the PD source. On the other hand, the AE sensor is located at the wall of the tank in which its more affected by external noise.
B. RF antenna versus AE sensor for PD measurement: The PD signal was measured using both the RF and AE sensors simultaneously in the 30x30x30 cm tank filled with new transformer oil. Typical PD signals using the RF antenna and the AE sensor are depicted in Figure 7, respectively. The following observations can be made about both the RF and AE PD signals: The magnitude of the RF signal is much larger than the AE signal. The AE signal is highly impeded in acoustic noise which makes its detection much harder than the RF signal.
(a)
(a) (b) Figure 8: peak Voltage Level: (a) RF Antenna, (b) AE sensor.
In addition, it can be noticed that the RF antenna detects the PD signal at lower voltage level than the AE sensor. This voltage at which the PD starts is called the inception voltage. The inception voltage for PD using both the RF and AE sensors was measured at different conditions and the results are shown in Table 3. It is evident from Table 3 that the RF antenna can detect the PD before the AE sensor. Such higher sensitivity of PD activities enables the RF antenna to detect the PD signal at an early stage.
(b) Figure 7: Typical PD signal in new oil at 4.5kV using (a) RF sensor, (b) AE sensor.
Table 3: PD voltage inception for different oil and measuring conditions
Oil condition and measuring case D=5cm new oil D=5cm old oil D=17cm new oil D=17cm old oil
PD voltage inception AE Sensor RF antenna 3.1kV 2.7kV 2.7kV 2.5kV 4.9kV 4.1kV 3.1kV 2.8kV
can say as \\ increases, the water content in the oil increases and the breakdown voltage decreases and hence, the oil is aging. Taking more samples, we can correlate the value of the dielectric constant with oil test results using artificial intelligent techniques. This will enable us to tell when we have to change the oil without taking a sample from a transformer and waiting for the results to come back from the laboratory. IV. CONCLUSION A PD detection system, utilizing an RF antenna, was designed and built. Results showed that this system is capable of online detection of PD with higher sensitivity than the commercially available AE sensor. The location of the partial discharge source was varied and in all cases, the designed RF antenna system detected the occurrence of the PD signal. Moreover, using the same RF antenna with a reflectometer, the state of oil was monitored. A frequency of 2.4GHz was found to be optimal in determining the state of oil. This is due to the fact that the water content in old oil increases and 2.4GHz corresponded to the resonant frequency of water. The potential of utilizing an RF antenna system to detect partial discharge and monitor the oil quality was clearly demonstrated and future work will aim at developing a first generation prototype to accomplish the task. REFERENCES
[1] B. H. Ward, "A Survey of New Techniques in Insulation Monitoring of Power Transformers", IEEE Electr. Insul. Mag., Vol. 17, No. 3, pp. 1623, 2001. M. Duval, A review of faults detectable by gas-in-oil analysis in transformers, IEEE Electr. Insul. Mag., Vol. 18, No. 3, pp. 8-17, 2002. J. Lopez-Roldan, T. Tang, M. Gaskin, Design and Testing of UHF Sensors for Partial Discharge Detection in Transformers, 2008 International Conference on Condition Monitoring and Diagnosis, 2008, pp. 1052 1055. Z. Tang, C. Li, X. Cheng, W. Wang, J. Li and J. Li, Partial Discharge Location in Power Transformers Using Wideband RF Detection, IEEE Trans. Dielectr. Electr. Insul. Vol. 13, No. 6, pp 1193-1199, 2006. R. Plumb, and H. Ma, H., Swept Frequency Reflectometer Design for in-situ Permittivity Measurements, IEEE Transactions on Instrumentation and Measurement, . Vol. 42, No. 3, pp 730 - 734, 1993. Y. D. He, and L. C. Shen, Measurement of Complex Permittivity of Materials using a Monopole Antenna, IEEE Transactions on Geoscience and Remote Sensing, Vol. 30, No. 3, pp 624 - 627, 1992. L. Aamodt, M. Manwaring, and K. Manwaring, In Vivo Brain Tissue Water Measurement, . 12th IEEE Symposium on Computer-Based Medical Systems, pp. 130-135, 1999. E. Al Murawwi, K. Al Mazam, A-R. Huwair, A. El-Hag and N. Qaddoumi, Partial discharge and oil quality monitoring using an RF antenna, Industrial application magazine, vol. 16, issue 3, pp. 57-59, 2010
D: location of the high voltage electrode with respect to the RF and AE sensors
C. RF Antenna for Oil Quality Monitoring: Using the un-calibrated system shown in Figure 2, a voltage related to the reflection coefficient from an infinite half space of oil, new and old was measured [8]. Measurements were conducted at frequencies from 1GHz to 7.5GHz in 0.1GHz step to find a frequency with high sensitivity to the state of oil. A frequency of 2.4GHz was found to have the maximum difference in measured voltage between the two oil samples. Figure 9 shows the difference between the measured voltages of the two oils as a function of frequency. This variation occurred because the water content in the old oil increases and 2.45GHz corresponds to the resonant frequency of water. In other words, as the oil ages, the water content in it increases. This result agrees with the laboratory results shown in Table 1.
[2] [3]
[4]
Figure 9: Measured Voltage difference for New and Old oil as a Function of Frequency.
[5]
After knowing our range of frequencies, we used the calibrated system shown in Figure 3 to estimate the oils dielectric property. To calibrate the system, a short circuit was used at the end of the slotted line to set the reference. The antenna was then connected and air was used as a second load. Finally, the antenna was immersed in oil (new and old) to measure standing waves to obtain the dielectric properties of both oils. The estimated values of both the real and imaginary parts of the dielectric constant of new and old oil are shown in Table 3.
Table III: Correlation between Dielectric Property and the Standard Oil Tests
[6]
[7]
[8]
New oil Old oil
Water Content (ppm) 12 30
Breakdown voltage (KV) 44.9 27.7
( \ -j \\) 1.6-j0.45 2.3-j0.67
As we can see from the table above that \\ in old oil increases to almost 1.5 times that of new oil. Also, we can correlate the standard oil tests to the dielectric property. In other words, we
IEC 61850 and Smart Grids

Ahmed Elgargouri
College of Electrical and Electronics Technology Benghazi, Libya elgargouri@gmail.com
Mohamed M. Elfituri
College of Electrical and Electronics Technology Benghazi, Libya melfituri@yahoo.com
Mohammed Elmusrati
Department of Computer Science University of Vaasa Vaasa, Finland moel@uwasa.fi
AbstractThis paper defines an introduction and overview of the technology standard IEC 61850. The IEC standard offers reliability, efficiency, flexibility and interoperability needed in smart grids. The improved performance and sustainability of nowadays smart grid led it to be the main used technology for distributed energy resources (DERs) and renewable energy resources, such as wind power, hydroelectric power and solar applications where smart methodologies of measuring and controlling, as well as an integrated communications are required. Virtualized model and data objects of IEC 61850 as well as its flexibility to cooperate with other important standards (e.g. 61131-3) made it one of the core technology standards in smart grids to guarantee their sustainability. An overview of IEC 61850s cooperation with other famous standards as well as few examples of its implementation in DERs and renewable energy resources are discussed within this paper. Keywords DER; IEC 61850; Interoperability; Smart grid
Nowadays, IEC 61850 is a very important topic for researches as the power system automation needs are rapidly increasing, especially with the wide use of smart grids, renewable energy resources and distributed energy resources (DERs). There are several updates and new researches regarding to IEC 61850 to study the opportunities of meeting the requirements of whole electrical energy supply chain [2] as well as implementing the standard in smart grids and green power applications as a new technology or based on another related standard, such as IEC 61131-3 or IEC 60870-5. II. IEC 61850 OVERVIEW
A. Communication Model
I.
INTRODUCTION
Abstract Communication
IEC 61850 is a worldwide-accepted standard for Ethernetbased communication in substations. The IEC 61850 international standard consists of 14 parts (10 main parts and 4 subparts). It was founded in 2003 by International Electrotechnical Commissions (IEC) Technical Committee 57 (TC57). This standard takes advantage of a comprehensive object-oriented data model and the Ethernet technology, bringing in great reduction of the configuration and maintenance cost. The standard achieves both vertical and horizontal communications, which are needed for a perfect performance inside the SCADA (Supervisory Control and Data Acquisition). IEC 61850 is designed to be capable for domains besides substation automation (SA). The abstract data models defined in IEC 61850 can be mapped to a number of protocols. Current mappings in the standard are to MMS (Manufacturing Message Specification), GOOSE (Generic Object Oriented Substation Event), SMV (Selectable Mode Vocoder). In addition, the standard has been recently mapped over web services based on the Devices Profile for Web Services (DPWS) [1]. These protocols can run over TCP/IP networks and/or substation LANs using high speed switched Ethernet to obtain the necessary response times of e.g. less than 4ms for protective relaying.
Stack Interface
ISO/OSI-Stack Hierarchical set of rules how information is coded for
* Generic Object Oriented
Fig. 1. IEC 61850 Communication Model [4]
Figure 1 describes the mapping of IEC 61850 data models over the mentioned communication protocols in part I. Application level (represented in green) is mapped to the communication level through Abstract Communication Service Interface (ACSI) (in cyan) then represented in red color the implementation through Specific Communication Service Mappings (SCSM). ACSI is a subject of IEC 61850 part 7 and its subparts. Abstract means the definition of the data and information to describe what the services provide. The implementation is done through the SCSM by mapping to e.g. MMS, TCP/IP and
Ethernet. Mapping IEC 61850 over DPWS is subject to part 82 [5]. B. IEC 61850 Application and Communication views IEC 61850 is an application layer protocol that can be useable only if it is mapped to a communication service, such as MMS and DPWS. Figure 2 shows application and communication views represented by the IEC standard parts.
Application View
Similar functionalities performed by different devices build a logical node and several logical nodes performed by different devices or by the same device build a logical device. Logical device represented in virtualized model does not usually signify one real device. It mostly represents a different aspects or different logical nodes from different real devices. A logical device is always implemented in one IED even though it is built by logical nodes from different real devices, which means that logical devices are not distributed. D. Logical Device(LD) As it was described previously, one physical device can be divided into one or more logical devices and the logical nodes are sorted as sub-functions in the logical devices. Every logical device (LD) consists of a minimum of three logical nodes.
Communication View
Service by which the information can be accessed or manipulated Communication Objects and Service according to 7-2 mapped to a
Network Fig. 2. Application and communication views of IEC 61850 [6]
C. IEC 61850 Virtualized Model The core of IEC 61850 is the interoperability between IEDs (intelligent electronic devices) from different vendors. In other word, interoperability between different functions performed by different Physical (real) devices. This is done by using data models and data exchange based on these models. Virtualization means that every physical device can be represented in a Virtual world and only aspects of a real device that are of interest for the information exchange with other devices are virtualized. This method is called distributed functionality and the involved devices in data exchanging are called distributed devices. Achieving the virtualized model when using IEC 61850 comes from the fact of mapping the standard over MMS where VMD (Virtual Manufacturing Device) model is used.
Process
Fig. 4. Physical and logical devices [7]
Each logical node contains a pre-defined set of data classes. Every data class contains many data attributes (status value, quality etc.). The logical devices, logical nodes and data objects are virtual parameters, they merely seem to exist. The logical nodes and the data contained in the logical devices are fundamental for description and information exchange inside the power station automation systems to reach interoperability. E. Logical Node (LN)
TABLE I. Group Indicator A C G I L M P R S T X Y Z TOTAL LN GROUPS. LNs groups Automatic Control Supervisory Control Generic Function References Interfacing and Archiving System Logical Nodes Metering and Measurement Protection Functions Protection Related Functions Sensors and Monitoring Instrument Transformer Switchgear Power Transformer Further (power system) Equipment Number 8 5 3 4 3 8 28 10 4 2 2 4 15 92
Fig. 3. Virtual world vs. real world [6]
As it can be seen in the previous table, logical nodes are grouped into 13 main groups; each group contains a specific number LNs. 92 LNs are covering the most common applications of power stations and feeder equipment. The names of logical nodes begin with the character representing the group to which the logical node belongs. IEC 61850 store for the future clear rules relating to extensions of the information models, including extensions to logical nodes, new logical nodes, expanded and new data attributes [8]. LNs are a kind of Folders which contain data that can be used or exchanged. For example, The XCBR LN implements the functionality of CB (Circuit Breaker) by grouping 16 data classes as shown in Figure 5. XCBR data set contains characters correspond to the used commands for the CB such as BlkOpn (block open operation), Beh (Behavior), etc.
G. Substation Configuration Language (SCL) SCL is introduced in Part 6 of the IEC 61850 standard. The main duty of SCL is to guarantee the interoperability by exchanging information between IEDs from different vendors and the station computer. It comes from the fact that each IED is configured by independent configuration tool developed by its manufacturer. This is done by using four types of SCL common files. These files are the IED Capability Description (ICD), Configured IED Description (CID), Substation Configuration Description (SCD) and System Specification Description (SSD) files. All ICD files get imported into the IEC 61850 system configurator, which allows the configuration of GOOSE messages by specifying the senders (publishers) and the receivers (subscribers) of messages. The system configuration tool creates the SCD file, which includes the one line diagram of the station and the description of the GOOSE messages. Each IED Configuration tool must be able to import an SCD file and extract the information needed for the IED then join all information in one SCL file to be sent to another IED or to the station computer and the control unit in the SA.
Fig. 5. XCBR (Circuit Breaker) Logical Node [6]
F. Generic Object Oriented Substation Event (GOOSE) The term GOOSE is not new and was used in UCA 2.0 protocol. However, the IEC61850 GOOSE is an advanced version of the UCA GOOSE [9]. The major difference is that an IEC61850 GOOSE is not a static number of bits or bit pairs (binary output to binary input); this version can exchange a wide range of common data according to the use of data sets in the IEC standard. A GOOSE message is used to exchange data between IEDs and is only one part of the new standard IEC61850. GOOSE communication represents the horizontal communication in the substation automation system by broadcasting the same GOOSE message to one or several IEDs at the same time, which allows faster communication, less wiring and lower cost.
Fig. 7. SCL files and interoperability [10]
III.
BENEFITS OF IEC 61850 IN SMART GRIDS
Benefits of IEC 61850 is measured according to the high requirements of IEDs inside the SA, such as high-speed IED to IED communication, guaranteed delivery times, multi-vendor interoperability, etc [11]. In addition, IEC 61850 is designed to satisfy the main requirements of smart grids, such as reliability, efficiency, flexibility and interoperability. The increment of IEC 61850 overall benefits is also related to the use of GOOSE messages, which brings a number of highly performed functions. One of the essential preconditions for using GOOSE is that it performs adequately compared with a hardwired solution. In addition, due to the non-deterministic nature of Ethernet, reliability is guaranteed under difficult communication load conditions [12]. Besides, using certain number of IEDs and fiber optic communications instead of a huge number of copper wires enhances the grid protection, monitoring and control functions reliability [3].
Fig. 6. Example of GOOSE messages exchange [9]
The added values to the smart grids effic ciency when using IEC 61850 are mainly linked to the red duced installation, transducer and commissioning costs, with hout forgetting to mention the reduction in configuration costs due to the objectoriented meaning of the standard. Moreover, lowered equipment migration, extension and integr ration costs when using the IEC standard result an increasing in i the smart grids efficiency. Another important benefit when using u IEC 61850 is the improved flexibility of the grid. This is due to the fact of the standards feature of implementing new w capabilities. The radical services and unique features of IEC 61850 enable new capabilities that are not viable with most legacy protocols. Wide area protection schemes that would normally be cost prohibitive become much more feasible. Be ecause devices are already connected to the substation LAN, th he incremental cost for accessing or sharing more device e data becomes insignificant enabling new applications th hat would be too expensive to yield. IV. STANDARDS BASED ON IEC C 61850
Logical Device (LD1)
Composite Function Block
Fig. 8. Implementation of LNs and logical devices using composite FB [13]
This part discusses the interplay betwee en IEC 61850 and some of the well-known substation automati ion standards. The purpose of this part is to investigate which h standards might interplay with IEC 61850 instead of replacin ng a current existed protocol with the IEC standard. A. IEC 61499 and IEC 61131 The idea of the interplaying between IE EC 61499 and IEC 61580 comes from the fact that their architec ctures have certain aspects in common. As IEC 61850 uses log gical nodes to sort data classes and data attributes, IEC 6149 99 uses a similar functionality which so called function block ks (FBs). LNs and logical devices with their functions can be im mplemented using a composite FB. Most of IEC 61850-7-4 LNs are possible to be modeled as FBs [13]. IEC 61131 has almost the same architecture as IEC 61499 since both hav ve been the main mission of TC56 WG6*. It is known that t IEC 61131 was defined for communicating the PLC inside the substation automation. IEC 61850 LNs can be mapped to IEC 61499/IEC 61131 FBs. Hence, IEC 61850 LNs can be e implemented on PLC systems FBs [8]. The concept of mappi ing LNs to a FB is shown in Figure 8. B. IEC 60870 As future demands of SA are incrementing, there are currently several researches investigating the possibilit ties of sending IEC 61850 GOOSE through wireless channels by using WLAN networks, which is also known as wireless Ethernet [14] and Ethernet-to-Modbus switches. For example, in 2011 it was the first project to send GOOSE messages over r WiMAX. So far, transferring GOOSE through wireless chann nel is still slow, a total end-to-end latency takes 30-50ms ov ver WiMAX and 800ms over GPRS [15]. Figure 9 shows a ba asic architecture of a WLAN communication for substation and distributed energy resources (DERs).
IECs Technical Committee number 56, Working Group p number 6.
Fig. 9. WLAN network for substation n and DERs [14]
IEC 60870 is another internati ional standard for power system automation that was defined by b IEC TC 57 WG. The same working group generated also parts 5 and 6 of the standard. IEC 60870-5 was developed for r transferring remote control and remote protection messages between two different networks. Few researches nowadays are e checking the possibilities of mapping IEC 61850 to IEC 60870-5. This will lead to enhance the speed of transferring wireless messages inside the SA. C. IEC 61400 In 2001, IEC TC88 introduc ced IEC 61400-25 based on IEC 61850 and as an extension in order to meet the requirements of wind power systems. First draft ft of the standard documents was published in 2006 and it was ready r to use in 2009. Like IEC 61850, IEC 61400-25 is also object-oriented, which allows the use of virtualized models (Logi ical devices, LNs, etc.). V. IEC 61850 AND SM MART GRIDS APPLICATIONS
The most interesting smart grids applications currently are wind power and solar applicati ions, but hydroelectric power is remaining under research and development d sensation. This part summarizes the current the im mplementation of IEC 61850 in the three mentioned fields.
A. Hydroelectric Power IEC TC57 WG18 has defined an extension of IEC 61850 for Hydroelectric power plants by introducing new models. WG18 defined additional documents to cover the typical required data of hydroelectric power plants, such as water level, dam gate and turbine control etc. Four different groups of data objects are defined for typical hydro power plant [2]: 1) Electrical functions: All needed LNs for this group are included IEC 61850 documents. 2) Mechanical functions: This group is specified for hydropower plants since it includes functions related to the turbine and its supplementary equipment. 3) Hydrological functions: This group includes the main functionalities to control the hydropower plant. It contains LNs to control water flow, dams, reservoirs etc. 4) Sensors: In this group, LNs for monitoring and measuring of other than electrical data are included. This group is also specified for hydropower plants. Table II below describes the LNs used for hydropower plants.
TABLE II.
LN AFCO AKVR ALCO AMWR ASPC CCGR DPC GAPC HBRAK HGOV HJCL HPPU MMDE MMDP MMFE MMHE MMLE MMPE MMRF MMSF PPAM PSDE PSDE PSDE RTEM RVIB VPCO WMET WHYD ZTCR
B. Wind Power As it is mentioned in part IV section C, IEC 61400-25 was issued in order to meet the new defined needs of wind power plants. In fact, IEC 61400-25-6 is not based on any of the IEC 61850 parts. It contains new defined LNs for monitoring and control of wind power plants. Table III lists all assigned LNs to wind power systems. First four LNs are for a whole wind power plant and the rest are specified for the wind turbine.
TABLE III. LN WALM WMET WAPC WRPC WTUR WROT WTRM WGEN WCNV WTRF WNAC WYAW WTOW WSLG WALG WREP Description Wind power plant alarm information Wind power plant meteorological information Wind power plant active power control information Wind power plant reactive power control information Wind turbine general information Wind turbine rotor information Wind turbine transmission information Wind turbine generator information Wind turbine converter information Wind turbine transformer information Wind turbine nacelle information Wind turbine yawing information Wind turbine tower information Wind turbine state log information Wind turbine analogue log information Wind turbine report information WIND POWER PLANT LNS.
HYDROPOWER PLANTS INFORMATION MODEL

Description Flow Controller Automatic Voltage/var regulator Level Controller Active Power regulator Speed controller Cooling group control Controllable double point Generic automatic process control Brakes Hydraulic Governing system Hydraulic Joint control High Pressure pumping unit Density logical node Dewpoint logical node Flow element logical node Humidity logical node Level element logical node Pressure element logical node Rainfall logical node Snowfall logical node Phase angle/out-of-step 100% stator earth-fault Directional earth-fault 100% stator earth-fault Temperature monitoring system Vibration monitoring system Valve opening position controller Meteorological station Hydrological station Thyristor controlled reactive component
C. Destributed Energy Resourses (DERs) In the beginning of 2012, document called IEC 61850-7420 DER Logical Nodes, which was a subpart of IEC 618507-4, was published by IEC TC57 WG17. The document defined some useful LNs to be applied on DERs and PV systems.
TABLE IV. DER TYPE SELECTION
Notice that M/O/C means the data is one of three possible sorts: M=Mandatory, O=Optional or C=Conditional. The applied IEC 61850-7-420 LNs in the system depends on which kind of resource is used. For example, if the energy resource is PV then the value of DERtyp (Type of DER) within the logical node DRCT should be set to the value 4 (as
shown in previous figure) at the DER unit controller and hence the LD at the DC converter level will be a DC-AC inverter and ZINV logical node will be used. D. Solar Applications IEC-TC 57 defined several LNs to fulfill PV Systems tasks within the DER Logical Nodes document. At the time this paper came to be ready, there was not any factual implementation of IEC 61850 to meet the requirements of the whole solar applications and PV inverters yet. The first official document by IEC regarding to PV systems was published in March 2013. Reference [17], which this paper was majorly based on it, is considered as one of first researches to investigate the implementation of IEC 61850 in solar applications. Below is a brief definition for each of them [16]: 1) Photovoltaic module ratings (DPVM): It describes the photovoltaic characteristics of a module. 2) Photovoltaic array characteristics (DPVA): It describes the configuration of the PV array. 3) Photovoltaic array controller (DPVC): It contains information required for remote monitoring of critical photovoltaic functions and states. 4) Tracking controller (DTRC): It provides overall information on the tracking system to external users. 5) Inverter (ZINV): ZINV logical node defines the characteristics of the inverter, which transforms DC to AC. The DC may be the output of the generator or may be the intermediate energy form after a generators AC output has been rectified. However, In case of using a rectifier within the solar system, the logical node ZRCT, which defines the characteristics of the rectifier, can be used. Otherwise, only ZINV is needed to be implemented. VI.
CONCLUSION
REFERENCES
[1] J. Schmutzler, S. Groning and C. Wietfeld, Management of Distributed Energy Resources in IEC 61850 using Web Services on Devices, Smart Grid Communications (SmartGridComm), 2011 IEEE International Conference on, Commun, Networks Inst. (CNI), Dortmund Univ. of Technol., Dortmund, Germany, pp. 315- 320, October 2011. K. Schwarz, IEC 61850 ALSO OUTSIDE THE SUBSTATION FOR THE WHOLE ELECTRICAL POWER SYSTEM, Schwarz Consulting Company (SCC), Karlsruhe, Germany, 2005. J. McGhee and M. Canada Goraj, Smart High Voltage Substation Based on IEC 61850 Process Bus and IEEE 1588 Time Synchronization, Smart Grid Communications (SmartGridComm), 2010 First IEEE International Conference on, RuggedCom Inc., Calgary, AB, Canada, pp. 489 - 494, October 2010 Klaus-Peter Brand, IEC 61850 Short Tutorial, ABB University Switzerland & ABB Power Technology Systems, Utility Automation, ABB Switzerland Ltd. p. 24, March 2005 IEC 61850 blog, Next Step towards a Web Service Mapping in IEC 61850, News on IEC 61850 and related Standards, October 2012. IEC-TC 57, Communication networks and systems in substations, IEC Standard IEC/TR 61850, 1st Ed, 2003. Z. Gaji, IEC 61850, ABB Power Technologies AB, Sweden, p. 20, 2005. J. Lidn, Design and Implementation of an IEC 61850 gateway for PLC Systems, Masters Thesis, KTH Electrical Engineering, Stockholm, Sweden, 2006. Neteon, What Is Goose, Industrial Networking Solutions.
[2]
[3]
[4]
[5] [6] [7] [8]
[9]
[10] R. Aguilar and J.Ariza, Experience with Testing and Configuration of IEC 61850 Multivendor Protection Schemes, Megger, p. 2, USA, 2010. [11] R. Mackiewicz, Overview of IEC 61850 and Benefits, IEEE: 1-42440493-2/06/$20.00, p. 1, 2006. [12] A. Hakala-Ranta, O. Rintamki and J. Starck, Utilizing Possibilities of IEC 61850 and GOOSE, 20th International Conference on Electricity Distribution. ABB Oy, Distribution Automation, Prague, Czech Republic, p. 2, 2009. [13] N. Higgins, V. Vyatkin, Nirmal-Kumar C. Nair and K. Schwarz Distributed Power System Automation With IEC 61850, IEC 61499, and Intelligent Control, p. 4-5, 2010. [14] P.P. Parikh, M. Kanabar and T.S. Sidhu, Opportunities and Challenges of Wireless Communication Technologies for Smart Grid Applications, p. 3, 2010. [15] M. Goraj, L. Lipes and J. McGhee, IEC 61850 GOOSE over WiMAX for Fast Isolation and Restoration of Faults in Distribution Networks, Dublin, Ireland, p. 4, June 2011 [16] IEC-TC 57, IEC 61850 Part 7-420 DER Logical Nodes, Final Draft International Standard (FDIS), IEC Standard IEC/TR 61850, 2009. [17] Ahmed Elgargouri, Implementation of IEC 61850 in Solar Applications, Masters Thesis, University of Vaasa, Faculty of Technology, Vaasa, Finland, April 2012
IEC 61850 has become a very important topic for researchers as the power system automation needs are rapidly increasing. This is in one hand. In the other hand, the requirements of smart grids applications, such as solar applications and wind power systems, are rapidly increasing at the same time. Hence, the need of implementing a compatible standard that meets these requirements is growing as well. Part III of this paper explained how IEC 61850 is covering the smart grids main features, such as: Interoperability, reliability, flexibility, efficiency and enhanced performance due to the use of GOOSE messages and SCL files. This however does not prevent the fact of IEC 61850 security issues. IEC 61850 is still not considered as one of the most secured standards. The standards security issues are expected to be solved by using the new standard IEC 62351. This new standard was defined by TC57 WG15. It contains almost all documents of IEC 61850 and it was developed to handle the security of all TC57 series including IEC 61850.
A dynamic model of a grid connected PV system based on outdoor measurment using Labview
Imane Drouiche
Laboratoire Instrumentation Electronique Universit des Sciences et de la Technologie Houari Boumediene, USTHB, 16111, Algiers, Algeria idrouiche@usthb.dz
Aissa Chouder
Centre de Developpement des Energies Renouvelable, CDER, 16340, Algiers, Algeria achouder@cder.dz
Samia Harrouni
Laboratoire Instrumentation Electronique Universit des Sciences et de la Technologie Houari Boumediene, USTHB, 16111, Algiers, Algeria sharrouni@yahoo.fr
Abstract One of the main task of the grid connected inverter is to derive the maximum available power from a photovoltaic plant, generally performed by a DC-DC converter, under variable irradiance and temperature. The purpose of this paper is to present a low complexity calculation procedure that allows to model the behavior of PV grid connected system at the DC side. The proposed procedure allows deriving dynamic locations of the maximum power point (MPP) of an entire PV plant under operation. Analytical approach combined with outdoor measurements is adopted in order to extract the main five parameters of the one diode I-V characteristic equation. Parameters extraction procedure and MPP derivation have been implemented in Labview and Matlab environment (co-simulation) where measured temperature and irradiance are taken as input variables. Finally, the dynamic model is evaluated under real PV system operation conditions and it has been found that it can be used as a monitoring system and detection of malfunction at the DC side. Keywords--PV module; Dynamic model; Static model; monitoring; Labview; One diode model; MPP; Grid connected system.
and reliability have been extensively published in several research works [1-12]. In addition, several software specialized in the engineering of photovoltaic systems are now available each with a certain degree of complexity and precision such as PVSYST [13], or general purpose software which can be used to solve many photovoltaic systems issues such as Matlab Pspice Labview [3-5]. The present work deal with using a co-simulation environment between Matlab and Labview to derive a dynamic locations of the maximum power point (MPP) of an existing photovoltaic power plant used as grid connected PV system. The adoption of such co-simulation environment has been primarily driven due the flexibility of Labview in hardware interfacing capabilities and power of Matlab in mathematical calculation [3-5]. The prediction model of the dynamic locations of MPP is mainly based on the a priori extraction of the unknown parameters of the one diode characteristic equation [4], [5]. First, data collection in real conditions of temperature and irradiance of I-V curves from the whole PV plant is performed in order to use their key points such as short circuit current, open circuit voltage, and Voltage and current at MPP to extract the main parameters with analytical procedure under Matlab-Labview environment. The inputs to the simulation environment are the measured temperature and in plane irradiance. The paper is organized as follows: section II reviews the one diode mathematical model of PV module, the five parameters extraction method is given in section III followed by PV module equation solving method in section IV. Section V shows the dynamic behavior of PV module and finally the proposed approach validation is given in section V. II. MATHEMATICAL MODEL OF PV MODULE The electrical power available from a photovoltaic (PV) device can be modeled with the well known equivalent circuit shown in Fig. 1. This circuit includes a series resistance and a diode in parallel with a shunt resistance. This circuit can be
I.
INTRODUCTION
In the last years, there has been a sharp decline of the total cost of photovoltaic modules mainly due to massive production and the concurrency of Chinese market. On the other hand, prices of fossil energies are still high in addition to the environmental issues which are thought will not be resolved in the near future. This context has encouraged to a large extent of deployment of a large scale power plants based on renewable energy sources especially integrated photovoltaic systems in high and low distribution network [15]. Modeling of photovoltaic systems in order to predict the productivity at a given sites and to improve their performance
used either for an individual cell, for a module consisting of several cells, or for an array consisting of several modules [2][9].
Outdoor I-V measurements
Extraction of: Voc, Isc, Impp, Vmpp Solving of the nonlinear system of equations [4] Newton Raphson Algorithm Reference values of five parameters:
Figure 1. One diode equivalent circuit
I0ref, Iphref, aref, Rsref, Rshref

T G
The current-voltage relationship at a fixed cell temperature and solar radiation for the circuit in Fig.1 is expressed in (1).
Real values of five parameters: I0(T,G), Iph(T,G), a(T,G),

Figure 2. Five parameter extraction flowchart
(1) IV.
They namely are: the photocurrent Iph in (A), the diode saturation current I0 in (A), the modified ideality factor a = nkT/q, k the Boltzmann constant (JK-1), q the electronic charge (C), T the cell temperature (K), Vt the thermal voltage (Vt = kT/q, V), the series resistance Rs () and the shunt resistance Rsh (). Five parameters (Iph , I0, a, Rs, Rsh) must be known in order to determine the current and voltage, and thus the power delivered to the load. III. PARAMETER EXTRACTION To evaluate the five parameters of (1), five independent pieces of information are needed. In addition, these five parameters depend on incident solar irradiance, cell temperature and on their reference values [4]-[9]. Generally these reference values are given in the datasheet of PV modules for specified operating condition known as STC (Standard Test Conditions) for which the irradiance is 1000 W/m and the cell temperature is 25C and the air mass AM is 1.5. However, these values provided by the manufacturers are unreliable given of the ideal conditions where these values were taken. In addition, these laboratory conditions are almost never achieved in real conditions. Therefore, to obtain the key point at standard conditions, we carry out outdoor measurement of several I-V curves and using translation equations to derive Isc0, Voc0, Impp0 and Vmmp0 respectively, where index 0 stands for standard value (1000w/m2 and 25C) [14]. Then, the system of equations given by A. Chouder et all is solved by using NewtonRaphson algorithm. The real values of the five parameters are then derived for any working condition of irradiance and temperature. The step by step followed procedure to extract the five parameters of (1) is given in the flowchart of figure 3. Where T and G are the measured temperature in plane irradiance respectively.
SOLUTION OF THE PV ARRAY CHARACTERISTIC

EQUATION
A closed-form exact solution of (1) for the unknown current I is not available. Hence numerical methods should be used to solve it. In this work the Newton-Raphson iterative method is exploited because it converges quickly [4], [5], [8]. Fig.2 shows the flowchart of solving (1). Begin measured T and G Real values of five parameters: I0(T,G), Iph(T,G), a(T,G), Rs(T,G), Rsh(T,G) Solving the non-linear equation I = f(I,V): Newton-Raphson method Draw curves: I-V and P-V End
Figure 3. Flowchart of solving PV array equation
V.
DYNAMIC BEHAVIOR
In order to validate the accuracy of the extracted parameters to imitate a real operating system ; for each measured temperature (T) and irradiance (G) the model generate the correspondent IV curve and then the maximum power point coordinates are extracted. On the hand, the same variables are measured in real time by an installed data acquisition [4-5]. A comparison between simulated and measured variables is then carried out according to the flowchart given in Fig.5.
Measurements from DAQ
VI.
MODEL VALIDATION
G T
Solving (1) for the given T and G Extract values from I-V curve: Pmax-c Vmp-c Imp-c Calculate and plot: Measured power (Pmax-m) and calculated power (Pmax-c) Measured voltage (Vmp-m) and calculated voltage (Vmp-c) Measured current (Imp-m) and calculated current (Imp-c) Calculate errors between calculated and measured data Save results in excel file
Figure 4. Dynamic model flowchart
Electrical variables at the input of the PV grid connected inverter (Vdc & Idc) and meteorological variables (T & G) are measured by a data acquisition where the time stamp is taken equal to one minute (1 min.). All the acquired are transferred to The main computer via a GPIB cable. A. Co-simulation program Co-simulation approach between Labview, Matlab and Excel link has been adopted to model the time evolution of the predicted DC voltage and current of the entire PV plant. On the other hand the same measured variables are processed by a Labview VIs. Figure 5 shows the code program of dynamic model which consist of two principal parts: a VI allowing the solution of (1) followed by maximum power point extraction. In fig. 6 it is shown the co-simulation program giving the solution of (1). The dynamic model can be used in real time as he can be used for processing saved files. Table. I gives the five parameters given by the manufacturer and those calculated by the parameter extraction procedure of Isofoton 12-106 PV module..
Pmax-m Vmp-m Imp-m
Figure 5. Labview general program of dynamic model
Figure 6.
Co-simulation program Labview-Matlab giving the solution of (1)
TABLE I.
FIVE PARAMETERS FOR REFERENCE AND REAL CONDITIONS Conditions

Manufacturer Evaluated
Five parameter
Manufacturer Evaluated Iphref: (A) I0ref: (A) Rsref: () aref Rshref: ()
To validate the dynamic model, we simulate a day with the couple (T, G) and drawing after the different curves of comparison between calculated and simulated results. Measured and predicted Impp curve: Fig.7 shows both of measured and calculated current at MPP. It is clearly shown the good agreement between measured and predicted current with an error less than 0.5 (A).
Isc 0: (A) Voc 0: (V) Imp0: (A) Vmp0: (V) Pmax: (Wp)
6,54 21,8 6,1 17,4 106,14
6,68 21,3 5,95 16,68 99,275
6,548 4,44
-9
6.69 1.097-5 0.157 1.601 200.371
0,23 1,033 199,771
Figure 7. Calculated current and measured current curve for one day.
Measured and predicted Vmpp curve: Generally different models can give a good agreement in current, but not in voltage because of its complexity. For the dynamic model presented in this work, we can see in
Fig.8 that simulated voltage is very near than real voltage of the PV system, with an error less than 2 (%) for the high irradiance .
Figure 8. Calculated voltage and measured voltage curve for one day.
Measured and predicted MPP curve: Fig.9 shows both of measured and calculated maximum power point
(MPP). It is clearly shown the good agreement between measured and predicted power.
Figure 9. Calculated MPP and measured MPP curve for one day.
Error of power E (%): Fig.10 shows the error between MPP measured and MPP calculated. As we can see
the error is less to 4.5 (%) for all the day, and in high irradiance it is on average close to 1 (%).
Figure 10. MPP error E (%) for one day.
B. Validation of the approach for variable weather conditions After testing the dynamic model for one day, here after we carry out simulation for several days where the weather condition are variable. Thus, in figure 11 we report the simulation results for 15 days of august versus measurements.
We notice here that the developed procedure gives a good prediction for both clear sky days and cloudy days. In addition, the model can inform us about the state of the overall system ; healthy or malfunction. It is clearly shown in figure 11, the case where one string of the total PV array is removed.
Figure 11. MPP current for 15 days of August 2012.
VII. CONCLUSION A dynamic model with MPP simple extraction method is simulated for a grid connected PV system with rated power of 3.2 Kwp, composed by a photovoltaic generator Isofoton 12106 and a single phase grid connected inverter. The simulation results are discussed demonstring the feasibility of the proposed method. A Labview program permits to test PV system behavior and to verify the system output under different conditions of temperature and irradiance. Finally, the dynamic model is evaluated under real PV system operation conditions and it has been found that it can be used as a monitoring system.
[8]
[9]
[10]
[11]
[12]
REFERENCES
[1] I. Kashif, S. Zainal, M. Saad, and S. Amir, Parameter extraction of solar photovoltaic modules using penalty-based differential evaluation, Appl Energy 2012; 99, pp. 297-308. V. L. Brano, A. Orioli, G. Ciulla, A. Di Gangi, An improved five parameters model for photovoltaic modules, Solar Energy Materials & Solar Cells, 2010; 94, pp. 1358-1370. A. H. Alqahtani A simpliphied and accurate photovoltaic module parameters extraction approach using matlab, IEEE, 2012. A. Chouder, S. Silvestre, N. Seddaoui, L. Rahmani, Modeling and simulation of a grid connected PV system based on the evaluation of main PV module parameters, Simulation Modelling Practice and Theory, 2012; 20, pp.46-58. L. Rahmani, A. Kessal, A. Chouder, Parameters extraction of photovoltaic module at reference and real conditions, 46th International Universities Power Engineering Conference, 5-8th Septembre 2011, Soest, Germany. W. De Soto, Improvement and validation of a model for photovoltaic array performence, M.S Thesis, Mechanical Engineering University of Wisconsin-Madison, 2004. C. Carrero, O. Ramirez, I. Rodrigez, C. A. Platero, Accurate and fast convergence method for parameter estimation of PV generators based on
[13] [14]
three main points of the I-V curve,Renewable Energy 2011; 36, pp. 2972-2977. F. Adamo, F. Attirissimo, A. Di Nisio, A. M. L. Lanzolla, M. Spadavecchia, Parameters estimation for model of photovoltaic panels, 15 IME KO World Congress Fundametal and Applied Metrology, September 6-11, 2009, Lisbon, Portugal. W. De Soto, S. A. Klein, W. A. Beckman, Improvement and validation of a model for photovoltaic array performence, Solar Energy, 2006; 80, pp. 78-88. M. A. De Blas, J. L. Torres, E. Prieto, A. Garcia, Selecting a suitable model for characterizing photovoltaic devices, Renewable Energy, 2002; 25, pp. 371-380. G. Farivar, B. Asaei, Photovoltaic module single diode model parameters extraction based on manufactorer datasheet parameters, IEEE International Conference On Power ond Energy (PE Con 2010), Malaysia. J. Wang, J. Shieh, Y. Su, K, Kuo, Y. Chang, Y. Liang, J. Chou, K. Liao, J. Jiang, A novel method for the determination of dynamic resistance for photovoltaic modules, Energy, 2011; 36, pp. 5968-5974. http://www.pvsyst.com. http://www.pvpmc.org/.
[2]
[3] [4]
[5]
[6]
[7]
Investigation on the performance of direct borohydride fuel cell by using statistical approach
Fatma Gl BOYACI SAN, Osman OKUR, idem YGN KARADA, Il IIK GLSA, Emin OKUMU
Energy Institute TBTAK Marmara Research Center Gebze/Kocaeli, Turkey
Abstract Fuel cells are electrochemical devices that convert
chemical energy to electrical energy as long as the fuel and oxidant is supplied. Direct borohydride fuel cells (DBFCs) are a promising option especially for portable applications. DBFCs are facing two main problems: NaBH4 cross over and hydrolysis. The borohydride oxidation pathway can change depending on the catalyst type and operation conditions. In this work, a DBFC employing PtRu/C as the anode and Pt/C as the cathode catalyst is used to control the hydrolysis of NaBH4 by operation conditions. The experimental design method is used and ANOVA contributing factors which influence on the power density of DBFC are identified with the help of DESIGN EXPERT software. The effects of cell temperature, borohydride concentration, flow rates of fuel and oxidant on the performance of DBFCs are investigated by using statistical approach in design and analyzing of experiments. The maximum power density is observed at 80 C fuel cell temperature, 1,5 M NaBH4 concentration, 5 cm3min-1 flow of anolyte and 0.2 dm-3 min1 flow of air. Keywords-component; direct borohydride fuel cell; statistical approach; operation conditions I. INTRODUCTION
for portable and mobile applications. Sodium borohydride (NaBH4), has a capacity value of 5.67 Ah g-1 and a hydrogen storage capacity (weight content of 10.6%). The solid sodium borohydride is chemically stable, non-toxic, non-combustible, and could be stored and distributed easily; the by product (NaBO2) is non-toxic and also can be recycled to generate borohydride and no pollutant gases are exhausted from the DBFC reactions [2-4]. In this type of fuel cell, sodium borohydride in aqueous alkaline medium is fed directly to anode compartment as a fuel. It converts chemical energy stored in a borohydride and an oxidant directly into electricity by oxidation of BH4 at the anode and reduction of an oxidant at the cathode [2,4]. A schematic diagram of a direct borohydride fuel cell employing oxygen/air as oxidant is given in Figure 1 [5]. In a DBFC, borohydride, BH4, is oxidized in strong alkaline media (pH >12) to metaborate, BO2, and water, generating eight electrons according to (1). Oxygen or air is electrochemically reduced at the cathode electrocatalyst and the eight electrons are consumed according to (2) [3-6]. Anode: BH4- + 8OH- BO2- + 6H2O + 8e- (-1.24 V) Cathode: 2O2 +4H2O + 8e- 8OH(0.40V) Overall: BH4 +2O2 BO2 +2H2O (1.64 V) (1) (2) (3)
Fuel cells are promising energy conversion devices, which directly produce electricity from the chemical energy of fuels. It is intrinsically energy efficient, non-polluting, silent, and reliable. Polymer electrolyte membrane fuel cells (PEMFCs) have the advantage of high power densities at relatively low operating temperatures. PEMFCs are at the brink of commercialization, however, several issues have to be resolved, such as anode poisoning, the supply, storage, and transportation of hydrogen, and cell stack cost. In order to overcome these difficulties, liquid fuel cells can be used instead to fuel PEMFCs for low power requirements. Also, direct use of liquid fuel in a PEMFC simplifies the engineering issues [1-5]. Several fuels such as methanol, ethanol, propanol, formic acid, dimethylether and ethylene glycol are fed to fuel cells. However, the liquid fuel cells have limitation of inefficient fuel conversion, fuel crossover and low power density. Direct borohydride fuel cells (DBFC) overcome the given problems. DBFCs are considered attractive energy converters, especially
Figure 1. Schematic diagram of a direct borohydride fuel cell employing oxygen/air as oxidant [5].
Two main problems are observed for DBFC. One is BH4crossover in the fuel cell system, which causes the deactivation of the cathode and low utilization of BH4-. Another problem is the poor anodic efficiency of BH4 due to the simultaneous hydrolysis of BH4- at anodic electrocatalyst surfaces during electrochemical oxidation by reaction (4). Hydrogen evolution from the hydrolysis not only decreases fuel efficiency but also causes problems in mass transport and system design [2-6]. BH4 +2H2O BO2 + 4H2 (-0.86 V) (4)
B. Experimental Design Since four input factors each at two levels are considered, 24 design can be used. For conducting 24 designs with a single replicate, 16 MEAs have to be produced, that is, 16 runs are required. In order to reduce the number of runs one-half fractional factorial design, 241 is selected which requires only 8 runs. The construction of the design is shown in Table 2 by considering the 241 design in four parameters (cell temperature, borohydride concentration, anode and cathode flow rates). C. Fuel Cell Performance Tests The anode electrocatalyst ink was prepared by mixing 2propanol with 5 wt% Nafion solution and 60 wt.% PtRu/C. Then, the ink was coated on a carbon cloth (Electrochem, Inc.), yielding a metal loading mass on the electrode of 0.89 mg cm2. The Nafion 117 membrane was cleaned by boiling in 3 wt.% H2O2 and 3 wt.% H2SO4 for 1 h, followed by boiling in ultrapure water for 2 h. The cleaned membrane was activated in 2 M NaOH solution for 1 h prior to use. The anode/membrane/cathode unit was compressed between two graphite blocks with pin type flow fields (Figure 2). Silicon gaskets were assembled between the electrode and the graphite block. The active area of the fuel cell was 25 cm2. Cell performance was tested against a 1 mg cm2 Pt/C cathode coated on carbon paper. Peristaltic pumps were used to feed the fresh anolyte (an aqueous solution of 1M NaBH4 and 6M NaOH) at a 3 cm3 min1 flow rate. The flow rate of oxidant at the cathode chamber was 0.2 dm-3 min1. The oxidant was humidified by passing through a bubbler at 65C. Cell performance data were obtained using an electrochemical fuel cell test system (Electrochem 400 W) and a computer controlled E-load system (ECL 150).
TABLE II. Run
Factor 1 Cell temperatu re
Pecsok first suggested use of alkaline borohydride solutions as the fuel in 1953 [2]. In the development of the DBFC, many researchers were focused on the investigation of reaction mechanisms, the development of catalyst and design of the system. Various anode electrocatalyst materials have been investigated, such as Pt, Pd, Au, Ag, Ru, Ni, Os and its alloy, ZrCrNi alloy, Ni, Cu, Zn, ZnNi, ZnMH non-noble metals and AB5, AB2 hydrogen storage alloys. The most of the metals and alloys studied in relation to BH4- oxidation, the complete eight-electron exchange is not achieved due to hydrolysis of borohydride[1-9]. The borohydride oxidation pathway can change depending on the catalyst type and operation conditions. In this work, a DBFC employing PtRu/C as the anode catalyst and Pt/C as the cathode catalyst is used to control of reaction (4) by operation conditions. The effects of cell temperature, borohydride concentration, flow rates of fuel and oxidant on the performance of DBHFCs are investigated by using statistical approach in design and analyzing of experiments. Their higher and lower values of input factors are given in Table 1. The effect of the significant parameters on the performance of DBFC is determined experimentally. II. EXPERIMENTAL
DESIGN TABLE FOR THE EXPERIMENTS
A. Materials and Chemicals The chemical reagents, including 2-propanol, NaOH, NaBH4, and 5 wt. % Nafion solution (Sigma-Aldrich), Nafion 117 membrane (DuPont, USA), 20 wt. % Pt/C on Vulcan XC72 carbon support (BASF Fuel Cell, Inc.) are all purchased. Homemade 60 wt.% PtRu/C electrocatalyst was used in the electrode preparation.
TABLE I. FACTORS AND LEVELS OF INPUT FACTORS FOR EXPERIMENTAL
DESIGN
Factor 2 Concentrati on
Factor 3 Anode flow rate
Factor 4 Cathode flow rate
1 2 3 4 5 6 7
1 -1 1 -1 1 -1 -1 1
1 1 -1 -1 1 1 -1 -1
1 -1 1 -1 -1 1 1 -1
1 1 -1 -1 -1 -1 1 1
Input Factors
Low
High
Cell temperature, oC Borohydride M concentration,

3 -1 -1
8 50 0.5 1.0 0.1 80 1.5 5.0 0.4
Anode flow rate, cm min

3
Cathode flow rate,dm min
According to Figure 3, the highest performance was obtained at Run 3 (80oC, 0.5 M borohydride concentration, 5 cm3min-1 anode and 0.1 dm3min-1 flow rates). On the other hand, the highest hydrogen production rate was obtained at Run 1 (80oC, 1.5 M borohydride concentration, 5 cm3min-1 anode and 0.4 dm3min-1 flow rates). The data were analyzed using DESIGN EXPERT software to find the effect of each factor on the output response. Based on percentage contribution values obtained, the main effects were ranked as anode feeding rate (49.33%), borohydride concentration (16.64%) , cathode feeding rate (11.76%), and temperature (6.2 %). Although single temperature effect is small, interaction effects of temperature with concentration have substantial influences on cell performance and because of supporting hierarchy, temperature effect is taken into consideration in model equation. According to the results obtained from the analysis of both the probability plot and ANOVA analysis the estimated maximum power density is given as;
Figure 2. The used flow field for DBFC performance test.
III.
Power Density = +39.65604 + 0.096417 x Stack Temperature + 7.15167 x Borohydride Conc. + 1.04812 x Anode feeding rate-6.82500 x Cathode Feeding Rate -0.14683 x Stack Temperature x Borohydride Conc. For hydrogen production, the most significant effect on hydrogen production is anode feed rate by 46.86 % contribution. Using the analysis results of both the probability plot and ANOVA analysis the estimated hydrogen production rate is given as; Hydrogen production rate = +155.87500 - 4.48125 Temperature - 186.50000 x Borohydride Conc. 20.77500 x Anode feeding rate + 5.87500 Temperature x Borohydride Conc.+0.81125 Temperature x Anode feeding rate x Stack x Stack x Stack
A comparative investigation of DBFC polarization behavior was carried out with respect to the effect of cell temperature, borohydride concentration, flow rates of fuel and oxidant. The higher and lower level values were used in a statistical analysis to determine significant parameters. In Figure 3, the polarization curves were given for different runs. The production rates of hydrogen via hydrolysis reaction of sodium borohydride were also measured. The power density of different runs at 0.5 V and hydrogen production rate were used as output response, which were given in Table III.
Figure 4a and Figure 4b show actual values versus those from the model for power density and hydrogen production, respectively. The figures confirm that the predicted data of the response from the model are in agreement with observed data.
TABLE III. Run
OUTPUT RESPONSE
Response 1 Power density, mwcm-2
Response 2 Hydrogen production rate, cm 3min-1
1 2 3 4 5 6 7 Figure 3. The polarization curves for different runs. 8
42.6 49.2 49.4 44.4 41.2 47.3 47.3 22.0
485 104.1 118 13.6 225 200 76 25
ACKNOWLEDGMENT The authors are highly grateful to Mr. Aydin Canbasa for his technical assistance throughout this study.
REFERENCES
[1] H. Cheng, K. Scott, and K. Lovell, Material aspects of the design and operation of direct borohydride fuel cells, Fuel Cells, vol. 6, pp. 367 375, 2006. B.H. Liu and Z.P. Li, Current status and progress of direct borohydride fuel cell technology development, Journal of Power Sources, vol.187,pp. 291297,2009. J. Ma, N. A. Choudhury and Y. Sahai, A comprehensive review of direct borohydride fuel cells, Renewable and Sustainable Energy Reviews, vol. 14, pp. 183199, 2010. C.Celik, F.G. Boyaci San and HI Sarac, Effects of operation conditions on direct borohydride fuel cell performance, Journal of Power Sources , vol. 185, pp. 197-201, 2008. I. Merino-Jimnez, C. Ponce de Len,A.A. Shah and F.C. Walsh, Developments in direct borohydride fuel cells and remaining challenges, Journal of Power Sources, vol. 219, pp. 339-357, 2012. D.M.F Santos and C.A.C Sequeira, Zinc Anode for Direct Borohydride Fuel Cells, Journal of The Electrochemical Society, vol.157, pp. B13B19, 2010. H. Wu, C.Wang, Z.Liu and Z. Mao, Influence of operation conditions on direct NaBH4/H2O2 fuel cell performance, International Journal of Hydrogen Energy, Vol. 35, pp. 2648-2651, 2012. L.Wang, C. Ma, Y. Sun and S. Suda, AB5-type hydrogen storage alloy used as anodic materials in borohydride fuel cell, Journal of Alloys and Compounds, vol. 391, pp. 318322, 2005. V. W. S. Lam, A. Alfantazi and E. L. Gyenge, The effect of catalyst support on the performance of PtRu in direct borohydride fuel cell anodes, Journal of Applied Electrochemistry, vol. 39, pp. 1763-1770, 2009.
[2] (a) [3]
[4]
[5]
[6]
[7]
[8] (b) Figure 4. The actual data versus the predicted data for power density (a) and hydrogen production rate (b).
[9]
Salt Fog Testing of Glass Insulators with Different Surface Conditions
Luiz Henrique Meyer; Wagner Werner Beyer

University of Blumenau: Dept of Electrical and Telecommunications Engineering Blumenau - BRAZIL meyer@furb.br
Fernando Hidalgo Molina

Centrais Eltricas de Santa Catarina: Dept of Maintenance and Technical Inspection Florianopolis - BRAZIL fernandohm@celesc.com.br
AbstractThis work explores the salt fog test to evaluate the performance of glass insulators used in a 230 kV transmission line. Three types of glass insulators were used, from the field with original contamination, a clean one, and a RTV coated one. The results have demonstrated that the pre-contamination have a role in salt fog performance and that the RTV coated insulator has shown least leakage current and was the last to fail in the salt fog. Keywords-salt fog; glass insulator; RTV coating; leakage current, contamination performance.
be analyzed in many different ways, like its peak and average values. Other workers have explored the spectral analysis, in terms of harmonics, wavelets, and other more sophisticated methods. In this work, the leakage current is investigated in terms of its moving average, which shows the general trend, filtering rapid variations of its RMS value. [5, 6] II. MATERIALS AND METHODS
I.
INTRODUCTION
Transmission lines installed under heavy fog conditions are subjected to the problem of sudden flashover, as reported by many researchers [1]. The problem is especially strong in the Nordic countries were research is being carried out [1, 2]. In Brazil, some utilities in the southern part also experience such a problem, where the protection of the transmission line actuates with no apparent reason, frequently and, sometimes, three times a day. The tripping normally occurs during overnight, typically a couple of hours before sunrise [3]. The problem is described as a sudden flashover, and is proposed to be due to the inhomogeneous humidity degrees among the insulators in the string. The electrical field enhancement in the live extremity of the string is believed to be causing the variation in the humidity, drying out the insulator pieces closer to the line [1]. Since the other insulators in the string are humid, they have more surface conductivity, bringing ground closer to the live conductor, thus providing conditions for a partial flashover on the dried out insulators. This partial flashover thus initiates the full flashover in the string. In order to try to provide some solution, this work evaluates the performance of the RTV glass coated insulators in the salt fog chamber, with very light contamination, in comparison to the common glass insulator typically used in the 230 kV transmission line. RTV glass coated insulators are being employed in many new transmission lines, providing superior performance under contamination [4]. To compare the performance of different insulator the leakage current in the salt fog chamber is recorded. The leakage current can provide very useful information about insulation performance, and can
A. Insulators Three insulators were tested in the salt fog chamber, all of them of glass. One insulator is an insulator coated with RTV, commercially available. The second insulator was taken from the line, with original contamination from the field deposited on its surface. The other sample was also taken from the line, but it was cleaned before testing. Table 1 summarizes some specifications of the tested insulators.
TABLE I. GLASS INSULATORS TESTED IN THE SALT FOG CHAMBER Specifications Sample # and description
Details Leakage distance (mm) Dry Supportable Voltage (kV)
1:Glass insulator with RTV, new.
365
60
2: Glass insulator taken from the field, cleaned before testing. 3: Glass insulator taken from the field, with original surface contamination.
320
80
320
80
These samples were tested in the salt fog chamber, as described in next section.
The test routine is described in Table II, which shows the main parameters of the salt fog testing. It can be seen in Table II that there are rest periods, reflecting the conditions that could be found in the field.
TABLE II. TEST CONDITIONS AND PROCEDURE Start time / time duration (h) 0:00 / 4:23 Flow Rate (ml/min) 0.1
B. Salt Fog Chamber The chamber used in work has dimensions of 2m x 2m x 2m, a spraying system, a high voltage supply and a data acquisition system, as shown in Fig. 1. The voltage applied was 13.8kV, +/- 5%, 60Hz, regulated manually through a variac. The spraying system consists of 4 nozzles located in the corners of the chamber, spraying during one minute, one at a time, in a cyclic way. The data acquisition runs at 6 ksamples/s allowing for real-time leakage current monitoring. Fig. 2 shows the interface developed to monitor and record the leakage current data.
Condition
Voltage (kV)
Conductivity (S/cm)
No fog - dry
13.8 15h rest
Clean fog
4:23 / 3:35
13.8 20h rest
1.6 7.8
0.1
7 kg/m3
7:58 / 3:57
13.8 2h rest
11480
0.1
14 kg/m3 Salt fog 28 kg/m3
11:55 / 3:17
13.8 112h rest
22100
0.1
15:12 / 1:25
13.8 1h30min rest
41600
0.1
56 kg/m3
16:37 / 0:19
13.8
77500
0.1
The leakage current recorded during the test in the salt fog chamber, according to the parameters used is shown in the next section.
Figure 1. Insulators arrangement in the salt fog chamber.
III.
Fig. 3 shows the leakage current of the three insulators tested, for the whole period of test in the salt fog chamber, while Figs. 4-9 shows each part of the test, separately.
8 7 Leakage Current (mA) 6 5 4 3 2 1 0 0 02:25 04:51 07:16 09:42 Time (h) 12:07 14:33 17:00 Disc 1 Disc 2 Disc 3
Figure 2. Screeshot of the Labview interface of the data acquisition system.
Figure 3. Moving average of the leakage current for the whole test, including dry period, clean fog, and salt fog with incresing salt concentration, as shown in Table II.
Identify applicable sponsor/s here. If no sponsors, delete this text box. (sponsors)
From Table II it can be seen that the total period of test was of 167 hours approximately, although the energized period was of about 17 hours. From Fig. 3, it can be seen that at every step of the test, there was an increase at the magnitude of the leakage current. Also, it can be observed that at higher conductivity, two of the insulators have registered currents higher than 200mA, which opens the protection fuse of the data acquisition system.
Fig. 4 shows the moving average of the leakage current in dry condition, with no fog applied. When the insulators are first energized in the chamber, the polluted one show an higher current than others, probably due to moisture accumulated in the pollution layer. After about 2h30min of testing time, the currents in all 3 disc have stabilized, the polluted one (disc #3) around 0.32mA, while other two discs have currents of about 0.28 mA.
0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 00:34 01:08 01:42 02:16 Time (h) 02:50 03:24 Disc 1 Disc 2 Disc 3 03:58 04:30 Leakage Current (mA) Leakage Current (mA) 2 Disc 1 Disc 2 Disc 3
1.5
0.5 0 00:35 01:10 01:45 Time (h) 02:20 02:55 03:30
Figure 4. Moving average of the leakage current for the first part of the test, in dry condition.
0.5 Disc 1 Disc 2 Disc 3
Figure 7. Moving average of the leakage current for the fourth part of the test, with salt concentration of 14 kg/m3, conductivity of 22100 S/cm.
3 2.5 Leakage Current (mA) 2 1.5 1 0.5 0 0 Disc 1 Disc 2 Disc 3
Leakage Current (mA)
0.45
0.4
0.35
0.3
0.25 0
00:34
01:08
01:42 02:16 Time (h)
02:50
03:24
04:00
00:17
00:34
00:51 Time (h)
01:08
01:25
01:45
Figure 5. Moving average of the leakage current for the second part of the test, in clean fog condition.
1.6 1.4 Leakage Current (mA) 1.2 1 0.8 0.6 0.4 0.2 0 00:32 01:04 01:36 02:08 Time (h) 02:40 03:12 03:44 04:15 Disc 1 Disc 2 Disc 3
Figure 8. Moving average of the leakage current for the fifth part of the test, with salt concentration of 28 kg/m3, conductivity of 41600 S/cm.
8 7 Leakage Current (mA) 6 5 4 3 2 1 0 0 00:03 00:06 00:09 00:12 Time (h) 00:15 00:18 00:21 Disc 1 Disc 2 Disc 3
Figure 6. Moving average of the leakage current for the third part of the test, with salt concentration of 7 kg/m3, conductivity of 11480 S/cm.
Figure 9. Moving average of the leakage current for the sixth part of the test, with salt concentration of 56 kg/m3, conductivity of 77500 S/cm.
After the dry period of testing, the specimens were subjected to clean fog in the chamber, as seen in Fig. 5. In this period, the current in disc #3 have been practically constant, around 0.37mA, while discs #1 and #2 have their currents raised during the test, where disc #2 actually developed higher current than disc #3, towards the end of the test. Fig. 6 shows the first period of salt fog, as shown in Table II. In this period the order of leakage current have reversed, and disc #1 have shown discharge activity and higher current than other discs. As the test progresses to the salt concentration of 14 kg/m3, as shown in Fig. 7, the order is reversed again, with disc #1 showing lower values of leakage current and disc #3 the higher ones, especially more at the end of test. At the fifth period of the test, shown in Fig. 8, disc #3 have failed at the very beginning of the test, once its fuse have registered currents above 200mA. In this same test, about 15 minutes later, the disc #2 have failed, and only disc #3 stayed energized. Finally, in the last part of the test, disc #1 was energized and worked for about 20min, when it also failed, as shown in Fig. 9.
ACKNOWLEDGMENT The authors would like to acknowledge the high voltage laboratory personnel at University of Blumenau and Eng. Leandro Puchale at CEEE Electrification Company for the provided insulators.
REFERENCES
[1] INMR Magazine: Research Provides Insight into Unexplained Line Outages. Issue 94, Quarter Four 2011, Volume 19, Number 4, pp 7884. Gorur, R.S.; de la O, A.; El-Kishky, H.; Chowdhary, M.; Mukherjee, H.; Sundaram, R.; Burnham, J.T.; Sudden flashover of nonceramic insulators in artificial contamination tests . IEEE Trans. On Dielectrics and Electrical Insulation, Vol.: 4 , Issue: 1, 1997, Page(s): 79 87. Mustafa, T.I.A.H. ; Meyer, L.H. ; Cabral, S.H.L. ; Almaguer, H.D. ; Puchale, L.B.; Study of the correlation between weather conditions and protection trips in a 230 kV transmission line in southern Brazil. IEEE Transmission and Distribution Conference and Exposition (T&D), 2012 pp. 1 5. Cherney, E.A. ; El-Hag, A. ; Li, S. ; Gorur, R.S. ; Meyer, L. ; Ramirez, I. ; Marzinotto, M. ; George, J.; RTV silicone rubber pre-coated ceramic insulators for transmission lines. IEEE Trans. on Dielectrics and Electrical Insulation, Vol. 20 , Issue 1, 2013, pp. 237-244. El-Hag, A.H. ; Jayaram, S.H. ; Cherney, E.A.; Fundamental and low frequency harmonic components of leakage current as a diagnostic tool to study aging of RTV and HTV silicone rubber in salt-fog. IEEE Trans. on Dielectrics and Electrical Insulation, Vol.: 10, Issue 1, 2003, pp. 128 136; 2003. Sato, M. ; Nakajima, A. ; Komukai, T. ; Oyamada, T.; Spectral analysis of leakage current on contaminated insulators by auto regressive method. Annual Report Conference on Electrical Insulation and Dielectric Phenomena. Volume: 1, 1998 , pp. 64 66.
[2]
[3]
IV.
CONCLUSION
[4]
Salt fog testing of disk type insulators have been carried out, employing a piece from the field, with original contamination, another similar one, but cleaned before testing and a glass insulator with RTV coating. The procedure of testing employed a crescent fog stress, starting from dry, clean fog and salt fog at four increasing levels of salinity, from 7 to 56 kg/m3. It can be concluded that RTV coated insulators had the best performance, having lower levels of leakage current, and lasting longer in the test than other two discs. Also, although the contamination of the place where the actual line runs is very light, still it has contributed to determine the worst performance of the three tested insulators.
[5]
[6]
A Novel Feed-Forward Controller for PMSMs

Yusuf ALTUN
Faculty of ElectricalElectronics, Control and Automation Engineering Yildiz Technical University, Istanbul, Turkey altunyusf@hotmail.com
Kayhan GULEZ
Faculty of ElectricalElectronics, Control and Automation Engineering Yildiz Technical University Istanbul, Turkey gulez@yildiz.edu.tr
Tarik Veli MUMCU

Faculty of ElectricalElectronics, Control and Automation Engineering Yildiz Technical University Istanbul, Turkey tmumcu@yildiz.edu.tr
M. zgr KIZILKAYA
Aeronautics & Space Tech Inst., Electronic Engineering Department, Turkish Air Force Academy, Istanbul, Turkey mkizilkaya@hho.edu.tr
.AbstractIn this study, the current tracking problem of PMSM is handled as static feed-forward controller design for Linear Parameter Varying (LPV) system. H infinity controller formulated by means of Linear Matrix inequalities (LMI) with bounded and measurable system parameters is used. Parameterdependent Lyapunov function is preferred to achieve robust stability against inconstant parameters, throughout the control process. The LPV controller is used in two degrees of freedom (2DOF), by jointly adding static LPV feed-forward controller on a PI feedback controller. Space vector pulse width modulation (SVPWM) is used to generate switching signals. Overall system is simulated by Matlab/Simulink Software to evaluate the validity of the method in transient and steady state response by computer simulation. Simulation results reveal that proposed method has good dynamic performance and robustness. KeywordsFeedforward Control, PMSM Current Control, H controller, Robust control, SVPWM, Matlab, Simulink. I. INTRODUCTION Permanent Magnet Synchronous Motor (PMSM) has been extensively studied, owing to its fast dynamic response, high power density and efficiency. PMSM has almost instant torque capability when adequate current tracking techniques are applied. The improvement on the performance of PMSM is always a hot topic. There are a few studies using feed-forward control to achieve robust and stable performance for varying parameters e.g. by means of parameter identification as in [1] or intelligent control as in [2] Linear Parameter Varying (LPV) feed-forward control is an effective technique to achieve stability and performance for non-linear systems with varying parameters. Method have been applied to different high performance systems e.g. wind turbine [3], aircraft [4], inverted pendulum [5], fuel cell power generator [6]. There are different approaches in literature to acquire LPV controller synthesis, Rangajeeva used polytopic presentation in [7] and Apkarian in [8]. Becker used single Lyapunov function as in [9]. Recently Gahinet and Wang Na used affine Parameter Dependent Lyapunov Function (PDLF) to counter preservation increase due to single Lyapunov function [10,11] The aim of the feed-forward controller is to improve the reference tracking or to suppress the observable and measurable disturbances of an existing or deficient system. The
joint usage of feed-forward controller with a feedback controller in a 2-DOF control is a common approach as in [1213]. The joint usage is either by simultaneous design of feedback and feed-forward controller as in [15] or feed-forward controller is designed on a feedback controller as in [14] as which feedback achieves stable and tolerable output, while feed-forward controller enhances the system performance. There are various studies to accomplish control design for LPV systems by using available state and output feedbacks as in [17-19], besides different feed-forward control approaches for LTI systems, are proposed as in [15,20 and 21].A simple static feed-forward control design, separated from feedback controller, for LPV system is studied in [16]. There exists limited study on H feed-forward controller design for LPV systems in the literature. In this design, a new static feed-forward controller for PMSM current control is proposed as in [22] we add a feedforward controller on an existing PI feedback controller in the 2-DOF structure. II. MATHEMATICAL MODEL OF PMSM For verification of the method a well-known, nonlinear mathematical model of PMSM in d-q axis rotating frame is used by the following equations
Vsd = Rs isd + d sd pm sq dt d sq + pm sd dt
(1)
Vsq = Rs isq +
(2)
disd 1 = (Vsd Rsisd + Lsq pmisq ) dt Lsd

disq dt = 1 (Vsq Rs isq Lsd pmisd pm ) Lsq
(3)
(4)
Te =
3 p isq + ( Lsd Lsq ) isd isq 2
(5)
d m 1 = (Te TL Bm ) dt J d m = m dt
(6)
(7)
Where Vsd and Vsq are stator voltages, isd and isq are stator currents defined on d q frame, Rs is the stator resistance, Lsd and Lsq are stator d and q axis inductances, Te is the motor torque, TL is the load torque, is the flux and is the flux linkage, wm is the mechanical speed of the rotor, m is the mechanical rotor angular position and B is the friction coefficient. The nonlinear model of PMSM in state space format, according to LPV structure can be defined by (8) where Ap, Bp Bf and Cp are system matrix, input matrix, disturbance vector and output matrix, respectively defined in (9-12). The state variables and the control signals are defined in (13).
x = Ap ( m ) x + B p u + B f ( m ) y = Cp x
Rs Lsd Ap (m ) = L pm sd Lsq Lsq pm Lsd R s Lsq
0 Ap1 = L p sd Lsq III.
Lsq Lsd 0
(15)
STATIC LPV FEED-FORWARD CONTROL OF PMSM
The general form of system G() used in LPV H controller design is defined in (16) in the state space form where x(t ) are the states, z (t ) are the controlled outputs, u (t ) are the control inputs, w(t ) are disturbance effects and y (t ) are the outputs of the system. All system matrices of state space representation are a function of ( t ) which is defined by (17).
x ( t ) = A ( ( t ) ) x ( t ) + B1 ( ( t ) ) ( t ) + B2 ( ( t ) ) u ( t ) z ( t ) = C1 ( ( t ) ) x ( t ) + D11 ( ( t ) ) ( t ) + D12 ( ( t ) ) u ( t )
(16)
(8)
y ( t ) = C2 ( ( t ) ) x ( t ) + D21 ( ( t ) ) ( t ) + D22 ( ( t ) ) u ( t )
D
(9)
{ R
: i i i , i = 1,..., n
R n : i i i , i = 1,..., n
(17)
1 L sd Bp = 0
0 1 Lsq
(10)
Theorem 1: Consider the LPV plant of the form (16) with parameter trajectories constrained by (17). If there exists the symmetric positive definite matrix Y ( ) defined by (18), there exists the parameter dependent static feed-forward controller in (20) for the closed loop system having L2 gain such that the infinite LMIs in (21) and (22) hold for all pairs
( , ) in .

Proof: see [22] (11)
0 B f ( m ) = mL sq
1 0 Cp = 0 1
(12)
In this case; the controller depended on parameters are obtained as in (19). When LMIs, which are in (21) and (22), are solved by minimizing in the multiconvexity conditions in [17], the controller matrix defined in (20) is obtained for the optimal control.
Y ( ) = Y0 + i Yi 0
i =1 n
(18) (19) (20)
Vsd isd x = , u = , i sq Vsq
(13)
K ff ( ) = K ( )
Ap , system matrix of PMSM is a function of mechanical
speed of rotor, where Ap 0 and Ap1 are defined by (14) and (15).
Rs L sd = 0 0 Rs Lsq
K ff ( ( t ) ) = K0 + i Ki 0
i =1
Ap 0
(14)
T Y ( ) AT ( ) + A ( ) Y ( ) + Y ( ) B1 ( ) + B2 ( ) K ( ) Y ( ) C1T ( ) Y ( ) C2 ( ) T T T T T T D11 ( ) + K ( ) D12 ( ) D21 ( ) + K ( ) D22 ( ) * I 0 I * * 0 * * * I
(21)
Y ( )
(22)
Static LPV feedback controller is capable to improve the performance of a closed loop, stable system. In this section novel static feed-forward LPV controller method formed by (23), is designed on a feedback controlled PMSM system. A joint 2-DOF controller is proposed. The block diagram of proposed overall PMSM drive system based on static feed-forward controller on a usual PI controller is shown in Fig. 1. In this controller design, Pr ( ) is a parameter dependent system to be controlled, a well-known PI feedback controller is fixed for stable compensation. The static feed-forward controller is built on, to attenuate the tracking errors. Even though this controller is also beneficial for disturbance rejection, in this study we did not concern this issue. This control system is presented in Fig.2.
Thus, the parameter dependent L2-gain controller is defined by (23) for LPV system as in [19].
K ff ( )
G ( )
Figure 2. The feed-forward LPV control scheme
IV. CONTROL SYSTEM SIMULATION AND RESULTS With the principle of above analysis, the suggested control system is implemented in Matlab/Simulink environment. A block diagram of the proposed PMSM drive system using SVPWM is shown in Fig 3, Simulation Parameters; PMSM Stator resistance Rs = 0.9585 , d and q axis inductance Lsd = 0.0025 H Lsq = 0.0025 H, Permanent Magnet Flux = 0.01827, SVPWM frequency = 3 kHz, DC bus voltage = 50 VDC The current tracking performance of system is compared with and without feed-forward LPV block The system performance is simulated by means of current tracking for step change of reference current and a trapezoidal waveform with increment from 1 to 6 Ampere in 200ms. The rotor frame q axis, Isq current reference was changed while d axis portion, Isd reference was kept at zero amper. Fig. 4 shows the performance comparison of system with and without feedforward controller. Solid lines are system outputs of LPV controller in 2-DOF structure, while dashed lines are using PI controller only.
u f = K ff ( )
K ff ( )
(23)
Pr ( )
Figure 1. Feed-forward Reference tracking control system
The solution is obtained to solve PDLF via LMI for the static LPV feed-forward controller design problem in Fig. 1.
Figure 3. Simulink Model of PMSM with LPV Control
14 12
Isd & Isq Current (A)
10 8 6 4 2 0 -2 0 0.2 0.4 0.6 0.8 1
current is corrupted in d axis due to filters selection. Thus, it is studied on it. The applicability of the method to a PMSM control system by means of SVPWM is also shown. The simulation results attract attention to the novel control design. REFERENCES
[1] Zhao S, Cui L, Liu G, Chen Y, An improved torque feed-forward control with observer-based inertia identification in PMSM Drives 2012, 15th International Conference on Electrical Machins and Systems (ICEMS), Oct. 2012 pp 1-6 Edwar, I. Feed forward FLC for PMSM drives, Vehicle Electronics Conference, IVEC 2001, Proceedings of the IEEE International, pages 21-26, Sep.2001 Shirazi, Farzad A. An integratd approach towards structural and LPV controller design in wind turbines, American Control Conference ACC,2012, Conferenece Publications, pages 5789-5794, June.2012 Zhu, Qi-Dan Switching LPV control for aircraft lateral-directional axis during powered approach, Information and Automation, ICIA 2012, Conference Publications, pages 827-831 June 2012 Kajiwara, Hiroyuki, Apkarian, Pierre; Gahinet, Pascal M, LPV techniques fo control of an inverted pendulum IEEE Control Systems, vol 19, no 1, pp 44-54, Feb 1999 Hernndez -Torres, David Sename, Oliver; Riu, Delphine M., "An LPV control approach for fuel cell power generator air supply system", American Control Conference, pp. 4299- 4304,29 June 2012. Apkarian, P.; Gahinet, P.; Becker, G.; "Self-scheduled H linear parameter-varying systems", American Control Conference, vol.1, pp. 856- 860, 1 July 1994. Ali, M.; Abbas, H.; Werner, H.; "Controller synthesis for input-output LPV models",49th IEEE Conference on Decision and Control (CDC), pp.7694-7699, 15-17 Dec. 2010. G. Becker, A. Packard, D. Philbrick and G. Blas, Control of parametrically-dependent linear systems: a single quadratic lyapunov approach, in Proc. ACC 1993, pp.2795-2799, 1993. Gahinet, P.; Apkarian, P.; Chilali, M.; , "Affine parameter-dependent Lyapunov functions and real parametric uncertainty",IEEE Transactions on Automatic Control, vol.41, no.3, pp.436-442, Mar 1996. Wang Na; Zhao Ke-You; , "Parameter-dependent Lyapunov function approach to stability analysis for discrete-time LPV systems", 2007 IEEE International Conference on Automation and Logistics, pp.724728, 18-21 Aug. 2007. Danapalasingam, K.A.; la Cour-Harbo, A.; Bisgaard, M., "Disturbance effects in nonlinear control systems and feed-forward control strategy",2009 IEEE International Conference on Control and Automation,. ICCA 2009, pp.1974-1979, 9-11 Dec. 2009. Stan H. van der Meulen, Rob L. Tousain, and Okko H. Bosgra, "Fixed Structure Feed-forward Controller Design Exploiting Iterative Trials: Application to a Wafer Stage and a Desktop Printer",J. Dyn. Sys., Meas., Control, vol.130, No.5, September 2008. Kose I. E. and Scherer C. W., "Robust 2-gain feed-forward control of uncertain systems using dynamic IQCs", International Journal of Robust and Nonlinear Control, vol.19, no.11, pages 12241247, 25 July 2009. S. Skogestad and I. Postlethwaite, Multivariable Feedback Control, Analysis and Design, John Wiley & Sons, 1996. Prempain, E.; Postlethwaite, I., "A feed-forward control synthesis approach for LPV systems", 2008 American Control Conference, pp.3589-3594, 11-13 June 2008. Bruzelius, F., Breitholtz, C. and Pettersson, S., "LPV-Based Gain Scheduling Technique Applied to a Turbo Fan Engine Model", Proceedings of the 2002 International Conference on Control Applications, pp.713- 718, 2002. Fan Wang; Balakrishnan, V.; "Improved stability analysis and gainscheduled controller synthesis for parameter-dependent systems", IEEE Transactions on Automatic Control, vol.47, no.5, pp.720-734, May 2002.
[2]
[3]
Time (s)
Figure 4. Current tracking performance joint and PI only controller, [4]

14 12
[5]
10 8 6 4 2 0 -2 0 0.2 0.4 0.6 0.8 1
[6]
[7]
[8]
[9]
Figure 5. Performance of the system using SVPWM

14 12
[10]
10 8 6 4 2 0 -2 0 0.2 0.4 0.6 0.8 1
[11]
[12]
[13]
Time (s)
Figure 6. System output with and without SVPWM
[14]
Fig.5 shows the performance comparison of the system with and without feedback controller using SVPWM. Fig.6 shows the output of the system with and without SVPWM, when feed-forward controller is used, performance of the SVPWM to follow the controller output Vd and Vq is visualized. V. CONCLUSION This study illustrates the static parameter dependent feedforward control simulation of PMSM, which is an LPV system. The simulation results validate that the novel feed-forward control strategy with PI has given a better response and performance, according to step and ramp references, with respect to PI feedback controller alone. However, the stator
[15] [16]
[17]
[18]
[19] Gahinet, P.; Apkarian, P.; Chilali, M.; "Affine parameter-dependent Lyapunov functions and real parametric uncertainty", IEEE Transactions on Automatic Control, vol.41, no.3, pp.436-442, Mar 1996. [20] Giusto, A.; Paganini, F.; "Robust synthesis of feed-forward compensators", IEEE Transactions on Automatic Control, , vol.44, no.8, pp.1578-1582, Aug 1999. [21] Scorletti, G.; Fromion, V.; "Further results on the design of robust H feed-forward controllers and filters",2006 45th IEEE Conference on Decision and Control, pp.3560-3565, 13-15 Dec. 2006. [22] Altun, Y.; Gulez, K.; Mumcu T.V., "Static LPV feedforward controller synthesis for linear parameter-varying systems", 2013 Asian Control Conference, 23-26 June. 2013, in press.
Analysis of Various Transmission Line Switching Overvoltage Limitation Techniques

Smko Golabi1, Shahab Tanhaeidilmaghani
Faculty of Electrical and Computer Engineering, University of Tabriz Tabriz, Iran golabismko@gmail.com, shahab.tanhaei@gmail.com
Abstract In this paper performance of several transmission line switching overvoltage limitation techniques are analyzed and compared with each other. Application of closing resistors, intermediate surge arresters and controlled switching methods are simulated and analyzed. The effects of simultaneous application of the mentioned techniques are also investigated. The results of this comprehensive study help the utilities to properly select the suitable switching overvoltage limitation technique. Simulations and analyses are performed using a real 400 kV transmission line. Keywords: Switching overvoltage, insulation coordination, closing resistor, surge arrestor, controlled switching
Heresh Seyedi
Faculty of Electrical and Computer Engineering, University of Tabriz Tabriz, Iran hseyedi@tabrizu.ac.ir the opened transmission line. Following this auto-reclosing operation, the transient overvoltages with extreme high amplitude and frequency are produced. These overvoltage waves may be hazardous to the system insulation. Therefore, possible overvoltages must be evaluated during the design stage. Conventionally, transient studies are executed on the transmission lines using the EMTP, in order to obtain the switching overvoltages [3], [4]. Once the amplitude of overvoltages is obtained, it is compared with the SIWL of insulator. If the overvoltage exceeds SIWL, a technique must be adopted to reduce the prospective overvoltages. The most effective and practical techniques to reduce the switching overvoltages, may be classified as follows: Application of pre-insertion resistors Intermediate surge arresters Controlled switching [2]. Conventionally, a resistor is inserted in the transmission line circuit breaker to reduce the switching overvoltages. Although it significantly reduces the overvoltages, its failure rate may be unacceptable. Therefore, this method is not suggested by the utilities and manufacturers any more [5]-[8]. Normally, surge arresters are placed at both ends of the transmission lines. In addition to these surge arresters, one or two sets of arresters may be installed at suitable places along the line, in order to reduce the switching overvoltage. This method sufficiently decreases the overvoltages [9]-[11]. Another effective method to reduce the switching overvoltage is controlled or point-on-wave switching. It can effectively diminish the switching overvoltage. Recently, this technique has been more noticeable [8], [12]-[15]. Although a large number of research studies, related to the switching overvoltages are found in the literature, no comprehensive study has been conducted, considering all available techniques. Therefore, in this paper, all the three mentioned switching overvoltage reduction methods are completely and distinctly investigated. Then, comprehensive simulations are performed to figure out the effect of each method on the switching overvoltage reduction. In addition, simultaneous application of the methods is also investigated. For this purpose, a real 400 kV transmission line with 463 km length from the power system of Iran is selected. Simulations
I.
INTRODUCTION
One of the power system fundamental components is the transmission line. Concordantly, in the design of transmission lines, several factors must be considered, in order for the power system to be economically commodious and also reliable in performance. One of those issues, in the transmission lines design, that must certainly be considered, is the overvoltage problem. The overvoltages significantly impress the system insulation withstand level, which is indirectly related to the system reliability performance and economic problems. The most prominent causes of the overvoltages are switching as internal source, and lightning as external source. As the nominal system voltage increases, the overvoltage caused by the switching is subsequently intensified. Therefore, in the systems with the rated voltage above 300 kV, switching overvoltage is the most important factor to be considered in the insulation design [1], [2]. The dielectric voltages of devices and insulators are known as the system insulation level. Accordingly, for each rated voltage level, a standard insulation level has been proposed. [1] The system insulation level for the switching overvoltage is known as the Switching Impulse Withstand Level. In order to prevent the breakdown of insulators, during transmission line switching operations, the consequent transient overvoltages must be less than the line SIWL. When a short circuit occurs in the system, the transmission line is disconnected by the operation of protective relays. Immediately, the auto-reclosing relays operate to re-connect
1. Atrin Industries Co. Tabriz, Iran & Erbil, Iraq
for various cases are executed using EMTP. From the obtained simulation results, the impact of each of these methods on the switching overvoltages may clearly be observed. II. SWITCHING OVERVOLTAGES
In general, the abrupt changes in the power system structure, such as sudden disconnection of the transmission lines or system equipments, result in the temporary phenomena that produce transient waves, subsequently. The impact period of these waves is roughly within the range of microseconds. These waves produce the transient overvoltage waves with high frequency within 50 Hz to 20 kHz ranges and much fast damping. Switching overvoltages appear in various forms such as oscillatory, aperiodic and repetitive wave shapes. They are produced due to a variety of reasons, such as capacitive circuit disconnection, inductive load separation, energization of the transmission lines, disconnection and removal of the short circuit, transmission line auto-reclosing, etc. However, in this work, the produced overvoltages ensuing the switching including fast auto-reclosing are analyzed. Whenever a fault occurs in the system, the transmission line is separated by the line circuit breakers. Immediately, closing commands are sent to the circuit breakers by autoreclosing relays and the transmission line is re-connected to the system. As a result of this auto-reclosing operation, transient overvoltages with extremely high amplitude and frequency are produced, which may result in the breakdown of insulators. Therefore, if the maximum amplitude of switching overvoltages exceeds the line SIWL, it should be in some way decreased to prevent the insulation breakdown. III. SWITCHING OVERVOLTAGE REDUCTION METHODS
with the main contact. After a few milliseconds, normally 8 to 12 milliseconds, the main contact is closed. Afterwards, the resistor is bypassed and then cut off in a very short time by the series contact. During the presence of resistor in the circuit, due to the voltage drop and discharge of transmission line trapped charge, the overvoltages caused by auto reclosing is extremely diminished. Hence, the closing or pre-insertion resistor considerably reduces the switching overvoltages. Of course, it should be mentioned that the installation of closing resistor in the circuit breaker structure, in addition to the performance complexity, increases both cost and failure rate of the circuit breaker.
Resistor contact Closing resistor
Main contact
Transmission line
AC
Voltage source
Figure 1. Application of closing resistor to the circuit breaker
Various studies, to reduce the switching overvoltages, have been investigated in the literature and a number of techniques have been proposed. The most effective and practical methods include the use of the pre-insertion or closing resistor, application of surge arresters and controlled switching. As follows, these methods are separately explained in three next sub-sections. A. Application of Pre-insertion Resistors Normal and auto reclosing operations of the UHV transmission line circuit breakers are frequently performed in power systems, resulting in the high amplitude transient overvoltage. Therefore, transmission line circuit breakers are conventionally equipped with a pre-insertion resistor. If this resistor is inserted in the circuit breaker to reduce the overvoltages caused by the closing of switch, it is known as closing resistor. The execution manner of the closing resistor is in accordance with Fig.1. Whenever the closing command is issued by the auto reclosing relay, first, the transmission line is connected through a series resistor. The resistor is in parallel
B. Application of Inetermediates Surge Arresters Surge arresters prepare the conditions to protect against transient overvoltages. They are employed to decrease the transient waves of both switching and lightning. The arresters are placed in parallel with the device under protection, or between the phase and ground. Their performance instruction is in the manner that, they exert high resistance for the normal voltages, while exert low resistance against the overvoltages. Thereby, they deflect the transient overvoltage from the device under protection, or transfer it to the ground. Normally, surge arresters are conventionally installed at both ends of the transmission lines, the sending and receiving sides. However, a proposed method to reduce the switching overvoltages is the installation of one or two sets of arresters at appropriate locations along the transmission line, according to Fig.2. Hereby, the switching overvoltages are decreased; also the profile of overvoltage along the transmission line becomes flatter.
Transmission line Main contact
M O V
First arrester
M O V
moveable
M O V
End arrester
M O V
AC
Voltage source
Intermediate arresters
Figure 2. Installation of intermediate arresters along the line
IV.
CONTROLLED SWITCHING
Circuit breakers in the power systems perform the opening and closing operations of transmission lines without consideration of their current and voltage wave shapes, which may sometimes cause transient overcurrents and/or
overvoltages. These transient oscillations caused by the switching, depend on the insulation and mechanical stress withstand strength of ultra-high voltage equipments in the power system and may result in gradual and impulse damages in the appliances. Controlled or point-on-wave switching is a technique to prevent dangerous overvoltages. It functions with the time control of the switching operations. The closing or/and opening commands at the circuit breaker for the connection or/and disconnection of the switch contacts are performed with a delay. Hereby, the switching operations are accomplished at the intended points and angles of the closing or/and opening. [16] Many of the short circuit faults are temporary and spontaneously removed. Therefore, in practice, the autoreclosing switches have been used in the power systems. After the short circuit occurrence, these switches are once or twice or more reclosed until being ensured of the obviation of the short circuit. After auto-reclosing, if the short circuit still exists, the switch would remain open. Sudden disconnection and interruption is the main factor to produce the trapped charge in each side of the transmission line. The existence of trapped charge may cause overvoltages up-to twice nominal voltage between circuit breaker contacts at the instant of closing. This may result in high amplitude transient overvoltage waves across the line insulators. These overvoltages may be as high as 3 p.u. [2]. The trapped charge in transmission line, after the disconnection of the line, is gradually discharged through insulators. The complete discharge may take between 0.1 to 0.5 seconds. One simple but effective method to reduce the transient overvoltage is the discharge of trapped charge. According to results of [17], usage of Inductive Voltage Transformers helps to discharge the trapped charge. Therefore, IVT will cause overvoltages to reduce. However, it is impractical in EHV and UHV transmission lines, due to economical restrictions. The Effect of controlled switching or point-on-wave switching, on the reduction of switching overvoltages, is also considered in [17]. It is obvious that, the most suitable closing instant for the circuit breaker is the time when the voltage difference between two contacts of the breaker reaches its minimum value. Therefore, it can be stated that an appropriate instant for switching is the zero-crossing time of voltage [12], [16]. However, switching operation can be performed at different points of the wave. When the switching is executed at instants, except for the peak points of the voltage cycles, these instants can be in the positive or negative ramp. The fact that the ramp of the switching instant can affect the overvoltage amplitude may seem unimportant, but the results of simulations demonstrate that the overvoltage amplitudes are different at the positive and negative ramps. Sometimes, it is required to achieve the unity overvoltage amplitude to determine the insulation coordination. For this purpose, (1) is used to combine the overvoltages, related to
positive slope and negative slope zero crossings. Of course, it should be emphasized that modern relays are working, based on this approach.[18]
0.50 1
1 2
.exp -
x2 x2 1 2 dx + 0.5 .exp 0 - dx = .98 2 2 2
(1)
Where, U0 is known as the statistical overvoltage. According to (1), 98 percent of the prospective overvoltage amplitudes will be less than U0. Also 1 and 2 are defined as follows:
1
2
U 0 1
(2) (3)
U 0 2
Assuming the overvoltage to be a random variable, and are the mean and standard deviation of the variable, respectively. The subscripts 1 and 2 are related to the positive and negative slope zero-crossings, respectively. Solving (1), requires numeral calculations which are executed by the MATLAB software. Both Uniform and Gaussian distribution functions may be used for the closing time instant [19]. The Uniform and Gaussian distribution functions are depicted in Figs.3 and 4, respectively. According to Fig.3, length of covered area by the Uniform distribution is equal to 2 3 .
Mean
Figure 3. Covered area in the Uniform distribution
The covered area by the Gaussian distribution is mathematically unlimited, based on Fig.4. However, the area beyond 3, may practically be neglected. Therefore, length of covered area by the Gaussian distribution is assumed to be 6.
Covered Area
Fig.4. Covered area in the Gaussian distribution
V.
SIMULATION RESULTS
In this section, simulations are performed for various cases using EMTP. For this purpose, a 463 km 400 kV transmission line of the Iran power system is selected. The line is connected between Tabriz and Rajaee 400 kV substations. Switching operations are executed from Tabriz side. The transmission line is divided into ten sections, each section 46.3 km, in order to obtain the overvoltage profile along the line. The JMARTI model which is an accurate frequency-dependent model is selected for these ten sections. Other transmission lines, connected to Tabriz substation are also simulated with the JMARTI model. The rest of the system is modeled with the Thevenin equivalent. Basic diagram of the simulated system under study is depicted in Fig.5. To find more information about the simulation details refer to the literature [11], [20][27]. Simulations for the different cases are described in following sections:
3PH-TC
first stage, a 100 resistor is used. Duration of the resistor presence is assumed 8 milliseconds. Similarly, simulation is repeated for the 300 resistor. The results of simul ations are shown in Fig.7. It can clearly be seen that closing resistors extremely reduce the maximum amplitude of switching overvoltages. Also, by increasing the resistor value, the overvoltage amplitude is decreased more rapidly.
2.9 2.8 2.7 2.6 2.5 2.4 2.3 2.2 2.1 2 1.9 1.8 1.7 1.6 1.5 1.4 1.3 1.2 1.1 1
Overvoltage(p.u.)
normal with R 100 with R 300 0 46.3 92.6 138.9 185.2 231.5 277.8 Length(km) 324.1 370.4 416.7 463
Figure 7. The profiles of the switching overvoltages for case b
TL CB ZE ZE ZE TL: Transmission Line SA: Surge Arrester ZE: Thevenin Equivalent Impedance 3ph-TC: 3Phase Trapped Charge CB: Circuit Breaker TL TL TL1 SA TL2 TL3 TL10 SA
Figure 5. Basic diagram of the simulated system using EMTP
2.9 2.8 2.7 2.6 2.5 2.4 2.3 2.2 2.1 2 1.9 1.8 1.7 1.6 1.5 1.4 1.3 1.2 1.1 1
Overvoltage(p.u.)
A. Base Case In this section, the switching is performed at Tabriz substation, with no overvoltage limitation technique. The overvoltage profile is depicted in Fig.6. Each point of this figure is a statistical overvoltage, obtained statistically after 100 consecutive switching operations. Maximum amplitude of the overvoltages is equal to 2.723 p.u. which occurs at 185.2 km, from Tabriz substation.
C. Application of Intermediate Surge Arresters In this section, according to the mentioned principles in section B, surge arresters are added to the base case. First, a surge arrester is placed in the middle of the line i.e. at 231.5 km from Tabriz. At the next stage, instead of a mid-point arrester, two sets of arresters are placed at the locations of 185.2 and 324.1 km from Tabriz. The results of all simulations are depicted in Fig.8 and compared with the base case. It shows that, the surge arrester reduces the maximum switching overvoltages. In addition, in case of using two arresters, the overvoltage reduction is more considerable and the profile is flatter.
2.9 2.8 2.7 2.6 2.5 2.4 2.3 2.2 2.1 2 1.9 1.8 1.7 1.6 1.5 1.4 1.3 1.2 1.1 1
Overvoltage (p.u.)
normal with one arrester with two arrester 0 46.3 92.6 138.9 185.2 231.5 277.8 324.1 370.4 416.7 463
Figure 8. the profiles of theLength(km) switching overvoltages for case c

normal 0 46.3 92.6 138.9 185.2 231.5 Length(km) 277.8 324.1 370.4 416.7
Figure 6. Switching overvoltages profile for case a
B. Application of Pre-insertion Resistors According to the mentioned principles in section A, in this stage, a closing resistor is applied to the circuit breaker. At the
D. Application of Controlled Switching In this section, simulations are executed for different cases of controlled switching, according to the principles, mentioned in section C. At the first stage, Uniform distribution function is used for closing operations. Simulations are performed for the 86.6% and 10% covered-zones around both peak and zero crossing
points. Fig.9 shows the results of the simulations. The abbreviations, used in Fig.9 are described as follows: 86.6% CP and 10% CP: it means that 100 consecutive switching operations are performed at random times. These random times belong to the 86.6% and 10% of the cycle around peak point of the voltage signal, respectively. 86.6% CZ and 10% CZ: it means that 100 consecutive switching operations are performed at random times. These random times belong to the 86.6% and 10% of the cycle around zero-crossing point of the voltage signal, respectively. At this stage, negative slope zero-crossing has been selected. According to the results of Fig.9, it is obvious that, maximum reduction of the overvoltages is related to the 10% CZ case. At the second stage, the Gaussian distribution function is selected for the closing operation times. The same simulations of the first stage are repeated at this stage. The results are depicted in Fig.10. Again, maximum reduction of the overvoltages is related to the 10% CZ case. At the third stage, since at the previous stages maximum reduction of the overvoltages was related to 10% CZ case, this case is simulated with the positive-slope zero-crossing. The results of this case are depicted in Fig.11. In this figure the results of negative-slope are shown again, just for the sake of comparison. Notice, in this figure, the abbreviations UD, GD and R stand for the Uniform Distribution, Gaussian distribution and Ramp (slope) of the zero-crossing, respectively.
2.9 2.8 2.7 2.6 2.5 2.4 2.3 2.2 2.1 2 1.9 1.8 1.7 1.6 1.5 1.4 1.3 1.2 1.1 1
2.9 2.8 2.7 2.6 2.5 2.4 2.3 2.2 2.1 2 1.9 1.8 1.7 1.6 1.5 1.4 1.3 1.2 1.1 1
Overvoltage(p.u.)
normal with 10% CZ, UD, R<0 with 10% CZ, GD, R<0 with 10% CZ, UD, R>0 with 10% CZ, GD, R>0 0 46.3 92.6 138.9 185.2 231.5 277.8 Length(km) 324.1 370.4 416.7 463
Figure 11. Profiles of the switching overvoltages for case d, stage 3
normal with 86.6% CP with 10% CP with 86.6% CZ with 10% CZ 0 46.3 92.6 138.9 185.2 231.5 Length(km) 277.8 324.1 370.4 416.7 463

2.9 2.8 2.7 2.6 2.5 2.4 2.3 2.2 2.1 2 1.9 1.8 1.7 1.6 1.5 1.4 1.3 1.2 1.1 1
E. Combination of Methods In this section, the simulations are carried out for different cases which apply two reduction methods of switching overvoltages, simultaneously. At the first stage, pre-insertion resistor and intermediate surge arresters are simultaneously utilized. First, while a 300 pre-insertion resistor is applied, an intermediate surge arrester is placed at the mid-point of the line, i.e. 231.5 km. Then, the same simulations are repeated for the placement of another arrester at a location close to the line end, i.e. 416.7 km. The results of the simulation are depicted in Fig. 12. According to the diagram, no considerable change in the overvoltage reductions is achieved, compared with the results of Fig.7. Notice the abbreviations R, MA and EA stand for Resistor, Middle Arrester and End Arrester, respectively. At the second stage, intermediate surge arresters and controlled switching are simultaneously utilized. First, while an intermediate surge arrester is placed in the middle of the line, 231.5 km, the 10% CZ case, introduced in case d, is simulated. Then, the same simulations are repeated for the placement of the two sets of arresters along line, at 185.2 and 324.1 km. The results of simulations are depicted in Fig. 13. According to the diagram, no considerable change in the overvoltage reduction is achieved, compared with the results of Figs.9-11. Of course, profile of the overvoltages becomes flatter. Notice the abbreviations R, CS, MA and EA stand for closing resistor, Controlled Switching, Middle Arrester and End Arresters, respectively.
2.9 2.8 2.7 2.6 2.5 2.4 2.3 2.2 2.1 2 1.9 1.8 1.7 1.6 1.5 1.4 1.3 1.2 1.1 1
Overvoltage(p.u.)
normal with 86.6% CP with 10% CP with 86.6% CZ with 10% CZ 0 46.3 92.6 138.9 185.2 231.5 277.8 Length(km) 324.1 370.4 416.7 463
Overvoltage(p.u.)
Overvoltage(p.u.)
normal with R and EA with R and MA 0 46.3 92.6 138.9 185.2 231.5 277.8 Length(km) 324.1 370.4 416.7 463
Figure 12. Profiles of the switching overvoltages for case e, stage 1
2.9 2.8 2.7 2.6 2.5 2.4 2.3 2.2 2.1 2 1.9 1.8 1.7 1.6 1.5 1.4 1.3 1.2 1.1 1
normal with MA and CS with IAs and CS 0 46.3 92.6 138.9 185.2 231.5 277.8 Length(km) 324.1 370.4 416.7 463
Figure 13. Profiles of the switching overvoltages for case e, stage 2
F. Comparison of Three Basic Methods Eventually, in this section, the simulation has been separately performed for four different cases. First, the switching operation is ordinary carried out without applying any of the reduction techniques. Then, each three reduction methods of the switching overvoltage is consecutively applied as follows: application of 300 pre-insertion resistor, application of intermediate surge arresters at the points of 185.2 and 324.1 km along the line and, finally, application of the 10% CZ controlled switching. The simulations are separately executed. The results of simulations are shown in Fig.14. It can clearly be seen that all three methods effectively reduce the switching overvoltages.
2.9 2.8 2.7 2.6 2.5 2.4 2.3 2.2 2.1 2 1.9 1.8 1.7 1.6 1.5 1.4 1.3 1.2 1.1 1
Overvoltage(p.u.)
increases, the overvoltage amplitude further decreases. Application of the intermediate surge arresters along the line relatively reduces the switching overvoltage, in this study up to 2.35 p.u. In other words the percentage reduction is 14%. Also, with larger number of intermediate arresters, the overvoltage is further decreased and its profile will be flatter. Application of controlled switching strategy considerably affects the switching overvoltage. It significantly reduces the switching overvoltage, in this study up to 2.27 p.u. value. In other words the percentage reduction is 18%. Of course at most of the locations along the line, the percentage reduction is much more than this value. Simultaneous application of any two methods does not considerably decrease the switching overvoltage. Although, simultaneous application of arresters and controlled switching, reduces the switching overvoltage more than those being applied individually. Generally, the variation rate of the switching overvoltage amplitude and profile is more sensitive to the factors change (such as distribution function type, covered area and crossing point of switching instant), of the controlled switching method, (see Figs.9 to11). In this study, the effect of pre-insertion resistor on the reduction of switching overvoltages is more significant than the effect of surge arresters and controlled switching. Despite the excellent performance of closing resistors, they are not applied by most utilities, due to their cost and high failure rates.
Overvoltage(p.u.)
normal with Closing Resistor with Intermediate Arrester with Controlled Switching 0 46.3 92.6 138.9 185.2 231.5 277.8 Length(km) 324.1 370.4 416.7 463
Figure 14. Profiles of the switching overvoltages for case f
G. Discussion According to the simulation results presented in section IV, the following points are concluded. In long UHV transmission lines, the maximum amplitude of switching overvoltage is significantly high, in this study 2.73 p.u., which may damage the transmission line insulation. A conventional method to reduce the switching overvoltage is the application of pre-insertion resistor. This method effectively reduces the switching overvoltage, in this study up to 1.97 p.u. In other words, 29% reduction of the overvoltage is obtained. In addition, as the value of the resistor
H. Conclusion Switching overvoltages are among the important factors to be taken into consideration in the design of ultra high voltage transmission lines. In this paper each of three practical techniques of the transmission line switching overvoltage reduction is separately studied. The simulations are comprehensively executed on an actual transmission line, which clearly indicate the effect of each method on the switching overvoltage reduction. According to the results of simulations, application of closing resistor has the most significant effect on the overvoltages, although, due to the reliability and economic constraints, this method is not applied anymore. REFERENCES
[1] [2] Insulation Coordination, part 2, IEC Standard 60071-2-1996. Ehsan Abbasi, Heresh Seyedi, and Kai Strunz. Simulation an d analysis of the effect of single-pole auto-reclosing on HV transmission lines switching overvoltages IEEE PES general meeting, 2009
H.W. Dommel, Electromagnetic Transients Program, Reference Manual (ATP Theory Book), Bonnovile Power Administration, Portland 1986. J. R. Marti, Accurate modeling of frequency dependent transmission lines in electromagnetic transient simulations, IEEE Trans. Power Apparatus and systems, vol. PAS-101, Jan. 1982. R. N. Yeckley, R. E. Friedrich, Michael E. Thuot. EHV breaker rated for control of closing voltage switching surges to 1.5 per unit IEEE Summer Meeting and International Symposium on High Power Testing, Portland, Ore., July 18-23, 1971 A.C.Legate, J.H. Brunke, J. Ray, E. J. Yasuda, Elimination of closing resistors on ehv circuit breakers, IEEE Transactions on Power Delivery, Vol. 3, No. 1. January 1988 T. A. Bellei, R. P. O'Leary, E. H. Camm, Evaluating capacitor switching devices for preventing nuisance tripping of adjustablespeed drives due to voltage magnification, IEEE Transactions on Power Delivery, Vol. 11, No. 3, July 1996 Controlled Switching Application Guide, ABB, 2004. Lennart Stenstrom, Minoo Mobedjina, Limitation of Switching Overvoltages by Use of Transmission Line Surge Arresters, in Proc. 1998 CIGRE SC 33 Interpower Conf., Zagreb. J. R. Ribeiro, M. E. McCallum , An application of metal oxide surge arresters in the elimination of need for closing resistors in emt breakers, IEEE Transactions on Power Delivery, Vol. 4, NO. 1, January 1989 H. Seyedi, M. Sanaye-Pasand, Application of Transmission Line Surge Arresters to Reduce Switching Overvoltage, IPST Conference 19-23 June 2005. A. Carvalho M. Lacorte 0. Knudsen, Improved EHV line switching surge control by application of MO-arresters and controlled switching, IEEE Catalogue No. 95TH8130 ,0-78032981-3/95/$4.00Ol995 IEEE Sang-Pil Ahn, Chul-Hwan Kim Aggarwal, R.K. Johns, An alternative approach to adaptive single pole autoreclosing in high voltage transmission systems based on variable dead time control , IEEE Transactions on Power Delivery, Volume: 16, 2001. K. Froehlich,C. Hoelzl, M.Stanek, A. C. Crarvalho, W. Hofbauer, P. Hoegg, B. L. Avent, D.F. Peelo, J.H.Sawada, Controlled switching on shunt reactor compensated transmission lines Part I: closing control device development, IEEE Trans. Power Delivery , vol. 12, pp. 734-740, Apr. 1997. YANG Linghui and ZHANG Jiamin, Research on transient over voltage during operation of 500kV GIS disconnecting switch, East China Electric Power, vol.32, pp.38-41. Jan 2004. Karcius M. C. Dantas, Student Member, IEEE, Damsio Fernandes Jr.,Member, IEEE, Washington L. A. Neves, Member,IEEE, Benemar A. Souza, Senior Member, IEEE and Luiz C. A.
[3]
[17]
[4]
[5]
[18]
[6]
[19]
[7]
[20] [21]
[8] [9]
[10]
[22]
[11]
[23]
[12]
[24]
[13]
[25]
[14]
[26]
[15]
[27]
[16]
Fonesca,"Mitigation of Switching Overvoltages in Transmission Lines via Controlled Switching " U.Samitz Ove, H.Siguerdidjane, F.Boudaoud, P.Bastard, J.p.Dupraz, M.Collet, J.Martin, T.Jung,"On controlled switching of high voltage unloaded transmission lines", Osterreischer Beitrag Fr die CIGR-konferenz 2002, vom 25. Bis 30 August 2002 in Paris M.Sanaye-Pasand , M.R.Dadashzadeh , M. Khodayar ,"Limitation of Transmission Line Switching Overvoltages using Switch-sync Relays" , Presented at the international Conference onPower Systems Transients(IPST ' 05) in Montreal, Canada on June 19-23 , 2005 Paper No. IPST05-087 J. Martinez, R. Natarjan and E. Camm, Comparison of statistical switching results using Gaussian, uniform and systematic switching approaches, in Proc. 2000 IEEE Power Engineering Society Summer Meeting, vol.2, pp. 884-889. A. Greenwood. Electrical transient s in power systems, WileyInterscience; 2nd edition (April 18, 1991) B. R. Shperling, A. Fakheri, Single -Phase Switching Parameters for Untransposed EHV Transmission Lines, IEEE Transactions on Power Apparatus and Systems, Vol.PAS-98, No.2 March/April 1979. B. R. Shperling, A. J. Fakheri, C. H. Shinh, B. J. Ware, Analysis Of Single-pole Switching Field Tests On The AEP 765 KV System, IEEE Transactions on Power Apparatus and Systems , Vol. PAS-100, No. 4, April 1981. Fakheri, Erling A. Ware, B. J. Shperling, B. R., Compensation Scheme for Single-Pole Switching on Untransposed Transmission Lines, IEEE Transactions on Power Apparatus and Systems, Volume PAS-97, pages 1421-1429, 1978. Aggarwal, R.K. Johns, A.T. Song, Y.H. Dunn, R.W. Fitton, D.S., Neural network based adaptive single pole autoreclosure technique for EHV transmission systems, Generation Transmission and Distribution, IEE Proceedings, Volume141, pages155-160, 1994. Thomann, G.C. Lambert, S.R. Phaloprakarn, S,R Non-optimum compensation schemes for single pole reclosing on EHV double circuit transmission lines, IEEE Transactions on Power Delivery, Volume 8, pages 651-659, 1993. Basu, K.P. Hamid, S. Hasan, S., Single -Pole Switching of EHV Transmission Line with Capacitance Grounding, Power Engineering Review, IEEE, Volume 22, pages 45-46, 2002. Balser, S.J. Krause, P.C., Single -Pole Switching Study of System Transients with Transposed and untransposed Lines, IEEE Transactions on Power Apparatus and Systems, Volume PAS-93, pages 1208-1212, 1974.
Analysis of the Coupling of a Lightning Wave with an Overhead Line
Mustapha. Boumaiza
Laboratory of Electrical Engineering Constantine Department of Electrical Engineering Faculty of Science and Technology University Constantine1 mustapha_boumaiza@yahoo.fr
Djamel. Labed
Laboratory of Electrical Engineering Constantine Department of Electrical Engineering Faculty of Science and Technology University Constantine1 djamel_labed@yahoo.fr
AbstractIn this work we analyze the electromagnetic coupling of a lightning wave with an overhead line. The analysis is conducted directly in time domain with hold in account the effect of a finite conductivity of the soil. This analysis calculates the currents and voltages induced at every moment and at every point within the line. To strengthen our theoretical work, we present a set of applications that will allow us to validate this analysis. Keywords- lightning; electromagnetic coupling; overhead line; transient ground resistance; induced voltage.
II. EQUATION COUPLING OF A LIGHTNING WAVE WITH AN OVERHEAD LINE A. Single-Wire Overhead Line without Loss of TimeDomain Equations coupling expressed by Taylor [1] are:
I ( x, t ) h e dU ( x, t ) = Bx +L ( x, z , t )dz t t 0 dx U ( x, t ) h e dI ( x, t ) = C Ez ( x, z , t )dz +C t t 0 dx
The boundary conditions are:
(1)
I. INTRODUCTION In the power transmission network for optimum distribution of power flows, maintaining the frequency, reactive power compensation, etc..., the real-time knowledge of the electrical characteristics of the latter ensures his control and command. This control function and controlling the flow of energy is provided by a set of electronic low levels. If we can say that these days, the energy carriers mastered the proper protection of the network against accidental faults, this is not the case for protection against lightning (natural defect), especially when an indirect impact. The lightning is a common phenomenon that behaves as a perfect current generator. The ionized channel lightning behaves as a long wire which radiates an electromagnetic field. This field induces in the large ground loop voltages that account kilovolts. These induced surges can cause damage, as well as the power network monitoring and control electronic networks and transport. This work aims at characterizing overvoltages and overcurrents caused by a wave of lightning on overhead lines after electromagnetic coupling. We model the coupling Lightning line by the theory of transmission lines. For electromagnetic excitation, which is the second member of the line equations, we use the formalism of dipoles with taking into account the ground effect for calculating the electromagnetic field radiated by the lightning channel.
(2)
U (0, t ) = Z A I (0, t ) U ( L, t ) = Z B I ( L, t )
(3) (4)
The coupling circuit equivalent described by the tow equation (1) and (2) is shown in figure 1.
e j By ( x, z , t )dz 0 Ldx + h
I ( x + dx) U ( x + dx) U ( L)
ZA
U (0) U ( x)
jC E ze ( x, z, t )dz
0 h
Cdx
ZB
x + dx
Figure 1. Coupling circuit differential equivalent.
B. Single-Wire Overhead Line with Loss of FrequencyDomain Equations coupling expressed by Taylor [1] are:
h dU ( x) e + ZI ( x) = j Bx ( x, z )dz dx 0
(5)
h dI ( x) + YU ( x) = jC E ze ( x, z )dz dx 0
(6)
U (0, t ) = Z A I (0, t ) U ( L, t ) = Z B I ( L, t )
(14) (15)
The longitudinal impedance matrix is given by [2].
Z = j L + Zw + Zg
Where: Zw is the per-unit-length internal impedance of the wire; Zg is the per-unit-length ground Impedance. Transverse admittance matrix by:
(7)
The coupling circuit equivalent described by the tow equation (11) and (12) is shown in figure2.
e By ( x, z , t ) dz t 0 +
h
I ( x, t ) Ldx
I ( x + dx, t )
Z A U (0)
I ( x, t ) g t h U ( x, t ) C E ze ( x, z , t ) dz t 0
C dx
U ( x + dx, t )
U ( L)
ZB
Y=
(G + jC )Yg G + jC + Yg
0 x
x + dx L
(8)
Figure 2. Coupling circuit differential equivalent.
C. Single-Wire Overhead Line with Loss of Time- Domain The coupling equation in the time domain is obtained by Fourier transformation of the equation (5) and (6).
C. Multiconductor Overhead Lines with Loss in TimeDomain The generalization of equations coupling (11) and (12) for the case of an overhead line with multiple conductors in time domain given by:
I ( x, t ) dU ( x, t ) +L + Z (t ) I ( x, t )d t dx 0 =
t
e By ( x, z , t )dz t
(9)
[ ]
(10)
dI ( x, t ) + Y ( )U ( x, t )d dx 0 = C E ze ( x, z , t ) t 0
The equations (9) and (10) can be written.
h
d [U i ( x, t )] [I i ( x, t )] + + Lij dx t h e gij [I i ( x, t )] = B y ( x, z, t )dz t t 0
[ ]
(16)
[ ]
d [I i ( x, t )] + C ij [U i ( x, t )] + t dx h Gij [U i ( x, t )] = [C ] E ze ( x, z , t )dz t 0
[ ]
(17)
I ( x, t ) I ( x, t ) dU ( x, t ) +L + g dx t t = e By ( x, z , t )dz t
Where: Lij , Gij (11)
[ ] [ ] and [C ] respectively matrices inductance,

ij
conductance and capacitance per unit length of the line;
U ( x, t ) dI ( x, t ) + GU ( x, t ) + C t dx (12) h e = C E z ( x, z , t )dz t 0 Where: is the product of the convolution and transient

resistance of the soil as defined [3]:
h e h e By ( x, z , t )dz and Ez ( x, z , t )dz are vectors of 0 0

the magnetic field and electric exciter
Ee Be
2a j
i
hi hj
j
k
rjk
g = F 1
Zg j
(13)
rij
hk
soil g , rg
Figure3. Geometric of multiconductor lines.
With boundary conditions:
The matrix elements
[ (t )]
sij
of the ground transient
resistance have a singularity when [4]. F. Rachidi and all [5] shows that the expressions of A. Timotin [4] and D. Orzan [6] are used with excellent approximation when
[U n+1 ] [U Kn+1 ] [L] I n+ 3 2 = K +1 k x t 1 g (t ) I kn+1 I kn 2
n+ 1 I k 2 + [L] t
t > t min =
0 rs and are simplified as follows: s

0 2 1 . sii t
(18)
] ([ ] [ ])
lim sii
t 0
lim sij
t 0
0 2
ij .cos 2 Tij t 1
1 n1 ( g (n j )t ) + g ((n + 1 j )t ) j +1 j 2 j =0 Ik Ik
n +1 z , k +1
[[ ] [ ([ ] [ ])
n 1 k n x, k g
]]
(21)
(19)
III. DISCRETIZATION OF THE EQUATIONS OF LINES BY FDTD The discretization of equations (16) and (17) by finite difference method FDTD. Consists in subdividing each conductor alternately nodes current and voltage. Two consecutive nodes of the same type are separated by an interval x in space and t/2 over time. The two ends of the line are defined as nodes voltage.
] [E ] + ([I ] [I ])[ (t )] [h] [E x [E ] + [E ]

n +1 k n +1 k ,k n +1 x ,k
2
B. Equations to Extremity of the Line
[C] [G] [C] [G] [U ] = + [U ] t 2 t 2

1 n+1 1
U1 I1
0
U2 I2
x / 2
U k max I k max 1
L
I3
1 n+ 2 1
Figure 4. Schematic spatial discretization of a conductor.
The time step and spatial step must verify the stability condition:
I n+1 n Ex ,1 + Ex,1 [G][h] 2 x n+1 n Ex,1 Ex ,1 [C][h] , k =1 t

1 n+ 2 0
n 1
[ ][ ]
(22)
[ ][ ]
1
Where: v is the speed of propagation of the wave on the line. A. Recurrence Equations for Current and Voltage The discretization of the equations of the lines by the FDTD method gives:
v x t
[C] [G] [C] [G] [U ] = + [U ] t 2 t 2

n+1 k max
n k max
[C] [G] [C] [G] [U ] = + [U ] t 2 t 2

1 n+1 k
n k
n+1 n Ex ,k max + E x,k max G h (23) [ ][ ] 2 n+1 n E E [C][h] x,k max x,k max , k = k max t
1 n+ 2 k max
I x
1 n+ 2 k max1
][
][
n+1 n+1 I k 2 Ik12 1 n Ezn,+ k + Ez,k G h [ ][ ] x 2
IV. APPLICQTIONS AND VALIDATIONS (20) A. Voltage induced on an overhead conductor To validate our theoretical developments for lightning-line coupling, we discuss an application whose results are published in [5]. Consider an overhead conductor 1km in length and 9.14mm radius at a height of 10m above a perfectly conducting ground and finite conductivity. The conductor is assumed to be terminated in its characteristic impedance at both ends. The point of impact is considered
[ ] [ ]
[E ] [E ] [C][h]
n+1 z,k n z,k
symmetrical at both ends and at a distance of 50m from the conductor. The electromagnetic field emitted by the lightning wave is calculated using the return stroke to the MTL model [7] with a typical value of v = 1.3.108m / s, a rate of decrease of intensity of current = 1700 km and expression of F.Heidler [8] for the current at the base of the channel with the data in Table 1.
TABLE 1. LIGHTNING PARAMETERS
same value for the characteristic impedance of the conductor (non-value communicated by F. Rachidi). B. Voltage induced on a three-phase overhead line In this application we consider a three-phase line length 1km water illuminated by a lightning wave as presented figure.8. The line is terminated at its ends by its characteristic impedance (Zc = 461.4). The point of impact is considered symmetrical at both ends is 50 m line. Note that the electromagnetic field emitted by the lightning wave is calculated using the return-stroke current to the MTL model [7] with a typical value the velocity v =1.9.108m/s, a decay rate of the intensity = 1700km current and F.Heidler expression [8] for the current at the base of the channel with the data of Table 1.
z
I01 (kA) 10.7
(s) 0.25
11
(s) 2.5
21
n1 2
I02 (kA) 6.5
(s) 2.1
12
(s) 230
22
n2 2
1000 m
ZC
50 m
Impact
Figure 5. Line studied.
ZC
x R
phase 1 phase 2
phase 3
z r
g= g= 0.001 S/m
ZC = L C
ZC =
L C
75
.
Figure 8. Configuration of studied.
50 I n d u c e d v o lta g e (K V )
25
Calculation results that we present in this section are those obtained for different configurations of the line (figure.9).
Cond 2 3.7m
Cond 2 3.7m 3.7m Cond 1 7.4m Cond3
Cond 1 3.7m 10m
Cond 2 3.7m
Cond 3
Cond 3 3.7m Cond 1
10m
10m
10m
10m
10m
10m
-25
4 Time(s)
8
horizontal vertical triangle
Figure 6. Voltages induced at the extremity of the conductor. Figure 9. Different configurations of the line.
50 phase2, g= 40 In d u c e d v o lta g e (K V ) phase2, g= 0.01 S/m
30
20
10
Figure 7. Voltages induced at the extremity of the conductor [5].
-10
4 Time(s)
10
Our calculation results (figure.6) correspond to those published by F. Rachidi [5] (figure.7). A slight difference which is quite predictable because we do not take the
Figure 10. Voltages induced at the extremity of the line horizontal line.
50 phase1,3, g= 40 I n d u c e d v o lt a g e (K V ) phase1,3, g= 0.01 S/m I n d u c e d v o lt a g e (K V )
50 phase1, g= 40 phase1, g= 0.01 S/m
30
30
20
20
10
10
-10
4 Time(s)
10
-10
4 Time(s)
10
Figure 14. Voltages induced at the extremity of the line vertical line.
50 phase2, g= 40 I n d u c e d v o lt a g e (K V ) phase2, g= 0.01 S/m
50 phase2, g= 40 I n d u c e d v o lt a g e (K V ) phase2, g= 0.01 S/m
30
30
20
20
10
10
-10
4 Time(s)
10
-10
4 Time(s)
10
Figure 12. Voltages induced at the extremity of the line triangle line .
50 phase1,3, g= 40 In d u c e d v o lt a g e (K V ) phase1,3, g= 0.01 S/m
50 phase3, g= 40 In d u c e d v o lt a g e (K V ) phase3, g= 0.01 S/m
30
30
20
20
10
10
-10
4 Time(s)
10
-10
4 Time(s)
10
Figure 13. Voltages induced at the extremity of the line triangle line .
In the case of the configuration horizontal and triangle, we see that the induced voltages at the extremity of the phase conductors 1 and 3 of the line are identical; this confirms our simulation, because the two phases are excited by the same field electromagnetic calculated middle position. In the configuration vertical we see an increase in the height of the phase conductors causes to an increase in the value of the induced voltages. We also note that the effect of the finite conductivity of the ground is manifested by the appearance of a negative peak and a reduction of the induced voltage due to losses incurred in the line by the finite conductivity of the ground. C. Influence of the finite conductivity of the soil To highlight the effect of the finite conductivity of the ground induced overvoltages, we do the calculations for two representative values. These values are: 0.01S/m et 0.001 S / m .
The results we obtain in Figure 17 and 18 show a decrease in the finite conductivity of the ground causes an increase in negative peak which leads to a significant reduction of the induced voltage. V. CONCLUSIONS This set of results, shows that it is possible to quantify by calculating the electromagnetic effect of lightning on an overhead line that is reflected in the apparition currents and induced voltages. This is an advantage for the insulation coordination and the proper choice of voltage protection. In this work we analyzed the electromagnetic coupling of a lightning wave with an overhead line. This analysis is developed in time-domain from transmission lines equations with second members, taking into account the finite conductivity of the ground, and our results are very satisfactory. Bin that our work is carried out in two stages (calculation of the electromagnetic field radiated by the lightning channel, and then solving the equations of couplings), its advantages are a simple computer implementation and computation time very low. REFERENCES
[1] C. D. Taylor, R. S. Satterwhite, W. Jr. Harrison, The Response of Terminated Two- WireTransmission Line Excited By a Nonuniform Electromagnetic field, IEEE Trans. On Antennas and propagation, vol.AP-13, pp.987-989, 1965. Rachidi F, Nucci C. A, lanoz M and Mazzetti C, Influence of a Lossy Ground on Lightning-Induced Voltages on Overhead Lines. IEEE Transactions on ctromagnetic compatibility, VOL.38, NO.3, AUGUST 1996, 250-264. Tesche F. M, On the inclusion of loss in time-domaine solutions of electromagnetic interaction problems, IEEE Trans. EMC 32 (1) (1990) 1-4. Timotin A, Longitudinal transient parameters of a unifilar line with ground return, Rev. Roum. Sc. Tech. Electrotech. Energie 12(4) (1967) 523-535 (Bucarest). Rachidi F, Loyka S. L, Nucci C. A, lanoz M, A new expression for the ground Power transient resistance matrix elements of muliticonductor overhead transmission lines, Electric Systems Research 65 (2003) 41-46. Orzan D, Couplage externe et interne entre un champ lectromagntique et un rseau de lignes multifilaires. Ph. D. dissertation, Ecole polytechnique fdrale de lausanne EPFL, 1998. C. A. Nucci, C. Mazzetti, F. Rachidi, M. Ianoz, On Lightning Return Stroke Models for LEMP Calculations Proc. 19th International Conference on Lightning Protection, Graz, 1988. F. Heidler, Analytische Blitzstrom-Funktion zur LEMPBerechnung , Paper1.9, pp. 63-66, Munich, September 16-20, 1985.
30 phase2,g= 0.01 S/m phase2,g= 0.001 S/m 20 I n d u c e d v o lt a g e (K V )
[2]
10
[3]
0
[4]
-10 0 2 4 Time(s) 6 8 10
[5]
35 30 25 I n d u c e d v o lt a g e (K V ) 20 15 10 5 0 -5 -10 0 2 4 Time(s) 6 8 10 phase1,3,g= 0.01 S/m phase1,3,g= 0.001 S/m
[6]
[7]
[8]
Matrix Converter based Unified Power Flow Controllers (UPFC)

Atheena A. Assistant Professor Sree Buddha College of Engineering, Pattoor. Kerala, India.
Abstract In this paper a new scheme of unified power flow controller (UPFC) with matrix converter is presented. Matrix converters (MCs) are power electronic ac/ac converters which accept electric power from one system and convert it to delivery to another ac system with waveforms of different amplitude, frequency and phase. It replaces a set of two voltage source inverters in conventional UPFCs. To reduce energy storage equipment, with consequent reduction of power losses AC-AC converters with minimum storage requirements are used. This also increases the lifetime of UPFC systems. The MC-UPFC provides active and reactive power control by selecting an appropriate matrix converter switching state guaranteeing good steady-state and dynamic responses. Simulations are carried out in MATLAB/Simulink, showing the effectiveness of the presented DPC method. Keywords Direct power control (DPC), Matrix converter (MC), Unified power-flow controller (UPFC).
I. INTRODUCTION Electrical energy is the back bone for the development of the society. With the industrial growth of a nation there is always an increased requirement of electrical energy. The increased demand for electric energy requires increasing the transmission capabilities. However, the inherent thermal, dielectric and stability limits of power system restrict the power transaction, leading to the underutilization of the existing transmission resources. The deregulation and competitive environment in the contemporary power networks will imply a new scenario in terms of load and power flow condition and so causing problems of line transmission capacity. But, nowadays some problems exist to change the present structure of transmission system. So, the need for new power flow controllers capable of increasing transmission capacity and controlling power flows through predefined transmission corridors will certainly increase. Thus came the Flexible AC Transmission Systems (FACTS) with power electronics based equipment which controls transmission lines power flow. UPFCs are the most flexible and complex FACTS allowing accurate and reliable control of both active and reactive power flow over the network. The UPFC concept was proposed by Gyugyi in 1991. The UPFC was devised for the real-time control and dynamic
compensation of ac transmission systems, providing multifunctional flexibility required to solve many of the problems facing the power delivery industry. UPFC is able to control simultaneously or selectively all the parameters affecting the power flow in transmission lines. The UPFC consists of two voltage sourced converters. These back to back converters share a common dc capacitor link. The converters are connected to the transmission line through coupling transformers one in shunt and other in series with the line. In this type of AC-AC converter power flow can takes place in either direction between the AC terminals of the two converters. Matrix converter replaces the two three-phase inverters and hence the DC link (bulk) capacitors are excluded, reducing costs, size, maintenance, increasing reliability and lifetime. In this paper a Matrix Converter based UPFC is recommended, using a direct power control approach (DPC-MC). This control method is based on sliding mode control techniques and allows real time selection of adequate state-space vectors to control input and output variables. Transmission line active power support and reactive power compensation can be directly provided using this approach and the performance is evaluated and discussed using detailed simulations. II. UPFC POWER SYSTEM A.General structure
Fig. 1 Matrix converter UPFC connected to transmission network.
A basic power transmission network using the suggested matrix converter UPFC is presented in Fig. 1, where VS and VR are, respectively, the sending-end and receiving-end sinusoidal voltages of the GS and GR generators feeding load ZL. The transmission line 2 is represented as a series inductance with series resistance (L2 and R2) to which the matrix converter is connected, through coupling transformers T1 and T2 . Three phase equivalent circuit of the matrix converter based UPFC is shown in Fig. 2. While modelling the system, the power sources and the coupling transformers are all considered ideal. Also, the matrix converter is considered ideal and represented as a controllable voltage source, with amplitude VC and phase . VR0 is the load bus voltage. The simplified elements are treated as disturbances.
B. Matrix Converter model
Fig. 3. Matrix Converter scheme.
S=[
(6)
The matrix converter topological constrictions implies

Fig. 2. Three-phase equivalent circuit of the transmission line and matrix UPFC.
Based on (6), the relationship between load and input voltages can be stated as [ ]T = S [ ]T (7)
By applying Kirchhoff laws to Fig. 2, the ac line currents are obtained in dq coordinates = + ( ) (1) = - + ( ) (2)
The sending end generators active and reactive power in dq coordinates are given by [ ] [ ][ ] (3)
The relationship between input phase currents and output phase currents can be stated as [ ]T = ST [ ]T (8)
VRod and VSd = Vd are assumed as constants. The rotating reference frame is synchronized to the VS source so that VSq = 0, active and reactive power P and Q are given by (4) and (5), respectively P= (4) Q= (5) Reference currents (Idref, Iqref) can be calculated from (4) and (5) for current controllers based on the desired active and reactive power (Pref, Qref). Three-phase matrix converter (fig. 3) which is denoted as an array of nine bidirectional switches S kj with turn-on and turn-off capability. The switches allow the link of each one of three output phases directly to any one of the three input phases. Matrix converter bidirectional switch Skj (k,j {1,2,3}) can take on two possible states: S kj=1 if the switch is closed or Skj=0 if the switch is open. The nine matrix converter switches can be represented as a 3*3 matrix (6)
From the 27 possible switching patterns, time variant vectors can be attained (Table I) representing the matrix output voltages and input currents in coordinates. The direct power control matrix converter will select one of these 27 vectors at any given time instant.
Fig. 4. a) Input voltages and their sector. b) Output voltage state space vectors when the input voltages are located at the sector Vi1.
III. DIRECT POWER CONTROL OF MATRIX CONVERTER BASED UPFC A. Line Active and Reactive Power Sliding Surfaces The DPC controllers for line power flow are derived here based on the sliding mode control theory. From Fig. 2, in steady state, Vd is imposed by source Vs. From (1) and (2), the transmission-line currents can be considered as state variables with first-order dynamics dependent on the sources and time constant of impedance L2/R2 . Therefore, transmission-line active and reactive powers present first-order dynamics and have a strong relative degree of one. From the sliding mode control theory, robust sliding surfaces to control the P and Q variables with a relatively strong degree of one can be obtained considering proportionality to a linear combination of the errors of the state variables. The active power error ep and the reactive power error eQ is the difference between the power references Pref , Qref and the actual transmitted powers P, Q respectively. = = (9) (10)
If , 0PPref, then choose a vector suitable to decrease P. If = 0, then choose a vector which doesnot significantly change the active power.
The robust sliding surfaces Sp(ep,t), and SQ(eQ,t), must be proportional to these errors, being zero after reaching sliding mode ( ( (11) (12)
The same procedure should be applied to the reactive power error. In order to ease vector selection (Table I), sliding surfaces and ) should be converted to coordinates . Six vectors of group I requires extra algorithms for calculating their time varying phases hence while designing the DPC control system they are not used. To guarantee the previously discussed required levels of VLd and VLq needed to fulfill the reaching conditions only the 12 highest amplitude voltage vectors from group II are used. The lowest amplitude voltages vectors, or the three null vectors of group III, could be used for near zero errors. Nine output voltage error combinations are obtained if the control errors and are quantized using two hysteresis comparators, each with three levels (-1, 0 and +1). If a twolevel comparator is used to control the shunt reactive power, as discussed in next subsection, 18 error combinations (9*2=18) will be defined, enabling the selection of 18 vectors. Since the three zero vectors have a minor influence on the shunt reactive power control, selecting one out 18 vectors is adequate. As an example, consider the case of C = >0 and C = < 0. Then, dP/dt>0 and dQ/dt<0 imply that dI/dt>0 and dI/dt>0. To choose the adequate output voltage vector, it is necessary to know the input voltages location [Fig. 4] as the output voltage vectors depend on the input voltages (sending voltage). Suppose now that the input voltages are in sector Vi2, then the vector to be applied should be +9 or -8. The final choice between these two depends on the matrix reactive power controller result, CQi. The same thinking is used for the left over eight active and reactive power error combinations and for all other input voltage sectors. These P, Q controllers were designed based on control laws not dependent on system parameters, but only on the errors of the controlled output to ensure robustness to parameter variations or operating conditions and allow system order reduction, minimizing response times. C. Direct Control of Matrix Converters Input Reactive Power In addition, the matrix converter UPFC can be controlled to ensure a minimum or a certain desired reactive power at the matrix converter input. Sliding surface ( ) will be a linear combination of reactive power error and its first order time derivative + ( (15)
Kp and KQ values are chosen to impose appropriate switching frequencies. B. Line Active and Reactive Power Direct Switching Laws Based on the errors ep and eQ the matrix converter switching states(vectors) are selected in real time. In order to guarantee the stability for active power and reactive power controllers, stability conditions given in (13) and (14) must be satisfied 0 0 (13) (14)
From these conditions if , then the value must be decreased, that is its time derivative should be negative 0. Similarly, if then 0 According to (11) and (13), the criteria to choose the matrix vector should be If , 0PPref, then choose a vector suitable to increase P.
TABLE I Switching combinations and output voltage/input current state-space vectors Group I I I I I I II II II II II II II II II II II II II II II II II II III III III Name 1g 2g 3g 4g 5g 6g +1 -1 +2 -2 +3 -3 +4 -4 +5 -5 +6 -6 +7 -7 +8 -8 +9 -9 Za Zb Zc A a a b b c c a b b c c a b a c b a c b a c b a c a b c B b c a c a b b a c b a c a b b c c a b a c b a c a b c C c b c a b a b a c b a c b a c b a c a b b c c a a b c VAB Vab - Vca - Vab Vbc Vca - Vbc Vab - Vab Vbc - Vbc Vca - Vca - Vab Vab - Vbc Vbc - Vca Vca 0 0 0 0 0 0 0 0 0 VBC Vbc - Vbc - Vca Vca Vab - Vab 0 0 0 0 0 0 Vab - Vab Vbc - Vbc Vca - Vca - Vab Vab - Vbc Vbc - Vca Vca 0 0 0 VCA Vca - Vab - Vbc Vab Vbc - Vca - Vab Vab - Vbc Vbc - Vca Vca 0 0 0 0 0 0 Vab - Vab Vbc - Vbc Vca - Vca 0 0 0 ia iA iA iB iC iB iC iA - iA 0 0 - iA iA iB - iB 0 0 - iB iB iC - iC 0 0 - iC iC 0 0 0 ib iB iC iA iA iC iB - iA iA iA - iA 0 0 - iB iB iB - iB 0 0 - iC iC iC - iC 0 0 0 0 0 ic iC iB iC iB iA iA 0 0 - iA iA iA - iA 0 0 - iB iB iB - iB 0 0 - iC iC iC - iC 0 0 0 Vo Vi -Vi -Vi Vi Vi -Vi - - - - - - - - - 0 0 0
ab -
0 i i+ i i+ i+ - i+ 0 0
ab
Ii iO iO iO iO iO iO iA - iA iA - iA iA - iA iB - iB iB - iB iB - iB iC - iC iC - iC iC - iC
i O -O - O+ 0+ o+ -O+
0
bc
0
bc
0
ca
0
ca
ab
ab
bc
bc
ca
ca
ab
ab
bc
bc
ca
ca
0 0 0
The gain KQi is selected to obtain a suitable switching frequency. The sliding surface is quantized only in two levels (-1 and +1) using one hysteresis comparator. The sliding mode is reached when vectors applied to the converter have the necessary iq current amplitude to satisfy stability conditions (14). Since the iq input current depends on the output currents, it is necessary to know the output currents location, to choose the most adequate vector in the chosen dq reference frame (Table I). Considering that the dq reference frame depends on the Via input voltage location, the sign of the matrix reactive
power Qi can be determined by understanding the location of the input voltages and the location of the output currents. Considering the previous example, with the input voltage at sector Vi2 and sliding surfaces signals and ( ) < 0, both vectors +9 or -8 would be suitable to control the line active and reactive powers errors (Fig. 4). However, at sector Io1, these vectors have a different effect on the value: if iq has a suitable amplitude, vector +9 leads to >0 while vector -8 originates <0.
So, vector +9 should be chosen if the input reactive power sliding surface is quantized as CQi = -1, while vector -8 should be chosen when is quantized as CQi = +1. When the power errors =0 and =0, the null vectors of group III, or the lowest amplitude voltages vectors at sector Vi2(-1,+7,-4,+1,-7,+4) could be used. These vectors do not produce significant effects on the line active and reactive power values, but the lowest amplitude vectors have a high influence on the control of matrix reactive power. Only the highest amplitude current vectors of sector should be chosen: vector +1 if is quantized as, CQi = -1 or vector -1 if is quantized as CQi = +1. Using the same thinking for remaining eight combinations at sector Vi2 and applying it for the other output current sectors, Table II is obtained.
TABLE II State-space vectors selection, for input voltages located at sector Vi2 CC (Vi2) Io12;Io1 CQi +1 -1 -1+1 -1 0 -1 -1 0 +1 0 0 0 -1 +1+1 +1 0 +1-1 -9+8 +3-2 -6+5 -9+8 -1+1 +9-8 -5+6 -3+2 -8+9 Io2;Io3 CQi +1 -1 -9+8 +3-2 +5-6 -9+8 +7-7 +9-8 +6-5 -3+2 -8+9 Io4;Io5 CQi +1 -1 -9+8 -2+3 +5-6 -9+8 -4+4 +9-8 +6-5 +2-3 -8+9 Io6;Io7 CQi +1 -1 +8-9 -2+3 +5-6 +8-9 +1-1 -8+9 +6-5 +2-3 +9-8 Io8;Io9 CQi +1 -1 +8-9 -2+3 -6+5 +8-9 -7+7 -8+9 -5+6 +2-3 +9-8 Io10;Io11 CQi +1 -1 +8-9 +3-2 -6+5 +8-9 +4-4 -8+9 -5+6 -3+2 +9-8 Fig. 5. Control scheme of matrix converter operating as a UPFC.
As shown in the block diagram (Fig.5), the control of the instantaneous active and reactive powers requires the measurement of Gs voltages and output currents necessary to calculate and and sliding surfaces. The output currents measurement is also used to determine the location of the input currents q component. The control of the matrix converter input reactive power requires the input currents measurement to calculate . At each time instant, the most suitable matrix vector is chosen upon the discrete values of the sliding surfaces. IV. SIMULATION RESULTS. The performance of the presented direct control system was evaluated with a detailed simulation model using the MATLAB/Simulink to represent the matrix converter, transformers, sources and transmission lines, and Simulink blocks to simulate the control system assuming that the UPFC is inserted in a transmission line of a power system. Ideal switches were considered to simulate matrix converter semiconductors minimizing simulation times.
Fig 6. MATLAB/Simulink model of Transmission network with MC-UPFC
Figure 6 shows a Matrix Converter based UPFC which is connected to the transmission network via three phase shunt transformer in the input side and three phase series transformer at the output side.
results for the series active and reactive power response and the shunt reactive power for Pref = 0.8 pu and Qref = 0.35 pu and Q iref = -0.07 pu. Both results clearly show that the active and reactive power is tracking the reference. Figure 9 shows the line current in steady state for P ref = 0.8 pu and Qref = 0.35 pu and Q iref = -0.07 pu. Currents are almost sinusoidal with small ripple content. Now the load power is changed to active power 1.5 kW, inductive reactive power QL = 900 VAR and capacitive reactive power QC = 1000 VAR. The simulation results are shown in figure 10 and figure 11. Both results clearly show that the active and reactive power is tracking the reference P ref = 0.8 pu and Qref = 0.35 pu and Q iref = -0.07 pu even after the change in load. Figure 12 shows the line current and it is almost sinusoidal with small ripple content.
Fig. 7 Active and reactive series power response for Pref=0.8 pu and Qref=0.35 pu.
Fig. 8. Reactive shunt power response for Q iref = -0.07 pu. Fig. 10 Active and reactive series power response for Pref=0.8 pu and Qref=0.35pu with P=1.5 kW, QL=900VAR, QC=1000VAR
Fig. 9. Line currents iA , iB , iC for Pref=0.8 pu and Qref=0.35 pu.
The load power is 1.5 kW and the transmission line is simulated as inductance L = 15 mH and series resistance R = 0.2. Figure 7 and figure 8 shows respectively the simulation
Fig. 11 Reactive shunt power response for Q iref = -0.07 pu with P = 1.5 kW, QL = 900 VAR, QC = 1000 VAR.
Fig. 12 Line currents iA , iB , iC for Pref=0.8 pu and Qref=0.35 pu with P = 1.5 kW, QL = 900 VAR, QC = 1000 VAR.
V. CONCLUSION This paper deals with how the FACTS device known as the Unified Power Flow Controller can be effectively used to maintain and improve power system operation and stability. A new matrix converter is connected to power transmission lines as UPFCs. A novel structure has been suggested for UPFC, with a view to eliminate the need for dc link capacitor. Presented simulation results show that active and reactive power can be advantageously controlled, thus confirming the expected performance of the presented nonlinear DPC methodology.
REFERENCES
[1]J. Monteiro, J. Fernando Silva, S. F. Pinto, and J. Palma, Matrix Converter-Based Unified Power-Flow Controllers: Advanced Direct Power Control, IEEE Trans. Power Del., vol. 26, no. 1, pp. 420 430, jan. 2011. [2]L. Gyugyi, C. Schauder, S. Williams, T. Rietman, D. Torgerson, and A. Edris, The unified power flow controller: A new approach to power transmission control, IEEE Trans. Power Del., vol. 10, no. 2, pp. 10851097, Apr. 1995. [3]L. Liu, P. Zhu, Y. Kang, and J. Chen, Power-flow control performance analysis of a unified power-flow controller in a novel control scheme, IEEE Trans. Power Del., vol. 22, no. 3, pp. 1613 1619, Jul. 2007. [4]M. Malinowski, M. Jasinski, and M. Kazmierkowski, Simple direct power control of three-phase PWM rectifier using space-vector modulation (DPC-SVM), IEEE Trans. Ind. Electron., vol. 51, no. 2, pp.447454, Apr. 2004. [5]A. Dasgupta, P. Tripathy, and P. Sensarma, Matrix converter as UPFC for transmission line compensation, in Proc. 7th Int. Conf. Power Electronics, Exco, Daegu, Korea,pp. 1050105, Oct. 2007.
Diagnosis-Oriented Sequence Circuit-Based Modeling of Eccentric Rotor Traction Induction Motors Accounting for Cage Damping and Split-P ase Currents
Claudio Bru!!ese
Dept" of Astronautical# Electrical# and Energ$ Engineering %ni&ersit$ of Rome - Sapien!a Rome# Ital$ claudio"'ru!!ese(uniroma)"it
AbstractA 1130kW squirrel-cage induction motor used for railway traction is analyzed in this paper in case of eccentric rotor. ince the motor has parallel-connected windings! the airgap un"alance makes rise un"alanced currents in the stator "ranches! and the split-phase current signature analysis can "e attempted for fault detection. #he split-phase currents are "etter analyzed when transposed in the comple$ domain! "y using symmetrical components. %oreo&er! the rotor cage currents must "e carefully taken in account in the model! since they de&elop a powerful damping effect on the signature of static eccentricity. #his paper shows that the faulted machine can "e represented "y the superimposition of fi&e equi&alent circuits corresponding to fi&e centered-rotor &irtual machines! useful for fault-related split-phase current calculation. #he mathematical modeling shown in this paper is useful to carry out a diagnostic method! as shown in a companion paper. #he latter also reports confirmatory simulations of the faulted 1130kW motor! "ased on winding function approach. KeywordsDiagnosis, eccentricity, equivalent circuit, induction motor, rotor, split-phase, symmetrical components, traction.
I" S,B2 e f fm Fn g2 kq Kp1 lstack LuS LkR L,qLm LSR n m

,q-
*OME*C+AT%RE
fault-related .q-pole /.0-frequenc$ comple1 current ma1imum flu1 densit$ in t e ealt $ mac ine column &ector of &oltages induced '$ t e rotor field fundamental frequenc$ ,3/.0rotor rotation mec anical frequenc$ ,3f/p4ortescue5s n1n generali!ed transformation matri1 minimum air gap lengt in t e ealt $ mac ine stator-rotor transformation ratio for t e q-t circuit 6inding factors for .,p7)--pole flu1 densit$ 6a&es mac ine iron stac8 lengt inductance of stator PP9s displaced of t e angle uS inductance of rotor loops displaced of t e angle kR q-t sequence inductance p ase magneti!ing inductance stator-rotor mutual inductance n1m matri1 num'er of elementar$ stator PP9s ,3:pnum'er of elementar$ rotor loops/'ars
p mac ine pole pairs RS# LSl stator p ase resistance and lea8age inductance RR)# LRl) rotor p ase resistance and lea8age inductance RSS# LSS stator resistance and inductance n1n matrices RRR# LRR rotor resistance and inductance m1m matrices Rbar# Lbar cage 'ar resistance and lea8age inductance RPPG# LPPG stator PP9 resistance and lea8age inductance Rring# Lring cage end ring resistance and lea8age inductance s mac ine slip sq slip of t e q-t sequence mac ine u# k su'script inde1es for stator and rotor circuits vS# iS real stator PP9 &oltage and current column &ectors vR# iR real rotor loop &oltage and current column &ectors v,q-# i,q- comple1 q-t sequence &oltages and currents line-line rms &oltage at t e mac ine terminals LL !S# !R comple1 unit &ectors ,3e"s# e"R# respecti&el$S# R displacement angle of consecuti&e stator,3.0/n- and rotor ,3.0/m- circuits #$ flu1 increment due to eccentricit$ % rotor rotation mec anical angle ,3t/p%S# %R angular a'scissas in stator and rotor frames &s# &' static and d$namic p"u" eccentricities# 2;&s<&'=) >s# >' angles of static and d$namic eccentricities p (( ) .p-pole# /.0-frequenc$ flu1 lin8age space &ector $2 ma1 PP9 flu1 lin8age in t e ealt $ mac ine m rotor mec anical speed ,3.0fm3,)-s-/p? comple1 con@ugate operator 5 transformed quantit$ A rotor quantit$ reported to t e stator '$ using kq
II"
I*TROD%CTIO*
RACTIO* motors used in rail6a$ applications are ea&il$ solicited electromec anical con&erters su'@ect to a num'er of faults" Bro8en 'ars B)C# B.C# rotor misalignments BDC# BEC# stator faults BFC# and gear faults B:C are not uncommon in traction motors# nonet eless t ese motors are required to offer an ig degree of afforda'ilit$# to a&oid ser&ice interruption# dela$s# and incidents" Bro8en 'ars in
squirrel-cage traction motors for ig -speed trains 6ere anal$!ed in B)C# B.C# BGC" Papers B:C reported a stud$ of t e gear dri&e train fault due to torsional &i'rations in rail6a$ dri&es" Stator faults a&e 'een anal$!ed in t eir general importance in BFC for medium &oltage motors# as for traction dri&es" T is paper studies static and d$namic rotor eccentricities ,SRE and DRE- in squirrel-cage induction motors 6it parallel arrangement of t e stator 6inding internal 'ranc es# as often &erified for motor used in rail6a$ traction dri&es B.C" As a case stud$# a ))D28H# )"IG8J# )22K!# four-pole cage induction motor used for ig speed trains is considered B.C" Ko6e&er# t e t eoretical stud$ is carried out in general terms and can 'e applied to an$ motor 6it s$mmetrical structure" T e effect of a general rotor eccentricit$ on t e split-p ase currents of s$nc ronous mac ines as 'een alread$ anal$!ed in pre&ious papers BIC-B))C" SRE and DRE ma8e rise additional .,p7)--pole flu1 densit$ 6a&es in t e a'normal air gap of a .ppole mac ine# 6 ic produce an un'alance in t e ot er6ise perfectl$ 'alanced currents in t e parallel 'ranc es of t e s$mmetric 6inding" T e un'alance in t e split-p ase currents can 'e con&enientl$ anal$!ed '$ using a s$mmetrical-component'ased transformation" Papers B)2C# B))C suggested t e use of fi&e equi&alent circuits of ealt $ centered-rotor s$nc ronous mac ines for calculation of t e un'alanced internal currents" A criterion 6as also gi&en for assessing t e fault 'ased on t e split-p ase current signature anal$sis ,SPCSA-" Ko6e&er# t e damping effect of t e cage on t e fault-related current signatures as 'een also pointed out" T e cage reacts mainl$ to t e SRE# '$ a counter-rotating field 6 ic appears at stand-still in t e stator frame# so as to damp t e air gap field distortion due to t e static rotor center displacement" T e damping effect is neglecta'le to some e1tent in s$nc ronous mac ines BIC# 'ut it is &er$ po6erful in cage motors due to t e &er$ lo6 'ar resistance# and cannot 'e neglected if a reasona'le diagnosis is required# as s o6n in B).C" T is paper resorts to a s$mmetrical component transformation of t e motor p ase-model including all t e stator split-p ase currents and rotor cage-loop currents BGC" T e anal$sis leads to fi&e equi&alent circuits useful for split-p ase current calculation and fault se&erit$ assessment# ta8ing in account t e cage damping" T e fi&e circuits correspond to fi&e 5&irtual5 centered-rotor mac ines# and t e superimposition of t eir currents $ields t e o&erall currents of t e fault$ mac ine" T e model is diagnosis-oriented# in t e sense t at it is aimed to esta'lis some quantitati&e lin8 'et6een fault gra&it$ and fault signatures t roug mac ine parameters" B$ follo6ing t e approac of B)2#))C# t e consequential step consists in e1ploiting t e useful relations ips ,ideall$# '$ a mat ematical in&ersion- to assess t e fault starting from signature measurements" T e latter step is illustrated in t e companion paper B).C# 6 ere simulations a&e 'een carried out '$ using t e 6inding function approac for t e ))D28H traction motor# in no-load and loaded conditions# for &arious SRE and DRE le&els# and 6it mi1ed fault" T e simulations in B).C confirm and &alidate t e proper motor modeling s o6n in t is paper" III" SEL%E*CE CIRC%ITS O4 A CA9E I*D%CTIO* MOTOR HITK PARA++E+ CO**ECTIO*S I* TKE STATOR ,KEA+TKM CASEBefore proceeding 6it t e anal$sis of t e faulted mac ine# 6e esta'lis ere t e model of a squirrel cage induction motor 6it parallel connections in t e stator" 4ig" )
s o6s t e s$mmetrical 6inding structure# decomposed in stator pole-p ase groups ,PP9s# or groups of series-connected coils of a p ase under a pole# called S)#"""#Sn in 4ig" )- and in rotor cage loops ,R)#"""#Rm in 4ig" )# using a classical mes model for t e cage and e1cluding t e end-ring-" 4ig" . s o6s t e electric sc eme of t e stator in case of parallel-connected PP9s for a four-pole mac ine"
(
'(
'3 ')
'1
1
4ig" )" S$mmetrical induction motor 6inding structure"

POLE-PHASE-GROUP CURRENT PRO"E
4 3 L2#
POLE-PHASE-GROUP ELEMENTARY CIRCUIT
L# 6
A'
e1 R s L0 !1
12 8 11 9 10
PHASE 'A' INTERNAL CONNECTIONS
4ig" ." 4our-pole t ree-p ase 6inding 6it parallel-connected PP9s"
*+ P,ase(-.'el T e stator and rotor electrical 'alances in t e time domain are 6ritten in ,)-# ,.- in matri1 form# assuming n circuits on t e stator and m circuits on t e rotor B.CN
v S = RSS i S ' ) 't S
,),.-
v R = R RR i R
' ) " 't R
T e flu1 column &ectors can 'e decomposed as follo6sN ) S =) SS ) SR = L SS iS L SR i R= L SSl L SSg iS L SR i R ) R=) RS ) RR= L RS i S L RR iR = L RS i S L RRl L RRg i R ,D,E-
6 ere t e su'scripts l# g indicate lea8age and air-gap magneti!ing flu1 inductance matrices" All t e matrices RSS# RRR# LSS# LRR# LSSl# LRRl# LSSg# LRRg are s$mmetric and circulant O
t at is# an$ ro6 is o'tained from t e precedent '$ a circular rig t-s ift" T erefore# onl$ t e first ro6s are reportedN
first ro6 of first ro6 of first ro6 of first ro6 of
RSS 3BRPPG 2
P -Rbar
2C ,F2 P 2 -RbarC ,:-
RRR 3B.,Rbar<Rring/m-
LSS 3BL2s Ls L.s PL,n0-)-s Ln0s L,n0-)-s P L.s LsC ,GLRR 3BL2R LR L.RPL,m011-R Lm0R L,m011-R""" L.R LRC ,I-
B+ Sequence(-.'el T e mac ine model ,)-# ,.- can 'e con&enientl$ simplified '$ a s$mmetrical component transformation# as s o6n in Appendi1 B" T e complete model is reported in BGC" All t e circulant matrices turn out in a diagonali!ed s ape in t is 6a$# and t e sequence circuits are easil$ decoupled" T e generic qt sequence stator and rotor equations can 'e o'tained from ,B)-# ,B.- '$ neglecting t e ig er-order armonicsN
' q ' i L q i q 't S 't SR R
6 ere nQ is n/.# for n e&enR for n odd# nQ is ,n<)-/.# and t e element LnQs disappears" T e same definition is &alid for mQ" T e elements of LSSl# LSSg are formall$ deri&ed from t e elements of LSS# and t e elements of LRRl# LRRg from t ose of LRR# as follo6sN Lus 3 Lus2l < Lus2g# LkR 3 LkR2l < LkR2g# u3)#"""# n5 k3)#"""#m5 ,S,)2-
v Sq = R Sq iSq LSq
,)E,)F-
v R = R R i R LR
q q
' q ' q q i L i " 't R 't SR S
o6e&er t e lea8age matrices LSSl# LRRl# can 'e e1plicated '$N

first ro6 of first ro6 of
LSSl 3 BLPPG
P 2 P 2
2C -LbarC
,)),).-
LRRl 3 B.,Lbar<Lring/m- -Lbar
T e s$stem ,)E-# ,)F- represents t e d$namics of a generic q-t 5space-sequence mac ine5 &irtuall$ em'edded in t e model of t e real mac ine" All t e sequence mac ines are electricall$ decoupled# and mec anicall$ coupled at t e s aft 6it a gi&en rotor speed m" It can 'e pro&en BIC t at# in case of parallel connection of t e stator# vS,q-32 for qTp" Moreo&er# vR,q-32 due to t e s ort-circuit rings" Eqs" ,)E-# ,)F- can 'e 6ritten for sinusoidal stead$-state 6it generic frequenc$ q/.0 and constant speed '$ using ,B.E-# and 8eeping in mind t e follo6ing mec anical constraintN
= m t = ) s t = ) s q q t p q
6 ere LPPG# Lbar# and Lring are t e lea8age inductances of a single PP9# of a rotor 'ar# and of t e end-ring# respecti&el$" T e actual distri'ution of t e electrical circuits and t e consequent space- armonics are accounted for '$ e1panding in s$mmetrical 'ilateral 4ourier series t e mutual stator-rotor inductances" T e mutual inductance 'et6een t e ut, stator PP9 ,Su- and t e kt, rotor loop ,Rk- is ,)D-# t at is t e ,u2k- element of t e matri1 LSR in ,D-" *ote t at LRS3LSRt" T e coefficients 3,,are s o6n in 4ig" D for t e ))D28H traction motor ,mac ine data are in Appendi1 A-"
,):-
6 ere sq is t e slip of t e q-t sequence mac ineN

sq = q / q m q / q
"
,)G-
In ,)G-# q/q is t e s$nc ronous speed of t e q-t sequence mac ine" T e stead$-state form of sequence &oltages and currents is as follo6sN vS = S e
v =2
q R
" q t
i S = 4S e i = 4 e
q R q R
"qt " sq q t
,)I" ,)S,.2,.)-
lSu # Rk , - =
, =
( ) cos( ,( ,u )-
,
+ ,k )- R ) )
,)D-
B$ using ,):--,)S- and ,B.E- in ,)E-# ,)F-# 6e o'tainN

q q q q q Sq = RSq 4 S " q LS 4 S " q - 4R
2 =
R q q 4 q " q LR 4 R " q - q 4 Sq " sq R
q R
An equi&alent circuit for t e stead$-state q-t sequence mac ine can 'e carried out from ,.2-# ,.)-# 6 en t e lea8age and air-gap inductances ,B):-# ,B)S- are ta8en in account# and '$ using t e perfect coupling e1pressed '$ ,B.:-N
q q q q q Sq = RSq 4 S " q L Sl 4 S " q L Sg 4m
,..,.D-
2 =
RR q ) q ) q q 4 q ) " q L Rl 4 R " q LSg 4m " sq R
q )
In ,.D- t e rotor quantities 6ere reported to t e stator '$N

4ig" D" Karmonic inductance coefficients 3,,- ,))D28H motor-"
q ) q q ) . q q ) . q 4 R = 4 R / k q # R R = k q R R # L Rl =k q L Rl
,.E-
6 ere t e stator-rotor q-t transformation ratio is introducedN

k q= L Sg / - = - / L Rg "
q q q q
6 ere B23;2-2/g2 is posed# and also %R3%S-mt" T e fundamental .p-pole air-gap flu1 densit$ 6a&e 6it amplitude B2 in ,D)- is esta'lis ed '$ t e magneti!ing current flo6ing in t e mac ine .p-pole sequence circuit as in 4ig" E for q3p" T e ot er ,fault-related- 6a&es in ,D)- appear in t e airgap due to t e geometric anomal$# and t e$ induce currents in t e ,p7)--sequence circuits# as s o6n in t e follo6ing" B+ Stat.r(flu6(linkages T e 6a&es ,D)- produce flu1 lin8ages 6it 'ot stator and rotor circuits" Ko6e&er# onl$ t e stator flu1 lin8ages 6ill 'e carried out in t e follo6ing# since t e rotor-lin8ed flu1es can 'e o'tained from t e former '$ using t e stator-rotor transformation ratio kq o'tained in t e pre&ious Section" T e calculation of t e stator-lin8ed flu1es is performed as alread$ e1plained in B)2C# so some inessential details are omitted ere" T e stator-lin8ed flu1es are as follo6sN ) k = 2 cos t p S k ) s 2 p ) cos t p ) S k ) s s 2 p ) cos t p ) S k ) s ,D. ' 2 p ) cos / p t p ) S k ) ' ' 2 p ) cos / p t p ) S k ) ' 6it flu1 amplitudesN s / ' 2 p )= & s / ' 2 K p ) " . ,DD-
,.F-
*ote from ,.F- t at eac sequence mac ine as a different transformation ratio kq" T e q-t magneti!ing current in ,..-# ,.D- is finall$ defined asN
q q q 4 m = 4 S 4 R /k q "
,.:-
In conclusion# t e ealt $ mac ine is represented '$ n/.<) sequence circuits as in 4ig" E# for q32# )# P# n/. ,t e remaining n/.-) circuits are redundant-" T e circuit for q3p is connected to t e feeding source and performs t e main electromec anical energ$ con&ersion# and it is actuall$ a per-pole per-p ase mac ine equi&alent circuit" T e parameters of t e p-t sequence circuit are related to t e p ase parameters of t e parallel-connected mac ine as follo6sN
L Sl =. pL Sl # L Rl =. pL Rl # L Sg = . pL m #
p p ) ) p
,.G,.I-
R = . pRS # R
p S
p ) R
= . pR # p= # s p = s "
) R
T e remaining circuits ,for qTp- are s ort-circuited and not operati&e in t e ealt $ mac ine" Ko6e&er t e$ are e1cited in t e fault$ eccentric-rotor mac ine# as s o6n in t e follo6ing"
RS
q
q L Sl
q 1 1 q q 4 R 51 q L Rl 5 R R 5 / s q 4S q 4m
q S
q L Sg
4ig" E" Equi&alent q-t sequence circuit"
In case of parallel-connected PP9s# t e flu1 amplitude $2 is 'ound to t e line-line &oltage LL t roug $23U. LL/UD"
IJ" SEL%E*CE CIRC%ITS O4 A CA9E I*D%CTIO* MOTOR HITK PARA++E+ CO**ECTIO*S I* TKE STATOR ,4A%+TM CASE*+ *ir(Gap(Flu6(7ensit8(in(9ase(.f(:ccentric1R.t.r A generic mi1ed-t$pe rotor eccentricit$ ,RE- is s o6n in 4ig" F" T e permeance function of t e a'normal air gap is usuall$ affected '$ a first-order space armonic as follo6s B)2CN P 2 ) & s cos S s & ' cos R ' " g2 ,.St is &ector is fi1ed ' 5 q5 t is &ector rotates 6it t e rotor
Vs3SC/g2
Vd3CR/g2
g2 3 gap of t e ealt $ mac ine
g2,)-Vs-VdWR >d
a
T e rotating MM4 6a&e produced '$ stator and rotor currents can 'e considered ere unaffected '$ t e REN - = - 2 cos t p S " ,D2>s
cage loopX)
WS
a1is of PP9X)
p 4 m # if t e -2 is proportional to t e magneti!ing current saturation in t e iron is neglected" T e air gap flu1 densit$ is t e product of ,.S- and ,D2-N
4ig" F" Rotor eccentricit$ sc eme and s$m'ols used in t e formulas"
B =B 2 cos t p S 2"F & s B 2 cos t p) S s 2"F & s B 2 cos t p) S s 2"F & ' B2 cos m t p) S ' 2"F & ' B2 cos m t p) S '
,D)-
9+ Sequence(9ircuits(.f(t,e(Fault,8(-ac,ine T e flu1es in ,D.- can 'e em'edded in t e mac ine sequence circuits '$ a preliminar$ transformation as s o6n in ,BD-# so o'tainingN
) Sg 5 = F n ) Sg =

2 p ) p ) ) )
m
T e flu1es ,DF-# ,D:- affect t e ,p7)--order circuits# so t e$ must 'e put in ,)E- for q3p7)N ,DE2= R S
p ) p) S
) " p ) p ) ) ) 2
m
LS
p )
' p ) ' p ) p) i L SR i R 't S 't ' p ) ' p ) ,DG ) ) 't 't

m
v S = RS i S L S 2= R S
p ) p) S
p p
6 ere )Sg is t e column &ector containing t e air-gap flu1es )k in ,D.-" T e sequence flu1es in ,DE- are as follo6sN ) =
p ) p )
' p ' p p i L i 't S 't SR R
,DI-
LS
p )
n
.
s 2 p )
" t s
,DF" ,D:-
' p ) ' p ) p) i L SR i R 't S 't ' p ) ' p ) ,DS ) ) 't 't

m
) =
m
n
.
,p-)-
' 2 p )
" m t '
H en 6ritten at stead$-state# t e fi&e equi&alent sequence circuits as in 4ig" : can 'e dra6n from ,DG--,DS-"
RS,p<)LSl
,p<),p<)-)
p<)#s-
RS,p-)-
LSl
-)S,,p -
R,s
,p-)-)
p-)#s-
p-)-) L,Rl RR((/sp-)#s
,p-)-)
<)S,,p -
R,s
p<)-) L,Rl RR((/sp<)#s
,p<)-)
,p-)m , -
,p<)m ,-
LSg
,p-)-
s p)# s = s
-)=S,,p -
s ) p
LSg
,p<)-
s p )# s = s
<)=S,,p -
s ) p
SRE(p-1)
RS,p-
LSl,p- < ,p- < ,p-) LRl,p-) R ,p-)/s R S ,R ,s<m,,p -
SRE(p+1)
,pS
LSg,pHEALTHY(p)
,p<)-)
p<)#',>m--
p-)RS,p-)- ,-m-LSl ,, S 1
,p-)-
R ,s
m-
,p-)-)
p-)#',1m--
,-m-LRl
m-
,p-)-)
RR((/sp-)#'
,p-)-)
p<)RS,p<)- ,<m-LSl ,, S >
,p<)-
R , s
m-
,<m-LRl R /s R(( p<)#'

m-
,p<)-) ,p<)-)
,p-)m ,1
,p<)m ,>
,-m-LSg
,p-)-
-)=S,p ,1
s s p)# ' = ) s ) p
m-
,<m-LSg
,p<)-
s p )# ' =
<)=S,p ,>
m-
s ) s p
DRE(p-1)
DRE(p+1)
4ig" :" Sequence circuits of t e fi&e centered-rotor &irtual mac ines actuall$ superimposed in t e model of t e real eccentric-rotor induction motor"
Eac &irtual mac ine in 4ig" : as a different slip# i"e" sp7)#s( and sp7)#' in case of SRE and DRE# respecti&el$" *ote t at t e currents in t e p7) circuits in 4ig" : a&e sign re&ersed 6it respect to t e con&ention used in 4ig" E# for con&enience ,t e s$m'ol * is used instead of 4 for t e fault-related currents-" B$ using t e time-deri&ati&es of flu1es ,DF-# ,D:-# t e &oltages =S,p7)- appearing in 4ig" : are carried out as follo6sN
) n " " Sp : = s 2 p ) e .
s
) n " " Sp : = m ' 2 p ) e .

m
'
"
,E)-
T e amplitude of &oltages ,E2-# ,E)- is proportional to SRE ,&s- and DRE ,&'- p"u" &alues# respecti&el$" J" CO*C+%SIO*
,E2-
T e main contri'ution of t is paper consists in t e fi&e sequence circuits of 4ig" :# representing fi&e &irtual ealt $ mac ines actuall$ superimposed in t e model of t e real fault$ eccentric-rotor motor" T e four fault-related ,p7)- circuits in
4ig" : esta'lis es t e desired quantitati&e lin8 'et6een p"u" eccentricit$ parameters &s# &' and stator currents S,p7)-# t e latter possi'l$ 'eing t e result of measurements on t e real mac ine" At t is scope# t e PP9 currents must 'e measured and transformed '$ ,BF-# and t en anal$!ed '$ 44T to o'tain t e /.0 and ,-m-/.0 frequenc$ components in t e ,p7)-order current &ectors similarl$ to 6 at done in BS# )2C" After6ard# t e circuits in 4ig" : can 'e used for calculation of &s# &'" T e procedure is furt er e1plained and &alidated in a companion paper B).C" Remar8a'l$# t e ,p7)- circuits in 4ig" : clearl$ s o6 t e role of t e cage in damping t e fault related current signatures" In fact# t e fault-related &oltage sources =S,p7)- actuall$ 5see5 t e stator and rotor circuits in parallel# and so t e cage finall$ 5su'tracts5 current to t e signature usefull$ measured in t e stator" Ko6e&er t e circuits in 4ig" : allo6 to ta8e in account t is p enomenon# so as to ad@ust t e diagnosis" APPE*DIY AN MACKI*E DATA Ta'le I reports t e ratings of t e ))D28H motor used in t e paper" 4ig" G s o6s a cross-section of t e motor"
TAB+E I" ))D2ZH I*D%CTIO* MOTOR RATI*9S
APPE*DIY BN MACKI*E MODE+ TRA*S4ORMATIO* T e motor model ,)-# ,.- is transformed as follo6sN ' ) 5 't S ' v R 5 = R RR 5 iR 5 ) R 5 't v S 5 =R SS 5 iS 5 6 ere t e transformed quantities areN ,B),B.-
v S 5 =F n v S # iS 5 = F n i S #
v R 5 = F m v R # i R 5 =F m i R #

)S 5 = F n ) S
) R5=F m) R "
,BD,BE-
T e transformation for stator currents iS 5 = F n i S is 6ritten do6n ere 'elo6# for clarit$N
+ameplate ,ata poles E line-line &oltage )IG2J frequenc$ )22K! current E2IA po6er ))D28H cosfi 2"S efficienc$ SF[ slip()22K! 2"GD[ ma1 speed E222rpm torque D:22*m %achine -eometry. tructure.Windings core leng t E)2mm stator stac8 diam" :SDmm air gap lengt .mm stator 'ore diam" EDImm Carter coeff" )"EIF s aft diameter ):Fmm Stator 6indingN t ree-p ase# four-pole# lap-6ound# dou'le la$er# 6 ole-coiled 6inding 6it :2 open slots# F slots/pole/p ase# F form-coils per PP9# I turns per coil# four parallel connected PP9s in eac p ase# slot angular s8e63D:2o/F:# and coil pitc reduction3. slots" Rotor 6indingN fa'ricated copper cage# F: 'ars# semi-closed slots# no s8e6" %achine /hase /arameters RS ,3RS,.-/E.F"DmO m LSl ,3LSl,.-/E2"I2FDmK ,.RRA ,3RR A/E.."GmO m LRlA ,3LRl,.-A/E2"EE.EmK Lm ,3LSg,.-/E.FmK %esh-%odel 0eakage and 'esistance /arameters RPPG )2)".mO m LPPG D"..)mK Rbar F.uO m Lbar )"):IuK Rring )D"GEuO m Lring 2")):IuK
2 i S ) i S . iS
) ) )
) S . S
) . S E S
iS
n )
n ) n ) ) . S S
i S1 i S? n ) . i S@ S n ) . i S Sn
) n ) S
" ,BF-
T e rotor current transformation i R 5 = F m i R is formall$ identical to ,BF-# 6 en !R is used in place of !S" T e sequence resistance matrices are as follo6sN R SS 5 = F n R SS F n= R SS = 'iag { RS }
R RR 5 = F m R RR F m= 'iag {R R }
q
,B:,BG-
6it
RS = R PPG
q
,BI,BS-
R R =. Rb ar ) cos q R R ring / m " T e inductance matrices are similarl$ transformedN

L SS 5 = F n L SS F n= 'iag { L S }
q
,B)2,B))-
L RR 5 = F m L RR F m= 'iag {L R } "
T e lea8age and air gap inductance matrices are transformed accordingl$N L SSl 5 = F n L SSl F n='iag {L Sl }
L SSg 5 = F n L SSg F n = 'iag { L Sg }
q
,B).,B)D,B)E,B)F-
L RRl 5 = F m L RRl F m= 'iag { L Rl }

4ig" G" ))D28H traction induction motor# four PP9s of one p ase are s o6n"
L RRg 5 = F m L RRg F m= 'iag { L Rg }
T e sequence inductances in ,B)2--,B)F- can 'e carried out more directl$ as follo6sN L S = L Sl L Sg L Sl = L PPG
q n q LSg = LSg = Lu u =2 n)
S
,B):,B)G2g
cos qu S
,B)I,B)S,B.2,B.)4ig" )2" Rotor inductance matri1# m3F: ,))D28H motor-"
L R = L Rl L Rg
q
L Rl =. L b ar ) cos q R L ring / m
q m q LRg = LRg = Lk k= 2 m )
R
2g
cos qk R "
T e stator inductances Lus#g and LSg,q- appearing in ,B)Iare s o6n in 4igs" I and S# respecti&el$" T e rotor inductances in ,B.)- are s o6n in 4igs" )2# ))"
4ig" ))" Rotor sequence inductance matri1# m3F: ,))D28H motor-"
T e inductances 6ere e&aluated '$ using 6inding functions" T e transformed stator-rotor mutual inductance matrices are as follo6sN
4ig" I" Stator inductance matri1# n3). ,))D28H motor-"
L SR 5 = F n L SR F m=ban' 'iag { L SR } L RS 5 =F n L RS F n= ban' 'iag { L RS }

q
,B..,B.D-
6 ere t e sequence mutual inductances LSR,q-# LRS,q- are comple1 functions of t e rotor rotation angle %N L SR = L RS = nm e
q q q "q
"
,B.E-
4ig" S" Stator sequence inductance matri1# n3). ,))D28H motor-"
Since t e matrices LSR5 and LRS53,LSR5-t? are usuall$ non-square matrices# t eir s ape is 'and-diagonal BGC as s o6n in ,B.Ffor LRS5 in case of four-pole mac ines ,n3).-" 4igs" ).# )D s o6 t e real and transformed mutual inductances# respecti&el$"
L RS 5 =
[ ]
2 L RS ) L RS . LRS D LRS E L RS F LRS : L RS G L RS I LRS S LRS )2 L RS )) L RS ). L RS )D L RS )E LRS )F LRS
"""
"""
"""
): LRS
)F LRS
"
,B.F4ig" )D" Stator-rotor mutual sequence inductance matri1# n3).# m3F:# t e amplitude -,q-3\LSR,q-\ of t e comple1 coefficients is reported ,))D28Hmotor-"
)E L RS
)D L RS
). L RS
)) RS
)2 LRS
S LRS
I L RS
RE4ERE*CES
B)C C" Bru!!ese# O" Konorati# E" Santini# ]Karmonic current side'and-'ased no&el indicators of 'ro8en 'ars for on-line e&aluation of industrial and rail6a$ cage motor faults^# 4:::(4nternati.nal(S8mp.sium(.n(4n'ustrial ( :lectr.nics2(4S4:(?AAB# Jigo# Spain# _un" E-G# .22G" B.C C" Bru!!ese# C" Boccaletti# O" Konorati# and E" Santini# ]Rotor 'ar 'rea8age in rail6a$ traction squirrel cage induction motors and diagnosis '$ MCSA tec nique" Part IIN T eoretical arrangements for fault-related current side'ands#^ in Pr.c+ ( 4::: ( S7:-P:7# Jienna# Austria# Sep" G-S# .22F# pp" .2S-.)E" BDC S" *andi# T" C" Ilamparit i# S" B" +ee# and D" K$un# ]Detection of eccentricit$ faults in induction mac ines 'ased on nameplate parameters#^ 4:::(Crans+(4n'+(:lectr.n+# &ol" FI# no" F# pp" ):GD-):ID# Ma$ .2))" BEC D" Morinigo-Sotelo# +" A" 9arcia-Escudero# O" Duque-Pere!# and M" Pere!-Alonso# ]Practical aspects of mi1ed-eccentricit$ detection in PHM &oltage-source-in&erter-fed induction motors#^ 4:::(Crans+(4n'+ ( :lectr.n+# &ol" FG# no" )# pp" .F.-.:.# _an" .2)2" BFC Pin@ia ` ang# Mi Du# Ka'etler# T"9"# Bin +u# ]A sur&e$ of condition monitoring and protection met ods for medium-&oltage induction motors#^ 4:::(Crans+(.n(4n'+(*ppl+2 &ol" EG # no" )# pp" DE O E:# .2))" B:C K" Kenao# S" K" Zia# and 9"-A" Capolino# ]Torsional-&i'ration assessment and gear-fault diagnosis in rail6a$ traction s$stem#^ 4:::( Crans+(4n'+(:lectr.n+# &ol" FI# no" F# pp" )G2G-)G)G# Ma$ .2))" BGC C" Bru!!ese# O" Konorati# and E" Santini# ]E&aluation of classic and inno&ati&e side'and-'ased 'ro8en 'ar indicators '$ using an e1perimental cage and a transformed ,n#m- comple1 model#^ in Pr.c+( 4:::(4S4:(9.nf+# Jigo# Spain# _un" E-G# .22G" BIC C" Bru!!ese# ]Stud$ of cardioid-s aped loop current space &ector tra@ectories for rotor eccentricit$ detection in po6er s$nc ronous mac ines#^ in Pr.c+( 4:::(S7:-P:7(9.nf+ # Bologna# Ital$# Sep" F-I# .2))" BSC C" Bru!!ese# ]A &irtual instrument for on-line e&aluation of alternator5s s aft misalignments t roug ICSJA ,Internal Current Space-Jector Anal$sis-#^ in Pr.c+( 4:::(S7:-P:7(9.nf+# Bologna# Ital$# Sep" F-I# .2))" B)2C C" Bru!!ese# ]4ield e1perience 6it t e split-p ase current signature anal$sis ,SPCSA-N Eccentricit$ assessment for a stand-alone alternator in time-&ar$ing and un'alanced load conditions#^ in Pr.c+ ( 4::: ( D:-797(9.nf+# Paris# 4rance# Mar" ))-).# .2)D# pp" .FF-.:I" B))C C" Bru!!ese# ]Eccentricit$ diagnosis in .p-pole alternators t roug superimposition of four .,p7)--pole &irtual mac ines^# in Pr.c+(4:::( 4nternati.nal ( 9.nference ( .n ( P.Eer ( :lectr.nics2 ( 7rives ( an' ( :nerg8 ( S8stems2(P:7:S# Bengaluru# India# Dec" ):-)S .2)." B).C C" Bru!!ese# ]Jalidation of sequence circuits useful for split-p ase current signature anal$sis ,SPCSA- and diagnosis of eccentric-rotor traction cage motors^# in Pr.c+ ( 4::: (:P:9S51@( 9.nf+# Instan'ul# Tur8e$# Oct" .-E# .2)D"
G LRS
: L RS
F L RS
E LRS
D LRS
. L RS
) L RS
4inall$# t e follo6ing important relations ip olds true 'et6een self and mutual inductances 'elonging to t e same sequence order ,note -,q-3Unm3,q- is posed for 're&it$-N
q L SR = - q = LSqg LRqg . .
,B.:-
6 ic defines a perfect coupling 'et6een t e q-t sequence circuits of stator and rotor t roug t e air gap magneti!ing flu1 densit$# 6 en ig er-order space armonics can 'e neglected ,t is is usuall$ t e case for q3p# p7)-"
4ig" )." Stator-rotor mutual inductance matri1# n3).# m3F:# %32 ,))D28H motor-"
Hybrid GM(1,1)-NARnet One Hour ahead Wind Power Prediction

Fatemeh Marzbani, Ahmed Osman, Mohamed Hassan
College of Engineering American University of Sharjah Sharjah, United Arab Emirates {g00046499, aosmanahmed, mshassan}@aus.edu
Aboelmagd Noureldin
Electrical And Computer Engineering Royal Military College of Canada Kingston, Ontario, Canada noureldin-a@rmc.ca
AbstractGrey system theory deals with systems characterized by the uncertainty of the partially known/unknown information. The traditional Grey prediction model GM(1,1) has been widely used in different short-term prediction applications including wind power forecast. However, it is proved that it cannot provide sufcient prediction accuracy. In this paper a new approach for short-term wind power prediction is proposed. The suggested technique is a hybrid method comprised of the GM(1,1) forecasting model and the Nonlinear Auto Regressive neural network (NARnet) method. The forecasting precision of the proposed method is examined by applying it to an actual wind power data set. The experimental results conrm that the proposed technique outperforms the traditional GM(1,1), GM(1,1)-ARMA, and the persistence method.
Keywordswind power forecast; grey theory; grey prediction model; neural networks
I. I NTRODUCTION When compared with other renewable energy sources, wind power utilization has witnessed a dramatic increase in the previous decades. It is demonstrated in the statistics of the Global Wind Energy Council (GWEC) that the total wind power capacity has increased 20% in 2012 [?]. As wind generation technology becomes more mature, it becomes more essential to be accommodated in power systems. To achieve this task, the demand-generation equilibrium should be maintained. However, the intermittent and highly uctuating nature of wind power typically introduce some challenges to the power grid operation. This includes a potential risk increase in the reliability of the electricity supply as well as the operating cost in addition to reserve requirements and limiting the maximum penetration of the wind power in the electricity grid [?], [?]. An accurate wind power prediction mechanism is expected to overcome the previously mentioned challenges. There are four different horizons for wind power prediction: long-term, medium-term, short-term, and very short-term. As a subclass of wind power prediction horizons, short term wind power forecast refers to 30 minutes up to 6 hours ahead prediction. It is mainly oriented to economic load dispatch planning and load increment/decrement decisions applications [?].
Wind power prediction methodologies can be divided into two main approaches. The rst approach is based on forecasting the wind speed and then utilizing the wind turbine power curves in order to convert it to wind power. In the second approach, the historical wind power data is directly employed in predicting the wind power. During the last decades several prediction approaches have been developed and utilized. Among the developed methods, time-series-based approaches are widely used. Time-seriesbased methods identify the historical data patterns and then predict the future values based on these obtained patterns. These methods include Auto Regressive (AR), Moving Average (MA), Auto Regressive Moving Average (ARMA), and Auto Regression Integrated Moving Average (ARIMA) models [?], [?]. There also exists Neural-Network-based prediction models [?] such as: Articial Neural Network (ANN) [?], Radial Basis Function (RBF) network, and Adaptive Networkbased Fuzzy Inference System (ANFIS) [?]. The Neural Network (NN)-based methods are employed for pattern identication and parameter estimations which typically require huge sets of historical data. In correlation-based prediction models, also known as spatial correlation models, the spatial correlation between wind power in different neighboring sites is exploited when predicting the wind power. In more details, in correlation-based models, wind speed/power at a given site is predicted using some knowledge about the wind speed/power of the neighbor sites [?]. A major difculty faced by such methods stem from the need of huge measurements of many spatial correlated sites in addition to the needed timely communications of the measured data. Such disadvantages also motivates the need for developing precise yet more simple prediction techniques. Wind power historical data usually contains some measurement errors. Consequently, such errors might lead to inefciently predicted values. Therefore, models with the capability of moderating the impact of measurement errors are needed. Grey models are powerful prediction techniques that can tolerate the impact of uncertain information. Hence, Grey models are good alternatives when compared with other shortterm predictor tools. Various applications benet from Grey prediction models. This includes transport planning, stock
price forecasting, annual maximum energy demand, etc. [?], [?], [?]. Grey models can be used to predict wind speed and wind power [?], [?]. In [?], the Grey model was improved to reduce the prediction error. Furthermore, hybrid Grey models such as GM-Weibull and GM (1,1)-Cloud model are discussed in [?], [?] [18]. In this paper, a hybrid Grey-based GM(1,1) prediction model is suggested to provide one hour ahead wind power forecast. The proposed hybrid predictor is comprised of a GM(1,1) integrated with Nonlinear Auto Regressive Neural Network (NARnet) model. The results obtained from the proposed model are compared with the corresponding results acquired by the traditional GM(1,1) alone then with those obtained by a GM(1,1)-ARMA model as well as the results of a persistent model. The rest of the paper is organized as follows. Section II describes the preparation procedure applied to the wind power time series under investigation. The general methodology to establish the traditional Grey forecasting model is dealt with in Section III. The scheme of the hybrid GM(1,1)ARMA prediction model is presented in Section IV. Section V describes the procedure for developing the GM(1,1)-NARnet forecasting model and the obtained results are presented. Finally, Section VI concludes the paper. II. P REPARATION OF THE WIND POWER TIME SERIES The objective of the present study is to propose an accurate yet simple short-term (one hour ahead) wind power forecast model. The proposed model is tested using a historical data set published by Alberta Electrical System Operator (AESO) [?]. The utilized wind power data set is recorded at 10 minutes intervals during January 2011 with a length of 4464 measurements. This data is presented in Fig.1. The prediction models investigated in this study are applied to two sets of wind power time series: unprocessed and processed data sets. The unprocessed time series refers to the original wind power as measured and recorded during January of 2011, while the processed time series is the hourly averages obtained as follows:
6
700
600
500
Wind Power(MW)
400
300
200
100
500
1000
1500
2000 2500 Number of samples
3000
3500
4000
4500
Fig. 1. Original wind power time series of January 2011.
known/unknown information [?]. Grey prediction models are also known by their high accuracy and the need for less historical data in addition to their simplicity of calculations. Among the family of Grey predictors, the most frequently used model is the GM(1,1) model. In order to establish a GM(1,1) model, four phases should be developed. The mathematical description and formulation corresponding to each phase are briey discussed in what follows. A. Accumulated Generating Operation (AGO) In this stage the rst order accumulated generating operator (AGO) is applied to the unprocessed time series P (0) (obtained through measurement). The obtained time series P (1) includes less randomness compared to the original data and can be mathematically described as follows:
k
P (1) (k ) =
i=1
P (0) (i), k = 1, 2, . . . , n,
(2)
where n represents the number of the measurements (n 4). B. GM(1,1) Differential Equation The mathematical formulation for the GM(1,1) differential equation can be expressed as follows:
Prcd (i) Pavg (j ) =

i=1
, j = 1, 2, . . . , 744,
(1)
P (0) (k ) + aZ (1) (k ) = b,
(3)
where Pavg (j ) represents the hourly average wind power during the hour j , and Prcd (i) refers to the ith recorded data point during the hour j, i = 1, . . . , 6. The proposed predictor is applied to both processed and unprocessed time series. For the unprocessed case, six steps ahead predictions are obtained; then they are averaged to provide one hour ahead prediction. The hourly average time series is directly applied to the predictor and one hour ahead predicted values are obtained. III. T RADITIONAL GM(1,1) PREDICTION M ODEL The Grey system theory has been originally introduced by Julong Deng in 1982 [?]. Grey systems refer to systems with the ability to deal with uncertainty in the partially
where the mean generated sequence of consecutive neighbors of P (1) represented by Z (1) is described as follows: Z (1) (k ) = 0.5 P (1) (k ) + P (1) (k 1) , k = 2, 3, . . . , n. (4) The Grey differential equations describe the relations between the dependent variables and independent ones. Specically, for the traditional GM(1,1), the dynamic model is characterized by one independent variable and no dependent variables. Therefore, the following equation states the whitening equation. dP (1) + aP (1) = b, dt (5)
where the independent variable is depicted by P and the Grey model coefcients are represented by a and b. C. Parameters Estimation The coefcients of GM(1,1) can be determined by the aid of the least square method. A = [a, b] = B B in which, B= Z (1) (2) Z (1) (3) . . . Z (1) (n) 1 1 . . . 1
T T T 1
(0) (k + 1) P (0) (k + 1) P n
RM SE =
k=1
(12)
B Y
(6)
The calculated RMSE for the GM(1,1) model using unprocessed and processed data are 2.7755 and 12.8134 MW, respectively. The two RMSE values are smaller than the RMSE of the persistence method of 19.88 MW. IV. C OMBINED GM(1,1)-ARMA FORECASTING MODEL
, (7)
Y = P (0) (2), P (0) (3), ..., P (0) (n) D. Calculating the forecasted values
(8)
In this stage the predicted values are obtained by solving the differential equation. According to (5), the AGO solution of P (1) (k ) is: (1) (k + 1) = eak P (0) (1) b P a b + , k = 1, 2, . . . , n. a (9)
E. Inverse Accumulated Generating Operation (IAGO) In order to obtain the original predicted time series it is required to utilize the inverse of the AGO (IAGO). The mathematical formulation of the original forecasted values can be written as: (0) (k +1) = eak P (0) (1) b P a (1 ea ) , k = 1, . . . , n.
(10) The GM(1,1) has a rolling mechanism in which each time the predicted data becomes an observation the original time series gets updated. This prediction model will be established by repeating the steps described in Sections III-A to III-E. Based on the Grey system theory, the dynamics of the wind power can be modeled by a Grey system. Thus, Grey forecasting models can be utilized in order to provide short-term wind power prediction. In this study, the rst ve measurements are taken as the training data and the GM(1,1) is applied to both unprocessed and processed wind power time series. In order to evaluate the accuracy of forecasting, the Relative Percentage Error (RPE) for the (k + 1) instant is calculated as follows: RP E (k +1) = (0) (k + 1) P (0) (k + 1) P P (0) (k + 1) 100, (11)
It is noteworthy that the GM(1,1) model is applicable to roughly exponential measurements and, therefore, when used alone it is difcult to precisely predict the wind power time series which typically includes non-stationarity. Autoregressive moving average (ARMA) models are simple though useful linear prediction models that enable the estimation of a future value of a time series. When ARMA models are used, the future value of the series is expressed as a linear combination of a number of immediate past values of the time series. An ARMA model is characterized by the order of the autoregressive model p and the order of the moving average model q and is typically written as ARMA(p, q ). A major issue with ARMA models lies in the proper identication of the parameters p and q in addition to the assumption about the full knowledge of the time series. In order to provide more precise short-term predictions and to take the advantages of GM(1,1) as well as the ARMA models, the hybrid GM(1,1)ARMA short-term forecasting model is proposed [?]. The algorithm of this predictor can be described in three steps. First, forecast the wind power time series by employing the GM(1,1) prediction model and obtain the corresponding residuals. Second, investigate the stationarity of the residuals time series to nd the proper order of ARMA(p, q ) model. Third, modify the Grey predicted values using the forecasted residuals. This forecasting technique is applied to both (processed and unprocessed) sets of the wind power time series. The corresponding RPE values are calculated and recorded in Table II. The obtained RMSE for unprocessed and processed time series are 2.1981 and 7.6240 MW, respectively. In comparison with the RMSE values obtained from GM(1,1), these values are smaller which proves the accuracy of the GM(1,1)-ARMA prediction model. As seen in Fig. 2, applying the average process to the predicted values leads to more efcient forecast. V. H YBRID GM(1,1)-NAR NET MODEL According to the Wold decomposition theorem a discrete stationary time series is comprised of Linear-Deterministic (LD), Nonlinear Deterministic (ND), and unpredictable components [?]. As previously mentioned, ARMA models are linear deterministic models that predict values of a given time series by utilizing the linear deterministic components whereas the nonlinear components will be left in the residuals. The wind power time series is a function of wind speed which is highly random data. Thus, in order to provide an accurate
where (k + 1) n. The obtained GM(1,1) RPE for both processed and unprocessed data are summarized in Table I. Moreover, the efciency of the proposed method is compared with the traditional GM(1,1) by the aid of calculating the Root Mean Square Error (RMSE) given by:
TABLE I O NE HOUR AHEAD GM(1,1) RPE USING UNPROCESSED W IND POWER TIME SERIES . Time (hr) Unprocessed Data Processed Data Persistence 1 0.7539 2.6801 0 2 0.9538 2.5571 5.2913 3 0.1667 1.7619 9.4806 4 0.1354 1.8092 3.8371 5 0.5143 3.2699 1.1482
TABLE II O NE HOUR AHEAD GM(1,1)-ARMA RPE USING UNPROCESSED W IND POWER TIME SERIES . Time (hr) Unprocessed Data Processed Data Persistence 1 0.4225 0.6552 0 2 0.0504 0.1539 5.2913 3 0.2865 2.5489 9.4806 4 0.7992 1.5003 3.8371 5 0.1829 1.3704 1.1482
prediction, it is essential to consider the nonlinear components. In the GM(1,1)-ARMA predictor, the nonlinear components are left in the residuals and are not used in the forecasting process. Hence, it may not be efciently able to describe the wind power time series. Nonlinear Autoregressive Network forecasting models predict future values of a time series based on the nonlinear combination of the previously measured data. Neural network models are capable of dealing with intricate systems. There are several types of neural architectures in the family of neural models. The Neural network architecture includes three or more layers: one input layer, one output layer and one or more hidden layers. The mathematical procedure of the network can be described by the following equation. Yj = f
i th
The same training set (including ve measurements) is used to nd the GM(1,1) coefcients and then, the Grey forecasted values are obtained. The corresponding time series is predicted by the aid of NARnet model. Finally, the Grey prediction is modied by the obtained NARnet residuals forecast. Figures 5 and 6 show that for both processed and unprocessed wind data, the obtained RPE for the GM(1,1)-NARnet is much smaller than that of the traditional GM(1,1) and GM(1,1)ARMA models. Fig. 9 presents a visual comparison between the predicted values as obtained by the GM(1,1)-NARnet, the traditional GM(1,1), and the GM(1,1)-ARMA. It is proved that for all prediction methods, averaging the predicted values results in more accurate forecast in comparison with applying the average process to the original data. VI. C ONCLUSIONS In this paper, we proposed a new prediction tool for one hour ahead wind power prediction based on the Grey forecasting model. This technique is a hybrid technique that benets from the advantages of Grey system theory in addition to the simplicity of ARMA models. In the proposed method, rst an equal dimension Grey prediction model is established using the wind power measurement time series. Next, NARnet model is applied to the obtained Grey residuals. The results afrm that the traditional Grey prediction model and the hybrid GM(1,1)-ARMA model improve the accuracy of the forecasted values. R EFERENCES
[1] http://www.gwec.net. [2] Y. Vardanyan and M. Amelin, The state-of-the-art of the short term hydro power planning with large amount of wind power in the system, in Energy Market (EEM), 2011 8th International Conference on the European, pp. 448454, 2011. [3] E. Ela, B. Kirby, E. Lannoye, M. Milligan, D. Flynn, B. Zavadil, and M. OMalley, Evolution of operating reserve determination in wind power integration studies, in Power and Energy Society General Meeting, 2010 IEEE, pp. 18, 2010. [4] S. Soman, H. Zareipour, O. Malik, and P. Mandal, A review of wind power and wind speed forecasting methods with different time horizons, in North American Power Symposium (NAPS), 2010, pp. 18, 2010. [5] Z. Huang and Z. Chalabi, Use of time-series analysis to model and forecast wind speed, Journal of Wind Engineering and Industrial Aerodynamics, vol. 56, no. 23, pp. 311 322, 1995.
Wij Xij
(13)
where the output of the j node is represented by Yj , the connection weight between node j and node i is given by Wij , Xij is the input signal form node i to node j , and f () is the nonlinear transfer function between layers. In this study, a hybrid method that improves the prediction accuracy and takes the advantages of the GM(1,1) and NARnet prediction models is proposed. The network architecture includes 4 layers: an input layer, an output layer and two hidden layers. The training algorithm employs the Levenberg-Marquardt back-propagation method constrained on the minimum mean square error. The GM(1,1)-NARnet algorithm can be described as follows: 1) Let P (0) represent the wind power measurements and select a proper window size of data as the training set. 2) Establish the GM(1,1) forecasting model and calculate the predicted AGO wind power time series. 3) Obtain the original predicted wind power time series by employing the IAGO. Then calculate the corresponding residuals et . 4) Build a proper NARnet predictor by nding the proper number of hidden layers and nodes, then, train the network and obtain the forecasting values of Grey residuals. 5) Modify the Grey wind power predicted time series by utilizing the values of NARnet model to the Grey residuals.
[b]0.5
1 0.9 0.8 Relative Percentage Error 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1 2 3 Time (hr) 4 5
GM(1,1) GM(1,1)ARMA GM(1,1)NARnet
560 Measurements Unprocessed data Processed data Persistence model
540
520 Wind Power(MW)
500
Fig. 5. One hour ahead GM(1,1)-NARnet RPE using unprocessed data.

3.5
GM(1,1) GM(1,1)ARMA GM(1,1)NARnet
480
Relative Percentage Error
3 2.5 2 1.5 1 0.5 0
460
440
3 Time (hr)
420 1
Fig. 6. One hour ahead GM(1,1)-NARnet RPE using processed data.
1.5
2.5
3 3.5 4 4.5 5 Time (hr) [6] A. Sfetsos, A comparison of various forecasting techniques applied to
[7] [8] mean hourly wind speed time series, Renewable Energy, vol. 21, no. 1, pp. 23 35, 2000. A. More and M. Deo, Forecasting wind with neural networks, Marine Structures, vol. 16, no. 1, pp. 35 49, 2003. M. Costa and E. Pasero, Articial neural systems for verglass forecast, in Neural Networks, 2001. Proceedings.IJCNN 01. International Joint Conference on, vol. 1, pp. 258262 vol.1, 2001. I. Damousis, M. Alexiadis, J. Theocharis, and P. Dokopoulos, A fuzzy model for wind speed prediction and power generation in wind parks using spatial correlation, Energy Conversion, IEEE Transactions on, vol. 19, no. 2, pp. 352361, 2004. M. Alexiadis, P. Dokopoulos, H. Sahsamanoglou, and I. Manousaridis, Short-term forecasting of wind speed and related electrical power, Solar Energy, vol. 63, no. 1, pp. 61 68, 1998. C.-I. Hsua and Y.-H. Wenb, Improved grey prediction models for the trans-pacic air passenger market, Transportation Planning and Technology, vol. 22, no. 2, pp. 87107, 1998. C. Li, Study on theory of the grey Markov chain method and its application, in Computational Engineering in Systems Applications, IMACS Multiconference on, pp. 17421746, 2006. H. Morita, T. Kase, Y. Tamura, and S. Iwamoto, Interval prediction of annual maximum demand using grey dynamic model, International Journal of Electrical Power and Energy Systems, vol. 18, no. 7, pp. 409 413, 1996. C.-Y. Huang, Y.-W. Liu, W.-C. Tzeng, and P.-Y. Wang, Short term wind speed predictions by using the grey prediction model based forecast method, in 2011 IEEE Green Technologies Conference (IEEE-Green), pp. 15, 2011. T. H. M. El-Fouly, E. F. El-Saadany, and M. M. A. Salama, Grey predictor for wind energy conversion systems output power prediction, IEEE Transactions on Power Systems, vol. 21, no. 3, pp. 1450 1452, 2006. T. H. M. El-Fouly, E. F. El-Saadany, and M. M. A. Salama, Improved
Fig. 2. Using unprocessed data.
[b]0.5
560 Measurements Unprocessed data Processed data Persistence model
[9]
540
[10]
520 Wind Power(MW)
[11]
[12]
500
[13]
480
[14]
460
[15]
[16]
440
420
[b]0.5
560
[17]
540
Measurements GM(1,1)NARnet GM(1,1)ARMA
[18]
520 Wind Power(MW)
[19] [20] [21] [22] [23]
500
grey predictor rolling models for wind power prediction, IET on Generation, Transmission and Distribution, vol. 1, no. 6, pp. 928 937, 2007. Z. He, X. Jianyuan, and W. Xiaoyu, The output forecast research of the wind farm based on GM-weibull wind speed distribution combined model, in 4th IEEE Conference on Industrial Electronics and Applications, ICIEA 2009, pp. 21372140, 2009. X. Han, F. Meng, Z. Song, and X. Li, Short-term prediction of wind power combining GM(1,1) model with cloud model, in 2012 IEEE International Conference on Automation and Logistics (ICAL), pp. 191 195, 2012. http://www.aeso.ca. D. Ju-Long, Control problems of grey systems, Systems and Control Letters, vol. 1, no. 5, pp. 288 294, 1982. K. Wen, Grey systems: modeling and prediction. R. Wang, L. Yao, and Y. Li, A hybrid forecasting method for dayahead electric-ity price based on GM(1,1) and arma, in GSIS 2009. IEEE International Conference on, pp. 557581, 2009. A. Wold and Wiksell., A Study in the Analysis of Stationary Time Series.
480
460
440
420
1.5
2.5
3 Time (hr)
3.5
4.5
Fig. 7. Using unprocessed data.
[b]0.5
540 Measurements GM(1,1)NARnet GM(1,1)ARMA GM(1,1)
520
500 Wind Power(MW)
480
460
Active Damping Control of DFIG Wind Turbines during Fault Ride Through
Hao Xu1,2, Honghua Xu1, Liang Chen3, Jan Wenske3
1 Institute of Electrical Engineering, Chinese Academy of Sciences, Beijing, 100190, China 2 University of Chinese Academy of Sciences, Beijing, 100049, China 3 Fraunhofer Institute for Wind Energy and Energy System Technology, Bremerhaven, 27570, Germany xuhaocas@gmail.com AbstractGrid voltage sag causes typical transients in the
generator electromagnetic torque, which result in significant stress on the wind turbine drive-train. In this paper, a combination of different simulation packages, namely FAST (Fatigue, Aerodynamics, Structure, Turbulence) and Simulink, is used to model the mechanical and electrical aspects of a 5-MW doubly-fed induction generator (DFIG) based wind turbine respectively. To reduce the mechanical loads induced by grid faults, a linearquadratic regulator (LQR) based generator torque controller is presented. Simulation results showed that the active damping controller can effectively suppress the vibration of drive-train during fault ride through (FRT). Index Termsactive damping control, drive-train, DFIG, grid fault, wind turbine
I.
INTRODUCTION
In the past few years many power system operators have revised their grid codes and set higher threshold for wind power integration [1]. Among these new requirements, voltage disturbance ride through capability is the most important issue, especially the requirements for wind turbines to stay connected to the grid during and after a voltage dip (fault ride through requirement). It implies potential challenges for both the electrical and mechanical systems of wind turbine. Variable speed wind turbines based on DFIG are currently the most popular wind energy generation system. The conventional way to protect this type of wind turbine from grid fault is by installing crowbar devices [2-4]. The crowbar is capable of limiting the over voltage and current within the generator. However, the rotor circuit of the DFIG has to be shorted by the crowbar during the grid fault which turns the DFIG into a conventional induction generator (IG). During generator mode transition, the generator torque begins to oscillate and the oscillation will be transferred to the drive-train. As a result, large mechanical loads within drive-train should be expected in the post-fault process. Especially in a worst-case scenario the turbine is loaded with rated torque before a FRT grid event. Drive-train damping filter (DDF) is a widely used application to reduce the loads on the drive-train [5, 6]. The
DDF is basically a band pass filter sitting on the drive-train first dynamic mode frequency. The wind turbine torque demand is slightly modified by the filter to increase drive-train damping. The DDF is normally tuned for wind turbine normal operation. However, the vibration induced by grid fault is much larger than the vibration induced by wind speed variation. The conventional DDF is not able to produce enough damping for wind turbine drive-train during grid fault [7]. In this paper, an alternative approach through LQR controller with state estimation is designed, which can be implemented in both normal and fault operation. To study the structural loads of wind turbines under grid fault, a complete model including both mechanical and electrical parts should be constructed. A drawback of wind turbine simulation is that either a simple mechanical component is used with a detailed electrical model, or a simple electrical model with a detailed mechanical model is used, which could not provide a thorough insight on the structural loads caused by sudden grid disturbances. In this paper, a combination of different simulation packages, namely FAST (Fatigue, Aerodynamics, Structure, Turbulence) and Simulink, is used to model the electrical and mechanical aspects of a wind turbine respectively. The paper will be organized as follows: In section , the complete wind turbine model is introduced in detail. Section shows the impact of voltage sag on drive-train. Section introduces the improved active damping control strategy. Sectionshows simulation results of different control strategies. II.
WIND TURBINE MODEL
A. Mechanical model based on FAST FAST is a comprehensive aeroelastic simulation code capable of predicting both extreme and fatigue loads of two- and three-bladed horizontal-axis wind turbines [8, 9], which is developed by NREL (National Renewable Energy Laboratory). The core of this code is an advanced model for flexible structure of the wind turbines, taking the flexibility of the tower, blades and other components of the wind turbines into account.
In this research, FAST has been used to model an full dynamic wind turbine, considering 16 DOFs (degree of freedom) including first and second flapwise blade mode, edgewise blade mode, drive-train rotational flexibility, yaw angle, first and second tower fore-aft bending mode, first and second tower side-side bending mode, and generator azimuth angle. To define a typical large wind turbine, we use properties from the NREL 5-MW baseline model [10], summarized in Table I. Although FAST contains thus detailed models for the aeroelastic and mechanical aspects in a wind turbine, the models for the electrical components and controller are typically very simplified. Moreover, FAST can be used to exchange data with Simulink. The controller and the electrical parts of a wind turbine can be implemented by Simulink blocks where the wind turbine is simulated by FAST, as shown in Fig. 1.
gen
stator, rotor resistance; Rs, Rr stator, rotor and magnetizing inductances; Ls, Lr, Lm s , r stator and rotor angular electrical speed; p number of pole pairs. All rotor-side variables and parameters above are referred to the stator. The parameters for DFIG used in this study are given in Table .
TABLE I. SUMMARY OF WIND TURBINE PROPERTIES 5 MW 3-bladed, upwind Variable speed, variable pitch 126m 90m 11.4m/s, 25m/s 12.1rpm 3.09107kgm2 97:1 8.676108Nm/rad 6.215106Nms/rad
Power rating Rotor Control Rotor diameter Hub height Rated, cut-out wind speed Rated rotor speed Total rotor inertial (J1+ J2) Gearbox ratio Equivalent shaft stiffness (K23) Equivalent shaft damping (D23) TABLE II.
PRAMETERS OF DFIG MODEL 5.0MW 960V 5.03106kgm2 2.1m 2.1m 0.153mH 0.149mH 4.26mH 3
Te* , Q*
1,2,3
Figure 1. Simulation structure of DFIG wind turbine
B. Electrical part model based on Simulink 1) DFIG subsystem The electrical model of the DFIG is basically a wound rotor induction machine. It is expressed in a reference frame (d-q frame rotating at synchronous speed. Stator and rotor equation are both based on motor convention, taking positive currents going into the machine. The induction machine modeling equations can be written as [11]: vds = Rs ids + (1) ds s qs
Rated power Rated voltage (line to line) Generator inertia (J3) Rs Rr Ls Lr Lm p
+ vqs = Rsiqs + qs s ds ( ) vdr = Rr idr + dr s r qr + ( ) vqr = Rr iqr + qr s r dr

The flux and current relations are given as ds = Ls ids + Lmidr
(2) (3) (4) (5) (6) (7) (8)
qs = Lsiqs + Lmiqr dr = Lr idr + Lmids qr = Lr iqr + Lmiqs

The electromagnetic torque is expressed as Te = 1.5 p(ds iqs qs ids )
(9)
where d, q s, r
direct and quadrature axis; stator and rotor variables;
2) Control system Wind turbine control system includes pitch controller and generator torque controller. The basic control objectives are to optimize the power production for below-rated wind speed and to limit the aerodynamic power for above-rated wind speed. For below-rated wind speed, the generator speed is varied by controlling the generator torque to follow an optimal torque versus speed curve. For above-rated wind speed, the generator torque is held constant at the rated torque and the rotor speed is controlled by limiting the aerodynamic torque by varying the st pitch angle. The pitch actuator is modeled by a 1 order lag and a limiter for both pitch angle and its rate of change. 3) Converter system A back-to-back converter is used to connect the rotor of the DFIG to the grid. The DFIG is controlled by the rotor side converter. This way the active and reactive stator power is controlled indirectly by means of the inner rotor current control loop. The control is performed field oriented, where the rotor current loop is stator flux orientated. The grid side converter controls the DC link voltage to a constant value and may also control the active and reactive power taken from or transferred to the grid. For this analysis the converter is not modeled with its switching function but as continuous sinusoidal voltage source. 4) Crowbar
In case of a rotor overcurrent due to a grid fault the rotor side converter has to be disconnected from the rotor. The rotor of the DFIG is switched from the rotor side converter to three phase resistor Rcrow. The external resistor Rcrow is applied to reduce the rotor current magnitude after switching the crowbar. So the generator with activated crowbar operates as an induction generator with rotor nearly in short circuit and with a large reactive current. According to [2], small resistance values cause higher current and torque transient peaks in the fault moment while a too high crowbar resistance can imply a risk of excessive rotor current, torque and reactive power transients when crowbar is removed. In this paper, a constant resistance value of 100Rr will be used as an appropriate trade-off. The impact of the crowbar injection on the drive-train will be discussed in Section IV. 5) Grid Model The grid is represented by a Thevenin equivalent, consisting of three constant voltage source connected in series with their impedance. To simulate the voltage dip, a voltage step with duration of 200ms, is applied to the voltage source.
active damping controller based on generator torque control will be presented in this paper.
Figure 2. Fault ride through without active damping controller
III.
IMPACT OF VOLTAGE SAG ON DRIVE-TRAIN
In order to show the impact of voltage sag on drive-train, we considered the worst-case scenario. (i.e., the turbine is loaded with rated torque within the drive-train before a FRT grid event.) Three dimensional turbulent wind has been used for simulating inflow turbulence environments which is produced by Turbsim [12]. In this study, the 3 phase voltage dip is simulated by a voltage step of the voltage source in the grid model at t=15s, with duration 200ms, which causes stator voltage sags to 20%. As shown in Fig. 2, the time domain simulation results from top to bottom are wind speed, generator speed, electrical torque, low-speed shaft (LSS) mechanical torque, pitch angle and active power output respectively. When voltage sag happens in t=15s, the rotor over-current is detected immediately and crowbar is engaged until the fault clearance. During the crowbar period, DFIG operates as a conventional IG and thus the electrical torque is decreased almost to zero. The generator switches back to DFIG mode after fault clearance and generator torque is controlled at the rated torque. The generator experiences large torque variation between mode switch which creates large speed oscillation and torque oscillation on the drive-train. Note that the torsional oscillations exited by the grid fault are only slightly damped still 10s after the grid fault event without damping controller, which would cause severe gearbox damage and may leads to rapidly to turbine shut-down. The power spectral density (PSD) of LSS torque illustrated in Fig. 3 shows that two dominant oscillation frequencies exist at 1.7Hz and 4 Hz, which should be suppressed through active damping controller. Although the pitch controller is activated in this case, it is not able to damp the oscillation using pitch controller due to the large time constant of pitch control. Thus
Figure 3. PSD of LSS torque without active damping controller
IV.
ACTIVE DAMPING CONTROL
Large drive-train torsional torque will be expected due to weak structure damping in case of grid fault. A LQR controller is designed to adjust the generator torque at the end of the drivetrain to remove the unwanted variations. A. Drive-train equivalent model In earlier work, a two-mass model was used to represent the drive-train mechanical system which takes into account only the shafts flexibilities and assumes that the blades are rigid. Nevertheless, in the study of electrical transient performance, it is also important to consider the rotor dynamics[13, 14]. The reason is that the blade edgewise symmetrical mode couples directly to the drive-train so if this mode is excited it will also lead to torsional vibrations. Hence a three-mass model which considers both drive-train and rotor dynamics is employed here,
as shown in Fig. 4. Where J1 represents the inertia of the effective flexible part of the blades, J2 represents the inertia of the hub and the rigid part of the blades, J3 represents the generator inertia. In Fig. 5 the used mathematical model is presented. Where K12 represents the effective blade stiffness, K23 represents the resultant of the low-speed and high-speed shafts stiffness, D12, D23 are the damping coefficients, N is the gearbox ratio, 1,2,3 and 1,2,3 represent the rotation angle and rotation angular speed of the three mass respectively, Twin and Te represent the aerodynamic torque input and electromagnetic torque input. The dynamic equations referred to the low-speed shaft (LSS) are given by (10-14): 1 = Twin K12 (1 2 ) D12 (1 2 ) (10) J 1 2 = K12 (1 2 ) + D12 (1 2 ) J 2 (11) K 23 ( 2 3 ) D23 (2 3 ) 3 = K 23 ( 2 3 ) + D23 (2 3 ) N Te J 3 (12) = (13)
1 2 1 2
= (14) 2 3 2 3 The natural frequencies of vibration (f1, f2) of the three-mass model are given by (15):
1 b b2 4c 0.5 1 b b2 4c 0.5 ( ) , f2 = ( + ) (15) 2 2 2 2 2 2 where 1 1 1 1 J + J2 + J3 b = [ K12 ( + ) + K 23 ( + )], c = K12 K 23 ( 1 ) J1 J 2 J2 J3 J1 J 2 J 3 Notice that the damping coefficients D12 and D23 are relatively small, and for the purpose of controller designing they are neglected considering the worst case. The generator inertia J3, the shaft stiffness K23, and the rotor total inertia J1+J2 are known variables. The two vibration frequencies have been obtained by simulation result in section as 1.7 Hz and 4 Hz. Substituting these known values in (15) and solving J1, J2 and K12 yield the following three-mass model parameters: J1=3.09107kgm2, J2=4.54106kgm2, K12=3.09107Nm/rad. f1 =
Figure 5. Simplified mathematical model of the mechanical drive-train
B. LQR design Fig. 6 is a schematic of the proposed control paradigm; the control scheme is modified to compensate for the above contingencies by a modification on the generator torque control reference. Te* = TBC + TLQR (16) where TBC is the basic torque control reference calculated from the torque-speed curve. TLQR is the control signal contribution by the LQR, which aims to add damping to the drive-train torsional mode and blade edgewise symmetrical mode without energy loss, using a state-space structure so that multiple control objectives can be considered simultaneously.
Twin
+ +
* e
yk
1 & 1
2 & 2
3 & 3
x
Figure 4. Equivalent tosional representation
T win
Figure 6. Active damping controller
A linearized model of a DFIG based wind turbine drive-train is developed from (10-14) and may be expressed in the general form: x = Ax + Bu + Bd ud (17) y = Cx Where x is a vector consisting of five system states: x = [1 2 1 2 2 3 2 3 3 ]T , Additionally, u represents the control input command signal ( Te ), ud represents the disturbance input signal ( Tin ).
A R 55 is the state matrix, B R 51 is the control input matrix, Bd R 51 is the disturbance input matrix and C R15 is the output matrix. Neglecting the damping coefficients D12 and D23, the various matrices in (17) are constituted as follows: K12 K12 K 23 0 0 0 J J J2 1 2 0 0 0 0 1 K K K12 0 23 23 0 , A = 0 J2 J2 J3 0 0 1 0 0 K 23 0 0 0 0 J3
C. State estimator The practical application of control techniques is usually limited by problems in obtaining measurements of the states needed in the controller. Extra sensors and measurements can add considerable cost and complexity to a wind turbine, while errors in the measurement of these states can cause poor controller behavior. Thus, a state estimator is employed, where unknown states are determined from just one measured generator speed. In order to achieve good state estimation despite having uncertainty in the model and measurement noise, a discrete-time Kalman filter (KF) is used here. A KF makes a one-step-ahead prediction of the states, and also estimates what the measured output would be. A correction updates the state estimates, taking into account the prediction error. Thus k +1 = x k + M ( yk y k ) x (20)
Where, assuming the stochastic distances acting on the system are Gaussian, the matrix M is computed from the system dynamics and the disturbances, subject to minimization of the expected sum of squares of the prediction error. The disturbance input ( Twin ) to the KF is not available and thus has to be estimated. In this paper, the aerodynamic torque estimate is through solving the differential equation given by (10) using the derivative of the generator speed [15]. In order to mitigate the problem of noise associated with derivatives, a lowpass filter was used. V. SIMULATION RESULTS The proposed active damping controller is tested by simulations in the same wind condition as Section . In this case, the generator torque is controlled by the modified torque command given by (16) instead of the basic torque control reference. Compared Fig. 7 to Fig. 2, the generator speed oscillation and torque oscillation within drive-train can be damped rapidly after 3 seconds. The price for the improved system dynamics is the increased electrical loading of the system when the damping torque is positive, as can be seen from power signal. Frequency converter is the critical component for overloading because its power electronic devices have the maximum voltage and current they can tolerate. Nevertheless, it is practicable because converter does allow certain overloading that is usually defined in a form of overpower as a function of time it can be withstood (e.g. 150% of rated power during 1 s). For safety, the overloading has been limited to 10% in this contribution. The power ripple in normal operation is rather small due to a trade-off between Q and R. A comparison of LSS torque PSD is shown in Fig. 8, it shows that significant reduction has been achieved at the resonant frequencies using active damping controller. That means the damping controller offered adequate damping to both mode frequencies.
B = 0 0
N J3
N , J3
T
1 Bd = 0 0 0 0 , J1 C = [0 0 0 0 1] . Using the state-space model in designing the LQR to effective optimal control, a performance index J defines the controller objectives. The quadratic function J is expressed by: J = (18) x T (t )Q x (t ) + u T (t ) Ru (t ) dt 0 Where Q is the symmetric, positive semi-definite weighting matrix on the states that satisfies the algebraic Riccati equation (ARE) and R is the symmetric, positive definite weighting on the control input. Often, it is adequate to let the two matrices simply be diagonal. In this study, the matrices are adjusted based on a trade-off between control performance (Q large) and low power fluctuation (R large). We first picked values for R and Q and then simulating the model against a step input. The resulting lowspeed shaft load was compared after each iteration until reasonable shaft load was obtained. The optimal control sequence minimizing the performance index is given by (19), which is the control component TLQR in (16), where the gain G can be obtained by simulating a MATLAB function lqr(). (19) u ( t ) = G x ( t )
using crowbars," in Power Electronics and Applications, 2007 European Conference on, 2007, pp. 1-8. L. G. Meegahapola, T. Littler, and D. Flynn, "Decoupled-DFIG fault ride-through strategy for enhanced stability performance during grid faults," Sustainable Energy, IEEE Transactions on, vol. 1, pp. 152162, 2010. A. Nasiri, E. Muljadi, G. Mandic, and F. Oyague, "Active Torque Control for Gearbox Load Reduction in a Variable Speed Wind Turbine," 2011. E. Bossanyi, "Wind turbine control for load reduction," Wind Energy, vol. 6, pp. 229-244, 2003. F. Zhang, W. Leithead, and O. Anaya-Lara, "Wind turbine control design to enhance the fault ride-through capability," in Renewable Power Generation (RPG 2011), IET Conference on, 2011, pp. 1-6. J. M. Jonkman and M. L. Buhl Jr, "FAST users guide," Golden, CO: National Renewable Energy Laboratory, 2005. R. Fadaeinedjad, M. Moallem, and G. Moschopoulos, "Simulation of a wind turbine with doubly fed induction generator by FAST and Simulink," Energy Conversion, IEEE Transactions on, vol. 23, pp. 690-700, 2008. J. M. Jonkman, S. Butterfield, W. Musial, and G. Scott, Definition of a 5-MW reference wind turbine for offshore system development: National Renewable Energy Laboratory Colorado, 2009. B. Wu, Y. Lang, N. Zargari, and S. Kouro, Power conversion and control of wind energy systems: Wiley-IEEE Press, 2011. B. J. Jonkman, TurbSim user's guide: Version 1.50: National Renewable Energy Laboratory Colorado, 2009. G. Ramtharan, N. Jenkins, O. Anaya-Lara, and E. Bossanyi, "Influence of rotor structural dynamics representations on the electrical transient performance of FSIG and DFIG wind turbines," Wind Energy, vol. 10, pp. 293-301, 2007. H. Li and Z. Chen, "Transient stability analysis of wind turbines with induction generators considering blades and shaft flexibility," in Industrial Electronics Society, 2007. IECON 2007. 33rd Annual Conference of the IEEE, 2007, pp. 1604-1609. K. Z. stergaard, P. Brath, and J. Stoustrup, "Estimation of effective wind speed," in Journal of Physics: Conference Series, 2007, pp. 1-9.
[4]
[5] [6] [7] [8] [9]
Figure 7. Fault ride through with active damping controller
[10] [11] [12] [13]
[14]
[15]
Figure 8. PSD of LSS torque for different controllers
VI. CONCLUSIONS In this paper, a LQR controller with state estimate is proposed to reduce the torque oscillations and improve drive-train reliability. A damping torque is added to the generator torque control loop to suppress the vibration of drive-train during FRT. Simulation results show the effectiveness of the proposed control strategy. Because no extra sensors are added, the controller can be easily implemented to the commercial wind turbine. Future work should include impact of voltage sag on other wind turbine mechanical components like blade and tower and also active damping control of these loads. VII.
[1] [2] [3]
REFERENCES
M. Tsili and S. Papathanassiou, "A review of grid code technical requirements for wind farms (Selected papers from EWEC 2008)," IET renewable power generation, vol. 3, pp. 308-332, 2009. A. D. Hansen and G. Michalke, "Fault ride-through capability of DFIG wind turbines," Renewable energy, vol. 32, pp. 1594-1610, 2007. R. Lohde, S. Jensen, A. Knop, and F. W. Fuchs, "Analysis of three phase grid failure and doubly fed induction generator ride-through
Improvement of Thermal Performance of 36 kV DAF30 Type Bushing

Ilim ERDEN1, Aydogan OZDEMIR2, Suat ILHAN3
Electrical Engineering Department Istanbul Technical University 34469 Maslak, Istanbul, Turkey 1 ilim219@hotmail.com, 2ozdemiraydo@itu.edu.tr, 3ilhansu@itu.edu.tr
AbstractBushings are one of the most essential parts of the electrical distribution and transmission systems since they enable high currents to pass through from one medium to another and isolate the high voltages from the grounded parts. In addition to electrical and mechanical stresses, they are also subjected to thermal stresses. This study presents temperature distribution of 36 kV DAF-30 wall type bushings used in Turkish National Power Distribution System. Several experimental tests and simulations were performed to determine the critical parts as well as to improve the thermal performance of the bushing. In addition, skin effect of the bushing conductor was investigated to propose appropriate thermal solutions for the bushing. Keywords-Bushing, temperature distribution, skin effect
limited studies for resin impregnated paper transformer bushings [7-9]. Some authors account for dielectric dissipations in order to improve the thermal models [4, 5, 8]. However, in most of the thermal models, dielectric dissipations were ignored because of their limited effects on temperature rises [6, 9, 10]. In addition, several theoretical studies were verified by experimental tests [7, 8, 10]. On the other hand, skin effect caused by alternating current was taken into consideration in some heat distribution studies [10, 11]. Most of the previous studies on heat distribution problem were concentrated on transformer type bushings, and little research was conducted for wall type bushings. Basic motivation behind this study was the overheating problems experienced in wall type bushings The aim is to determine the temperature distribution and to determine the critical (thermally overstressed) regions for 36 kV wall type bushings; namely DAF-30 type bushing with 1000 A current rating., Both the simulation and experimental studies were performed and possible solutions were proposed to improve the thermal performance of the bushing. Skin effect was effectively taken into account in the simulations. II. MATHEMATICAL MODEL In this section, mathematical model is briefly illustrated. First, electrical and thermal models of high voltage bushings are constructed. Bushing conductor is modeled as an heat source (resistive losses) due to high current flows through it. Dielectric losses were ignored since they are much less than resistive losses. A. Electrical and Thermal Models The dominant heat source in a wall type bushing is the resistive losses of the conductor. Depending on the current level and on the properties of the conductor material, resistive losses can either increase or decrease. Dissipated energy is defined as: where, : I: R: Electrical energy which is converted to thermal energy (W.h) Current passing through the conductor (A) Resistance of the conductor () (1)
I.
INTRODUCTION
Bushings provide current flow through their conductors while isolating the electrical parts from ground. IEC 60137 is an international standard and specifies the characteristics and the tests for insulated bushings whose ratings are above 1000 V and operate at the power frequencies of 15 Hz up to and including 60 Hz [1]. It classifies the high voltage bushings with respect to their production materials, to application areas and to several other technical properties. Liquid-filled bushings, gasfilled bushings, ceramic, glass or analogous inorganic material bushings, plug-in type bushings and outdoor-indoor bushings are some well known high voltage bushing types [1]. DAF-30 family includes outdoor-indoor type high voltage bushings which are commonly used in Turkish National Power Distribution System. They are known as wall type bushings in industry. DAF-30 signifies 30 kV voltage level and it covers 400, 600, 1000 and 1250 A rated current levels. This family is expected to be extended by 1600, 2000, 2500 and 3150 A current levels in the future. In addition to electrical and mechanical stresses, thermal stress is another problem for high voltage bushings because of the heavy currents flowing through their conductors. Past laboratory experiences showed that especially 1000 A and 1250 A bushings had some thermal problems. Moreover, bushing failures caused by thermal stresses have frequently been observed in Turkish Transmission and Distribution System. Temperature distribution in oil impregnated paper bushings have been analyzed in the past [2-6]. However, there are
Turkish Ministry of Industry and Technology, and KALESERAMIK Inc. are the main sponsors of this study.
: Time interval (h). Figure 1 shows the 2D axe-symmetrical model of the bushing.
1 2
where, P: : J: V: Electrical power (W) Electrical conductivity in (S/m) Current density (A/m2) Volume of conductor (m3)
Symmetry axis
Figure 1. Simulation model of the bushing,:1: Porcelain dielectric, 2: Flange, 3: Bushing conductor
Thermal energy flows from the heat source (conductor) to porcelain and to the surrounding medium. Equation 2 shows the heat conduction equation in cylindrical coordinates [12].
It is more convenient for electrical engineers to make use of the analogy between the electrical circuits and thermal networks [13]. In thermal networks heat flows similarly as electric current flows in electrical circuits. Electric current in an electrical circuit can be defined using well known Ohms Law as:
(6)
where, (2) where, k: T: W: : c: t: Thermal conductivity (W/(m.oK)) Temperature (oK) Heat source density per volume (W/m3) Density of the material (kg/m3) Heat capacity of material (J/kg.oK) Time (h) I: Electric current (A) : Electrical Potential (V) R: Electrical resistance () Heat flow can be defined in the same manner as the current flow as:
(7)
where, : Heat flow (W/m) : Thermal resistance (oK.m/W) Comparison and the analogy of the elements in thermal networks and electrical circuits can be found in [10]. Alternating current mainly passes through the skin depth, known as . Skin dept of a conductor is determined by using Equation 8. (8) where,
Thermal convection equation known as Newtons Cooling Law can be given as: (3) where, q: T: Heat flux (W/m2) Temperature (oK)
: Temperature difference (oK)
Electrical conductivity is an important parameter in power loss calculations and it varies with the temperature. So electrical conductivity can be given as: (4) where, Electrical conductivity in (S/m) : Resistivity at reference temperature (.m) : Temperature coefficient of resistivity (1/oK) T: Temperature in Kelvin (oK) Using electrical conductivity, joule dissipation at the conductor can be defined as temperature dependent value as given below: (5) :
: : f: : :
Skin depth (m) Electrical conductivity in (S/m) Power frequency (Hz.) Permeability of free space (410 H/m ) Relative permeability of conductor III. EXPERIMENTAL STUDIES AND SIMULATIONS
DAF-30, 1000 A wall type outdoor-indoor bushing is used in experimental and simulation studies. Its conductor is made of 32 mm cylindrical aluminum. There is an 8 mm axial air gap between the central conductor and the porcelain dielectric. DAF-30 bushing and its basic dimensions are given in Figure 2.
t this bushing (flange point) did 70 oC. This result showed that not satisfy the temperature rise e conditions although the hottest point temperature is less than th he upper limit of 90 oC.
Figure 2. DAF-30 Type high voltage bushing an nd its dimensions.
A. Experimental Studies Experimental test results of the bushings s are considered to satisfy the conditions stated in IEC 60137 standard [1]. The m the hottest main task of the experimental test is to measure points temperatures as well as to deter rmine the critical temperature rises over the ambient temper rature. The hottest point should not exceed 90 oC and the maximum steady state rature should not temperature rise over the ambient temper exceed 60 oC. [1]. ction were used to Aluminum bus bars of 4x2 cm2 cross sec connect the bushing and current transforme er. 1000 A, 50 Hz power frequency test current was passed thro ough the aluminum conductor of the bushing until reaching the st teady state thermal conditions. The experimental test setup is giv ven in Figure 3.
Figure 4. Temperature measurement points on bushing.
In addition to exceeding al llowable temperature rise limits, the end point temperature rises s of the conductor were near the temperature rise limits. In sum mmary, the flange was the worst region exceeding the thermal limits and conductor terminals were near the limits from the highest temperature rises point of o the project was focused on the view. Therefore, the first task of solution to alleviate the thermal l problem at flange.
90
Temperature [Degree]
75
60 Point_A Point_B Point_C Point_D Point_E 15 0 1 2 3 4 5 6
45
30
Time, hours
Figure 5. Temperature me easurements of critical points.
Figure 3. Experimental test setup ph hoto.
Figure 4 illustrates the critical locations where continuous temperature measurements were performed. A and B represents the measurement points on the outdoor condu uctor side, D and E represent the measurement points on the indo oor conductor side, and finally C represents the measuremen nt points on the grounded flange surface. In addition, several temperature measurements were taken along the conn nection busses to ensure that the test setup satisfies the condi itions described in the standard [1]. Special thermocouples s were used in temperature rise test. t it takes about 6 The experimental study has shown that hours to reach the steady-state condition. Temperatures T were measured, and stored continuously at 1 min nute time intervals. Figure 5 shows the variation of the tempera atures measured at the points shown in Figure 4. After reachin ng the steady state condition, ambient temperature was measur red to be 16.4 oC, and the maximum temperature which occu urs on the flange surface (point C) was recorded as 84 oC. In addition, a maximum temperatures at the two terminals of the bus shing were around
This abnormal temperature condition at the flange was rrents flowing along the flange indeed the result of eddy cur surface. We have verified that case, by removing the porcelain ange temperature in free air insulator and testing the fla conditions. We have attempted several methods to decrease the flange ctromagnetic theory. The first temperature using basic elec possible solution seemed to use e a non-magnetic material for the flange. This case was tested for r aluminum flange (ring) and the results have shown that the flange temperature decreased ratures. The usage of magnetic almost to the ambient temper material with lower magneti ic permeability will obviously decrease the temperature of the e flange. However, we were not sure if this decrease would be sufficient enough to satisfy the f into two parts and insert a conditions. We also sliced the flange dielectric material between th he two parts. In this case, the flange temperature decreased to o 58 oC. We are still studying on flange geometry and the dimensions to provide some additional perature. decreases in the maximum temp Experimental studies were also conducted for the bushings without flange metals to see its effect on the temperature rise of
the other parts. The studies showed that excessive heating of the flange metal was limited effect on the temperature rise of the indoor/outdoor conductor regions. B. Simulation Studies Finite element method was used to solve the thermal balance equations [14, 15]. 2D axe-symmetrical model as shown in Figure 1 was used for the simulations. At simulation studies, only steady state conditions were considered for the sake of less computational time. Simulation-based thermal distribution studies required some material parameters as an input data. These were thermal conductivity, material density, heat capacity of air porcelain and aluminum. Some of those parameters are temperature dependent and this dependence may not always be linear [12]. Those parameters are given in Table 1. In addition to the material properties, boundary conditions are also required for the solution. Initial temperatures of the bushing and the ambient temperature were taken as 16.4 oC, as in the experimental studies, and the ambient temperature was assumed to be constant throughout the test period.
TABLE I. MATERIAL PARAMETERSS USED IN THE SIMULATIONS k (W/oK.m) Air Porcelain Aluminium Copper k(T) 1,5 k(T) k(T) (S/m) 10 10
-15 -12
90 75 Temperature [Degree] 60 45 30 15 0 5 5.5 6 6.5 h [W/(m.K)] 7
Point_A Point_B Point_C Point_D Point_E
Figure 6. Critical point temperatures versus different h values.
r
1 4,2 1 1
(kg/m3) (T) 2200 (T) (T)
(J/kg.oK)
Cp(T) 920 Cp(T) Cp(T)
Figure 7 shows the temperature distribution of the bushing for steady-state conditions. As explained previously, origin of the excessive heating at the flange region comes from the induction heating. Due to the modeling limitations, temperature of the flange surface was assigned to be 84 oC as a boundary condition. Except from the flange region, maximum temperatures occur at the bushing conductor because of the resistive losses. However, there is not a significant thermal stress at the porcelain dielectric due to the good thermal properties of the porcelain.
o
(T) (T)
Induction heating of the flange region was not considered during the simulation studies. Instead, flange surface was assumed to be at 84 oC constant temperature condition. Convective cooling is a very complex phenomenon that requires lots of calculations to find temperatures of surfaces which interact with liquid or gas medium. Heat transfer coefficient known as h, can vary between 2-25 (W/K.m2) for free convection of gases, and 10-1000 ((W/K.m2)) for free convection of liquids [12]. Smaller h value means less convective cooling and higher surface temperature for bushing in our case. Figure 6 shows the effect of h coefficients on the temperatures at the critical regions. It is clear that the critical point temperatures besides the flange temperature decrease with increasing h values. In this research, h parameter is assumed to be 5 for the simulations because of considering the worst operating conditions. Actual temperatures and temperature rises are therefore expected to be less than the ones determined by simulations. Using the data in Table 1 and h=5 for heat transfer coefficient, simulation based temperature distribution results for the bushing were given in Figure 7.
Figure 7. Temperature distribution of bushing for the given parameters.
It is well known that the skin effect is an important factor affecting the thermal performance of current carrying conductors, especially under high frequency conditions. Figure 8 shows the current density distribution on a conductor cross section which carries 1000 A, 50 Hz alternating current.
A/mm2
x106 A/mm2
Figure 10. Current density of 1mm m copper coated aluminum conductor. Figure 8. Skin effect of 50 Hz. alternating curre ent on conductor.
Temperature distribution inside the conductor c is not uniform because of the skin effect. The part ts of the conductor which are close to the outer surface have high her current density than the other parts. C. Simulation Based Improvement Studies DAF-30 type bushing's conductor operates near the sired phenomenon maximum temperature limits. It is an undes that causes power losses and decreases the material life R losses are effectively because of high temperatures. Resistive very important factor to alleviate the ther rmal problems on current carrying conductors. Skin effect is another important factor affecting the thermal performance. We focused on the conductor material to decrease the mperatures. It can resistive losses by reducing the hot spot tem be seen from Figure 8 that the current density increases from rface. This means the center of the conductor to the outer sur that outer parts of the conductor causes higher h temperature rises than the other parts. Coating the outer surface with ermal conductivity copper, which has higher electrical and the than aluminum, would be a possible solutio on to increase the thermal performance. Figure 9 shows the cross c section of a conductor which consists of copper coated c aluminum conductor.
Figure 11. Current densities for severa al copper coated aluminum conductors.
It can be seen that a huge e amount of alternating current passes through the copper coat ting. This means that there will be a lower temperature and lo ower resistive losses on copper coated aluminum conductors. Full F copper conductor is another alternative; however it is not t an appropriate solution from economical point of view. fi the temperatures of points A In addition, we wanted to find and E and of the center point t of the conductor after coating conductor with different thickne esses of copper. Resistive losses are calcula ated for both the aluminum and the copper parts at 16.4 oC. Calculation C results are given in Table 2. Increasing copper laye er thicknesses decrease the total power losses as expected. At next step, these resistive losses were entered as heat sources for both aluminum and copper ratures were calculated for each regions, and maximum temper case. Ambient temperature was s taken 16.4 oC and heat transfer coefficient was taken as h=5 in the simulations. The aim of this perature decreases caused by the study was to find relative temp copper coating. Calculated temperature values are given in Table 3. e Table 3 that copper-coated It can be seen from the aluminum conductor shows better thermal performance than uncoated aluminum conductor. It is evident that increasing the electrical conductivity results in decreasing the skin depth; s through the more conductive therefore, more current passes region of the conductor. As sh hown in Table 3, increasing the copper thickness decreases the maximum m temperatures.
rCu d rAl
Figure 9. New conductor geometry to improve the th hermal performance.
First of all, it would be useful to investiga ate the distribution of the current density for the new design. Figure 10 shows the r coated by 1 mm current density for the aluminum conductor thickness copper, for 1000 A, 50 Hz current t. Figure 11 shows the variation of the current densities alon ng the radius for different designs.
TABLE II. RESISTIVE LOSSES FOR DIFFERENT CASES
= (3,774x107)[S/m] ; AC Resistive Losses at 20 , =(5,99x107)[S/m], I = 1000 Arms Cases Full Aluminum Full Copper d=1mm Copper d=2mm Copper d=3mm Copper d=4mm Copper d=5mm Copper d=6mm Copper d=7mm Copper d=8mm Copper TABLE III. Cases Full Aluminum Full Copper d=1mm Copper d=2mm Copper d=3mm Copper d=4mm Copper d=5mm Copper d=6mm Copper d=7mm Copper d=8mm Copper Aluminum Losses [W] 35,30 0 25,95 19,38 14,64 11,14 8,50 6,46 4,86 3,60 Copper Losses [W] 0 24,18 7,16 11,80 14,90 17,07 18,65 19,85 20,79 21,57 Total Losses [W] 35,30 24,18 33,11 31,17 29,55 28,21 27,14 26,30 25,65 25,16
Simulation studies were repeated for several conductor cases, such as full aluminum, full copper and copper-coated aluminum. To alleviate the heating problem of the conductor, copper-coated aluminum conductor design was discussed and focused on changing conductor design of the bushing. Simulation studies showed that using copper-coated aluminum will reduce joule dissipations to some extend depending on the copper thicknesses. ACKNOWLEDGMENT Authors wish to acknowledge Turkish Ministry of Industry and Technology, and KALESERAMIK Inc. for their financial support and encouragement. REFERENCES
[1] [2] IEC 60137 Insulated bushings for alternating voltages above 1000 V, 2003-08. D. O. Craghead, J. K. Easley, Thermal Test Performance of a Modern Apparatus Bushing, IEEE Trtansactions on Power Apparatus and Systems, vol. PAS-97, no.6, pp. 2291-2299, Nov/Dec 1978. P. P. Hebert, R. C. Steed, A High Voltage Bushing Thermal Performance Computer Model, IEEE Transactions on Power Apparatus and Systems, vol. PAS-97, no.6, pp. 2219-2224, Nov/Dec 1978. H. K . Youssef, R. Hackam, Computerized Thermal Analysis of High Voltage Bushings, IEEE Transactions on Power Delivery, vol. PWRD2, no. 4, pp. 1145-1151, Oct 1987. H. K. Youssef, M. Abdel Aziz, R. Hackam, Steady State Temperature Distribution of High Voltage Bushings-Analysis and Measurements, IEEE Transactions on Power Systems, vol. 3, no.1, pp. 278-285, Feb. 1988. S. Zhang, Evaluation of Thermal Transient and Overload Capability of High-Voltage Bushings With ATP, IEEE Transactions on Power Delivery, vol. 24, no. 3, pp. 1295-1301, July 2009. D. Zeng, A Study of Current-Carrying Capacity of Drawlead Cable in Condenser Bushing, IEEE Transactions on Power Delivery, vol.15, no. 3, pp. 925-930, July 2000. N. S. Jyothi, T. S. Ramu, M. Mandlik, Temperature Distribution in Resin Impregnated Paper Insulaton for Transformer Bushings, IEEE Transactions on Dielectrics and Electrical Insulation, vol.17, no. 3, pp. 931-938, June 2010. N. S. Jyothi, T. S. Ramu, A Model for the Temperature Distribution in Resin Impregnated Paper Bushings, Power and Energy Engineering Conference (APPEEEC) , Asia Pasific , pp. 1-4, 27-29 March 2012. C. Ko, S. Zhang, Improved Transient Hot-Spot Temperature Calculation Method of High Voltage Bushings, Transmission and Distribution Conference and Exposition:Latin America, 2008 IEEE/PES, pp. 1-5, 13-15 Aug. 2008. D. Zeng, An Improved Method for Estimating Temperature Rise of a Bushing Loaded Above Nameplate Rating, IEEE Transactions on Power Delivery, vol.14, no. 3, pp. 986-995, July 1999. A. engel, A. J. Ghajar, Heat and Mass Transfer Fundamentals and Applications, 4th ed., Mc Graw Hill Press, 2011. W. J. McNutt, J. K. Easley, Mathematical Modelling - A Basis for Bushing Loading Guides, IEEE Transactions on Power Apparatus and Systems, vol. PAS-97, no.6, pp. 2393-2404, Nov/Dec 1978. I. Kocar, A. Ertas, Thermal Analysis for Determination of Current Carrying Capacity of PE and XLPE Insulated Power Cables Using Finite Element Method, IEEE MELECON 2004, Dubrovnik, Croatia, pp. 905-908, 12-15 May 2004. C. C. Hwang, Y. H. Jiang, Extensions to the Finite Element Method for Thermal Analysis of Underground Cable Systems, Elsevier Electric Power Systems Research, vol. 64, pp. 159-164, 2003.
[3] TEMPERATURE CALCULATONS FOR EACH CASE E [ 59,2 48,5 57,1 55,2 53,6 53,1 51,3 50,5 49,9 49,4 Center point of the conductor [ ] 65,6 52,1 62,9 60,6 58,6 57,9 55,8 54,6 54,0 53,4 A [ ] 57,5 47,5 55,6 53,8 52,3 51,8 50,1 49,3 48,7 48,3 Reduction % 0 0 4,12 7,62 10,67 11,74 14,94 16,77 17,68 18,60 [9] [8] [6] [4]
[5]
[7]
IV.
CONCLUSIONS
[10]
Simulation and experimental based temperature distributions at the DAF-30 outdoor-indoor wall type were presented in this study. 1000 A, 50 Hz test current was passed through the bushing until the steady state conditions. Simulation studies were conducted by using 2D axesymmetrical model, including porcelain, bushing conductor and flange. Following summarizes the research results: The test results have shown that maximum temperature occurs on the flange surface due to the electromagnetic heating caused by the eddy currents. Maximum temperatures on the conductors end points were measured to be close to the limits described in the standard. However, temperature on the flange surface exceeds the limit temperature values.
[11]
[12] [13]
[14]
[15]
Hardware-in-Loop Real-Time Simulation of a Battery Storage System in a Wind Generation Scheme

Damon Bazargan
ECE, University of Manitoba, Winnipeg, Canada umbazarg@cc.umanitoba.ca
Abstract This paper develops a hardware-in-loop simulation test bench to simulate an actual battery embedded in a type 1 wind generation system. This provides an opportunity for studying batteries under realistic charge/discharge profiles as they undergo such stress cycles in a wind application. A Shepherd battery model is developed for two battery chemistries and is compared to real batteries in the loop of the real-time simulation. The outcome of this study is to show the difference in the final value of the state of charge of the batteries in a purely mathematical simulation using a battery model and in the real-time battery-in-loop simulation. Keywords: Wind energy, hardware-in-loop simulation, battery energy storage, induction generator, Shepherd battery model.
Shaahin Filizadeh
ECE, University of Manitoba Winnipeg, Canada shaahin_filizadeh@umanitoba.ca environment interfaced with a real piece of hardware. Hence, the external hardware becomes part of the closed loop system. Since battery models show deviations from the actual battery behaviour this application of HIL is particularly appealing for the simulation of systems with embedded batteries. Batteries are electrochemical devices and show nonlinear characteristics. This is particularity true for an aged battery, and is hard to predict. Corrosion, Shedding, irreversible sulphation, active mass degradation and electrode stratification are recognized as the roots of the aging phenomena in a lead acid battery [10]. This is an important challenge in the use of batteries in power system applications; this is particularly true in case of keeping batteries in the cycle for many years or using aged batteries. It is predicted that a large number of replaced vehicular lithium-ion batteries will be available in the near future for stationary storage applications. Although these batteries are no longer efficient for the use in a vehicle, they can still be used in a wind farm to store energy. This paper describes a hardware-in-loop (HIL) real-time simulation setup in the context of a wind generation system. The objective of this paper is to explore possible differences between a pure simulation and an HIL simulation of a wind generation system with emphasis on the role of the battery as the storage device. The simulation is performed once with the embedded battery model in software and once with an HIL configuration and the results will be compared. Additionally since the HIL simulation emulates the application in a wind farm, such a setup can also be used to expose an actual battery to different operating conditions, stress levels, and usage profiles to assess its performance under conditions that may be prohibitively difficult or costly to replicate in practice. The setup is particularly useful for simulation-based sizing [11] to prevent over-sizing of the battery energy storage system. II. DESCRIPTION OF THE HIL SYSTEM Hardware-in-loop (HIL) simulation is normally used in the second stage of the development of a complex system, following software simulations; its purpose is to obtain more accurate test results by incorporating actual hardware, rather than mathematical models, in the simulation. Compared to a complete physical prototype of a system or performing tests on the actual system, HIL yields benefits such as cost effectiveness, rapid
I.
INTRODUCTION
Wind power has seen growing interest as a source of clean energy in the last two decades [1]. Wind has a major proportion of the installed capacity of the renewable energy sources. Since wind power is a renewable source of energy and prevents production of greenhouse gases, it reduces environmental concerns. Lack of reliability is a major drawback of most renewable energy sources such as wind or solar [2]. Since there is limited control on such source, it is demanding to predict the produced power in such plants. For example the changes in the wind speed cause fluctuation of the produced power in a wind farm. Different methods are suggested to overcome the consequences of the varying nature of the wind speed in a wind farm [3]. Battery storage systems are suggested to harvest the maximum possible energy from wind and to alleviate the effect of the wind speed variations [4-8]. Batteries store the extra produced power and use it to provide a smooth output power profile. Due to the complex structure of wind farm equipment, in the early stages of the development, mathematical simulation is inevitable. The major aim of the simulation is to tune the parameters and to prevent over-sizing of the equipment. The simulation environment provides an approximation of the actual system performance and its accuracy depends on the level of accuracy and validity of the models used. Real-time hardware-inloop (HIL) is proposed to decrease the level of independence of a purely computer-based simulator on mathematical models [9]. HIL simulation is composed of a real-time simulation
The financial and technical support from the Manitoba HVDC Research Center and from MITACS Accelerate program is greatly acknowledged.
prototyping, repeatability and safety, to name a few [12]. The HIL simulation can consequently be viewed as a bridge between software simulation and physical system prototyping. Real-time simulation has been used in the past for the study of wind energy systems with storage models [13]. In this study, however, a power HIL (PHIL) configuration [14] is developed, in which electric power is exchanged between the external hardware, i.e. the battery, and simulation environment via a gridconnected interfacing actuator (amplifier). In particular the external hardware is a battery, connected to a simulation model of a wind energy system. Fig. 1 shows the schematic diagram of a PHIL configuration. The actuator is used to mimic the effect of the simulated subsystem on the hardware under test. In this study a 1kW AE7224 Techron amplifier is used as the actuator. The amplifier is used to sink and source power from/to the battery and exchange it with the ac network it is fed from. Measurements taken on the external hardware are then fed back to the real-time simulator. The amplifier is a linear device (at least over an adequately large range of frequencies, depending on the application), whose role is to draw a specified amount of current from the battery. Since most amplifiers are controlled voltage sources, a closed loop current controller must be deployed to craft the desired current by manipulating the voltage of the amplifier (as shown in Fig. 4). The rating of amplifier must be more than or equal to the power rating of the external hardware (a battery in this case). This may in fact require excessively large amplifiers if the external battery has large voltage, current or power ratings. To address this limitation, reduced-scale HIL is proposed [15]. It reduces hazard of electric shock in high power HIL simulations and is easier to implement. In a reduced scale PHIL, a sufficiently scaled down hardware is interfaced with the simulator. The effect of the hardware subsystem under test will be amplified by proper power amplification (PA) factor. In this context, the tested battery represents a larger battery storage system. The numerical value of the battery voltage is multiplied by the number of batteries (m) in series and the numerical value of its current is multiplied by the number of batteries (n) in parallel. This emulates a network of (mn) batteries. RTDS hardware is used as the real-time simulator in this study to simulate a wind generator with a battery energy storage system (BESS) for a specific wind profile. The RTDS performs electromagnetic transient power system simulations with a typical time step of 50 s. RSCAD is a particularly designed GUI interface for development of circuit schematics and monitoring of the run-time results of the RTDS [16]. Fig. 2 shows a screenshot of the RSCAD environment, in which different real-time parameters of the system such as state of charge (SoC) of the batteries, real and reactive power and the contribution of the BESS are monitored concurrently.
A. Description of the simulated wind generator configuration A type 1 wind generator is simulated in this study [1, 17]. It is composed of a squirrel cage induction generator (SCIG) directly connected to the grid via a transformer. This configuration is shown in Fig. 3. Capacitor banks are used for reactive power compensation as the SCIG draws reactive power from the grid. A soft-starter is usually used for smoother grid connection [18]. The soft-starter connects the generator to the grid smoothly.
Real Simulation
Actuator Hardware Under Test
PA Interface Simulated Software
Fig. 1: PHIL schematic diagram.
BESS enable switch
Battery Voltage Script Panel Battery Current DCLINK Voltage
Available Racks
SoC BESS Reactive Power Total Wind BESS Active Turbine Active Power Active Power Power Wind Turbine Reactive Power Total Reactive Power
Fig. 2: RSCAD screen shot.
Gear
SoftStarter SCIG Capacitor Bank VSC Battery
Grid
Fig. 3: Schematic diagram of a type 1 wind generator.
To smooth out the real power output of the wind generator (which varies as the wind speed changes), one may consider using a battery storage system. A static synchronous
compensator (STATCOM) with a hybrid BESS/capacitive dc side [19] is a candidate configuration for this purpose. The capacitor bank will then be used to provide the base reactive power in steady state. The role of the hybrid STATCOM is therefore adjustment of the real power exchange rather than the reactive power. Fig. 4 shows the configuration of the simulated system. Wind profile is given as the input to the wind turbine block, which acts on the generated power and the wind profile to adjust the pitch angle of the blades. Mechanical torque on the generator shaft is the output of this block. A 1.32 GW SCIG is used as the generator. It starts up in speed control mode as a motor and then switches to torque control mode and becomes a generator. The line voltage of the generator output is set to 690 V. A step up transformer increases the voltage up to 13.8 kV on the high voltage side. The BESS is connected via a 3-phase voltagesource converter to the point of common coupling (PCC). The voltage of the dc link of the VSC is maintained at 1.6 kV. The dc link is connected to a bidirectional dc-dc converter that maintains the dc voltage constant. The other side of the dc-dc is connected to the actual battery via the interfacing amplifier. V1 is the voltage across the real battery. V2 is a scaled-up version of V1 by the factor of m. The current I2 is the current passing the voltage source V2. I1 is the scaled-down version of I2 by a factor of n. This resembles connection of the dc-dc converter to a network of mn batteries. For this study m and n are selected as 32 and 380, respectively. The generated wind power is compared to the expected dispatched power. Any deviation from the targeted dispatched power should be compensated by BESS in this configuration. Hence, The VSC provides or absorbs the difference out/in from its dc-side capacitor to regulate the output power of the wind generator. Importing or exporting real power changes the voltage of the VSCs dc capacitor. The dc-dc converter is used to keep the voltage of the dc link constant by charging or discharging the battery accordingly. B. Description of the Control Strategy
To prevent immature aging of the battery cells in the BESS, it is suggested to avoid complete depletion [21]. In case of successive discharge rather than charge the SoC decreases gradually. The BESS does not provide energy in case its SoC is less than 20% in this study and only accepts energy to have its SoC increased gradually. Since the potential charging periods are intermittent and of variable and unknown length, it is suggested to charge the batteries with the highest allowable current to harvest maximum energy. This is essentially charging at constant current at the highest permitted rate [22].
Point of Common Coupling (PCC) IM
Transformer
Power System
Torque Wind Turbine Model Transformer
R L
Wind Speed profile
Voltage Source Converter
Capacitor
Capacitor
Bi-directional DC-DC Converter
V2 I1 V1 Battery
I2
Interfacing Hardware and Software Using an Amplifier
Lead Acid Battery
A more detailed description of the control strategy is presented in Fig. 5. A cascade control loop with proportionalintegral (PI) controllers is used to appropriately operate the dcdc converter. The reference voltage of the dc link (1.6 kV) and the reference current of the battery are the main and the intermediate set points of this control loop. A pulse-width modulation (PWM) technique is used for the switching purpose of the converter. The dq0 control algorithm [20] is selected to control the real and reactive power exchange of the VSC. The dq0 method decouples the control parameters. Therefore the exchanged real and reactive power are independently controlled.
Fig. 4: Schematic diagram of the simulated BESS embedded type 1 wind generation system.
III. DESCRIPTION OF THE BATTERY MODEL This section describes the shepherd battery model for the mathematical description of the batteries used. Since this application of the battery is stationary, cost prevention is a more important parameter for battery type selection than the energy density and specific energy [23]. Consequently lead acid batteries are strong candidates for this application. Nonetheless, since the application of used lithium-ion batteries is a future
possibility, this type of the battery chemistry is also considered in this study. An 18Ah PC625 odyssey lead-acid and a LPF26650P lithium-ion battery are used for this purpose. Specifications of these batteries are listed in Table I.
BESS Vdcref + Vdc Idref + Iqref IBattref PI + Conteroller IBatt Decoupled Vsq dq Vsd control PI Conteroller PWM Signal DC-DC Converter Vcap PWM Signal abc/dq PLL
Ia Ib Ic
Figs. 6 and 7 show the verification of the developed Shepherd model for the lithium ion and the lead acid battery. It depicts the discharge traces of the battery (from a fully charged state) at a constant rate of 0.2 C. The figure shows the output of the Shepherd model as well as the recorded terminal voltage of the actual battery. IV. RESULTS AND DISCUSSION As is described in [1] the produced wind power is highly dependent on the cube of the numerical value of the wind speed, turbine parameters, etc. For the purpose of this study a typical wind profile with an average speed of 12 m/s is produced using the Rayleigh probability density function (pdf). Rayleigh PDF described in (2) is commonly used for the purpose of estimation of the wind behavior in the analysis of the wind power where c is the average wind speed and v is the wind speed.
+ -
dq/abc Iq Id
VSC
Qwind Qref + - 1.5 Vm + 1.5 Vm Pref Pwind
L R
Fig. 5: Schematic control diagram of the simulated BESS embedded type 1 wind generation system. TABLE I. BATTERY SPECIFICATIONS Parameter Nominal capacity (Ah) Average voltage Weight Pulse discharge (A) Charge current (A) Max. charge voltage (V) Li-ion (specification of 1 cell out of 4 in series) 2.6 3.2 80.52 (g) 26 2.6 3.65 Lead-acid 16 13.5-13.8 6 kg 265 (CCA) No limit 14.4-15
2v ( ) F (v ) 2 e c c
13
v 2
(2)
12.5
12
Model
Voltage (V)
11.5
Actual SoC=35-65%
11
10.5
The shepherd model is composed of a variable dc voltage source and a constant resistance, which represents the internal resistance of the battery [24]. The variable dc voltage source is a function of the state-of-charge (SoC) of the battery as described in (1). An enhanced coulomb counting [25] method is used for tracking the battery SOC. Q (1) V V0 K A exp( B i(t )dt ) Q i(t )dt
10
5000
Time (s)
10000
15000
Fig. 6: Verification of the developed Shepherd model for the lead acid battery.
Where V0 is the nominal voltage and Q is the battery capacity in Ah. The values of constants parameters A, B and K in (1) are shown in Table II for both types of the batteries considered. These values are identified using the method described in [23].
TABLE II. MODEL PARAMETERS Parameter A B K R V0 Li-ion 0.55 83 0.12 0.3 13.8 Lead-acid 0.42 1.1 0.48 0.2 12.87
Fig. 8 shows that the utilization of the BESS (solid black profile) has decreased deviations of the produced power (grey profile is without compensation). Meanwhile during the simulation the lead-acid battery cell is continuously being charged and discharged (Fig.9). Traces for the Li-ion cell are not shown for brevity. Depending on the variations of the wind speed, produced wind power and the dispatched power set point, the SoC of batteries might increase or decrease during the simulation. The SoC is traced individually for the real battery and the model assuming the coulomb efficiency equals 100%. As shown in Figs. 10 and 11, in this study the battery is more discharged than charged, consequently the SoC decreases. Although the remaining SoC of the lead acid battery in the HIL simulation is 42.9 percent, this value is 44.5 percent for the battery model. Since the demanded power is the same for the battery model and the real battery, less SoC implies that more current is drawn out of the battery; or the average voltage of the
actual battery was less than the average of the battery model voltage during the simulation. For the lithium ion battery the remaining SoC are 45.1 and 45.9 for the real battery and the model, respectively. As is shown in Figs. 6 and 7, the voltage of the lithium ion battery around 50% SoC is more than the voltage of the lead acid battery around the same SoC range. Consequently the remaining SoC of the battery is more for the lithium ion battery.
14 13.5 Model 13 12.5
0.51 0.5 0.49 0.48 0.47 0.46 0.45 0.44 0.43 0.42 0 1000 2000 3000 Time (s) 4000 5000 6000
Actual
Voltage (V)
Fig. 10: Lead acid battery SoC trend during the simulation (actual cell).
0.51 0.5 0.49
SoC
12 SoC=35-65% 11.5 11 10.5 10 0
2000
4000
6000
8000
10000
12000
14000
16000
18000
Time (s)
Fig. 7: Verification of the Shepherd model for the lithium ion battery.
SoC
0.48 0.47 0.46 0.45 0
1000
2000
3000 Time (s)
4000
5000
6000
Fig. 11: Li-ion battery SoC trend during the simulation (actual cell).
V. CONCLUSIONS This paper developed a real-time battery-in-loop simulation setup in the context of a type 1 wind generation system. All required power electronic equipment and the related control blocks were developed in the RSCAD real-time simulation interface. A Shepherd battery model was developed for the lithium-ion and the lead acid chemistries. Afterwards a pure software simulation using battery models was compared with the battery-in-loop simulation of the same system driven by the same wind profile. The simulation results showed that the numerical SoC value of the battery models and the real battery were not exactly the same at the end of simulations. This is due to the fact that battery models were predicting a larger average voltage for the batteries in comparison to the real battery profile within the operating range. The inaccuracy of a purely mathematical simulation of renewable energy systems with embedded batteries in comparison to a battery in loop simulation of the exact system is shown in this paper. This is due to the fact that HIL simulation is the second best option to actual field experiment after prototyping. This is especially true in case of battery embedded
Fig. 8: Comparison of produced power with and without BESS compensation. Gray trace: without compensation; Black trace: with compensation
5 4 3 2
Current (A)
1 0 -1 -2 -3 0 1000 2000 3000 Time (s) 4000 5000 6000
Fig. 9: Lead acid battery current profile (actual cell).
systems, in which nonlinearity of the batteries should be considered. The HIL simulation can also be used to perform other studies on nonlinear batteries. It can mimic exact effect of the system on batteries as they are in a real field test.
VI. REFERENCES [1] T. Ackermann, Wind power in power systems, John Wiley & Sons, 2005. [2] A. Roy, S. B. Kedare, S. Bandyopadhyay, Optimum sizing of wind-battery systems incorporating resource uncertainty, Journal of Applied Energy , vol. 87, no. 8, pp. 27122727, Aug. 2010. [3] X. She, A. Huang, F. Wang, R. Burgos, Wind energy system with integrated functions of active power transfer, reactive power compensation, and voltage conversion, IEEE Trans. on Industrial Electronics, forthcoming. [4] V. Virulkar, M. Aware, M. Kolhe, Integrated battery controller for distributed energy system, Journal of Energy, vol. 36, no. 5, pp. 2392-2398, May. 2011. [5] S. Teleke, Optimal control of battery energy storage for wind farm dispatching, IEEE Trans. on Energy Conversion, vol. 25, no. 3, pp. 787 - 794 Sep 2010. [6] S. Teleke, M. E. Baran, A. Q. Huang, S. Bhattacharya, L. Anderson, Control strategies for battery energy storage for wind farm dispatching, IEEE Trans. on Energy Conversion, vol. 24, no.3, pp. 725-732, Sep. 2009. [7] T. K. A. Brekken, A. Yokochi, A. V. Jouanne, Z. Z Yen, H. M. Hapke, D. A. Halamay, Optimal energy storage sizing and control for wind power applications , IEEE Trans. on Sustainable Energy, vol. 2, no. 1, pp. 69-77, Jan. 2011. [8] X. Y. Wang, D. M. Vilathgamuwa, S. S. Choi, Determination of battery storage capacity in energy buffer for wind farm , IEEE Trans. on Energy Conversion, vol. 23, no. 3, pp. 868 - 878, Sept. 2008. [9] A. Bouscayrol and J. Irwin, Control and mechatronics, CRC Press, 2011. [10] E. I. Baring-Gould et al., Detailed evaluation of renewable energy power system operation: A summary of the European union hybrid power system component benchmarking project, National Renewable Energy Laboratory, May 2005. [11] E. Tara, S. Shahidinejad, S. Filizadeh, E. Bibeau, Battery storage sizing in a retrotted plug-in hybrid electric vehicle , IEEE Trans. on Vehicular Technology, vol. 59, no. 6, pp. 2786-2794, July. 2010. [12] L. Gauchia, A per-unit hardware-in-the-loop simulation of a fuel cell/battery hybrid energy system , IEEE Trans. on Industrial Electronics, vol. 57, no. 4, pp. 1186-1194, Apr. 2010.
[13] W. Li, G. Joos, J. Belanger, Real-time simulation of a wind turbine generator coupled with a battery supercapacitor energy storage system , IEEE Trans. on Industrial Electronics, vol. 57, no. 4, pp. 1137-1145, Apr. 2010. [14] P. Kotsampopoulos, V. Kleftakis, G. Messinis, N. Hatziargyriou, Design, development and operation of a PHIL environment for distributed energy resources, in Proc. of 38th Annual Conference of the IEEE Industrial Electronics Society - IECON2012, pp. 4765 - 4770, Oct. 2012. [15] A. Allegre, Reduced-scale-power hardware-in-the-loop simulation of an innovative subway, IEEE Trans. on Industrial Electronics, vol. 57, no. 4, pp. 4765 - 4770, Apr. 2010. [16] RTDS Technologies: http://www.rtds.com [17] T. Mostafa, Wind farms production: control and prediction, Ph.D. dissertation, University of Waterloo, 2007. [18] G. Bhuvaneswari, S. Charles, M.G. Nair, Power quality studies on a soft-start for an induction motor , in Proc. of IEEE Transmission and Distribution Conference and Exposition, Apr. 2008. [19] S. Mohod, M. Aware, A STATCOM-control scheme for grid connected wind energy system for power quality improvement, IEEE Journal of Systems, vol. 4, no.3, pp. 346-352, Sep. 2010. [20] S. Filizadeh, Electric machines and drives: principles, control, modeling, and simulation, CRC Press, 2013. [21] J. Dogger, B. Roossien, F. Nieuwenhout, Characterization of Li-ion batteries for intelligent management of distributed grid-connected storage, IEEE Trans. on Energy Conversion, vol. 26, no. 1, pp. 256-262, Mar. 2011. [22] Y. H. Liu, Y. F. Luo, Search for an optimal rapidcharging pattern for Li-Ion batteries using the Taguchi approach, IEEE Trans. on Industrial Electronics, vol. 57, no. 12, pp. 256-262, Dec. 2010. [23] D. Bazargan, Hardware-in-Loop simulation of battery storage systems for power system applications, M.Sc. dissertation, University of Manitoba, Nov. 2012. [24] O. Tremblay, A generic battery model for the dynamic simulation of hybrid electric vehicles, in Proc. IEEE Vehicle Power and Propulsion Conference, 2007. [25] K. Ng, C. Moo, Y. Chen, and Y. Hsieh, Enhanced coulomb counting method for estimating state-of-charge and state-of-health of lithium-ion batteries, Journal of Applied Energy, 2008.
Stability Analysis of Adaptive method for speed sensorless associated with the IFOC Control of Induction Motor
S. Berrissoul , A. Bouhenna , A. Mansouri , M. Chenafa
LAAS Laboratory Department of Electrical Engineering, ENSET of Oran, BP 1523 El MNaouer, Oran, Algeria
e-mail: berrisou@yahoo.com
Abstract: : This paper presents a sensorless speed observer method of an induction motor drive. The drive uses an adaptive ux observer for speed estimation. Many previously developed observer based speed sensor in very low and zero speeds of IM was not satisfactory. The stability study is based of estimation speed of induction motor drive in which the speed-adaptive law is given. By using robust control theory, the constant observer gain is obtained by solving two bilinear matrix inequalities.this method is compared to the adaptive full-order ux observer is used on determinant of matrix and a linearized model. The Simulation results of indirect eld orientation IM control without speed sensors are presented. keywords: Induction motor, Indirect vector control, adaptive observers, stability 1. INTRODUCTION The speed-sensorless control of induction motor drives have developed signicantly during the last number of years have been proposed (1; 2; 4; 5; 6). Most control systems use a speed sensor, which imposes a increasing the cost and complexity of the drive system and reduces the robustness of the overall system. It therefore becomes necessary to call for technical control and observation without sensor based on the use of observability of system properties. Nevertheless, there are problems related to unobservable and instability at low speed. The performances are considered satisfactory for large and medium speeds, by cons at low speeds, observability problems appear. performance degradation strongly limits the safe applicability of sensorless controllers. As a matter of fact, not only reduced speed tracking accuracy and load torque rejection capability aect the sensorless approach, but instability phenomena can also occur in the low-speed region and in the regeneration mode. However, a few studies have highlighted this problem of unobservability. In (7), the "observer controller" was tested on a "benchmark." The trajectories of this benchmark are chosen to evaluate the IM sensorless algorithm under conditions of observability and unobservability. Since the observability properties are lost at very low speed, it is well known that it is impossible to reconstruct the state of the IM that asymptotically converges to indistinguishable trajectories (3). However, under these trajectories, it is possible to design an observer whose performances is acceptable even if the asymptotic stability cannot be guaranteed. In this paper, we analyze and compare two dierent rotor speed observers for induction motors. We rst a method for rotor speed. The classical stator current and rotor ux induction motor model is considered. The rotor speed is regarded as timevarying of induction motor. By using Lyapunov stability theory, a new adaptive state observer and a rotor speed estimation scheme are proposed. The constant observer gain is obtained by solving two bilinear matrix inequalities (BMIs) using the LMI Toolbox in MATLAB (6; 11), and the stability of the speed adaptive observer is guaranteed by the gain in the whole speed range. The control system has been designed, the proposed the rotor speed estimation algorithms are veried by some simulation results. The other observer, proposed by Bouhenna et al.(5) where describes the design of an adaptation law that minimizes the instability region of an adaptive speed estimator. This paper is organized as follows. In the rst section, we present the dynamic model of the induction motor in the dq reference. In the second section the observer ZHIWU et al .(By using Lyapunov stability theory, solving two bilinear matrix inequalities (BMIs) using the LMI Toolbox in MATLAB). The third section presents the observer Bouhenna et al. (model, linearization of the system, stability criterion, gain matrix proposed). Simulation results are illustrated and discussed in section 4.
2. DYNAMIC MODEL OF INDUCTION MOTOR For the induction machine, if the stator currentis and rotor ux r are selected as the state variables, the state equations can be expressed as (1) in the stationnary frame: dis B1 is A11 A12 dt + us (1) d = A A 0 r r 21 22 dt o is = [ isd , isq ]is the stator current, r = [ rd , rq ]is the rotor ux and us = [ usd , usq ] is the stator voltage. Rs , Rr stator and rotor resistance Ls , Lr stator and rotor self inductance M mutual inductance = 1 (M )2 /(Ls Lr ) leakage coecient
Stability Analysis of Adaptive method for speed sensorless associated with the IFOC Control of Induction Motor Tr = Lr /Rr r rotor time constant motor angular velocity
can be guaranteed, the state error equation is asymptotically stable. Thus the state observer is asymptotically stable. The adaptive law is given by d T T A11 = {Rs /(Ls ) + (1 )/(Tr )} I s J = I s J r = x A P e + eT P A x (8) dt 2 A12 = M/(Ls Lr ) {(1/Tr )I r J } = (K/Tr )I Kr J the adaptive scheme for rotor speed estimation is given by A21 = (M/Tr )I t T T T T = K x A P e + e P A x + K x T AT r p i P e + e P A x A22 = (1/Tr )I (s r )J
0
B1 = 1/(Ls )I, C = [I 0] K = M/(Ls Lr ), I = 10 ,J = 01 0 1 1 0 5. OBSERVER GAIN COMPUTATION
(9)
3. SYNTHESIS OF THE OBSERVER BASED ON THE METHOD OF LMI (LINEAR MATRIX INEQUALITY) The model of induction motor is rewritten in the (d, q) as follows: x = [A + r A ]x + [B1 0]T Us is = Cx where A = and A1 = A r A The state observer, which estimates the state current and the rotor ux together, is given by the following equation: dx = [A1 + r A ] x + Bus + G( is i s ) (4) dt where denotes the estimated values and G is the observer gain matrix which is choosed so that the stability of can be guaranteed. 4. ADAPTIVE LAW FOR SPEED ESTIMATION In order to derive the adaptive scheme, Lyapunov theorem is utilized. From and , the estimation error of the stator and rotor ux is described by the following equation. de = [A1 + GC + r A ]e r A x (5) dt where e = x x , r = r r Now, let us dene the following Lyapunov function (? ) ( )2 (6) V (e, r r ) = eT P e + where P is a symmetric positive denite matrix with 4 x 4 dimensions. The time derivative of V (e, r r ) along error equation becomes: dV = eT (A + GC )T P + P (A1 + GC ) + r (AT P + P A ) e dt 2 dr T + r r x T AT P e + e P A x dt Let dr 2 T = r [ xT AT ] d r d P e + e P A x dt If we select observer gain matrix G so that the validity of the inequality (A1 + GC )T P + P (A1 + GC ) + r (AT P + P A ) 0 (7) M 0 A1 , A1 = J, A2 = J 0 A2 Ls Lr (2) (3)
By the Lyapunov stability theory, the state observer is asymptotically stable if there exist symmetric positive denite matrix P and matrix G such that hold. Suppose is a upper bound of permitted rotor speed. can be determined by experiment or calculated from other parameters of induction motor. For ensuring system with a certain reliability and robustness, is selected larger than rated speed of IM in general. When the rotor speed changed in the interval [-, ], the robust stability of error equation can be guaranteed by the following two matrix inequalities: (A + GC )T P + P (A + GC ) + (AT P + P A ) 0(10) (A + GC )T P + P (A + GC ) (AT P + P A ) 0(11) Inequality and are bilinear matrix inequalities (BMIs) about matrix variables P and G. if xed matrix G, BMIs and will become linear matrix inequalities (LMIs) about matrix variable P if xed P, BMIs and will become linear matrix inequalities (LMIs) about matrix variable G. So, the feasible solution of BMIs and can be obtained by using the following alternate algorithm (6) and LMI toolbox in MATLAB (4; 10). Denote B1 (P, G) = (A + GC )T P + P (A + GC ) + (AT P + P A ) B2 (P, G) = (A + GC )T P + P (A + GC ) (AT P + P A ) B (P, G) = diag (B1 (P, G), B2 (P, G)) and "(A)" is notated the maximum eigenvalue of matrix A. Now, let us present the algorithm for calculating matrix P and G. Initialize: Given a arbitrary matrix P0 0, let i = 0 Repeat: let i = i + 1, solve optimization problem min (B (Pi , G))
G
to obtain Gi : Solve optimization problem min (B (P, Gi ))

P
to obtain positive denite matrix Pi . Until:Matrix inequality (B (Pi , Gi )) 0 is satised. Let P0 = I . By using the algorithm stated above section, we obtain a feasible solution of BMIs and : 1.8060 1.8663 1.8663 1.8060 G= 0.1792 0.0028 0.0028 0.1792
Stability Analysis of Adaptive method for speed sensorless associated with the IFOC Control of Induction Motor us induction motor r C is r + is +
Induction motor drive parameters Stator resistance Rotor resistance Stator inductance Rotor inductance Mutual inductance Pole pair Le nombre de paire de ple Viscous friction coecient Rs = 9.65 Rr = 4.3047 Ls = 0.4718H Lr = 0.4718H M = 0.4475H J = 0.0293kg.m2 p=2 f=0.0038 kg.m2 /s
+ ++ r
I/s A
Table 1. Parameters of the induction motor. adaptive scheme of speed Figures (6, (4)and the Figure (5) show simulation results of a speed, load torque and error of the speed respectively. We nd that the instability phenomenon appears at low speed generator mode from the moment 7sec. For this we propose another approach to eliminate this problem of instability.
120 100 80 60 speed(rad/s) 40 20 0 20 40 60 80 0 2 4 6 Time(sec) 8 10 measured reference estimated
G Fig. 1. Block diagram of speed adaptive observer 0.0010 0 0.0352 0 0.0010 0 0.0352 0 P = 0.0352 0 2.6181 0.0044 0 0.0352 0.0044 2.6181 5.1 Simulations Results We have performed simulations using MatlabSimulink on the benchmark of Fig. (2) and the motor parameters given in Table 1. We studied the performances of the adaptive observers in closed loops associated with FOC control of the induction motor.
Speed(rad/s)
100 50 0 0 2
Flux(Wb)
Fig. 3. Estimated speed, measured and reference
12 Ce* Cr 10
10
1 0
torque(N.m)
10
2 0 1 2 3 4 5 6 Time(sec) 7 8 9 10 11
Torque(N.m)
10 5 0
Fig. 4. Electromagnetic torque and load torque. 6. THE ADAPTIVE OBSERVER (A.F.O)OF FLUX As in the observer of Kubota et al., Bouhenna et al.5 . We now describe representation of the model that Bouhenna et al. (5) use when constructing their observer. Let d ( r , s ) x + Bus + G( is is ) (12) x =A dt (13) is = C x
10
Fig. 2. Reference trajectories. For the following PI gains of the speed adaptive scheme are selected Ki = 200, Kp = 3000.
Stability Analysis of Adaptive method for speed sensorless associated with the IFOC Control of Induction Motor where means the estimated values and G = [G1 G2 ]T is the observer gain matrix. 6.1 linearized model us Induction Motor is
x is +
The nonlinear and complicated dynamics of the speed adaptive observer can be studied via small + C B signal linearization. It is useful to proceed with a + + local analysis based on the principle of stability in the rst approximation (8). We will choose the A particular form G2 = g2 I22 where I22 is the r identity matrix and G1 = 022 . The complete Speed adaptive observer may be written as an equation adaptive (15). The motor model (12) may be written as law 1 1 d s ) is + K is = ( + j j r r + u Tr Ls s dt G M d 1 = + g e i + j 2 i sl s r dt r Tr Tr Fig. 6. Speed adaptive observer d d r = kp (eid rd rq eiq rd ) + ki (eid rq eiq ) dt dt (14) 1 1 d is = ( + js )is + K jr r + us dt T L r s M d 1 dt r = Tr is Tr + jsl r d r = 0 dt (15) We investigate the stability of the observer by linearizing the two systems (14) and (15) around an equilibrium operating point. Dening the new state vectors x = x0 + x with x0 = [is0 r0 r0 ]T x = [is r r ]T and r 0 ]T , x = + x = [ with x i x =x
0 0 s0
[ is r ]T . r The reference frame is synchronized with the rq0 = 0) and estimated rotor ux ( s = s
40 30 20 10 error of speed(rad/s) 0 10 20 30 40 50 60
r0
6 Time(sec)
10
(8). En Preserving only dynamic parts, the two systems (14), (15) become after linearization,: 1 d is = ( + js0 )is + K j0 r dt Tr 1 + u jK0 r ji0 s Ls s M d 1 i + jsl r j0 sl = dt r Tr s Tr d r = 0 dt (16) 1 d s0 ) is + K is = ( + j j 0 r dt Tr 1 0 u jK + s r j i0 Ls s d 1 + g2 eei = M + j sl0 is r r dt T T r r K d 0 + 0 r = Kp (K0 rd rq dt Tr s) s0 0 isd 0 isq K 0 eiq ) + eiq0 + Ki (eid0 rd rq 0 (17) Dening e = [e ei . er ], the system describing the estimation error is as follows: d 1 ei = ( + js0 )ei + K j0 e dt T r je jK e jKe 0 r jKer0 i0 r 0 r d M 1 dt e = Tr g2 ei Tr + jsl0 e r + j 0 e je0 sl + je0 r d K 0 e + 0 eq e = Kp K0 d dt r Tr 0 e 0 e K 0 e s0 id iq + eiq + eiq0 Ki eid0
rq rd 0
Fig. 5. The speed estimation error.
(18) Separating each state in d and q components, we 1 : obtain the corresponding state matrix A
s0 M g2 Tr s0 0 K Tr Kr0 1 Tr Kr0 K Tr sl0 0 0 K 0
speed (rad/s)
1 = A
1 M 0 g2 sl0 0 Tr Tr 0 0 KKp K Kp Kp s0 0 (Ki p )0 KKp r 0 0 Tr (19) 1 by Note the dependency of the dynamic matrix A the operating conditions. In order to obtain analytic conditions about the local stability using the necessary condition for stability based on the determinant of (19) (9), it is possible to obtain a relevant result as reported in the next section 6.2 Stability criterion: We use the following property:
5
120 100 80
60 40
estimated reference measured
20
0 20 0 2 4 6 Time(sec) 8 10
1 ) = det(A
1
(20)
1 . The determiwhere i are the eigenvalues of matrix A 1 is: nant of matrix A 2 Ki s0 ((s0 0 )Ls Lr Tr 1 ) = M det(A 0 + M 2 0 M 0 g 2 T r + s0 Ls Lr )/((Ls Lr )2 Tr ) 1 ) = 0 we obtain two solutions. si det(A (22) g 2 M + Rs Lr s0 = 0 (23) Rr Ls + Rs Lr In the plane torque / speed, this is equivalent to a region of instability dened by two lines (5). The principle of the instability reduction proposed here consists in the calculation of the feedback gain so that the unstable region will be limited to the inobservability. We can note that, whatever the structure of the matrix G, this line is always dened by s0 = 0. From equation (21), we can write the condition s0 = 0 in equation (24) s0 = 0 (s0 0 )Ls Lr Tr + M 2 0 M 0 g1 Tr + s0 = 0 (24) This provides the gain g2 of the gain matrix G that stabilizes the observer and therefore IRFOC at low speeds and in generator mode. Lr Rs g2 = (25) M 6.3 Simulations Results The simulation results of this approach are obtained by applying the same benchmark(2). We found that problems occurred with the rst approach (AFFO) have disappeared. This type of observer proposed by Bouhenna et al.(5) whose gain is calculated by an original method (the determinant criterion) can stabilize the system at low speed and generator mode. Figures (7), (8) and (9) are a perfect illustration of the performance of this observer. (21)
Fig. 7. Estimated speed, measured and reference.

10 Ce* Cr 8
6 Torque(N.m)
6 Time(sec)
10
Fig. 8. Electromagnetic torque and load torque.

20 15 10 error of speed(rad/s) 5 0 5 10 15 20
6 Time(sec)
10
Fig. 9. The speed estimation error.

7. CONCLUSION In this paper, we compare the performance of two approaches in estimating the speed of induction motor using adaptive observers. the rst method uses the adaptation law proposed by Kubota et al. (2), but the second method of calculating the gain matrix of the observer is dierent. The second approach has helped to stabilize the control system in generator mode at low speeds and thus preserving the high dynamic performance of the sensorless control.
REFERENCES C. Schauder Adaptive speed identication for vector control of induction motors without rotation transducers, in Con Rec. IEEWIAS Ann. Mtg., 1989, pp. 493-499. [2] H. Kubota, K. Matsuse, and T. Nakano DSP-based speed adaptive ux observer of induction motor, IEEE Transactions on Industry Applications, vol. 29, No.2, pp. 344-348, Mar./Apr. 1993. [3] S. Ibarra-Rojas, J. Moreno, and G. Espinosa, Global observability analysis of sensorless induction motors, Automatica, vol. 40, no. 6, pp. 10791085, Jun. 2004. [4] Boyd, S.,Cgaoui, L. E., Feron, E., Balakrishnan, V. Linear matrix inequalities in system and control theory, 1994, Philadelphia, SIAM. [5] Abderrahmane Bouhenna, Abdellah Mansouri, Mohammed Chenafa, Abdelkader Belaidi Feedback Gain Design Method for the Full-Order Flux Observer in Sensorless Control of Induction Motor Int. J. of Computers, Communications Control, ISSN 1841-9836, E-ISSN 1841-9844 Vol. III (2008),No.2, pp.135-148 [6] Goh, K. C., Turan, L. Safonov, M. G. et al. Bane matrix inequality properties and computational methods,Proceedings of the American Control Conference, Baltimore, Maryland, 1994, 850-855. [7] M. Ghanes, J. de Leon, and A. Glumineau, Novel controller for induction motor without mechanical sensor and experimental validation, in Proc.IEEE Conf. Decision Control, San Diego, CA, 2006, pp. 40084013. [8] M. Hinkkanen and J. Luomi Stabilization of regenerating-mode operation in sensorless induction motor drives by full-order fux observer design, IEEE Trans. Ind.Electron., Vol. 51, pp. 1318-1328, 2004. [9] E. Etien and N. Bensiali and C. Chaine and G. Champenois Adaptive Speed Observers For Sensorless Control Of Induction Motors: A New Criterion Of Stability, International review of electrical engineering, Vol. 1, pp. 36-43, 2006. [10] Gahinet, P., Nemirovski, A., Laub A., Chilali, M. The LMI control toolbox, 1994, The math Works, Inc. [11] B. Dobrucky, F. Chaloupka, P. Kucer, and V. Racek, A neural fuzzy control with stator resistance estimator for IM drives with DSC, in Proc. Europe. Power Electron. Applicat. Conf., 1995, pp. 1.789-1.794. [1]
Design and Analysis of Composite Filter for Power Quality improvement of Electric Arc Furnace
Deepak Chhaganrao Bhonsle
Electrical Engineering Department Maharaja Sayajirao University of Baroda Vadodara, INDIA dcbhonsle@gmail.com
Abstract Electrical Arc Furnace (EAF) is one of the responsible cause for deteriorating power quality in the distribution network by, introducing harmonics, propagating voltage flicker and causing unbalance in voltages and currents. This paper presents design, analysis and simulation of composite filter for improving power quality of distribution network. The distribution network consists of non-linear load as an Electric arc furnace. Passive filters have been preferred for power quality improvement due to low cost, simplicity, reliability and control-less operation. At first passive filter has been designed to compensate current and voltage harmonics. The drawback associated with the passive filter is tried to solve by connecting a series active power filter along with the passive filter thus forming a composite filter. Performance of composite filter and passive filter is compared and analyzed to improve the power quality and at point of common coupling. The composite filter is consisting of a shunt LC passive filter connected with a lower rated voltage source PWM converter based series active power filter. Simulation for a typical distribution network with a composite power filter has been carried out to validate the performance. The simulations have been carried out in MATLAB environment using SIMULINK and power system block set toolboxes. Keywords- Power quality, EAF, Hyperbolic, Exponential, Voltage flicker, Harmonics, Passive Filter, Series Active Filter, Composite filter
Dr. Ramesh B. Kelkar

Electrical Engineering Department Maharaja Sayajirao University of Baroda Vadodara, INDIA : e-mail address if desired conductance =Total Current Harmonic distortion THDV = Total Voltage Harmonic distortion Q =Reactive power to be generated by the filter at
THDI g min =Minimum
fundamental frequency. V =Voltage at which filter is to be installed f1 =Fundamental frequencies

f r , f L =Tuning q =Quality
frequency
factor II. INTRODUCTION
I.
NOMENCLATURE
i =Arc current v =Arc voltage g =Arc conductance E0 =Momentarily constant steady state arc voltage =Arc time constant 0 =Constant 1 = Constant = Constant P0 =Momentarily power loss I 0 =Transition current
The increasing popularity of EAF in metallurgical industries to melt scrap causes significant impacts on power system and electrical power quality. EAF is one of responsible source for deteriorating the power quality in the connected network. The EAF is inherently non-linear, time-variant load and it can cause power quality problems such as current-voltage harmonics, voltage flicker and voltage unbalance. Odd and even harmonic currents are generated by EAF operations. These harmonic currents, when circulated in the electric network can generate harmonic voltages which in turn can affect other users connected in the distribution network. Flicker is the sensation that is experienced by human eye when subjected to changes in the illumination intensity. The maximum sensitivity to change in illumination is in the frequency range of 5 to 15 Hz [4]. As EAF is a large source of flicker, causes voltage fluctuation in the connected electric network which is a major power quality issue. This in turn affects operation of other connected load also. Hence, modeling of EAF has attracted attention of power system engineers to solve these problems of power quality issues pertaining to EAF. An electric arc furnace (EAF) is chosen as an industrial non-linear load to demonstrate power quality problems. Literature Survey shows that majority efforts are put to model EAF either from voltage flicker point of view or from harmonics point of view. Few models are found with combination of harmonic as well as voltage flicker. The important issue in the modeling of the EAF is the simulation of arc. There are several methods used to describe the electric
arc [2-4, 7-8]. The balanced steady state equations are used in [7]. The time domain methods based on the differential equations are also presented [2]. Other methods such as frequency response, V-I characteristic are employed to analyze the behavior of the EAF [8]. Comparison of EAF modeling in time domain and frequency domain shows that the time domain models are more useful in studying the EAF [2, 4, and 8]. The above methods suffer from limitations such as knowledge of initial conditions for the differential equations, balanced situation of thee phase currents and use of complicated mathematical equation for the modeling of EAF. Next section presents mathematical modeling of two EAF time domain models. First model is known as Cassie-Mayer model [1, 5]. Second model is proposed model, which is a combination of Hyperbolic and Exponential EAF models. The main feature of the proposed model is modeling of the explained method with a good approximation without need of initial conditions of the EAF. Also, the proposed method can be used to describe different operating situations of the EAF and its effect of the connected electric network. III. EAF MODELING AS NON-LINEAR LOAD
The V-I characteristic of the proposed model is described by following equation with tabulated values in Table II:
V at v com (i ) = V at
i 1 e I o C + D+i
for higher arc current for lower arc current

(4)
TABLE II.
EAF MODEL PARAMETERS
EAF Model 2 Proposed

Parameter Description Parameter Value
Threshold voltage Arc power Arc current Current constant
Vat C D Io
180 V 19 kW 5 kA 25 kA
IV.
DYANAMIC BEHAVIOR
A. Model:1 Cassie-Mayer Mathematical model of Cassie-Mayer EAF model expressed as in [1, 5]:
i2 i2 i 2 v i dg g = g min + 1 exp 2 + exp dt I 0 P0 I 0 E0
= 0 + 1 exp ( i )
v= i g
(1) (2)
Dynamic EAF model is required for real time analysis of the effect of the arc. The dynamic arc characteristic is simulated by varying arc voltage. In general the variation is of random nature. However two types of variation are considered for the study-sinusoidal variation and random variation. In order to study the effect of voltage flicker on the power system of EAF, Vat is varied sinusoidally and randomly. In this regard Vat is modulated as follows: A. The sinusoidal variation The sinusoidal variation is assumed as [6],
(3)
v at (t ) = Vat 0 1 + m sin f t
)]
(5)
Typical values of and E0 , 0 , 1 , , P0 , I 0 , and g min are tabulated in Table 1[4-6].

TABLE I. EAF MODEL PARAMETERS
In (5) m is modulation index and f is a flicker frequency. B. The random variation The random variation is assumed as [6],
EAF Model 1 Cassie-Mayer

Mimimum arc conductance Tansition current Momentarily constant state arc voltage Momenttarily power loss Time Consatnat Time Consatnat Constant steady
g min I0 E0 P0
0.008 10 A 250 V 110 W 110 s 100 s 0.0005
In (6) N (t ) is a band limited white noise with zero mean and unit variance. The parameters used for sinusoidal variation and random variation are tabulated in Table III.
TABLE III. PARAMETERS FOR EAF DYNAMIC BEHAVIOR Sinusoidal variation
v at (t ) = Vat 0 [1 + m N (t )]
(6)
0 1
Initial threshold voltage Modulation index Flicker frequency
Vato
200 V 0.9 4 hz
B. Model:2 Proposed The proposed model is a combination of two existingExponential and Hyperbolic models of EAF.
f
Random variation
Vato
Initial threshold voltage
200 V
Sinusoidal variation
Modulation index Band limited white noise
m
N (t )
filtering. Passive filter (PF) is shown in Fig. 2 consists of passive parameters L f and C f .The passive filtering is the simplest conventional solution to mitigate the harmonic distortion.
0.9 4-14 Hz
V.
EAF MODELING WITH POWER SYSTEM
The EAF is connected to power system as per single line diagram shown in Fig. 1.
Figure 2. Power quality improvement using Composite Filter Figure 1. EAF connected in power system
The system parameters along with proposed EAF Model are tabulated in Table IV [7].
TABLE IV. PARAMETERS DISTRIBUTION NETWORK Sinusoidal variation
Source voltage Supply frequency Sorce impedance Apparent power Transformer T1 (HV/MV)
Vs
f Zs S
115 kV 50 Hz 12.19 /82.9 deg 100 MVA 110/13.8 kV 30 MVA 13.2/0.55 kV 42 MVA 0.38 m
As the regulatory requirements become more stringent, the passive filters might not be able to meet future revisions of a particular Standard. This may required a retrofit of new filters. A Series active voltage source power filter (SeAPF), is shown in Fig. 2. It typically consists of a three phase pulse width modulation (PWM) inverter. This combined topology is known as composite filter shown in Fig. 2, where the series active filter acts as controlled voltage source. Voltage v c is the voltage that the said inverter should generate to improve the power quality.
A. Passive Filter Modeling Equations (8) to (12) show design procedure of passive filters: 1) Single Tuned Filter
Q C= V 2 2f
VP / VS
S
Transformer T2 (MV/LV)
VP / VS
S Rc
(8) (9) (10) (11)
Lc 8.589 H Voltage flicker assessment is also one of the important aspects of power quality study. The assessment of voltage flicker involves the derivation of system RMS voltage variation and the frequency at which the variation occurs. The voltage flicker usually expressed as the RMS value of the modulating waveform divided by the RMS value of the fundamental value, as follows [17-19]: V +V % Voltage Flickr = 2 P 1P (7) V2 P V1P Equation (7) is useful for estimating voltage flicker. A variety of perceptible/limit curves are available in published literature which can be used as general guidelines to verify whether the amount of flicker is a problem [17].
Cable
1 L= 2 C (2f r ) R = q 2f r L 1 CL = L (2f )2
2) Double Tuned Filter

Q CH = V 2 2f
(12) (13) (14)
1 LH = C (2f )2 1 H 1 CL = L (2f )2 L L
VI.
POWER QUALITY IMPROVEMENT
Harmonic distortion in power distribution network can be suppressed using two approaches namely, passive and active
B. Series Active Filter Modeling Various control strategies for series active filter control are surveyed in the literature [12, 13 and 15]. In this paper control strategy based on the dual formulation of the electric power vectorial theory for non-sinusoidal and unbalanced voltage is proposed. In [16] the same theory is implemented for balance and resistive load. Here an attempt is made to apply the same theory for unbalanced and non-sinusoidal voltage conditions for randomly varying load as an electric arc furnace (EAF). Under ideal condition of balanced and sinusoidal voltages, load current at point of common coupling (PCC) will be proportional to the supply voltage:
v = Re i
By means of current vectors and the real-imaginary power, the voltage vector can be calculated as:
v = p
2 i
i +
q
2 i
(21)
For balanced and sinusoidal currents, average power supplied by the source will be [16]:
q P p v c = L L i L i 2 I 2 i2 i 1
(22)
(15)
Equation (22) gives reference voltage vector for the output of the series active filter under balance and sinusoidal conditions. For unbalanced and non-sinusoidal currents of EAF Re can be refined as:
Re =
+2 I1
PL
(23)
Where, Re =equivalent resistance, v =voltage vector, i =load current vector. Further set of supply quantities-voltage and current vectors of (15) can be defined as:
va v = vb = [va vc ia i = ib = [ia ic vb vc ]T
+2 Where, I1 =square RMS value of norm of the positive +2 sequence fundamental component. Substituting I1 instead of
I1
in (23) yields: (24)
(16)
P p q v c = L L i L i 2 2 2 + I i i 1 v P pL L c = +2 2 2 v i + i c I1
ib
ic ]T
(17)
abc
i i qL 2 2 i i + i i
Now for three phases three wire system, transformation coordinates obtained:
v = v 1 1 va 2 1 2 2 v 3 b 3 0 3 v 2 2 c 1 1 ia 2 1 2 2 i 3 b 3 0 3 i 2 2 c
to
(18)
PL p L i q L i 2 2 2 2 +2 v I1 + i + i i i c = qL pL vc PL + i i +2 2 2 2 2 I1 + + i i i i
( (
(25)
i = i
Equation (25) gives reference signals in coordinates for series active filter. Further simplifying: (19)
P pL L vc = +2 2 2 I1 + i i
Active instantaneous active and reactive power can be expressed in vectorial form by means of dot product [16]:
T p i = T v q i
) )
i qL i 2 2 i + i
(26) (27)
P pL L vc = +2 2 2 I1 i + i
i + qL i 2 2 i + i
(20)
Taking to abc transformation of (27), we get:

v ca vcb = v cc 1 2 1 3 2 1 2 0 3 vc 2 vc 3 2
In the plane ,
and i vector establishes two
(28)
coordinates axes. The voltage vector v can be decomposed in its orthogonal projection on the axis defined by the currents vectors as shown in Fig. 3 [16].
The load and series active filter will behave as a resistor with Re value for the reference signal supplied by (28). Reference signals are obtained by means of reference calculator shown in Fig. 4. The input signals are the voltage vector measured at the load side and the current vector measured at the source side. By means of calculation block, v and i vectors in coordinates can be calculated. An average value of real instantaneous power is obtained with low pass filter (LPF), this is the power required by the
Figure 3.
Voltage vector decomposition
load. 2nd order LPF with 45 Hz of cutt-off frequency is used. The average power so obtained is then divided by the average value of the norm of the positive sequence fundamental component of the current vector. The positive sequence component of the current vector is implemented by:
+ +2 +2 +2 I1 = i1 a + i1b + i1c
(29)
+ + + Where, i1 a , i1b and i1c are fundamental components of the
positive sequence currents of phases a , b and c respectively.
electric network. The distribution network consists of EAF-1Cassie-Mayer model and EAF-2-Proposed model. EAF is modeled as a non-linear time varying voltage controlled source using subsystem/MATLAB. The arc current is taken as the input parameter to this function and the output is nonlinear time varying voltage. The fundamental harmonic of current positive sequence component for each phase is determined as shown in Fig. 5. Each component of the source current vector is multiplied by sin t and cos t , where t is the fundamental frequency in rad/sec. The average values are obtained using low pass filters (LPFs). The outputs are again multiplied by sin t and cos t , and then multiplied by 2. This allows the fundamental harmonic of current positive sequence component to be obtained.
Figure 6. Complete MATLAB simulation file Figure 4. Reference signal calculator
The EAF along with the electric system shown in Fig. 1 has been simulated using Simulink/MATLAB platform shown in Fig. 2. EAF is modeled as a non-linear time varying voltage controlled source using subsystem/MATLAB. The arc current is taken as the input parameter to this function and the output is non-linear time varying voltage. VIII. SIMULATION RESULTS, ANALYSIS AND DISCUSSION Performance analysis of an EAF includes-analysis in refining cycle, analysis in melting cycle (sinusoidal flicker), analysis in melting cycle (random cycle) and unbalanced voltage analysis. The performance of EAF in various cycles is discussed in the following sub-sections:
A. Analysis in Refining Cycle The total harmonic distortion of voltage (THDV) observed at PCC is 14.43 % without filter in refining cycle of EAF. This value is violating IEEE 519-1992 limits. It should be below 5%. That is improved to 1.45 % after passive filter application as shown in Fig. 7(b). An application of series APF along with passive filter changes THDv to 1.85 %. The total harmonic distortion of current (THDI) observed at PCC is 9.09% without filter in refining cycle of EAF. Similarly, it is improved to 0.63 % and then changes to 4.03 % after passive filter and composite filter application respectively. Harmonics spectrum quantified detail is tabulated in Table V. It can be noted that 5th, 7th, 11th, 13th and 17th are the major harmonic components present in voltage at PCC. Harmonic components, H5 and H7, are violating IEEE 519-1992 limit of harmonic distortion. H11 is quite near to that limit where as H13 and H17 are moderate.
Figure 5. Fundamental positive sequence component calculator
The real instantaneous power is divided by i as shown in Fig. 4. The result is multiplied by the current vector i , which allows the first term in the compensation voltage in (26) to be determined. On the other hand, the imaginary instantaneous power is obtained and divided by i and then multiplied by the current vector i . This determines the second term in the compensation voltage (27). VII. SIMULATION The EAF along with the electric power system shown in Fig. 2 is simulated using Matlab-Simulink platform to verify the proposed control. Each power device has been modeled using the SimPowerSystem toolbox library. Fig. 6 shows complete Simulink/MATLAB file of EAF connected to the

B. Analysis in Melting Cycle (Sinusoidal Flicker) The total harmonic distortion of voltage (THDV) observed at PCC is 5.83 % without filter in melting cycle considering sinusoidal flicker condition of EAF, which is not objectionable. That is improved to 1.32 % then to 2.32 % after passive filter and composite filter application respectively. The total harmonic distortion of current (THDI) observed at PCC is 0.81% without filter in melting cycle considering sinusoidal flicker condition of EAF. THDI is below 5% as per the IEEE limits though after passive filter and composite filter application, it has changed to 0.32 % then to 1.61 % respectively. Table VI shows harmonic spectrum details of EAF current at PCC for various EAF cycles. It can be observed that H5 and H7 are the major harmonic components
present in EAF current at PCC. H5 is more than IEEE 519-92 limit of 5 % of harmonic distortion, where as H7 is moderate.
C. Analysis in Melting Cycle (Random Flicker) The total harmonic distortion of voltage (THDV) noted at PCC is 15.19 % without filter in melting cycle considering random flicker condition of EAF. That is improved to 5.38 % then to 1.85 % after passive filter and composite filter application respectively. The total harmonic distortion of current (THDI) observed at PCC is 3.35 % without filter in melting cycle considering random flicker condition of EAF. THDI is below 5% as per the IEEE limits though after passive filter and composite filter application, it has changed to 2.94 % and then to 4.03 % respectively.
Parameter
Without Filter
TABLE V VOLTAGE HARMONIC ANALYSIS Voltage harmonic analysis at point of common coupling (PCC) Refining Cycle Melting Cycle (Random Flicker)
With Passive Filter With Composite Active Filter Without Filter With Passive Filter With Composite Active Filter Without Filter
Melting Cycle (Sinusoidal Flicker)

With Passive Filter With Composite Active Filter
Vpeak (V) Vrms (V) THDv (%) H3 (%) H5 (%) H7 (%) H11 (%)
10130 7166 14.43 0 10.78 7.24 4.04
9433 6670 1.45 0 1.29 0.53 0.08
11400 8064 1.85 0 0.64 0.44 0.23
7306 5166 15.19 1.85 8.74 6.62 2.88
7777 5499 5.38 3.15 0.75 0.79 0.06
11400 8064 1.85 0.01 0.64 0.44 0.23
5312 3756 5.83 0.09 4.34 2.84 1.68
5599 3959 1.32 0.25 0.71 0.24 0.03
11180 7906 2.32 0.06 0.17 0.15 0.09
Parameter
Without Filter With Passive Filter
TABLE VI CURRENT HARMONIC ANALYSIS Current harmonic analysis Refining Cycle

With Composite Active Filter Without Filter
Melting Cycle (Random Flicker)

With Composite Active Filter Without Filter

With Passive Filter
Ipeak (A) Irms (A) THDi (%) H3 (%) H5 (%) H7 (%) H11 (%)
977.4 691.1 9.09 0 8.04 3.86 1.37
1459 1032 0.63 0 0.6 0.17 0.02
150.5 106.4 4.03 0 3.53 1.74 0.56
1932 1366 3.35 0 2.38 1.29 0.36
1848 1307 2.94 1.97 0.36 0.19 0.12
150.5 106.4 4.03 0.2 3.52 1.74 0.56
2374 1679 0.81 0.07 0.7 0.31 0.1
2284 1615 0.32 0.1 0.11 0.04 0.01
210.6 148.9 1.61 0.43 0.73 0.4 0.05
Parameter
Without Filter With Passive Filter
TABLE VII POWER ANALYSIS Power analysis at point of common coupling (PCC) Refining Cycle Melting Cycle (Random Flicker)
With Composite Active Filter Without Filter With Passive Filter With Composite Active Filter Without Filter

P (KW) Q (KVAr) PF
5249 2720 0.89
5863 3808 0.84
527 123.5 0.97
5249 2719 0.89
5863 3808 0.84
527.8 118.1 0.98
3732 2923 0.79
4574 3917 0.76
404.1 106.3 0.97
Table VII shows variation of active-reactive power and power factor after passive and composite filter application. It can be observed from Table VII that better power factor of the
order of 0.97-0.98 is obtained by composite filter application in various cycles of an EAF. Fig. 7 shows three phase voltages at PCC under sinusoidal flicker condition of EAF.
(a)
(b)
(c)
Figure 7. Three phases VPCC for sinusoidal flicker (a) Without filter (b) With Passive Filter (c) With Composite filter
(a) Figure 8.
(b)
(c)
Three phases VPCC for random flicker (a) Without filter (b) With Passive Filter (c) With Composite filter
(a)
(b)
(c)
Figure 9. Three phase voltage unbalance (a) Without filter (b) With Passive Filter (c) With Composite filter
%Voltage flicker for sinusoidal variation is estimated using equation (10) for passive filter and composite filter application as shown in Table VIII.
TABLEVIII VOLTAGE FLICKER ANALYSIS Sinusoidal voltage flicker
Parameters Without Filter With Passive Filter With Composite Filter
perceptible criteria. Fig. 8 shows three phase voltages at PCC under random flicker variation of EAF. It clearly indicates better performance of composite filter in removing random flicker than passive filter.
D. Analysis of Voltage Unbalanced Unbalance voltage condition is simulated by varying values of E 0 in case of Cassie-Mayer EAF model where
Voltage Measurement 5000 5100 12000 12000 % Flicker Calculation 41.18 40.35 % Voltage Flicker V1P (Volts) V2P (Volts)
11500 11800 1.29
as Vat 0 in case of the proposed EAF model as per Table XI. Simulated three phase voltage at PCC for voltage unbalance case is shown in Fig. 9. The reflected phase voltage values both-peak and RMS are tabulated in Table IX and Table X. From these values, average voltage deviation and per phase voltage deviation is calculated for peak and RMS values. It can be observed from Table IX and Table X that , In Case 1 voltage unbalance has improved from 1.81 % to 0.12 % where as in Case 2 it has improved to 0.27 % from 4.69 % on an average.
Tabulated results in Table VIII and Fig. 7 clearly show that the passive filter fails to clear sinusoidal voltage flicker where as composite filter reduces voltage flicker from 41.18 % to 1.29 %. For 4 Hz of frequency pulsation applied and for 1.29 of %voltage pulsation (% voltage flicker) with composite filter, the operating point lies in non-perceptible [9]. Which means composite filter brings voltage flicker in within non-
TABLE IX VOLTAGE UNBALANCE ANALYSIS EAF voltage unbalance (case 1) Without Filter
Parameter Peak RMS Voltage Unbalance (Peak) % Unbalance (Peak)
TABLE XI VOLTAGE UNBALANCE CONDITION PARAMETERS

Parameter Model 1 (Cassie-Mayer) Model 2 (Proposed)
Voltage a-Phase Voltage b-Phase Voltage c-Phase Voltage (Average) Voltage (Average Deviation) Voltage (% Average Deviation) Voltage a-Phase Voltage b-Phase Voltage c-Phase Voltage (Average) Voltage (Average Deviation) Voltage (% Average Deviation)
Parameter
8585 8426 8163 8391.33 152.22 1.81
6071 5958 5772 5933.67 107.78 1.82
102.31 100.41 97.28
2.31 0.41 -2.72
E0=Vat0 for a-Phase E0=Vat0 for b-Phase E0=Vat0 for c-Phase
250 200 180
250 200 180
IX.
PERFORMANCE ANALYSIS
With Passive Filter 9060 6406 8813 6232 8615 6092 8829.33 6243.33 153.78 108.44 1.74
Peak
102.61 99.82 97.57
2.61 -0.18 -2.42
Performances of passive filter and composite filter for various operational cycles of EAF have been analyzed in this section. Table XII shows percentage improvement in the respective voltage harmonic components after application of passive filter and composite filter. It can also be noted that performance of composite filter is better than the passive filter for H5 and H7 components, which are the major concern.
TABLE XII FILTER PERFORMANCE IN VOLTAGE HARMONICS
1.74
RMS Voltage Unbalance (Peak) % Unbalance (Peak)
Parameter
Voltage a-Phase Voltage b-Phase Voltage c-Phase Voltage (Average) Voltage (Average Deviation) Voltage (% Average Deviation)
With Composite Filter 11380 8046 11350 8026 11350 8024 11360 8032 13.33 9.33 0.12 0.12
100.18 99.91 99.91
0.17 -0.07 -0.10
H5 (%) H7 (%) H11 (%) H13 (%) H17 (%) THDV (%)
Filter performances in voltage harmonic distortion Refining Cycle Melting Cycle Melting Cycle (Random Flicker) (Sinusoidal Flicker) With With With With With With Passive Composite Passive Composite Passive Composite Filter Active Filter Active Filter Active Filter Filter Filter 88.03 94.06 91.42 92.68 83.64 96.08 92.68 93.92 88.07 93.35 91.55 94.72 98.02 94.31 97.92 92.01 98.21 94.64 98.42 94.32 97.07 95.60 95.00 95.00 92.49 95.77 88.32 93.43 95.59 88.24 89.95 87.18 64.58 87.82 77.36 60.21
TABLE X VOLTAGE UNBALANCE ANALYSIS Eaf voltage unbalance (case 2) Without Filter
Parameter Peak RMS Voltage Unbalance (Peak) % Unbalance (Peak)
TABLE XIII FILTER PERFORMANCE IN CURRENT HARMONICS

Parameter Current harmonic distortion improvement Refining Cycle Melting Cycle Melting Cycle (Random Flicker) (Sinusoidal Flicker) With With With With With With Passive Composite Passive Composite Passive Composite Filter Active Filter Active Filter Active Filter Filter Filter 92.54 56.09 84.87 -47.90 84.29 -4.29 95.60 54.92 85.27 -34.88 87.10 -29.03 98.54 59.12 66.67 -55.56 90.00 50.00 93.07 55.67 12.24 -20.30 60.49 -98.77
Voltage a-Phase Voltage b-Phase Voltage c-Phase Voltage (Average) Voltage (Average Deviation) Voltage (% Average Deviation) Voltage a-Phase Voltage b-Phase Voltage c-Phase Voltage (Average) Voltage (Average Deviation) Voltage (% Average Deviation) Voltage a-Phase Voltage b-Phase Voltage c-Phase Voltage (Average) Voltage (Average Deviation) Voltage (% Average Deviation)
9140 8661 7993 8598.00 403.33 4.69
6463 6124 5652 6079.67 285.11 4.69
106.30 100.73 92.96
6.31 0.73 -7.03
H5 (%) H7 (%) H11 (%) THDI (%)
With Passive Filter 9632 6811 8946 6326 8498 6009 9025.33 6382.00 404.44 286.00 4.48 4.48
106.72 99.12 94.16
6.72 -0.88 -5.84
With Composite Filter 11420 8078 11350 8025 11350 8022 11373 8041.7 31.11 24.22 0.27 0.30
100.41 99.79 99.79
0.45 -0.21 -0.24
Table XIII shows percentage improvement in the respective current harmonic components after application of passive filter and composite filter. It can also be noted that performance of composite filter is better than the passive filter for H5 and H7 components, which are the major concern. Table XIV shows performance of passive filter and composite filter under voltage unbalanced condition. Tabulated results show that passive filter alone fails to clear voltage unbalance where as composite filter performs better than passive filter. Voltage unbalance during arcing is the one of the important phenomenon in EAF. The designed composite filter has capability of clearing this voltage unbalance as per the tabulated results in Table XIV.
TABLE XIV FILTER PERFORMANCE IN VOLTAGE UNBALANCE Filter performances in voltage unbalance
[2] [3] [4]
Peak Voltage Unbalance (%) RMS Voltage Unbalance (%) Peak Voltage Unbalance (%) RMS Voltage Unbalance (%)
Case 1 1.81 1.82 Case 2 4.69 4.69
3.87 4.40 4.48 4.48
93.37 93.41 94.24 93.60 [6] [7] [5]
Table XV again confirms better performance of composite filter in reducing voltage flicker than passive filter alone.
TABLE XV FILTER PERFORMANCE IN REDUCING VOLTAGE FLICKER Filter performance (voltage flicker)
[8]
[9]
Sinusoidal Voltage Flicker 41.18 40.35 % Voltage Flicker 2.01 % Improvement
1.29 96.87
[10] [11]
X.
CONCLUSION
[12] [13]
This paper describes the design, analysis and simulation of composite filter for power quality improvement of electric arc furnace distribution network. First, new EAF model is proposed by combining exponential and hyperbolic models using transition function. The proposed model does not require any initial conditions for modeling the arc. Distribution network is simulated using Cassie-Mayer and the proposed EAF models. The combined model connected distribution network describes most of the specifications and operational characteristics of EAF. The simulated EAF distribution network is used for power quality analysis-voltage-current harmonics, voltage flicker and voltage unbalance. Next, a control strategy, based on the dual vectorial theory of power, for a composite filter connected in parallel with the unbalance, non-sinusoidal and randomly varying EAF is proposed. Finally, performance of passive filter and series active power filter is compared for various operation cycles of EAFs connected distribution network. Performance comparison shows that, the proposed composite filter performs better than the passive filter alone for harmonic compensation, voltage flicker mitigation, and for clearing voltage unbalance. REFERENCES
[1] Tavakkoli, M. Ehsan, S. M. T. Batahiee and M. Marzband, A SIMULINK Study of Electric Arc Furnace Power Quality Improvement by Using STATCOM, IEEE International Conference on Industrial Technology 2008, ICIT 2008, 21-24 April 2008, pp. 1-6.
[14]
[15]
[16]
[17]
[18]
[19]
Golkar M. A and Meschi S., MATLAB Modeling of arc furnace for flicker study IEEE Conference on Industrial Technology (ICIT), pp. 1-6, 2008. K. Anuradha, B. P. Muni and A. D. Raj Kumar, Modeling of Electric Arc Furnace & Control Algorithms for voltage flicker mitigation using DSTATCOM, IPEMC, 1123-1129, 2009. Mahdi Banejad, Rahmat-Allah Hooshmand and Mahdi Torabian Esfahani, Exponential-Hyperbolic Model for Actual Operating conditions of Three Phase Arc Furnaces, American Journal of Applied Scinces 6 (*):1539-1547, 2009. Mokhtari H. And Heiri M., A New Three Phase Time-Domain Model for Electric Arc Furnace Using MATLAB, Transmission and Distribution Conference and Exhibition 6-10 October 2002: Asia Pacific, IEEE/PES, Vol. 3, pp. 20787-283 Rahmatallah Hooshmand, Mahdi Banejad and Mahdi Torabian Esfahani, A New Time Domain Model for Electric Arc Furnace, Journal of Electrical Engineering, Vol. 59, No. 4, 195-202, 2008. Zheng T., Makram E. B. And Girgis A. A., Effect of different arc furnace models on voltage distortion, IEEE Transactions , International Conference on Harmonics and Quality of Power, 1418 October 1998, Volume 2, pp. 1079-1085 Haruni A. M. O., Muttaqi K. M. And Negnevitsky M., Analysis of harmonics and voltage fluctuation using different models of Arc furnace, IEEE Transactions, Power Engineering Conference, 9-12 December 2007, AUPEC 2007, Australasian Universities, pp. 1-6. A. Cano Plata and H. E. Tacca, Arc Furnace Modeling in ATPEMPT, International Conference on Power Systems Transients (IPST05), Montreal, Canada, 19-23 June 2005, Paper No. IPST05-067. A. Alzate, A. Escobar and J. J. Marulanda, Application of a DSTATCOM to Mitigate Arc Furnaces Power Quality Problems, 2011 IEEE Trondheim power Tech, pp. 1-6. Douglas Andrews, Martin T. Bishop and John F. Witte, Harmonic Measurements, Analysis, and Power Factor Correction in a Modern Steel Manufacturing Facility, IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 32, NO. 3, MAY-JUNE 1996, pp. 617-624. G. Carpinelli, Member and A. Russo, Comparison of some Active Devices for the Compensation of DC Arc Furnaces, IEEE Bologna Power Tech Conference, June 23-26, Bologna, Italy. G.-Myoung Lee, Dong-Choon Lee and Jul-Ki Seok, Control of Series Active Power Filters Compensating for Source Voltage Unbalance and Current Harmonics, IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 51, NO. 1, FEBRUARY 2004 pp. 132-139. Janusz Mindykowski, Tomasz Tarasiuk and Piotr Rupnik, PROBLEMS OF PASSIVE FILTERS APPLICATION IN SYSTEM WITH VARYING FREQUENCY , 9th International Conference, Electrical Power Quality and Utilization, 9-11 October 2007, Barcelona. Juan W. Dixon, Gustavo Venegas and Luis A. Moran, A Series Active Power Filter Based on a Sinusoidal Current-Controlled Voltage-Source Inverter, IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 44, NO. 5, OCTOBER 1997 pp. 612-620. P. Salmern and S. P. Litrn, Improvement of the Electric Power Quality Using Series Active and Shunt Passive Filters", IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 25, NO. 2, APRIL 2010, pp. 1058-1067. S. R. Mendis, M. T. Bishop and J. F. Witte, Investigations of Voltage Flicker in Electric Arc Furnace Power Systems, IEEE Industry Applications Magazine, January/February 1996, pp. 2834. Z. Zhang, N. R. Fahmi and W. T. Norris, Flicker Analysis Methods for Electric Arc Furnace Flicker (EAF) Mitigation (A Survey), IEEE Porto Power Tech Conference (PPT 2001), 10th 13th September 2001, Porto, Portugal. M. Walker, Electric Utility Flicker Limitations, IEEE Transactions on Industry Applications, Vol. 1A-15, No. 6, November/December 1979..
Deepak Chhaganrao Bhonsle Dr. Ramesh B. Kelkar Graduated from Birla Vishvakarma Mahavidyalaya, Sardar Patel University, Vallabh Vidyanagar, Anand, Gujarat, INDIA in 1998. Post Graduated in Cultural Studies from Shree Tattvajnana Vidyapith, Thane, Maharashtra, INDIA in 2000. Post Graduated in Electrical Power Engineering from Maharaja Sayajirao University of Baroda, Vadodara, Gujarat, INDIA in 2008. Currently pursuing Ph. D. from Maharaja Sayajirao University of Baroda, Vadodara, Gujarat. . His research interests are-electrical power quality, active filters. Graduated from Science Gwalior. M. E. from College of Engineering, Pune, Maharashtra and Ph. D. from Maharaja Sayajirao University, Vadodara, INDIA. At present he is working as Professor in Electrical Engg. Department of Maharaja Sayajirao University. His area of interest is Economic Operation and Electrical drives and Power systems.
Modeling of PV Based Distributed Generation in LV Feeders for Diverse Load Patterns Considering Reverse Power Flow Control
Mehmet H. Cintuglu, Member IEEE
Department of Electrical and Computer Engineering Florida International University Miami, USA mcint015@fiu.edu
Abstract Solar based distributed generation resources are the most promising renewable energy technology in low voltage grid networks. Distributed generations are installed to supply local customer load demand as main purpose with a permanent percent power level of connected transformer. Load patterns show variation according to location and utilization habit of the customers such as residential, industrial, agricultural watering purpose. As a result of different load profiles, generated power and the local load demand might not match depending on the pattern. If injected power from distributed generation to the grid exceeds connected customer local load, this will result reverse power flow into MV level, which is generally prohibited situation since it introduces unexpected voltage variations and unreliable generation forecast. In order to avoid reverse power flow, the injected power to the grid by distributed generation should be controlled locally. In this paper we studied injected power control scheme to avoid reverse power flow in a part of generic Turkish distribution network consisting of MV line equivalent model, transformer, LV loads and inverter based distributed generation systems with proper protection and specifications comply with local regulatory rules and IEEE 1547 standards. Proposed reverse power flow controller method was performed based on digital computer simulation approach by using Matlab/Simulink software package. Keywords- distributed generation, solar energy, reverse power flow
Osama A. Mohammed, Fellow IEEE

Department of Electrical and Computer Engineering Florida International University Miami, USA mohammed@fiu.edu
I.
INTRODUCTION
In last decade, with the idea of penetration of distributed generation into existing grid, electricity network has been facing a deep change that never happened before from invention of the electricity. Distributed generation can provide many benefits including reduced distribution costs, flexible operation and enhanced reliability. Local generation can reduce the energy losses and delay required infrastructure investment for transmission and distribution lines. However, besides many benefits of distributed generation, implementation of this systems can be quite complex into the existing grid networks. Voltage regulatory problems, voltage flickers, reverse power flow and islanding are major problems related to distributed resources.
Basically, electricity network structured with one way power flow: generation, transmission and distribution. However with penetration of distributed generation, this one way power flow rule will be no longer valid. In contrast to conventional generation structure, distributed generation enables electricity generation locally mostly very close to dedicated loads. Once feeding the local load demand, exceeding power will be transfer to infinite electricity grid. In that sense, power flow direction will change. This conflict and unpredicted situation would result operation problems inside the grid. Inside the LV feeder zone, reverse power flow might be tolerated. However, flow exchange from distribution layers like LV to MV and MV to HV would result congestion and complicated network profile. Grid connected distributed resources are mostly used for peak shaving and local load demands. According to IEEE 1547 standards [1], they do not contribute voltage or frequency control regulation. Moreover, they need to operate at unity power factor or a range between 0.95 leading lagging. Active power contribution is basically the only requirement of distributed generation plants. Apart from other complexities, even only active power injection from remote distributed generation systems is enough to create unexpected power flows in existing conventional grid structure. Therefore regulations limit the distributed generation connection power to each transformer by allocable capacity. Table 1 shows allocable capacity rating in Turkish electricity grid. However, this limitation will not be enough to prevent from reverse power flow in some load patterns.
Table 1. Allocable capacity for DG Transformer Capacity Allocable Capacity Allocable Capacity for (kVA) For DG (kW) Each Individual (kW) Demand < 100 7.5 100 < Demand < 1000 Demand x 0.3 Demand x 0.1 Demand > 1000 100 kW
SCADA systems may not be available in far terminals to control remotely. Control of the generated power from distributed resource should be carried out locally and automatically with proper power management intelligence, since it is not possible to deploy a technician or an engineer
responsible for operation to each generation unit. So far, many researches have been done regarding prevention of reverse power flow. Battery storage and heat pump water heaters [2-4] were proposed to use the opportunity of additional power when generation exceeds demand. Battery storage is a promising way to store energy however heat pump water heaters may not be necessary in distributed generation systems which are connected to illumination and agricultural watering transformers. Furthermore, optimal sizing and location of a PV system was proposed in MV distribution feeders [5] to avoid reverse power flow along the feeder. The objective of this paper to present appropriate power control scheme in order to avoid reverse power flow from LV to MV by controlling inverter based distributed generation power output considering different load profile feeders. Section 2 deals with the load pattern analysis for various type utilization purposes. Section 3 analyzes typical distribution architecture in Turkish electricity network grid transformer model in rural area supplying local loads and comprising solar based distributed generation connected to MV feeder. Section 4 introduces simulation system of the proposed model. Conclusions are stated in Section 5. II. LOAD PROFILES OF DIFFERENT FEEDER PURPOSES
According to regulations, the maximum allocated generation capacity to 50 kVA transformer is limited to 15 kW. In this load pattern generation never exceeds load demand. In this case, 0.3 capacity factor for connection of distributed generation units is feasible. B. Commercial Load Profile Working hours result high peak between 8-21 hours in commercial load pattern customers. This period of time is obviously when solar energy generation is most available. Implementation of distributed generation units on commercial load profile figure 2(a)-(b) will contribute peak shaving significantly. In this case, since solar generation is available mostly from 8 am to 5 pm, 0.3 capacity factor for connection of distributed generation units is feasible.
Fig. 2(a). Capacity factor for commercial load profile
In power networks, different feeders exist in terms of load pattern according to the utilization purpose. Load demands might show variation according to their customer behavior. Load profiles are mainly divided into five different categories: Industrial, commercial, residential, agricultural watering and illumination. Following figures show actual load profile data collected from a utility zone in a city from Turkey depending on customer behavior [6]. If we will assume each of the profile connected to a 50 kVA transformer, demanded actual load also could be available. A. Industrial Load Profile Industrial load profile has the most uniform distribution among the load profiles during the day. Utilization decreases between 18-22 hours, this is because to balance residential peak hours. Figure 1(a)-(b) shows the distribution of demand in 24 hours for weekdays and weekends.
Fig. 2(b). Demand as per 50 kVA transformer for commercial load profile
C. Residential Load Profile Due to permanently running home appliances in residential customers, load profile shows flat demand with exception peak hours noon and evening. Figure 3(a)-(b) apparently shows capacity factor 0.3 is a good approximation.
Fig. 3(a). Capacity factor for residential load profile Fig. 1(a). Capacity factor for industrial load profile
Fig. 1(b). Demand as per 50 kVA transformer for industrial load profile
Fig. 3(b). Demand as per 50 kVA transformer for residential load profile
D. Agricultural Watering Purpose Load Profile As seen from figure 4(a)-(b), agricultural watering load profile shows significant variations during the day. Proper implementation of distributed resources would relieve utility supply by mitigation of fluctuation on demand. Since load pattern is quite complex, if inverters will be injecting power to the grid when solar irradiation is most available and presence of no local load, this situation will obviously lead to reverse power flow. Furthermore, many agricultural watering purpose transformers are located in same MV feeder. Many generation plants would cascade and create even a high amount of reverse power flow from MV feeder up to HV/MV substation.
III.
DISTRIBUTION NETWORK STRUCTURE
In this paper, Turkish grid structure and load profiles are taken as reference for study model. MV distribution feeders have 33 kV, 50 Hz operating condition. The studied transformer is 33/0.4 kV 50 kVA standard oil insulating rural type with manual tap changer positions.
33 kV MV Feeder Equivalent Circuit
Over/Under Frequency
81 O/U
CB
Transformer 50 kVA 33 / 0.4 kV
Under Voltage
27
Over Voltage
59
Anti - Islanding
AI
Fig. 4(a). Capacity factor for agricultural load profile
Battery Charge/Local Load
Battery Charge/Local Load
CB
Transmissin Line
CB
Transmissin Line
CB CB
Transmissin Line
TRIP CB CB
CB
Fig. 4(b). Demand as per 50 kVA transformer for agricultural load profile
E. Ilumination Load Profile Illumination purpose transformers are operating almost at zero duty factors during the day. However at night and until early morning, high supply is required due to high demand as seen in figure5 (a)-(b). Charging during the day and using stored energy at night is the most feasible way for illumination purpose feeders. A special control algorithm and necessary storage should be implemented to inverter based distributed resource system.
10 kVA 10 kVA Load Load 0.95 Pf 0.95 Pf
5 kVA Load 0.95 Pf
Battery Storage
7.5 kW Local Load
7.5 kW Inverter DG
Battery Storage
7.5 kW Local Load
7.5 kW Inverter DG
Fig. 6. Transformer and distributed generation model
Rural overhead transmission lines are mostly Swallow or Pigeon conductors. In LV distribution feeders, Aster, Pansy and Rose conductors are widely used. Reactance and resistance values for these conductors are given in Table 2.
Table 2. Transmission line impedance table Conductor Type Reactance Resistance Aster Swallow 3AWG 0.420 /km 0.4245 /km 1.0742 /km
Fig. 5(a). Capacity factor for illumination load profile
Fig. 5(b). Demand as per 50 kVA transformer for illumination load profile
In our study system, since distribution transformer is 50 kVA rated, according to table 1, we consider two individuals connected 7.5 kW power with total power 15 kW. Remaining capacity is allocated as global load connected to transformer. The inverters can feed either load or charge the battery according to control methodology. Protection model involves over/under voltage, over/under frequency and anti-islanding relays. Since inverters in our model are grid connected, antiislanding scheme with ROCOF (Rate of change of frequency) detection has been implemented to avoid energizing the islanded area when utility does not exist.
IV.
SIMULATION OF THE PROPOSED NETWORK AND POWER TRACKING ALGORITHM
Since agricultural watering load pattern requires the most complicated control, in this paper we studied that profile in this paper. However, this control technique should be implemented for each utility feeder with examining the special situation of that kind of feeder. 50 kVA transformers with MV line connection, loads and feeders are modeled in simulation environment. Two 7.5 kVA, grid connected, hysteresis current controlled inverter based distributed resources are modeled in Simulink. Current control method of the inverters is shown in figure 7.
ia*
A very similar load pattern agricultural watering purpose load pattern was constructed by switching the loads in simulation as seen in figure 9. The voltage profile in the main feeder is also in limits.
P ref* Q ref*
d ref* q ref*
dq
ia ib*
Grid
abc
ic*
PLL
ib
ic
Fig. 7. Grid connected hysteresis current controlled inverter control
Fig. 9. Agricultural watering load pattern
Figure 8 shows the Simulink expression of study system in figure 6, which consists MV equivalent circuit, transformer, transmission lines, loads, inverters, and protection and control algorithm structures.
Discrete, Ts = 5e-005 s. pow ergui
Load switching pattern is defined for each load and inverter power tracking algorithm is modeled by inside the Matlab function block. Related flow chart of the implemented algorithm is show in figure 10.
Generation is assumed to be avaliable Direct Supply or Battery
36 kV MV Feeder Equivalent Circuit

C A B
OU_Frequency1
local_load1
local_load1
HV Side Measurement
A B C
U_Voltage1 Trip1 OR O_Voltage1 ROCOF1
inverter_output1
inv erter_output1 local_load2
local_load2
a b c
fcn
global_load
36 / 0.4 kV 50kVA Transformer
Anti-Islanding OU_Frequency2
Yes
Check if Local Load Exists
No
global_load
inverter_output2
inv erter_output2 allowance
LV Side Measurement
B A C
U_Voltage2 Trip2 OR O_Voltage2 ROCOF2
Conn1 Conn2
Conn3 Conn4
Main Grid Side Measurement
Conn6 Conn5
LV Transmission Line Equivalent
Anti-Islanding
Feed Local Load with Avaliable Power (Power Tracking) Yes
- Check if Global Load Exists -Is it allowed to supply global load by regulations?
No
Distributed Generation Measurement

Conn1 Conn2 Conn3 Conn4 Conn6 Conn5 Conn1 Conn2 Conn3 Conn4 Conn6 Conn5
Distributed Generation Measurement
Feed Global Load with Avaliable Power (Power Tracking)
Yes
No Check if Battery Full Charged
Conn1 Conn2
Conn3 Conn4
Conn6 Conn5
Conn1 Conn2
Conn3 Conn4
Conn6 Conn5
com
com
com
com
Trip2
b b b b a a a a c c c c B C B A A C
Trip1
B A C com com
Cease Generation
Charge Battery
com
com
Fig. 10. Flowchart of system model
b B
PWM Control1
Out1 Out1
PWM Control
Inverter
+ -
Inverter
-
Storage1
Storage
The power tracking algorithm according to the flowchart system was constructed Matlab function to follow required power delivery to the system. Without power tracking algorithm if inverters are operating all day, we observe reverse power flow between 5-7 seconds, 13-15 seconds and 19-22 seconds where no load demand. 1315 seconds are important since the solar generation is highly active during these hours.
Vabc
Vabc
Iabc
Iabc
b B
V in
I in
V in
I in
Vdc+
Vdc+
Fig. 8. Simulink expression of study system
Vdc-
Vdc-
Fig. 11. Feeder profile without power tracking scheme
REFERENCES Figure 11 shows feeder profile including main grid power, voltage and injected power from inverter based distributed generation. Although global load feeding is allowed to the generator, when there is no load demand inside the transformer area, inverters still injecting power to the grid. This situation cause reverse power flow from LV to MV side. The power tracking scheme is implemented and results can be seen in figure 12. Proposed algorithm successfully eliminates reverse power flow when local load and global load do not exist in the system. In that case, inverter will continue to charge battery or cease generation. V. CONCLUSION
[1] [2] IEEE Application Guide for IEEE std 1547, IEEE Standard for Interconnecting Distriubted Resources with Electric Power Systems Hiroyuki Hatta, Satoshi Uemura, and Hiromu Kobayashi, " Cooperative Control of Distribution System with Customer Equipments to Reduce Reverse Power Flow from Distributed Generation," in Power and Energy Society General Meeting, 2010 IEEE , July 2010 Asari, M. ; Kobayashi, H., " Method of controlling reverse power flow of PV system with heat pump water heater," in Innovative Smart Grid Technologies (ISGT Europe), 2012 3rd IEEE PES International Conference and Exhibition on , October 2012 Inoue, K. , Iwafune, Y., " Operation of heat pump water heaters for restriction of photovoltaic reverse power flow," in Power System Technology (POWERCON) 2010 International Conference on, October 2010 Hatta, H. , Asari, M., Kobayashi, H. " Study of energy management for decreasing reverse power flow from photovoltaic power systems," in Sustainable Alternative Energy (SAE), 2009 IEEE PES/IAS Conference on, September 2009 Load profiles. Retrieved April 30, 2010, from http://www.akedas.com
[3]
[4]
[5]
In this project, utility connection of PV based distributed generation resource in low voltage feeders was proposed. An important concern about the penetration level of distributed resources is investigated. A proper model with an energy management algorithm for diverse load patterns considering reverse power flow control. Proposed algorithm and connection properties of distributed resource were tested in simulation environment. The simulation result shows reverse power flow control can be controlled by proper implementation of distributed resources by means of power rate and operation modes.
[6]
Reliability Assessment of Some High Side MOSFET Drivers for Buck Converter
Vahid Javadian
Center of Excellence in Power System Management and Control Department of Electrical Engineering Sharif University of Technology Tehran, Iran javadian@alum.sharif.edu
Shahriyar Kaboli
Center of Excellence in Power System Management and Control Department of Electrical Engineering Sharif University of Technology Tehran, Iran kaboli@sharif.edu
MTBF Ron WL Vf HTOL Ea KB Tuse Tstress Y N t h Mean time between failures On-state MOSFET drain-source resistance Inductor weight (lbs) On-state forward diode voltage High Temperature Operating Life Acceleration Test Activation energy in electron-volts (eV) Boltzmanns constant (8.6173 10-5 eV/0K) Normal condition temperature Stressed condition temperature Number of failures (normally zero) Confidence level (normally 90%) Sample size Total test time Harmonic number
Abstractnowadays power electronic devices have a wide usage in the industries and different electrical equipment for power conditioning. As a point of view, reliability is one of important figure of merits that should be considered to have long life time and also have more probability to do exactly the proposed mission. In this paper some methods are presented to improve the reliability of power electronic components. Also some methods of system reliability improvement are reviewed. One group of converters is DC-DC step down converter which is used in different applications such as battery charger and voltage regulator. In this paper as an example, reliability of a DC-DC step down converter with Buck topology is evaluated and some kinds of switching networks including P and N channel MOSFET and proper driver topology are compared based on reliability specifics. It should be noted that the main reference used to reliability evaluation is MIL-HDBK-217 which is a military handbook to evaluation of power electronic components failure rate. Keywordscomponent; Reliability, Failure Rate, MIL-HDBK217, DC-DC Converter, P-channel MOSFET, N-channel MOSFET, Switch driver b FIT TA Tj THS Q E T S C C C V A R(t) MTTF MTTR Nomenclature Failure rate in useful life period Base failure rate in MIL-HDBK-217 Failure in Time (Failures per 109 Hours) Ambient temperature ( C) Junction temperature ( C) Hot spot temperature ( C) MIL-HDBK-217 quality factor MIL-HDBK-217 environment factor MIL-HDBK-217 temperature factor MIL-HDBK-217 electrical stress factor(diode) MIL-HDBK-217 capacitance factor MIL-HDBK-217 diode contact construction factor MIL-HDBK-217 inductor construction factor MIL-HDBK-217 capacitor voltage stress factor MIL-HDBK-217 application factor Reliability as a function of time Mean time to failure Mean time to repair
I.
INTRODUCTION
Reliability is the probability of performing adequately to achieve the determined aim for the system [1]. This can be mentioned as a time dependent equation [1, 2]. Reliability concept has more importance in specific applications such as space and military equipment that mission can hardly be replaced or be done by one other system instead of failed part. For improving the reliability of a system, different researches have been done and several methods have been introduced to enhance system reliability [3]. Each system has some components that one method to enhance reliability is improvement of components reliability. This goal may be achieved by component specifics derating [4] or by component specifics improvement. The other method is using redundant or fault tolerant systems which after one system or part fault, the rest of system can work adequately to achieve the goal of whole system [4-8]. One problem of using redundant system is load current balancing. So, different researches have been done in current sharing and current control [9-11]. The other method that may be useful is comparison of different topologies or different operation conditions to choose a proper state to achieve the goal of design [7, 12, 13]. One other method which is reported for power distribution systems [14] or Megawatt
drive systems [15] is maintenance. But it can rarely be used in power electronic systems because these systems usually are not available to be repaired or maintained and also this method is not meaningful for each power electronic component. For example just a fan can be maintained to work properly and doing its duty to cooling heat sink, but for a MOSFET switch or film capacitor, maintenance do not have a practical meaning. There are some technical standards to numerical evaluation of power electronic components failure rate. MIL-HDBK-217 [16] and FIDES [17] are among them. FIDES last update is at 2009 but MIL-HDBK-217 has not been updated since 1995 but nevertheless FIDES has more complicated parameters and has not been widely used or cited yet. MIL-HDBK-217 is simple to use and as in [18-24] more papers have been citing to this military standard and some manufacturers such as MINMAX have been evaluating their products reliability based on this standard. Consequently, in this paper MIL-HDBK-217 is used for numerical reliability evaluation. In researches related to the reliability of power electronic systems, failure rate of some components have been reported to be higher than others [19-24]. Electrolytic capacitor and switch are the power electronic components which have high failure rate based on empirical reports and MIL-HDBK-217 reliability evaluation. As a general approach, component stress reduction can enhance reliability of component. Rather than MIL-HDBK-217 there is some accelerated test or aging test that manufacturers use them to evaluate the reliability of their products [2]. In this method, at each test one or more parameters of environmental conditions are stressed more than typical state to reduction of test time less than real state. Then, there are some determined relations between this accelerated test results and typical condition results that are used to finding failure rates and reliability evaluation. There is one other method for dynamic and time dependent reliability assessment named Morkov Chain that in power system analysis has been used for many years [14] but has not been much progress by using this method in the field of power electronic reliability analysis [25, 26]. In this method by using redundancy component it will be more important that failing mode of component, how damages whole converter operation. For example, if there are 2 parallel output capacitors, open circuit or short circuit failing of one capacitor, has different effects on system reliability [26]. In this paper you can see some efforts to evaluation and improvement of reliability of power electronic components and dc-dc converter. Also some comparisons between different switching networks of Buck converter are presented to select the more reliable option to achieve the design goal. Different switching network options consist of P and N channel MOSFETs and proper drivers for these switches. In section II some details about Buck converter and different switching networks are mentioned. In section III some concepts of reliability field and numerical evaluation which is related to this paper are presented. In section IV equations of components reliability and methods to improve reliability of power electronic components are mentioned and in section V
mentioned switching networks are assessed as reliability point of view. II. BUCK CONVERTER There are two options for type of MOSFET switch in Buck converter. One option is N-channel MOSFET and the other is P-channel MOSFET. Since the N-channel MOSFET source in the buck topology is not connected to the circuit ground or a constant voltage reference, MOSFET gate driver has more complexity than the other topologies in which MOSFET source is connected to circuit ground. The case of Buck switch driver is named high-side or floating gate driver [27]. Based on MOSFET gate drivers, 4 topologies for Buck converter are presented in Figure 1-4. Using N-channel MOSFET needs isolation between MOSFET gate signal and MOSFET source pin. This isolation, in Figure 1 is established by bootstrap technique which exists in IR2113 [27] and in Figure 2 is prepared by using optocoupler. Whereas, using P-channel MOSFET does not need this kind of isolation. Because Pchannel MOSFET source will be connected to input voltage and consequently voltage of MOSFET source is constant and there is no problem for providing gate signal without isolation. It should be noted that there may be some other methods for gate isolation but in this paper just these methods are considered.
Figure 1 Buck converter with N-channel MOSFET and IR2113 driver
Figure 2 Buck converter with N-channel MOSFET and optical feedback isolation
Figure 3 Buck converter with P-channel MOSFET and auxiliary switch
Figure 4 Buck converter with P-channel MOSFET without auxiliary switch
P-channel MOSFET driver can be set up by 2 resistors connected to gate pin [28]. In Figure 3-4 MOSFET switch is P-channel type. When current passes through R1, voltage dropped between gate and source goes up and MOSFET turns on. This method can be used only for input voltage range under 50 volts. Because higher input voltage results in increment of power losses of R1 and R2 while increment of R1 and R2 resistance to diminish power losses, results in bad MOSFET switching performance. In Figure 3 auxiliary switch is used to enhance current injection to MOSFET gate which leads to better performance. Nevertheless, in Figure 4 controller IC provides the whole current needed for MOSFET drive. In Figure 1, MOSFET is N-channel type and driver IR2113 which has bootstrap structure is used to charge gate capacitor. In Figure 2, optical isolation between controller and driver is designed to separate signal and power zones. ICL7667 has a floating ground point that needs an isolated power supply. There are some DC power supplies with isolation between input and output zones that are proper for this propose. III. RELIABILITY CONCEPT AND NUMERICAL EVALUATION Failure rate is portion of failures per unit time to the number of components exposed to failure [1]. In MIL-HDBK-217, failure rates of different components are mentioned [16]. Failure rate for almost every electronic component has a bathtube curve that is indicated in Figure 5 [2]. This curve consists of three time intervals. First interval is about debugging or defective production failures in company. Second interval is for useful operating time that is interval of performing adequately and usually has a constant failure rate. The last interval has an increasing failure rate that occurs due to aging of component. Failure rates in MIL-HDBK-217 are for second interval that has a constant rate and in this paper this failure rate is evaluated. MTTF is mean time which is expected the component or system to work adequately before failure. It can be deduced that MTTF is equal to the integral of reliability of component as shown in equation 1 [1] (1) MTTF = R (t ) dt
Figure 5 Bath-tube curve for electronic components failure rate [2]
In general, the relation between reliability and failure rate can be deduced as equation 2 [1]
( ) d (2) R (t ) = e 0 Equation 2 for electronic components that has constant failure rate reduces to equation 3 and MTBF will be derived as equation 4: (3) R (t ) = e t
t
MTBF =
In DC-DC converter there are some components such as switch, diode, inductor, capacitor and controller IC, which are in series or parallel relation as reliability point of view. For evaluation of converter reliability, failure rate of components must be determined. For a series system, total failure rate is summation of failure rate of all components and system reliability is multiplication of components reliability [1]: (5) system = component For a parallel system, the system reliability cannot be expressed by one exponential term but the unreliability of system is multiplication of components unreliability [1]: (7) Qsystem ( t ) = 1 Rsystem (t ) = 1 Rcomponent (t )
t R system ( t ) = Rcomponent (t ) = e
(4)
(6)
For a more complex system, the system should be divided in to simple subsystems. There are more system topologies and solution techniques which are described in [1]. IV. COMPONENT RELIABILITY IMPROVEMENT
A. Switch In MIL-HDBK-217 equation 8 is determined for switch failure rate: (8) p = b T A Q E Failure / 10 6 Hours
T is temperature factor that deduces from equation 9:

t = exp 1925 (
1 1 ) T j + 273 298
MTTR is mean time which takes to repair the failed component that in electronic reliability evaluations is generally negligible with respect of MTTF. MTBF is mean time between failure that is summation of MTTF and MTTR, which can be assumed equal to MTTF.
(9)
In equation 9, Tj is junction temperature that is related to switch power loss and thermal resistance. So, for determination of T, one important step is switch power loss evaluation that is represented in appendix A. Reduction of power loss results to decrement of T and therefore failure rate
comes down. Choose a MOSFET switch which has lower Ron, means to have lower failure rate [19]. Q that is quality factor is determined by packaging and covering method that for commercial usage it is lower quality and is equal to 5.5. Other quality factors that have less value are for specific applications such as military and aerospace applications. For example, using JANTX package quality which has Q equal to 1.0 results to lower switch failure rate [23].
B. Diode Equation 10 expresses diode failure rate: p = b T S C Q E Failure / 10 6 Hours
Consequently, one method to failure rate reduction is using capacitor which has higher rated voltage [4]. Applied voltage to capacitor (19) S= Rated voltage S 5 (20) V = ( ) +1 0.6
E. Controller IC In MIL-HDBK-217, section 5 has some standard formulas to determine ICs failure rate but it is too complicated to compare different ICs failure rate based on this formulas. In some websites of manufacturers there are reliability accelerated test results. Component failure rate can be deduced from these results [2]. For example, one of these tests is HTOL that is based on Arrhenius acceleration model. In this test, component will be tested in temperature which is stressed more than normal use condition [2]. Acceleration factor is defined by equation 21 and then failure rate will be evaluated by equation 22 [2]. which is evaluated by equation 22 is upper bound failure rate that is expected to be measured. So, for different ICs which are used in this paper, failure rate is evaluated by using results of tests which are done by manufacturers or with FIT data which is presented directly in websites of manufacturers. 1 E 1 (21) AT = exp a K T T B use stress 2 ( , 2 Y + 2) (22) = 2 N AT t
(10)
T that deduces from equation 11has relation to diode power loss. Lower Vf results to lower power loss and consequently failure rate comes down. Diode power loss evaluation is represented in appendix B.
t = exp 3091(
1 1 ) T j + 273 298
(11)
S is electrical stress factor that is related to rated and applied voltage. Whatever the portion of applied voltage to rated voltage becomes smaller, this factor becomes smaller too. So, selection of diode with higher rated voltage, results to lower failure rate [4]. Equation 12 indicates this relation: Applied voltage 2.43 (12) S = ( ) Rated voltage
C. Inductor For inductor, equation 13 is used to calculate failure rate: (13) p = b C Q E Failure / 10 6 Hours
V.
MOSFET DRIVERS COMPARISON
b is not constant and is expressed in equation 14: T + 273 15.6 (14) ) b = 0.000335 ( HS 329 1.1 11.5 PLoss (15) TH S = T A + W L0.6766 As it is clear in equation 13-15, failure rate has direct relation with power losses. However, inductor failure rate is much lower than other components and has not been seen any reports that inductor failure rate be the reliability problem of power electronic converters.
D. Capacitor The other component is capacitor that equation 16 expresses failure rate for it: p = b T C V Q E Failure / 10 6 Hours (16)
After requirements that were obtained before, whole converter reliability can be evaluated. Proposed Buck converter specifics are presented in Table 1. Operating point is located in CCM.
TABLE 1 PROPOSED BUCK CONVERTER SPECIFICATION
PO 60 W
Vin 30 V
VO 15 V
L 375uH
C 1.5uF
D 0.5
Fs 50 KHz
T is determined by ambient temperature: 0.15 1 1 ( ) T = exp 15 T A + 273 298 8.617 10
(17)
(18) C = C 0.09 ( C is in f unit ) As it is presented in equations 19 and 20, greater rated voltage leads to smaller V which results in lower failure rate.
In each of 4 converter topologies that are presented in Figure 14, it is assumed that different drivers do not have important effects on converter components operation just on MOSFET switch operation and switch loss. One important difference between N-channel and P-channel MOSFET is input capacitor values. Input capacitor in products of some manufacturer has been assessed and it is deduced that the average of input capacitor in P-channel MOSFETs is more than N-channel MOSFETs. Whatever input capacitor is higher, charge interval will takes more time and consequently MOSFET switching loss becomes higher. Higher loss results in higher switch failure rate. For example, input capacitor of IXYSs MOSFETs which have 100 volt and 200 volt rated voltage is shown in Figure 6 and Figure 7.
switch equals to 4.15 failures per million hours. In this case, power loss of driver is very high to reduce main switch failure rate. However, as can be seen in Table 2 , the overall failure rate of converter using P-channel MOSFET is still higher than using N-channel MOSFET and IR2113 driver. The difference of total failure rate in the cases that the resistance of R1 and R2 are greater than 15 ohm will be greater. Consequently, it is obvious that using N-channel MOSFET results in lower failure rate for whole converter failure rate.
Figure 6 Input capacitor of IXYSs MOSFET 100 volt rated voltage
Figure 8 Failure rate of different Buck MOSFET drivers Figure 7 Input capacitor of IXYSs MOSFET 200 volt rated voltage
In Figure 8 failure rate of different drivers which are used in Figure 14 are shown. These failure rates are calculated based on data of manufacturers for the failure rate of ICs and failure rate prediction of MIL-HDBK-217 for diodes, capacitors and resistors which are used in driver circuits. Details of driver circuits can be seen in application notes [27, 28]. As it is seen in Figure 8 failure rate of drivers for P-channel MOSFET are lower than N-channel MOSFET drivers. For reliability comparison between 4 topologies that are presented, 2 switches of types P-channel and N-channel are selected to use datasheet parameters and spice simulation to calculate loss and then evaluate switch failure rate. It should be noted that to use available spice models, 2 switches of International Rectifier are selected for spice simulation. IRF540N and IRF5210 are switches which used in converter simulation. Other power electronic components are same and gate signal is ideal. In spice simulation, power loss of Pchannel MOSFET is about 0.99 watt but power loss of Nchannel MOSFET is 0.67 watt. Therefore, it is evident that switch failure rate for N-channel MOSFET is less than Pchannel MOSFET. As already mentioned, resistance of R1 and R2 have effect on MOSFET switching performance. In 4 cases, spice simulation for P-channel MOSFET with various R1 and R2 were done. Whatever these resistors have more resistance, power loss of these resistors are less but switching performance gets worse and power loss of main MOSFET increases that results in higher switch failure rate. In Figure 9 power loss of R1, R2 and main P-channel MOSFET in 4 cases of various resistors of driver are shown. Also, failure rate of Pchannel MOSFET is evaluated. For each of R1 and R2 equal to 15 ohm, power loss of switch is 1.475 watt and failure rate of
Figure 9 Power loss and failure rate in varied resistors of P-channel driver TABLE 2 TOTAL BUCK CONVERTER FAILURE RATE (Failures/million hours)
PMOS NMOS
Q1 4.15 3.83
D 0.047 0.047 VI.
C 0.0583 0.0583
L 0.002 0.002
Driver 0.237 0.435
Total 4.494 4.372
CONCLUSION
In this paper there are some efforts to evaluation and improvement of dc-dc converter reliability. Some methods to have more reliable converters are presented. Using more reliable components results in have a more reliable system. This goal maybe achieved by component specifics derating, that practical methods were presented in section IV. The other method is comparison and assessment of different topologies to achieve the goal of design and selection of optimum option. For the goal of this study, that was selection of switching network for Buck converter, as reliability point of view, N-channel and P-channel MOSFET drivers were compared. Failure rate of N-channel MOSFET driver is higher than P-channel MOSFET driver but failure rate of N-channel
MOSFET is lower than P-channel MOSFET. Input capacitor of P-channel MOSFET is larger than N-channel MOSFET and this is an important reason to higher switching power loss of P-channel MOSFET which results in higher failure rate. Consequently, it was obtained that whole Buck converter failure rate with N-channel MOSFET is lower than Buck converter failure rate with P-channel MOSFET. One other method is system or component redundancy in reliability point of view which is proposed for future works. REFERENCES
[1] G. Yang, P. W. M. John, M. Rausand, and A. Hyland, "System Reliability Theory: Models, Statistical Methods, and Applications," Journal of quality technology: A quarterly journal of methods applications and related topics, pp. 84-87, 2005. D. Crowe and A. Feinberg, Design for reliability: CRC, 2001. Y. Song and B. Wang, "Survey on reliability of power electronic systems," Power Electronics, IEEE Transactions on, vol. 28, 2013. S. V. Dhople, A. Davoudi, P. L. Chapman, and A. D. DomnguezGarca, "Reliability assessment of fault-tolerant Dc-Dc converters for photovoltaic applications," in Energy Conversion Congress and Exposition, ECCE 2009, pp. 2271-2276. B. Ray, "High-reliability space power converters: design and analysis issues," in Energy Conversion Engineering Conference, IECEC 2002, pp. 242-247. A. Chen, L. Hu, L. Chen, Y. Deng, and X. He, "A multilevel converter topology with fault-tolerant ability," Power Electronics, IEEE Transactions on, vol. 20, pp. 405-415, 2005. G. Graditi, D. Colonnese, and N. Femia, "Efficiency and reliability comparison of DC-DC converters for single phase grid connected photovoltaic inverters," in Power Electronics Electrical Drives Automation and Motion (SPEEDAM),, 2010, pp. 140-147. S. Ceballos, J. Pou, E. Robles, I. Gabiola, J. Zaragoza, J. L. Villate, and D. Boroyevich, "Three-level converter topologies with switch breakdown fault-tolerance capability," Industrial Electronics, IEEE Transactions on, vol. 55, pp. 982-995, 2008. Y. J. Cheng and E. K. K. Sng, "Transient analysis and fault compensation during module failure in paralleled power modules," Industry Applications, IEEE Transactions on, vol. 42, pp. 591-601, 2006. R. P. Singh and A. M. Khambadkone, "Current sharing and sensing in N-paralleled converters using single current sensor," in Industry Applications Society Annual Meeting, IAS'08. IEEE, 2008, pp. 1-7. J. Wang, "Design a parallel buck derived converter system using the primary current droop sharing control," Power Electronics, IET, vol. 4, pp. 491-502, 2011. M. Aten, G. Towers, C. Whitley, P. Wheeler, J. Clare, and K. Bradley, "Reliability comparison of matrix and other converter topologies," Aerospace and Electronic Systems, IEEE Transactions on, vol. 42, pp. 867-875, 2006. F. Blaabjerg, K. Ma, and D. Zhou, "Power electronics and reliability in renewable energy systems," in Industrial Electronics (ISIE), 2012 IEEE International Symposium on, pp. 19-30. F. Aminifar, M. Firuzabad, and R. Billinton, "Extended reliability model of a unified power flow controller," Generation, Transmission & Distribution, IET, vol. 1, pp. 896-903, 2007. R. D. Klug and A. Mertens, "Reliability of megawatt drive concepts," in Industrial Technology, 2003 IEEE International Conference on, 2003, pp. 636-641. Military handbook Reliability Prediction of Electronic Equipment1991. FIDES. Reliability Methodology for Electronic Systems. Available: http://www.fides-reliability.org/
[18]
[19] [20]
[21]
[22]
[2] [3] [4]
[23] [24] [25]
[5] [6] [7]
[26]
[8]
[27] [28]
[9]
G. Chen, R. Burgos, Z. Liang, F. Lacaux, F. Wang, J. van Wyk, W. Odendaal, and D. Boroyevich, "Reliability-oriented design considerations for high-power converter modules," in Power Electronics Specialists Conference. PESC 04. 2004 IEEE 35th Annual, pp. 419-425. H. Calleja, F. Chan, and I. Uribe, "Reliability-oriented assessment of a Dc/Dc converter for photovoltaic applications," in Power Electronics Specialists Conference. PESC 2007, pp. 1522-1527. A. Ristow, M. Begovic, A. Pregelj, and A. Rohatgi, "Development of a methodology for improving photovoltaic inverter reliability," Industrial Electronics, IEEE Transactions on, vol. 55, pp. 25812592, 2008. T. Kaipia, P. Peltoniemi, J. Lassila, P. Salonen, and J. Partanen, "Power electronics in Smartgrids-Impact on power system reliability," in SmartGrids for Distribution,. IET-CIRED. CIRED Seminar, 2008, pp. 1-4. B. Abdi, A. H. Ranjbar, G. B. Gharehpetian, and J. Milimonfared, "Reliability considerations for parallel performance of semiconductor switches in high-power switching power supplies," Industrial Electronics, IEEE Transactions on, vol. 56, pp. 21332139, 2009. F. Chan and H. Calleja, "Reliability estimation of three singlephase topologies in grid-connected PV systems," Industrial Electronics, IEEE Transactions on, vol. 58, pp. 2683-2689, 2011. X. Tian, "Design-for-reliability and implementation on power converters," in Reliability and Maintainability Symposium. Proceedings. Annual, 2005, pp. 89-95. J. Wang, M. Ding, and S. Li, "Reliability Analysis of Converter Valves for VSC-HVDC Power Transmission System," in Power and Energy Engineering Conference (APPEEC), Asia-Pacific, 2010, pp. 1-4. R. Argile, B. Mecrow, D. Atkinson, A. Jack, and P. Sangha, "Reliability analysis of fault tolerant drive topologies," in Power Electronics, Machines and Drives, PEMD 2008. 4th IET Conference on, pp. 11-15. H. V. F. M. O. S. G. D. ICs, "Application Note AN-978," Rev D, 2007. O. Semiconductor, "TL494, NCV494 SWITCHMODE. pulse width modulation control circuit [E]," Semiconductor Components Industries, LLC, 2005.
Appendix
A. Switch Power Loss
Pc o n d u c t . = R o n 2 rm s , Psw itc h ., C C M = PL o s s , S w itc h = Pc o n d u c t . + Psw itch ., C C M 1 V o ff o n ( t o n + t o ff ) f s 2
[10]
[11] [12]
B. Diode Power Loss

Pcon d u ctio n = V f (1 D ) o , Psw itchin g = n eglig ib le PL os s , D io d e Pcon d u ctio n
[13] [14] [15] [16] [17]
C. Inductor Power Loss

Pco re = 1 .5 K f s V e , Pco p p er = R d c I d2c + R ac ,h = bh = l
(R
h =1
ac ,h
2 I rm s ,h )
Ah
l cu
cu [ rw2ire ( rw ire h ) 2 ]
sin [ h (1 D )
, h =
0 .0 6 6 h fs
( 1) h i p ea k p ea k
h2 (D D 2 ) 2 b I rm s , h = h 2 PL o ss , In d u cto r = Pco re + Pco p p e r
Validation of Sequence Circuits Useful for SplitPhase Current Signature Analysis (SPCSA) and Diagnosis of Eccentric-Rotor raction Cage !otors
Claudio "ru##ese
Dept$ of Astronautical% Electrical% and Energy Engineering Uni&ersity of Ro'e - Sapien#a Ro'e% (taly claudio$)ru##ese*uniro'a+$it
AbstractA squirrel-cage three-phase induction motor with eccentric rotor and parallel connections in the stator can be represented by five equivalent sequence-circuits of virtually centered-rotor machines, as mathematically proved in a companion paper. The five circuits allow the calculation of stator and rotor fault-related currents, in form of 2( p !"-pole space vectors for 2p-pole machines. The formal calculation of faultrelated currents is very useful for defining fault indicators and tools for diagnosis, and in case of parallel-connected stators (li#e in many traction motors" the split-phase current signature analysis becomes possible. This paper validates the sequencecircuit modeling by comparing the theoretical results with simulations of a !!$%#& traction induction motor with static, dynamic, and mi'ed faults in both no-load and full-load conditions. The simulations are carried out by using a full meshmodel of the motor, implemented by winding functions. A good match is obtained between theoretical calculations and simulations. (oreover, the role of the cage damping on the eccentricity-related current signatures is evidenced. )t is shown that the cage damps not only the signature of static fault but also that of dynamic fault, and that the damping itself is loaddependent in the latter case. KeywordsDiagnosis, eccentricity, equivalent circuit, induction motor, rotor, split-phase, symmetrical components, traction.
u% k su)script inde5es for stator and rotor circuits vS% iS stator PP: real &oltage and current colu'n &ectors vR% iR rotor loop real &oltage and current colu'n &ectors v(q)% i(q) co'ple5 q-th sequence &oltages and currents VLL line-line r's &oltage at the 'achine ter'inals S% R co'ple5 unit &ectors (2ejs% ejR% respecti&ely) flu5 incre'ent due to eccentricity air gap field angle on the direct a5is (po7er angle) rotor rotation 'echanical angle (2!0p) "s% "# static and dyna'ic p$u$ eccentricities% 6<"s="#>+ ?s% ?# angles of static and dyna'ic eccentricities p % $$ /p-pole% 0/1-frequency flu5 lin9age space &ector 6 'a5 PP: flu5 lin9age in the healthy 'achine m rotor 'echanical speed (2/1fm2(+-s)0p) @ co'ple5 conAugate operator (($ (,
R-DUC (-,
($
(q)
,-!E,C.A
URE
S() fault-related /q-pole 0/1-frequency current f funda'ental frequency (20/1) Fn 3ortescue4s n5n generali#ed transfor'ation 'atri5 g6 'ini'u' air gap length in the healthy 'achine Kp1 7inding factors for /(p8+)-pole flu5 density 7a&es L(q) q-th sequence inductance Lm phase 'agneti#ing inductance LSl%LRl+ phase lea9age inductances (LRl+ is reported to stator) n nu')er of ele'entary stator PP:s (2;p) m nu')er of ele'entary rotor loops0)ars p 'achine pole pairs RSS%LSS stator resistance and inductance n5n 'atrices RPPG resistance of a stator pole-phase-group (PP:) RS% RR+ phase resistances (RR+ is reported to the stator) sq slip of the q-th sequence 'achine
BE PARA..E. connection of the stator 7indings is a constructi&e option &ery popular in 'any applications% 7hen the 'achine &oltage rating 'ust )e 9ept )elo7 specified li'its$ Another reason is that a thinner conductor insulation can )e used% and a larger slot filling factor is gained$ he highest po7er tur)ogenerator stators are usually parallelconnected% and this% in addition to the use of Roe)el )ars% aids the designers to )ound the &oltage 7hile the 'achine output rating is raised )y designing a longer iron core C+D$ ,a&al synchronous generators are nor'ally lo7-&oltage high current 'achines 7ith parallel connected stators C/D$ raction 'otors are another e5a'ple CED% and this paper focuses on three-phase squirrel cage induction 'otors used in rail7ay traction dri&es$ he currents flo7ing in the parallel )ranches of sy''etrical 7indings should )e identical in principle$ Bo7e&er% any fault creating an asy''etry in the air gap flu5 distri)ution affects the )alance of the split-phase currents% as in case of static and dyna'ic rotor eccentricities (SRE and DRE% respecti&ely) C/D$ Another cause of split-phase current un)alance is the insulation fault% 7ith turn-turn% phase-phase% or phase-ground fault CFD$ (n fact% the split phase currents are so'eti'es 'onitored in large 'achines )y differential relays for protection against the stator faults C+D$ Also )ro9en )ars 'ay produce definite un)alances in the parallel currents CGD$
Paper C;D proposed a sy''etrical co'ponent-)ased 'odeling of squirrel-cage parallel-connected induction 'otors 7ith REs% 7hich allo7s 'achine decoupling in fi&e equi&alent circuits (sequence circuits) of fi&e &irtual centered-rotor 'achines$ hese circuits are useful for calculation of fault related stator currents% ta9ing in account the &ery i'portant da'ping effect de&eloped )y the rotor cage$ he 'odel proposed in C;D is &alidated in this paper for a practical ++E69H 'otor used for high-speed rail7ay traction dri&es$ he fault-related currents calculated )y using the sequence circuits theoretically deri&ed in C;D are co'pared 7ith the &alues directly o)tained )y dyna'ic si'ulations 7ith a co'plete 'esh-'odel )ased on 7inding function approach (H3A)% at no-load and full-load conditions% and 7ith &arious co')inations of SRE and DRE$ A good 'atch is o)tained% 7hich 'a9es the sequence-circuit 'odeling suita)le for further future de&elop'ents of fault indicators useful for 'achine condition 'onitoring$ ((($ SEIUE,CE C(RCU(
S -3 BE
7here the su)scripts l% g indicate lea9age and air-gap flu5 inductance 'atrices$ All the 'atrices RSS% RRR% LSS% LRR% LSSl% LRRl% LSSg% LRRg in (+)% (/) are sy''etric and circulant L that is% any ro7 is o)tained fro' the precedent )y a circular rightshift$ heir e5plicit for' is reported in C;D$ !oreo&er% they are assu'ed constant$ he stator-rotor 'utual inductance 'atrices LSR in (+) and LRS2LSRt in (/) instead depend on the rotor position % and their ele'ents can )e 'odeled )y )ilateral 3ourier e5pansions 7ith coefficients 0(*) C;D$ he 'odel (+)% (/) is con&eniently transfor'ed )y using t7o 3ortescue4s 'atrices% i$e$ F+/% FG; for stator and rotor% respecti&ely$ he transfor'ation of stator currents i S 1 = F +/ i S is e5plicitly 7ritten here )elo7 for clarity (the rotor current transfor'ation i R 1 =F G; i R has the sa'e for'al structure)K
++E6JH !-
-R
&' Resume$(f$)*e(+,$F+(m$-./ he 'achine 'odel introduced in C;D descri)es a squirrel cage induction 'otor 7ith parallel connections in the stator and generic rotor eccentricities$ 3ig$ + sho7s the sy''etrical stator 7inding structure% 'ade up of pole-phase groups (PP:s% or groups of series-connected coils of a phase under a pole% called S+%$$$%Sn in 3ig$+ of C;D)$ he rotor cage is deco'posed in ele'entary loops R+%$$$% Rm for'ed )y couples of consecuti&e )ars and )y the Aoining ring seg'ents ('esh-'odel)$ he PP:s are parallel-connected in each stator phase$ he fourpole ++E69H induction 'otor considered in this paper (see Appendi5 A for data) has n2+/ stator PP:s and m2G; rotor cage loops$ Benceforth% the 'odel is tailored on this 'achine$
POLE-PHASE-GROUP CURRENT PRO"E

6 iS + iS / iS
q q q q q q
i S+ iS / + = i + / F // S S S SE +/ / ++ // ++ iS ++ ++ + iS S S S
+ +
+ S
+ / S
+ ++ S
(E)
he generic ter' iS(q) in (E) is the q-th-order space-sequence current% also referred to as /q-pole current space-&ector$ he transfor'ation puts the circulant 'atrices in a diagonal for'% and the equations turn out decoupled 7hen the higher space har'onics are neglected$ he generic q-th sequence stator and rotor equations are as follo7sK v S = RS i S L Sl L Sg
q q
# q # q q i L i #! S #! SR R # q # q q i L i $ #! R #! SR S
(F) (G)
v R = R R i R L Rl L Rg
q q
4 3 L2#
POLE-PHASE-GROUP ELEMENTARY CIRCUIT
L# 6
A'
e1 Rs L0 !1
Eqs$ (F)% (G) represent the dyna'ics of a generic q-th 4sequence 'achine4 &irtually e')edded in the 'achine 'odel$ All the sequence 'achines are electrically decoupled% and 'echanically coupled at the shaft$ (t can )e pro&en CMD that% in case of parallel connection of the stator% vS(q)26 (short-circuited stator) for qNp$ !oreo&er% vR(q)26 for any q% due to the shortcircuit rings$ 3or p2/% the four-pole &irtual 'achine 'anages the 'ain electro'echanical energy con&ersion$ 2' 1134k5$677en!+i78R(!(+$9(!(+$Sequen7e$:i+7ui!s (n a four-pole 'otor% a rotor eccentricity (3ig$ /) 'a9es rise additional /-pole and ;-pole flu5 density 7a&es in the air gap C;D$ hese 7a&es e5cite the /-pole and ;-pole &irtual 'achines% respecti&ely$ he additional E!3s 'ust )e put in (F) for q2+ and q2E as follo7sK 6 = R S i S L Sl L Sg
+ + + +
12 8 11 9 10
PHASE 'A' INTERNAL CONNECTIONS
3ig$ +$ 3our-pole three-phase 7inding 7ith parallel-connected PP:s$
he stator and rotor phase-'odels in the ti'e do'ain are 7ritten in (+)% (/) in 'atri5 for'% assu'ing n2+/ circuits on the stator and m2G; circuits on the rotor C;DK # L SSl L SSg i S L SR i R #! # v R =R RR i R L RS i S L RRl L RRg i R #! v S = R SS iS (+) (/)
# + # + + i L i #! S #! SR R # + # + % % #! #!
m
(;)
v S = RS i S L Sl L Sg
/ / / / /
# / # / / i L i #! S #! SR R
(M)
6 = RS iS L Sl L Sg
E E E E
# E # E E i L i #! S #! SR R $ # E # E % % #! #!
m
RS = R S = R S = R PPG = F RS
(S) (+6)
(O)
L = L = L = L PPG = F L Sl $
+ Sl
/ Sl
E Sl
he sa'e E!3s appear on the rotor in (G) through a proper stator-rotor transfor'ation ratio$ At this point% the equations can )e 7ritten at steady-state and fi&e equi&alent sequence circuits of the eccentric-rotor 'achine can )e carried out as done in C;D$ hese fi&e circuits are reported in 3ig$ E% 7here they ha&e )een adapted to the four-pole ++E69H 'achine$ :' :;l7ul;!i(n$(f$!*e$1134k5$9(!(+$Sequen7e$P;+;me!e+s he sequence para'eters appearing in 3ig$ E ha&e )een nu'erically carried out for the ++E69H 'otor )y using the guidelines e5plained in C;D% as follo7s$ he stator sequence resistances and lea9age inductances are o)tained fro' the 'achine phase para'eters in a)le ((( (see ("O)%("+M) in C;D)K
4 q4 this &ector rotates 7ith the rotor
he rotor sequence resistances and lea9age inductances are directly o)tained fro' ("S)% ("/6) of C;DK
R R = / R < ;+ + cos q E;6 / G; R +ing / G; L Rl =/ L< ;+ + cos q E;6 / G; L +ing / G;

q
(++) (+/)
7ritten for q2+%/%E$ After7ard% they are reported to the stator )y using the stator-rotor transfor'ation ratio (/G) of C;DK
q q k q= L Sg / L Rg $
(+E)
Ps2SC0g6
Pd2CR0g6
g6 2 gap of the healthy 'achine
he 'agneti#ing inductances LSg(+)% LSg(/)% LSg(E) of 3ig$ E and also the rotor-related inductances LRg(+)% LRg(/)% LRg(E) to use in (+E) ha&e )een already calculated in C;D )y using the H3A$ All the para'eters are reported in a)le ($ ,ote finally that LSg(/)% LRl(/)+% RR(/)+ 'atch the 'achine phase para'eters Lm% LRl+% RR+ of a)le ((( respecti&ely% 'ultiplied )y F (see (/O) of C;D)$
A".E ($ ++E6JH (,DUC
(-,
g6(+-Ps-Pd) QR ?d
a
!-
-R
SEIUE,CE PARA!E
ERS
this &ector is fi5ed + first cage loop
?s
QS
a5is of PP:R+
3ig$ /$ Rotor eccentricity sche'e and sy')ols used in the for'ulas$
parameters of sequence no. ! circuit (p-!" RS(+)2FRS +6+$/'T LSl(+)2FLSl E$//+'B LSg(+) RR(+)+2k+/RR(+) FO$E'T LRl(+)+2k+/LRl(+) MSGuB LRg(+) k+ /6G$/G parameters of sequence no. 2 circuit (p" RS(/)2FRS +6+$/'T LSl(/)2FLSl E$//+'B LSg(/)2FLm RR(/)+2k//RR(/) S6$O'T LRl(/)+2k//LRl(/) +$MM'B LRg(/)2FLm0k// k/ +GE$MO parameters of sequence no. $ circuit (p-!" RS(E)2FRS +6+$/'T LSl(E)2FLSl E$//+'B LSg(E) (E)+ / (E) RR 2kE RR FO$OE'T LRl(E)+2kE/LRl(E) +$6FF'B LRg(E) kE OM$OGE
+MO$M'B F$/F/uB +66$/'B F$/EMuB E/$;F'B F$//SuB
RS(+)
LSl
(+)
S(+) ()
R(s
(+)+
+%s)
L(+)+ Rl
RR$$0s+%s
(+)+
RS(E)
LSl
(E)
S(E) ()
R(s
(E)+
E%s)
L(E)+ Rl
RR$$0sE%s
(E)+
(+) m ()
LSg
*TAT.+ /+01203,4 5267 3.-8.A97 !%%:; /288-8.A97 !%%:;
(+)
s + s+% s= 6$G /
+.T.+ /+01203,4 %.<(s-!"5267
(E) m ()
LSg
*TAT.+ /+01203,4 5267 3.-8.A97 !%%:; /288-8.A97 !%%:;
(E)
s E% s=
E s + 6$G /
3.-8.A97 <%.%%=:; >S(+) () /288-8.A97 <%.$>:;
3.-8.A97 -=?.?@@:; >S(E) () /288-8.A97 -=@.?!:;
+.T.+ /+01203,4 %.<($s-!"5267
SRE(p-1=1)
RS(/)
LSl(/)
(/) S ()
=R (s) =m(/) ()
(/)+
LRl(/)+ R (/)+0s R
SRE(p+1=3)
(/) S
*TAT.+ /+01203,4 5267 3.-8.A97 !%%:; /288-8.A97 !%%:;
LSg
(/)
+.T.+ /+01203,4 s5267 3.-8.A97 %.%%@:; /288-8.A97 %.>$:;
S ((+=s)0/) RS(+) U(+=s)LSl

(+) (+)+ R(s) U(+=s)LRl RR$$0s+%# (+) m ((+=s)0/) (+)+ (+)+
(+)
HEALTHY(p=2)
S ((E-s)0/) RS(E) U(E-s)LSl

(E) (E)+ R(s) U(E-s)LRl RR$$0sE%# (E) m ((E-s)0/) (E)+ (E)+
(E)
U(+=s)LSg
*TAT.+ /+01203,4 %.<(!-s"5267 3.-8.A97 <%.%%=:; /288-8.A97 <%.$>:;
(+)
s+% # =
/s / s + s
U(E-s)LSg
*TAT.+ /+01203,4 %.<($-s"5267 3.-8.A97 !=?.??A:; /288-8.A97 !=?.A$:;
(E)
s E% # =
/s / s / E E s
+.T.+ /+01203,4 s5267 >S(+) ((+=s)0/) 3.-8.A97 %.%%@:; /288-8.A97 %.>$:;
>S(E) ((E-s)0/) 3.-8.A97
+.T.+ /+01203,4 s5267 %.%%@:; /288-8.A97 %.>$:;
DRE(p-1=1)
DRE(p+1=3)
3ig$ E$ Sequence circuits of the fi&e centered-rotor &irtual 'achines actually superi'posed in the 'odel of the real eccentric-rotor induction 'otor$ he circuits are dra7n for the ++E69H 'otor 7ithK pole pairs p2/% feeding frequency 0/12+66B#% no-load slip s,.26$66OV% full-load slip (rated slip*+66B#) s3.26$MEV$
(V$
C-!PAR(S-, -3
BE-RE (CA. A,D
S(!U.A
(-,
RESU.
&' F(+m;l$:;l7ul;!i(n$(f$F;ul!8Rel;!e#$:u++en!s he fault-related currents flo7ing in the p8+ circuits of 3ig$ E are carried out in a general for'al 7ay as follo7s$ he generic q-th circuit is sol&ed for so'e generic frequency q0/1 and slip sq$ Since the 'agneti#ing i'pedance qLSg(q) is 'uch greater than )oth resistances and lea9ages reactances% the 'agneti#ing )ranch can )e thought as short-circuited$ he generic 'agneti#ing current is thereforeK
q m &
q
related &ector% 7hereas the /- and ;-pole &ectors are faultrelated$ he &ectors 7ith O and +6 poles are also reported% for co'pleteness$ he rotor transfor'ation yields 'uch 'any independent &ectors (/S &ectors)% 7hich cannot )e all reported for reason of space$ So only the &ectors of order +-G are reported in 3igs$ F-S for the rotor% 9eeping in 'ind that each rotor &ector is al7ays strongly coupled to the stator &ector of the sa'e order$ So'e o)ser&ation can )e dra7n as follo7s$ Hith centered rotor% 3igs$ F% O sho7 that only the loadrelated &ectors (F-pole &ectors for the F-pole 'achine) ha&e rele&ance% in )oth no-load and full-load conditions$ he other &ectors are usually &ery s'all and neglecta)le$ Bo7e&er% the load &ectors are not sensiti&e to REs% 7hereas the other &ectors )eco'e larger (3igs$ G% ;% M% S)$ (n particular% the /-pole and ;pole &ectors assu'e &ery large a'plitudes% as theoretically pre&ie7ed$ he /-pole and ;-pole stator &ectors in 3ig$ G are single-frequency rotating &ectors% in case of pure DRE (+6V)$ he correspondent rotor &ectors are &ery s'all% as e5pected$ Also 3ig$ ; sho7s the /- and ;-pole stator &ectors as singlefrequency rotating &ectors% in case of pure SRE% )ut their a'plitude is 'uch s'aller than in case of DRE of the sa'e entity (+6V)$ (n fact% the rotor reaction is 'uch stronger in case of SRE% as pro&ed )y the larger /- and ;- pole rotor &ectors$ his pro&es the i'portance of ta9ing in account properly the cage reaction$ (n case of 'i5ed RE (3ig$ M)% t7o-frequency /- and ;-pole &ectors appears on )oth stator and rotor% 7hich are quite e5actly o)tained )y superi'position of the &ectors of 3ig$ G% ;$ (n particular% 3ig$ M-e sho7s the typical trochoidal closed traAectory generated )y t7o &ectors 7ith frequencies gi&ing a rational ratio$ Bo7e&er% 7ith loaded 'achine (3ig$ S)% the t7o frequencies of each )i-periodic &ector are no 'ore in a rational ratio due to the slip% and the traAectories are no 'ore closed$ (n 3igs$ S-a% )% e% f% the last rotation period of the &ector is dra7n 7ith a )old line for clarity$ he a'plitudes can )e so directly co'pared 7ith those in 3igs$ M-a% %)% e% f% for the sa'e RE le&els at no-load% and so'e difference can )e e&idenced$ (n particular% note that the /- and ;-pole rotor &ector traAectories in 3igs$ S-)% f are not centered as in 3igs$ M-)% f$ (n fact% an additional slip-frequency rotating co'ponent appears in 3igs$ S-)% f in loaded condition% 7hich is due uniquely to the DRE$ his pro&es the e5istence of a strongly load-dependent cage da'ping 7ith respect to the DRE% 7hich cannot )e neglected$ he cage da'ping 7ith respect to the SRE is% instead% quite independent on load$ :' Au;n!i!;!ive$:(mp;+is(n$(f$)*e(+,$;n#$Simul;!i(ns 1B :;l7ul;!i(n$(f$Cp1B8sequen7e$7u++en!s$in$7;se$(f$SR6D$ a)le (( reports the currents &S(p8+)% &R(p8+) ()old &alues) calculated for +6V of SRE )y using (+;)% (+M)$ ,ote that the SRE-related circuits in 3ig$ E are quite independent fro' the 'achine load% so sX6 is used$ he &alues 'ust )e 'atched 7ith those in 3ig$ ;-a%-)%-e%-f$ A fair agree'ent is o)tained$ EB :;l7ul;!i(n $ (f $ Cp1B8sequen7e $ 7u++en!s $ in $ 7;se $ (f $ @R6D a)le (( also reports &S(p8+)% &R(p8+) for +6V of DRE at noload% 7hich 'ust )e &alidated in 3ig$ G-a%-)%-e%-f$ hese currents change on-load% as sho7n in a)le (($ (n case of 'i5ed fault% all the a'plitudes should )e superi'posed in 3ig$ M (at no-load) and in 3ig$ S (full load)$ A good agree'ent is al7ays o)tained fro' these quantitati&e chec9s$
Sq 6
q
q j q LSg
(+F)
and the 'agneti#ing )ranch can )e treated as a current source feeding the stator and rotor )ranches$ he &oltage >S(q)(q) is proportional to the eccentricity as e5plained in Appendi5 "$ he current (+F) is then split in t7o co'ponentsK
q m q q + & = & S & R s
q q q q
(+G)
depending on the relati&e a'plitude of the stator and rotor longitudinal i'pedances$ he stator and rotor currents are easily carried out as follo7sK
q S & =
q
q + R ? q q + &m ? ? R q S
q
(+;) (+M)
+ R q & s =
q q
Sq ? q m & Sq ? Rq + ?
q
7here the i'pedances of stator and rotor )ranches areK

q q q ? S = R S j q L Sl q + ? R =
(+O) (+S)
RR q + j q LRl $ sq
q +
Eqs$ (+;)% (+M) clearly e5plain the da'ping effect of the cage$ (n fact% in presence of fault the 'agneti#ing current of the generic (p8+) circuit is proportional to the RE gra&ity due to ("F)% ("G)$ Bo7e&er% only the stator co'ponent in (+;) can )e practically 'easured% and so the rotor )ranch actually drains the current (+M) a7ay fro' the useful co'ponent% and the 'easura)le fault signature is 7ea9ened$ his leads to an hea&y underesti'ation of REs% if the cage da'ping effect is not ta9en in account% especially in case of static RE$ 2' 1134k5$9(!(+$@,n;mi7$Simul;!i(ns$<,$5F& 3igs$ F-S sho7 si'ulations of the ++E69H 'otor in no-load and full-load conditions% 7ith &arious co')inations of SRE and DRE$ A 'odel 7ritten in !at.a) and e')edding the 'achine dyna'ic equations is used% 7ith inductances e&aluated )y H3A as sho7n in C;D$ he 'achine currents ha&e )een transfor'ed )y using (E) for the stator split-phase currents and the analogous transfor'ation i R 1 =F G; i R for the rotor cage-loop currents$ he stator transfor'ation yields se&en independent space-&ectors (7ith sequence orders 6% +% /% W% ;)% )ut the &ectors of orders 6 and ; are al7ays neglecta)le% and they are not reported$ he F-pole &ector (order /) is the load-
G6$66FB#
6$66OB#
& S(0/)(+)2+$FOA a) +66B# )) 6$66OB# a) +66B# ))
& R(s)(+)2G$+A
6$66OB#
F S()(/)2F6A c) d)
F R(s)(/)2EMEA c)
F S()(/)2F6A d) +FS$SS;B#
F R(s)(/)2EMEA
6$66OB#
& S(E0/)(E)2E$/A e) f) e) f)
& R(s)(E)2E$+A
g)
h)
(mA) g)
(mA)
h)
i)
A)
i)
A)
3ig$ F$ a)-A)K Stator and rotor current &ectors% no-load% 6VSRE and 6VDRE$
3ig$ G$ a)-A)K Stator and rotor current &ectors% no-load% 6VSRE and +6VDRE$
+66B#
G6$66FB#
G6$66FB# +66B#
6$66OB# G6$66FB#
& S()(+)26$EGA a) +66B# ))
& R(0/)(+)2/F6A a) 6$66OB#
& S(0/)(+)=& S()(+)2+$OFA )) +66B#
& R(s)(+)=& R(0/)(+)2/FFA
6$66OB#
F S()(/)2F6A c) +66B# d)
F R(s)(/)2EMEA c) -FS$SOOB#
F S()(/)2F6A d) +FS$SS;B# +66B#
F R(s)(/)2EMEA
6$66OB# -FS$SOOB#
& S()(E)26$SA e) f)
& R(-0/)(E)2//6A e)
& S(E0/)(E)=& S()(E)2F$+A f)
& R(s)(E)=& R(-0/)(E)2//EA
g)
h)
g)
h)
i)
A)
i)
A)
3ig$ ;$ a)-A)K Stator and rotor current &ectors% no-load% +6VSRE and 6VDRE$
3ig$ M$ a)-A)K Stator and rotor current &ectors% no-load% +6VSRE and +6VDRE$
G6$EMB# +66B#
& R(s)(+)=& R(0/)(+)2EG;A
6$MEB# G6$EMB#
& S(0/)(+)=& S()(+)2+$;A a) +66B# )) 6$MEB# a) +66B# )) 6$MEB#
F S()(/)2/E6A c) d)
F R(s)(/)2EM9A c)
F S()(/)2/E6A d) +FS$;EB# +66B#
F R(s)(/)2EM9A
6$MEB# -FO$SB#
& S(E0/)(E)=& S()(E)2E$M+A e) f) e) f)
& R(s)(E)=& R(-0/)(E)2/S;A
g)
h)
g)
h)
i)
A)
i)
A)
3ig$ O$ a)-A)K Stator and rotor current &ectors% full-load% 6VSRE and 6VDRE$
3ig$ S$ a)-A)K Stator and rotor current &ectors% full-load% +6VSRE and +6VDRE$
A".E (($
RS(+) LSl(+)
++E6JH (,DUC
(-,
!-
-R
SEIUE,CE C(RCU( CA.CU.A

&m()(+)
(-,
!%B*+0, (p-!C!"-sequence circuit, no-load5full-load (sD%, s!,sD%.<"

LSg(+) LRl(+)+ /RR(+)+ 6S()(+) &S()(+) &R(0/)(+)+ &R(0/)(+)
6$+6+T /$6/T ++/T 6$GT 6$6SMT +OMV +$;;;A %.$$<A +$EEEA 2>=A !%B*+0, (p-!C$"-sequence circuit, no-load5full-load (sD%, s$,sD-%.<"
RS(E) LSl(E) LSg(E) LRl(E)+ -/RR(E)+ 6S()(E) &m()(E) &S()(E) &R(-0/)(E)+ &R(-0/)(E)
APPE,D(Y " 3-R!A. CA.CU.A (-, -3 (,DUCED V-. A:ES >S he fault-related &oltages >S(p8+) appearing in 3ig$ E are generally e5pressed as follo7s (see C;D for their deri&ation)K
+ ; p j j Sp 6 = s G p+ e /
s
6$+6+T /$6/T /6$GT 6$;G;T -6$6SOT ;/$EV E$6FA %.><2A /$EA !%B9+0, (p-!C!"-sequence circuit, no-load (sD%, s!,dD%.%!AB"
RS(+)
2%2A
ULSl(+) ULSg(+) ULRl(+)+ RR(+)+0s+%# 6S(0/)(+) &m(0/)(+) &S(0/)(+) &R(s)(+)+ &R(s)(+)
("+)
#
6$+6+T +$6+T G;$/T 6$/GT E6/T SE$GV +$;;;A !.AA<A G$;'A !.!<A !%B9+0, (p+!C$"-sequence circuit, no-load (sD%, s$,dD%.%%<B"
RS(E)
E E E ULSl(E) U LSg(E) U LRl(E)+ RR(E)+0sE%# 6S(E0/)(E) &m(E0/)(E) &S(E0/)(E) &R(s)(E)+ &R(s)(E)
6$+6+T E$6FT E6$OT 6$SOFT S+;T SE$GV E$6FA $.%=A +6'A %.@?A !%B9+0, (p-!C!"-sequence circuit, full-load (s %.>$B, s!,dC!.=AB"
RS(+) ULSl(+) ULSg(+) ULRl(+)+ RR(+)+0s+%# 6S(0/)(+) &m(0/)(+) &S(0/)(+) &R(s)(+)+ &R(s)(+)
+ ; p j j Sp 6 = m # G p + e /
m
("/)
7hereK s / # G p += "s / # 6 K p + $ / ("E)
6$+6+T +$6+T G;$/T 6$/GT E$E+T SE$GV +$;;;A !.<2A 6$FMA ?<.>A !%B9+0, (p+!C$"-sequence circuit, full-load (s %.>$B, s$,dD%.=?B"
RS(E)
E E E ULSl(E) U LSg(E) U LRl(E)+ RR(E)+0sE%# 6S(E0/)(E) &m(E0/)(E) &S(E0/)(E) &R(s)(E)+ &R(s)(E)
6$+6+T E$6FT E6$OT 6$SOFT +6T
SE$GV E$6FA 2.@!A 6$OGA >=.AA
V$ C-,C.US(-, his paper &alidates )y si'ulations a sequence-circuit)ased 'odel of an eccentric-rotor cage 'otor 7ith parallelconnected stator$ he 'odel is &ery useful to de&elop a 'ethod for fault diagnosis$ Eqs$ (+;)% (+M) can )e for'ally re&ersed to o)tain eccentricity p$u$ &alues starting fro' the 'easure'ent of split-phase currents and passing through the transfor'ation (E)$ Bo7e&er an o)stacle 'ust )e o&erco'e% i$e$ the calculation of the sequence para'eters needed in (+;)% (+M) starting fro' the 'achine phase para'eters$ A si'ilar pro)le' has )een already resol&ed )y the author for /-pole COD and 'ultiple-polepair CMD synchronous 'achines$ his &ery interesting tas9 7ill )e de&eloped in future papers$ 3uture 7or9 also 7ill )e de&oted to e5peri'ental tests on a la)oratory 'achine$ !i5ed series-parallel stator connections can )e analy#ed )y follo7ing the guidelines gi&en here and in C;D$ he diagnosis 'ethod is useful for ()ut not li'ited to) traction 'otors$ /-pole 'otors requires a different approach% follo7ing the guidelines of COD$ APPE,D(Y AK !ACB(,E DA A a)le ((( reports the ratings of the ++E69H 'otor$
A".E ((($ poles frequency po7er efficiency 'a5 speed ++E6JH (,DUC
(-,
(n case of parallel-connected PP:s% the flu5 a'plitude 6 in ("E) is )ound to VLL through 62[/VLL0[E$ "y using the latter in ("E) and putting ("E) in ("+)% ("/) 7e o)tain the a'plitude of the fault-related &oltagesK 6 S
p + + Sp =6 = " s
pV L L K p +
/
("F) +
p + + Sp 6 S =6 = " #
m m
p V L L K p +
/
+ s $ p
("G)
,ote that )oth the SRE- and DRE-related &oltages ("F) and ("G) are su)stantially independent of the 'achine load$ he coefficients Kp8+ are su)-har'onic 7inding factors calculated in CMD$ RE3ERE,CES
C+D :$ Jle'pner and ($ Jers#en)au'% H;n#<((k$(f$L;+ge$)u+<(8Gene+;!(+ $ Ipe+;!i(n $ ;n# $ 9;in!en;n7e% (EEE Po7er Engineering Series% Hiley (nc$% ,e7 \or9% /66O$ C$ "ru##ese% ]3ield e5perience 7ith the split-phase current signature analysis (SPCSA)K Eccentricity assess'ent for a stand-alone alternator in ti'e-&arying and un)alanced load conditions%^ in P+(7' $ F666 $ 569@:@$:(nf'% Paris% 3rance% !ar$ ++-+/% /6+E% pp$ /GG-/;O$ C$ "ru##ese% ]Analysis and application of particular current signatures (sy'pto's) for cage 'onitoring in non-sinusoidally fed 'otors 7ith high reAection to dri&e load% inertia% and frequency &ariations%^ F666$ )+;ns'$Fn#'$6le7!+(n'% &ol$ GG% no$ +/% pp$ F+EM-F+GG% Dec$ /66O$ PinAia _hang% \i Du% Ba)etler% $:$% "in .u% ]A sur&ey of condition 'onitoring and protection 'ethods for 'ediu'-&oltage induction 'otors%^ F666$)+;ns'$(n$$Fn#'$&ppl'G Vol$ FM % ,o$ +% pp$ EF L F;% /6++$ C$ "ru##ese% -$ Bonorati% E$ Santini% ]Bar'onic current side)and-)ased no&el indicators of )ro9en )ars for on-line e&aluation of industrial and rail7ay cage 'otor faults^% F666$Fn!e+n;!i(n;l$S,mp(sium$(n$Fn#us!+i;l $ 6le7!+(ni7sG$FSF6$E44J% Vigo% Spain% ùne F-M% /66M$ C$ "ru##ese% ]Diagnosis-oriented sequence circuit-)ased 'odeling of eccentric rotor traction induction 'otors accounting for cage da'ping and split-phase currents%^ in P+(7' $ F666 $6P6:S113$ :(nf'% (nstan)ul% ur9ey% -ct$ /-F% /6+E$ C$ "ru##ese% ]Study of cardioid-shaped loop current space &ector traAectories for rotor eccentricity detection in po7er synchronous 'achines%^ in P+(7'$ F666$S@69P6@$:(nf' % "ologna% (taly% Sep$ G-O% /6++$ C$ "ru##ese% ]/-pole tur)o-generator eccentricity diagnosis )y splitphase current signature analysis%^ in P+(7'$ F666 $ S@69P6@ $ :(nf'% Valencia% Spain% Aug$ /M-E6% /6+E$
C/D
!-
-R
RA
(,:S
3ameplate 9ata F line-line &oltage +OM6V +66B# current F6OA ++E69H cosfi 6$S SGV slip*+66B# 6$MEV F666rp' torque E;66,' (achine Eeometry5*tructure5&indings core lenght F+6'' stator stac9 dia'$ ;SE'' air gap length /'' stator )ore dia'$ FEO'' Carter coeff$ +$FOG shaft dia'eter +;G'' Stator 7indingK three-phase% four-pole% lap-7ound% dou)le layer% 7hole-coiled 7inding 7ith ;6 open slots% G slots0pole0phase% G for'-coils per PP:% O turns per coil% four parallel connected PP:s in each phase% slot angular s9e72E;6o0G;% and coil pitch reduction2/ slots$ Rotor 7indingK fa)ricated copper cage% G; )ars% se'i-closed slots% no s9e7$ (achine Fhase Farameters RS (2RS(/)0F) /G$E'-h' LSl (2LSl(/)0F) 6$O6GE'B RRZ (2RR(/)+0F) //$M'-h' LRl+ (2LRl(/)+0F) 6$FF/F'B Lm (2LSg(/)0F) /G'B (esh-(odel 8ea#age and +esistance Farameters RPPG +6+$/'-h' LPPG E$//+'B R<;+ G/u-h' L<;+ +$+;OuB R+ing +E$MFu-h' L+ing 6$++;OuB
CED
CFD
CGD
C;D
CMD
COD
Modeling and Simulation of Rotor Bar Skew in Induction Motor Based on Modified Winding Function Method
M.Harir1, A.Chaouch2, A.Bendiabdellah1 and N.Benouzza1
L.D.E.E Laboratory, Department of Electrical Engineering, Faculty of Electrical Engineering, University of Sciences and Technology of Oran, Algeria 2 Signals and Systems Laboratory, Department of Electrical Engineering, Faculty of Sciences and Technology, University of Mostaganem, Algeria, 1 bendiazz@yahoo.fr 2 ikchaouchdz@yahoo.fr
1
Abstract This paper deals with the modeling and simulation of the squirrel cage induction motor by taking into account all space harmonic components as well as the introduction of the bars skew in the calculation of the linear evolution of the magnetomotive force (MMF) between the slots extremities. The model used is based on multiple coupled circuits and the modified winding function approach (MWFA). The effect of skewing is included in the calculation of motors inductances with an axial asymmetry in the rotor. The simulation results show the effectiveness and merits of the model and the error that may be caused if the skew of the bars are neglected. Keywords Inductances, Modeling, MWFA, Skew, Squirrel cage induction motor.
I. INTRODUCTION
TODAY
squirrel cage induction motors are widely used in variable speed industrial drives, so that the speed control, the optimal design and the faults diagnosis of these motors are very important [1,2,3,4,5,6,7]. For more information on the diagnosis of faults and the optimal design, authors [8,9,10] tend to move towards the introduction of the skew of bar in their motors modeling. Skew is applied to electrical machines in order to reduce undesirable effects such as cogging torques, higher-harmonic air-gap fields, torque ripple, vibrations and noise. The squirrel cage of an induction machine is skewed as to filter out the first significant field harmonics due to the slotting of the machine. The skewing of induction motors induces an electrical field in the azimuth direction, which can lead to additional currents, if the rotor bars are not sufficiently insulated with respect to the lamination. Several authors take this skew effect in their models according to the application to be used on one hand and the way the inductance variation is to be taken into account on the other hand. Simulation of induction motors with skew effects, using a 2-D time stepping finite element method (FEM) has been developed [11]. The 2-D FEM equations of each slice are then coupled together and solved simultaneously for
further refinement by authors in [12]. To consider the inter-bar currents in the skewed rotor bars, a multi-slice-networkcoupled FEM has also been proposed [13]. However, in all these formulations the axial length of each slice along the motor axis has to be equal. . G. Barakat [14] has also analysed the effect of the static eccentricity including the effect of the skewed slots of the rotor and the stator in the permeance function obtained by means of simulation using FEM. In this paper, the dynamic performance of the squirrel cage motor has been calculated by considering the various stator and rotor magnetizing inductances as well as the mutual inductances between the stator phases as variables and not constants. The variation due to the skew of bar is taken into account by the axial asymmetry in the rotor, thus using the modified winding function approach (MWFA) for our motor modeling. The simulation results obtained illustrate the effectiveness of the model and point out to the necessity of considering skew effects in some applications related to faults diagnosis and design optimization in induction squirrel cage motors. II. THE SQUIRREL CAGE INDUCTION MOTOR MODEL Consider an induction motor having m stator circuits and n rotor bars. The cage can be viewed as n identical and equally spaced rotor loops [2]. Voltage equations for the motor can be written in vector-matrix form as follows:
Vs Rs I s d s
dt
(1)
Vr Rr I r d r
dt
where
(2)
Vs v1s v s2 ...vsm
, V 0 0 ...0
T
r
(3)
s Is i1
r s r r is 2 ...i m , I r i1 i 2 ...i n
T
and the stator and rotor flux linkages are given by
s Lss I s Lsr I r r Lrr I r Lrs I s
X-Z axes orthonormal reference, as shown in Figure.1, where (4) it is possible to imagine a planar representation of the machine. It is clear that x, in this case, correctly represents a linear displacement along the arc corresponding to the angular opening and similarly for the case of xr [5]. (5) Knowing that N is the MMF per unit current, the flux seen (6) by the turns of a coil due to the current through another coil is given by the expression:
Lss is an m*m matrix with the stator self and mutual inductances, Lrr is an n*n matrix with the rotor self and mutual inductances, Lsr is an m*n matrix composed by the mutual inductances between the stator phases and the rotor loops, Lrs is an n*m matrix composed by the mutual inductances between the rotor loops and the stator phases and Lsr = Lrst. The mechanical equations for the machine are
BjAi 0
2 r l
N x, z, x n x, z, x g x, z, x i
1 Ai r Bj r r 0 0
Ai
dzdx
(12) The inductance between any two coils will be:
d r J TL Te dt d r r dt
(7)
LBjAi xr 0
2r l
N x, z, x n x, z, x g x, z, x dzdx
1 Ai r Bj r r 0 0
(13) In order to design for subroutines to calculate for the (8) various inductances of the machine, it is then possible to derive a new expression that may be easier to be translated
1
where r is the rotor position, is the angular speed, J is the rotor-load inertia and TL is the load torque. The machine electromagnetic torque Te can be obtained from the magnetic co-energy as:
into algorithm. Let us put P g x, z , xr , defined as the air gap permeance. In this case, the expression (13) of the mutual inductance can be rewritten as follows:
Wco Te r Is , Ir
LBjAi xr 0
2r l
(9)
Pn n x, z , x n x, z, x dzdx P P
Ai Ai r Bj r 0 0
The magnetic co-energy is the energy stored in the magnetic circuits and can be written as:
LBjAi x, z , xr 2rl 0 pn Ai nBj 2rl0
Pn Ai PnBj P
(15)
(14)
With nAi (x ,z, xr) being the distribution function of coil Ai (10) of winding A, and n (x ,z, x ) being the distribution function Bj r of a rotor mesh [4]. In the same way as in reference [6], and The precise knowledge of the inductances making up the according to the type of connection between the coils, the matrices in (5) and (6) is essential for the analysis and inductance is obtained by the sum of all mutual inductances simulation of the motor. In the next section a method for the between p and q winding coils A and B, respectively [7,8,9 ] .
Wco
1 t I s Lss I s I st Lsr Ir I rt Lrs I s I rt Lrr I r 2
calculation of such inductances while taking into account the skewing of rotor bars is being presented. III. THE INDUCTANCES CALULATION All obtained inductances of the system, are calculated by using the modified winding function approach (MWFA) [15]. If we consider the modified winding function given by the following expression:
q p Pn A PnB LBA x, z , xr 2rl 0 pn Ai nBj i1 j 1 P

(16)
PnAi nrj
1 2r lg 0 1 2r lg 0
2r l
n x, z, x n x, z, x dzdx
Ai r rj r 0 0
N,r n,r
1 n,r g1,r d (11) 1 2 g ,r 0
x2 j z2 j x
(17)
Ai r rj r
Then by carrying out the variable change
x / r and
PnAi nrj
x1 j z1 j x
n x, z, x n x, z, x dzdx
(18)
r xr / r , this will be equivalent as if one is referred to an
which gives :
LBA xr LBjAi xr
i 1 j 1
B. Inductance calculation with skewing bars (19) It is assumed that the machine always has a uniform gap, but having an axial asymmetry in the rotor translated by skewing bars. Knowing that the distribution functions in Figure.2 can be defined in two dimensions as follows:
A. Inductance calculation without skewing bars
N s x1i x x2i , 0 z x l (20) n Ai x, z , xr 0 in the remaining interval

1 x1 j x x2 j , 0 z x l nrj x, z , xr 0 in the remaining interval
Z nAi (x) z = l NsA
r
N x x x2i , 0 z x l nAi x, z, xr s 1i 0 in the remaining interval
(22)
(21)
1 x1 j x x2 j , z1 j x z x z2 j x nrj x, z , xr 0 in the remaining interval

(23) with
z=0 x1i
x1j
x2j
ri
, x1 j x x1 j r 0 z1 j x l r x x1 j r , x1 j x x2 j l x x1 j , x1 j x x1 j r z2 j x r 0 , x1 j r x x2 j
(24)
x2i (25)
Fig.1 Distribution function of the stator coil Ai
Analytical integration gives us the results consigned in Table 1. For each given interval, there is a value of the mutual inductance which depends on r.
TABLE I Mutual inductance without skewing bars
Such that is the angle of skew of the bars, and i x2i x1i / r represents the opening of the stator coil phase. When the mesh is partially in the field of the coil, the range of integration is reduced to the common area 'D' between the projection surface of the rotor mesh and that of the coil (the gray portion of Figure.2). Referring to expressions (22) and (23), it is clear that the integral in the remaining field is zero.
r (rad)
0 r
Lai r (H)
2
i1 r
0 rl N s r g0
xr
i1 r r i1 i1 r i 2 r i 2 r r i 2
0 rl N s r i1 r g0
zl
r
z1 j x
0 rl N s r g0
z0
x1i r s ri
z2 j x
x1 j
0 rl N s r i 2 2 r g0
rl 2 0 N s r g0
(D) x2i
x2 j
Fig.2 Skewing bar presentation.
i 2 r 2
The analytical integration gives us the mutual inductance between a stator coil and a rotor mesh. The results are illustrated in Table 2.
TABLE III Mutual inductance with skewing bars
r
0 r
(rad)
Lai r (H)
2
i1 ( r )
0 rl N s r g0
i1 ( r ) r i1 i1 r i1 r i1 r r i1
0 rl 2 2 N s r r i1 2 i1 r i1 r 4 r 2g 0
0 rl r N s r i1 r g 0 2
2 2 0 rl r r i1 Ns i1 r r i1 r r r g 0 2 2
i1 r i 2 ( r ) i 2 ( r ) r i 2
0 rl N s r g0
2 2 0 rl i 2 r r Ns r i 2 r i 2 r r r g 0 2 2
i2 r i2 r i 2 r r i 2
0 rl r N s r i 2 r 2 g 0 2
2 2 0 rl i 2 r Ns i 2 r i 2 2 r g 0 2
i 2 r 2
0 rl N s r g0
After the analytical calculation of the mutual inductance between all coils of a stator phase with a rotor mesh, we can plot their forms. In Fig.3.a, we can see the shape of
Lr1 A r which is far from a sinusoid. It should be noted also
that an analytical calculation of the first derivative of
Lr1 A r (in red color) may lead to a discontinuous function. While in Fig.3.b, we can see the effect caused by the inclusion of skewing bars. These effects occur in a sensible change in the form of the mutual inductance, and hence a clear change in its first derivative, particularly in terms of its form rather than in terms of its maximum value.
IV. SIMULATIONS RESULTS Once the squirrel cage induction motor model was obtained, a simulation program written in Matlab language was then developed. The proposed program enables us to highlight the behaviour of a squirrel cage induction motor and to compare its performance for both cases: with and without considering the bars skew effects. The simulation model of the induction motor gives us the electromagnetic torque curves shown in Figs. (4.a, 4.b) and the stator currents in Figs. (5.a, 5.b). From these curves, it can be easily noticed the improvements brought by the bars
(a)
(b)
Fig.3 Mutual inductance between the stator phase A and rotor mesh r1 (blue) and its first derivative with respect to r (red). (a)- without skew, (b)- with skew
(a)
Electromagnetic Torque (N.m) Electromagnetic Torque (N.m)
(b)
Time (s)
Time (s)
Fig.4 Electromagnetic torque. (a)- without skew, (b)- with skew
Current (A)
Current (A)
Time (s)
Time (s)
Fig.5 Current of stator phase.(a)- without skew, (b)- with skew
conditions. On the other hand, the steady state current skewing, in terms of peaks reduction in the transient mode and oscillations reduction in the start-up and steady state current fluctuations are less important with the introduction of the skewing bars effects and the shape of its waveforms is more sinusoidal. V. CONCLUSION In this paper, a proposed model of a squirrel cage induction motor is being presented. To take into account some of the secondary effects existing in a real motor, the model has included these effects by firstly considering all space harmonics due to the non-sinusoidal distribution of the air gap MMF and secondly all the effects produced by the skewing of bars as in our case. The simulation results obtained, enabled us to compare the motor performance for both cases; that is; with and without skewing effects and showed us clearly the effectiveness and merits of including skew effect in the model particularly for applications related to fault diagnosis and design optimization of motors. REFERENCES
[1] X.Liang ILochonwu.O, Induction Motor starting in Practical Industrial Applications, IEEE Trans. Ind. App, vol. 47, pp. 271-280, 2011. [2] A.Anugrah, R.Omar, M. Sulaiman and A.Ahmad, Fuzzy Optimization for Speed Controller of an Indirect Vector Controlled Induction Motor Drive using Matlab Simulink, JATIT, vol. 28, no.2, June 2011. [3] I.P.Georgakopoulos, E.D.Mitronikas and A.N.Safacas, Detection of Induction Motor Fault in Inverter Drives Using Inverter Input Current Analysis, IEEE Trans. Ind. Electronics, vol. 58, pp. 4365-4373, Sep.2011. [4] A. Braham and Z. Lachiri," Diagnosis of Broken Bar Fault in Induction Machines Using Advanced Digital Signal Processing", IREE, vol.5, pp. 1460-1468, July-August.2010. [5] S.Laribi and A.Bendiabdellah," Stator Short Circuit And Broken Bar Faults Diagnosis Of An Indirect Vector Control Squirrel Cage Induction Motor, IREE, vol.5, pp. 2088-2094, September-October.2010. [6] A.Chaouch and A.Bendiabdellah, " Mixed Eccentricity Fault Diagnosis in Saturated Squirrel Cage Induction Motor", International Review on Modelling and simulations (IREMOS) , Vol. 5, No. 3, pp.1216-1223, 2012 [7] A. Bendiabdellah, N.Benouzza, D.Toumi. Cage motor faults detection algorithme using speed estimation and current analysis. Acta Electrotechnica et Informatica No. 2, Vol. 7, 2007. [8] M. G. Joksimovic, D. M. Durovic and A. B. Obradovic, Skew and Linear Rise of MMF Across Slot Modeling-Winding Function Approach. IEEE Transactions on Energy Conversion , Vol. 14, no. 3, pp. 315-320, Sept.1999. [9] G. Bossio, C. De Angelo, J. Solsona, G. Garca and M. I. Valla, A 2DModel of the Induction Motor: An Extension of the Modified Winding Function Approach, 28th Annual Conference of the IEEE Industrial Electronics Society IECON 2002 , Sevilla Spain, 5-8 Nov. 2002. [10] G. Bossio, C. De Angelo, J. Solsona y G. Garca, Modeling of Induction Machines with Axial Non Uniformity, Congresso Basileiro de Electrnica de Potncia, COBEP2001 , Vol.2, pp 630-634. Florianpolis, Brasil, 2001. [11] A. Barbour and W. T. Thomson, Finite Element Study of Rotor Slots Design with Respect to Current Monitoring for Detecting Static Air-Gap Eccentricity in Squirrel-Cage Induction Motor, Proceedings of the IEEE-IAS Annual Meeting Conference, New Orleans, LA, Oct. 5-9, 1997, pp. 112-119.
[12] S. Williamson, T. Flack and A. Voschenk, Representation of Skew in Time_Stepped Two-Dimensional Finite Element Models of Electrical Machines, IEEE Transactions on Industry Applications , Vol. 31, No. 5, pp. 1009-1015, 1995. [13] C. McClay and S. Williamson, The Variation of Cage Motor Losses with Skew, IEEE Transactions on Industry Applications, Vol. 36, No. 6, pp. 1563-1570, 2000. [14] G. Barakat, G. Houdouin, B. Dakyo and E. Destobbeleer, An Improved Method for Dynamic Simulation of Air-Gap Eccentricity in Induction Machines, IEEE Sdemped 2001, pp. 133- 138.
Integrated planning model of generation and transmission expansion for an offshore oil-field power system
Dawei Sun, Xiaorong Xie
State Key Lab. of Power Systems Department of Electrical Engineering, Tsinghua University Beijing 100084, China
Jianfeng Wang, Che Wei, Guofeng Liu, Qiang Li

China National Offshore Oil Corporation CNOOC Beijing, 100027, China 4) Unlike inland system of relatively separate generation and transmission, OPPS has its generation and transmission closely connected which is built and operated by a single corporation. With all the characters above, the planning model of OPPS should be an integrated one, incorporating both the generation and transmission expansion planning. However the traditional decomposition of GEP, TEP and DEP is normally performed to make the very highly complex combined problem possible [3]. Studies of generator expansion planning models have been investigated largely in [4-7]. In [8], WASP-IV is presented to be powerful software developed by International Atomic Energy Agency (IAEA) in which a dynamic programming approach is employed to find an overall optimal required generation capacity for the network. In using WASP-IV, a single-node load center is assumed which is not obviously a valid assumption. In other words, while WASP-IV is capable of predicting the overall generation capacity requirements for the grid, it is unable to geographically distribute and allocate the capacities among the areas [9]. TEP has been investigated extensively in [10-12]. In [13], the fuzzy mathematic theory is proposed to model the transmission expansion process and Chaotic Particle Swarm Optimization Algorism is applied to calculate the problem. However, without generator expansion scheme, the transmission expansion planner cant make sure the generation information of buses. And this increases the difficulty of the problem and the probability becomes more inclined to make wrong decision. The paper is organized as follows. In section 1, the mathematic models of generators expansion, transmission expansion and integrated planning are presented. The proposed model is exemplified by an application to KL-BZ OPPS in section 2. Finally, conclusions are drawn in section 3. II. MATHEMATICAL MODEL
AbstractAn integrated model is presented in this paper to solve the planning problem of offshore oil-field power system. With the integrated model, planning of the generators and transmission system is derived simultaneously, which counteracts the disadvantages of separate planning, such as problem of uncertain power input for transmission system planning and unreasonable distribution of generators. And the model comprehensively considered the building cost, operation cost, loss and reliability in index to make optimal programming of power plants and transmission lines. Comparing with traditional integrated planning model, the proposed integrated model plans the system in a new order of firstly checking generators constrictions in order to save needless computer internal storage and calculates time. Certain cases have been suggested to verify the effectiveness of the model. Keywords- Power generation planning; Power transmission planning; Power system planning; Offshore oil field power system; Decision-making; N-1 principle
I.
INTRODUCTION
As a capital-intensive and primary-energy-intensive industry, electric power has a significant effect on national economic development. Since power system is regarded as the concrete representation of electric industry, one inappropriate system planning may result in unrecoverable loss of economy and time. On the contrary, a reasonable planning can make enormous economic and social benefit. Unlike the traditional power system planning, including generation expansion planning (GEP), transmission expansion planning (TEP), and distribution expansion planning (DEP) [1], is used for large-scale power systems on land. However the model proposed in the paper is to solve the planning problem of relatively smaller and islanded offshore oil-field power systems (OOPS). In comparison to power grid on land, the OOPS owns the following distinct characteristics [2]: 1) Unlike the planning of large land system which aims at improving the competitiveness, the planning of OOPS should pay more attention to consider problem of reliability. 2) OOPS can avoid problem of dimension disaster because of its fewer buses, therefore integrated optimization planning of generators and transmission lines is feasible. 3) Because of the high cost of transmission lines such as cables, the loop network is difficult to be formed and redial structure is commonly be used.
The proposed integrated model is based on two basic models, namely the generation expansion model (GEM) and the transmission expansion model (TEM). A. Generator expansion model GEM is specified by its planning variables, objective
This work was supported by the National High Technology Research and Development Program of China (863 Program) (Grant No. 2012AA050216).
function and various constraints. 1) The objective function The objective function is the sum of fixed investment, operating cost and penalty function, which is formulated as (1). S K I P C

(1)
Notes: max number of generators can be built in power plant i in year t. 4. Power balance constraint The power balance constraint requires the total power provided by all generators must meet the margin of all loads. X W C

It is the fixed investment formulated by (1b), which represents the cost of building new generators. The operating cost Ct must be calculated for all working generators including the built generators and the new ones. There are lots of items which should be considered in operating cost such as fuel, water, repair, salary and so on. The model simplifies those items into two large categories which are variable operating cost and fixed operating cost, which is shown in (1c). The variable operating cost consists of fuel and other material costs while fixed operating cost contains maintenance and repair expense. The penalty function Pt is formulated by (1d). The function is the link between reliable factors and objective function because the time of constraint violations, which is considered in the penalty function, reflects the reliability of the power system.
(5a)
C D 1 R N Z W
(5b)
Notes: Wi: unit capacity of power plant i; Dmt: load power forecast of year t; R: ratio of spare power;
5. Electric capacity balance constraint The constraint requires that the electric capacity demand of all loads can be satisfied by all the generating units. X W T E

(6a)
Notes: r: discount rate of money.
2) The constraints 1. All the decision variables are positive integers Obviously the generator expansion planning is an integer optimization problem and the new generators in plan Xti and the retired generators in plan Zij must be positive integer, which is formulated as follows: X 0, Z 0 (2) 2. Limit of total generating units in plan There must be a limited number of generators which can be built in each power plant: X Nm

(1d) P F V In the benefit evaluation of the project, the investment in construction cannot be compared with the operating cost by simple addition because the construction investment is a present value and the operation cost is an equivalent annual value. In order to compare the values, the coefficient ratio between present value and equivalent annual value is used to transform the value of construction investment into annual value with the following formula [18]. r1 r (1e) K 1 r 1
C b X b N Z

I a X

(1b)
E E 1 Er N Z W T
(6b)
Notes: Ti: annual utilization hours of power plant i; Emt: load electric capacity forecast of year t; Er: ratio of spare capacity.
(1c)
B. The transmission expansion model Regarding new lines as decision variables, traditional transmission expansion theory uses the investment as the only objective function to model the problem of planning transmission lines. Similarly, the paper proposes the basic transmission expansion model as follows. 1) The objective function Objective function of basic transmission expansion model is the sum of construction investment and operating cost, which is formulated as (7) FK
,
Note: F is the total cost of transmission expansion; P is same as Pt in (1d), which is the penalty function; cmn: construction cost of new line m-n; NAL: set of all lines should be built; NL: set of all lines; f: the coefficient of electricity price.
c n
,
S 8760 f P
(7)
2) The constraints 1. Power balance at each node Power generated at the node demand at the node + net power input of the node =0: P P P 0 (8a) P e G e B f

(3)
Notes: Nmi: max number of generators can be built in power plant i.
3. Limited number of generators can be built in a year Because of the construction and manufacturing capacity, infinite generators cannot be built during limited period. (4) X M
Notes: Gmn: conductance of line m-n; Bmn: susceptance of line m-n; en,fn two components of voltage of bus n.
f G f B e
(8b)
2. Power balance at each node under N-1 secure
P P P 0
condition
P
e G e B f

(9a)
Notes: Pmnmax: upper bound on real power flow over line m-n; Pmn: is the power flow of line m-n in N-1 secure test.
3. Limits on power flow over transmission lines: |P | P (10a) m, n N (10b) P P m, n N 4. Limits on lines built in the corridor 0 n n m, n N (11)
f G f B e
(9b)
Notes: nmnmax: upper bound on lines built in corridor m-n.
N-1 principle is used to determine the reliability of power system, which is also known as a single fault safety guideline. There are two N-1 secure test methods proposed in the paper. The first one is the traditional N-1 test, which checks the constrictions under N-1 situation. The traditional N-1 test will be passed only if the power system is of high reliability, which usually requires looped network. However, in offshore oil field system, its too expense to form ring because of the exorbitant construction cost of submarine cables. Therefore, a new N-1 secure test method is proposed for offshore oil field systems. The new N-1 secure test permits that power shortage can appear at only one bus under N-1 situation, while the other constructions such as power flows limit cant be violated. Aiming at load of different importance, the decision maker can choose different N-1 secure test methods for diverse requirements of system reliability.
3) The N-1 principle
Fig.2 Flow chart of the integrated model
C. Integrated planning model Based on the above two basic plan models, the objective
of integrated model is to get the building ordering of units and lines in the most economical way. The flow chart of the model
is shown in Fig.2. Compared with the traditional integrated model of generators and transmission system, the model plans units and lines in a better order. In traditional integrated theory [14], In planning theories, algorisms like GA and PSO are applied to the mathematic model. These algorisms generate many feasible solutions first, and then chose the better ones to optimize and finally get the optimal solution. In traditional integrated theory [14], variables and constrictions of generators and lines are created and considered simultaneously during the calculating process. And this may generate unfeasible solutions. Though the proposed model arranges solving order as the Fig.2 shows, which is to check the generators constrictions before transmission lines variables generation. Thus, unfeasible solutions that violate generation constrictions are impossible to appear. III. CASE STUDY
A. Overview of the studied system To explain the influence of the factors, a practical engineering case, which is called KL-BZ system, is studied with the proposed integrated model. The system is an offshore oil field power system with many drilling platforms. The newly-formed small power system consists of 7 buses, which is shown in Fig.3. All information of drilling platforms location, existed power plants and transmission lines is provided in Fig.3, which includes five existing platforms and two others to be constructed. Table 3 lists the allowable maximum number of new generates units at each bus.
Fig.3 Existing offshore oil field power system

Table 3 Allowable maximum number of new generate units Platform Bus number Maximum number of units BZ34-6/7 1 0 KL3-2 2 0 KL10-1(CEP+WHPB) 3 5 KL10-1WHPA 4 5 BZ29-4 5 0 BZ35-2(CEP+WHPA) 6 1
Fig.4 Final transmission system calculated by integrated model Notes: All new lines need to be built in 2013. Table 4 presents the optimal solution of generators expansion. It is obvious that they have the same total costs; the three results are only examples of many different solutions with the same costs. Therefore, further planning is needed. Fig.4 presents the transmission expansion result of the proposed integrated model, which is same as the original transmission plan, figure Xs dotted lines, given by KL-BZ system planner: China National Offshore Oil Corporation (CNOOC). Table 5 presents the effect of separate consideration of generators and transmission lines planning in integrated model. The unordered model, which is based on the traditional integrated theory, needs much more time to calculate the optimal solution compared with the proposed model in the paper. The two models obtain the same solution though the ordered model that firstly checks generators constrictions saves 150mins more than the unordered model. Table 6 presents the comparison of the integrated model solution and primary plan that proposed by CNOOC. As discussed above, the transmission expansion solution of integrated model is the same as the primary plan, though the result of generators planning is not the same. Result of integrated model adds two more buses and one more generate unit than original plan which doesnt consider the N-1 secure test as the other traditional integrated models. The solution of integrated model passes the new N-1 secure check, discussed in 1.2.3, which is proposed firstly for small scale and newly formed system, while the primary plan doesnt pass it because of inadequate generate units and power supply buses. Table 4
Comparison of generators expansion plans Plan result Before of generators coefficients Solution 1 Bus number 3 4 6 2013 2 2 0 2014 0 0 0 2015 0 1 0 2016 0 0 1 Total Cost/million 364.55 considering Solution 2 3 4 6 3 1 0 0 0 0 0 0 1 0 1 0 364.55 two Solution 3 3 4 6 3 1 0 0 0 0 0 1 0 1 0 0 364.55
B. Result analysis Solutions of the proposed integrated plan model of generators and transmission lines are presented in Table 4-6 and Fig.4.
Table 5 Comparison of traditional integrated model and the proposed one New Convergence Transmission Total lines time/min lines cost cost/million number /million Unordered 6 333 519.84 145 Ordered 6 183 145 519.84 Table 6 Comparison of final generators expansion plan Result of integrated model Primary plan
Bus Number year 2013 2014 2015 2016 Bus Number year 2013 2014 2015 2016
[4]
[5]
[6]
3 4
3 0
0 2
0 0
0 0
3 4
4 0
0 0
0 0
0 0
[7]
IV.
CASE STUDY
[8]
This paper proposes an integrated optimal planning model of generators and transmission expansion. It is appropriate for small scale and new-formed power system planning. And superiority of firstly checking generators constrictions in integrated model has been demonstrated.. The concept of N-1 secure test is firstly considered in integrated planning of generators and transmission system and a new N-1 test method is applied for OOPS. And the proposed integrated model provides the function of taking retirement into account at the same time. Finally, a practical engineering example, KL-BZ offshore oil field power system, was used to explain the superiority of integrated generation and transmission expansion planning model as compared to the separate approach and traditional integrated theory. REFERENCES
[1] [2] [3] X.Wang, J. R. McDonald. Modern Power System Planning[M].New York: McGraw-Hill, 1994. Zhang Qiang, Li Qiang. Research on Intelligent Power Grid of Offshore Oil Field[J].Automation Application, 2012, 7: 81-83. Mohammad Sadegh Sepasian, Hossein Seifi. A Multiyear Security Constrained Hybrid Generation-Transmission Expansion Planning.
[9] [10]
[11]
[12]
[13]
Referring to the document of traditional planning theories[J].IEEE TRANSACTIONS ON POWER SYSTEMS,2009,24(3):1609-1618. Elissa, Title of paper if known, unpublished. Mehmet Yildirim,Kadir Erkan,Senira Ozturk et al. Power generation expansion planning with adaptive simulated annealing genetic algorithm[J].International Journal of Energy Research,2006,30(14):1188-1199. J. Sirikum,A. Techanitisawad. Power generation expansion planning with emission control: a nonlinear model and a GA-based heuristic approach[J].International Journal of Energy Research,2006,30(2):81-99. Clodomiro Unsihuay-Vila,J.W. Marangon-Lima,A.C. Zambroni de Souza et al. Multistage expansion planning of generation and interconnections with sustainable energy development criteria: A multiobjective model[J].International journal of electrical power and energy systems,2011,33(2):258-270. F. Delgado,A. Ortiz,C.J. Renedo et al. The influence of nuclear generation on CO_2 emissions and on the cost of the Spanish system in long-term generation planning[J].International journal of electrical power and energy systems,2011,33(3):673-683. J. Sirikum, A. Techanitisawad, and V. Kachitvichyanukul. A new efficient GA-Benders decomposition method: For power generation expansion planning with emission controls[J].IEEE Trans. Power Syst., 2007, 22:10921100. Wei Benning. Method study of Power Plan Based on the Theory of Goal Programming[D].North China Electric Power University (Beijing), 2004 Ge Erli,Kazuhiro Takahasi,Luonan Chen et al. Transmission expansion cost allocation based on cooperative game theory for congestion relief[J].International Journal of Electrical Power & Energy Systems, 2005, 27(1): 61-67. Rong-Ceng Leou. A multi-year transmission planning under a deregulated market[J].International journal of electrical power and energy systems,2011,33(3):708-714. Enrique B. Cedeno,Sant Arora. Performance comparison of Transmission Network Expansion Planning under deterministic and uncertain conditions[J].International journal of electrical power and energy systems,2011,33(7):1288-1295. Yu Qiuxia. Study on Modeling and Decision-Making for Transmission Expansion Planning[D].Huazhong University of Science and Technology, 2008. H.M.D.R.H. Samarakoon,R.M. Shrestha,O. Fujiwara et al. A mixed integer linear programming model for transmission expansion planning with generation location selection[J].International Journal of Electrical Power & Energy Systems,2001,23(4):285-29.
Voltage Stability Assessment of a Real Power System Using a Detailed Dynamic Load Model
S. A. Al Dessi
Power Transmission Planning Department Dubai Electricity & Water Authority Dubai, UAE salha.aldessi@dewa.gov.ae
A. H. Osman
Department of Electrical Engineering American University of Sharjah Sharjah, UAE aosmanahmed@aus.edu
A. I. Ibrahim
Ex-Vice President- Power Transmission Planning Dubai Electricity & Water Authority awadai@yahoo.com
AbstractVoltage stability problem has become one of the major concerns for power utilities in recent years. This is due to the exponentially growing demands and the associated stress on the power transmission resources. Moreover, voltage instability has been responsible for severe network collapses world-wide and subsequently, the possible threat of voltage instability is becoming more pronounced in power utilities. Having an accurate load model capable of capturing load behavior during system disturbances is crucial in voltage stability assessment. This paper comprehensively assesses the voltage stability of a real power system using a detailed dynamic load model. Keywords-voltage stability; dynamic analysis; steady-state analysis; load model.
system disturbances, such as faults, transmission voltages may drop below certain thresholds resulting in either motor stalling or tripping. Excessive motor tripping or stalling may result in either voltage collapse or cascaded generator tripping, especially if the reactive power compensation facilities of the power system are not adequately sized. Therefore, proper modeling of power system loads, especially induction motor driven loads, is essential for voltage stability assessment of a power system. This paper comprehensively assesses Dubai Power Grid voltage stability margin against the increasing use of power transmission resources, growing demand and associated stress on available and planned active and reactive power resources; with a special attention to the special load nature of the city. II. LOAD MODEL REPRESENTATION
I.
INTRODUCTION
Currently, most of the power systems around the world are being operated under much more stressed conditions than were usual in the past. Environmental pressure on transmission expansions, exponentially growing demands and penetration of new types of loads, such as inverter-based appliances, at demand side are some of the responsible factors for these stressed conditions. Under such stressed operational conditions, a power system may exhibit instability behaviors that are characterized by either slow or sudden voltage drops, i.e. voltage instability. Under certain conditions, voltage instability may escalate to a form of voltage collapse which intimidates system security. This was evidenced in several network collapses and blackouts world-wide. Subsequently, voltage stability has become a major concern for power system utilities [1]. Dubai Power Grid, being planned and operated by Dubai Electricity and Water Authority (DEWA), is undergoing similar circumstances that made voltage stability a critical issue. The growing demands and associated stress on the available and planned system resources as well as the relatively limited geographical area of the city had resulted in allocating most generation stations at one side of the city, hence, feeding the load centers through long transmission circuits. Additionally, the electrical load of Dubai has a particular nature; a significant amount of the supplied load, especially in summer, is dominated by induction motor driven loads, specifically Air Conditioning (AC) appliances. During
Load model uncertainty was proved, in many publications to be the major source for simulation inaccuracy. Therefore, in any system stability study, especially voltage stability studies, it is necessary to model loads accurately. Having accurate load models capable of capturing load behavior during system disturbances enhances power system planners ability to anticipate potential risks and design power systems more precisely [2]. In the past two decades, load modeling has received a great deal of attention. Research results have shown that the voltage-dependent characteristics of bulk loads not only affect the voltage secure operating region of a system but also influence the speed of voltage collapse [3, 4, 5, 6, 7]. The observed impact of load-to-voltage dependency on power system voltage stability motivated further interest on the load modeling work. Yet, load is still considered as one of the most uncertain and difficult components to model due to the large number of diverse load components, to its high distribution, variable composition with time and weather, and also because of lack of precise information on the composition of the load [8]. There are two main approaches for load modeling and identification of load model parameters, namely, ComponentBased Approach and Measurement-Based Approach. The Component-Based Approach involves developing a composite
load model from information on its constituent parts, i.e., mix of classes at the substation, composition of each of those classes, and main characteristics of each single load component as illustrated in Fig. 1 [9]. On the other hand, the Measurement-Based approach involves direct measurements at representative substations and feeders. The data is obtained from Power Quality (PQ) monitors, Digital Fault Recorders (DFR), etc., and includes voltage and frequency variations, and the corresponding variations in active and reactive load, either to intentional disturbances (test measurements) or to natural events (normal system variation). By fitting the measured data to a model, the parameters of that load model are identified as illustrated in Fig. 2 [10].
This was done at transmission system level since the purpose of the load model is to assess the overall system voltage stability. However, for special load zone studies, load classification needs to be done at distribution system level. DEWA Electricity Statistics by year 2012 shown in Fig. 3 [11] has been used to perform load classification.
Figure 3. Number of Consumers of Energy Consumption for different Consumer Categories in Dubai by Year 2012
2) Identification of Load Composition: Load composition is identified by realizing the percentage of different component in each load class/mix. Resistive, Small Motor (SM), and Large Motor (LM) are identified using data obtained from published typical models. Unless better information is available, this typical data can be used in deriving the overall load model composition. Table I is an example of typical load compositions that utilities serve [8].
Figure 1. Component-Based Approach for Load Modeling TABLE I. TYPICAL LOAD COMPOSITION FOR DIFFERENT LOAD CLASSES Load Class Resistive Small Motor Large Motor Discharge Lighting Load Composition (%) Residential 25 75 0 0 Commercial 14 51 0 35 Industrial 5 20 56 19
In this study discharge lighting class was combined with resistive load class into a more general class called static load class. The identified load composition for Dubai Power Grid at transmission system level was identified as shown in Fig. 4.
Figure 2. Measurement-Based Approach for Load Modeling
In order to get a better description of the load characteristic, it would be optimal to combine both approaches. Therefore, a detailed load model of Dubai Power Grid was developed using a hybrid methodology that combines Component-Based and Measurement-Based load modeling approaches. The steps of the implemented hybrid load modeling methodology are briefly explained as follows: A. Component-Based Approach Processes Component-based approach was implemented on Dubai Power Grid by performing the following three processes: 1) Classification of Load into Classes and Mixes: The existing load substations were classified into classes and mixes including: Residential, Commercial and Industrial.
Figure 4. Load Composition for Dubai Power Grid-Transmission Level
3) Selection of Aggregate Load Model Structure: Due to the large proportion of induction motor loads, the selected aggregate load model structure for parameter
estimation consists of a static part (Constant Impedance, Constant Current Constant Power (ZIP)) in parallel with dynamic part (induction motor) as shown in Fig. 5.
The recorded incidents collected from the available DFRs were used to identify load tripping rules with respect to voltage variation. The amount of tripped load during voltage variations was correlated to the magnitude of the voltage dip (Fast Tripping) as shown in Fig. 8, and to the dip duration (Slow Tripping) as shown in Fig. 9. After identifying all load model parameters, the load model was validated against recorded large voltage variation incident and found capable of representing the dynamic load behavior during voltage disturbances. Thus, the developed load model is used for performing dynamic voltage stability analysis of Dubai Power Grid.
Figure 5. The Selected Aggregate Load Model Structure
B. Measurement-Based Approach Processes The use of these actual measurements was deployed in two processes [12]: 1) Aggregate Load Model Parameter Estimation: An extensive set of monitored natural load versus voltage variations was collected from the available Digital Fault Recorders (DFRs) installed at different load buses were used for parameter estimation of the selected aggregate load model structure. The measured instantaneous three phase voltage and current signals of a sample recorded incident are shown in Fig. 6. The instantaneous samples of voltage and current signals are first converted into phasor form, corresponding to their fundamental frequency, using Discrete Fourier Transform (DFT). Then, the positive sequence voltage and currents are calculated using Fortescue transformation. After that, the positive sequence active and reactive power were used to estimate load model parameters using least square method. Fig.7 shows a sample of the estimated versus measured active and reactive power for the same incident.
Figure 8. Fast Load Tripping Scheme
Figure 9. Slow Load Tripping Scheme
III.
Figure 6. Measured Instantaneous Voltage and Current Signals
VOLTAGE STABILITY ASSESSMENT METHODOLOGY
Figure 7. Measured Versus Estimated Power and Active Power
2) Identification of Load Tripping Rules:
The comprehensive methodology for voltage stability assessment of power systems using modern analytical tools addressed in [13] was implemented to assess the voltage stability of Dubai Power Grid. The methodology endorses performing both steady-state and dynamic voltage stability analyses, with high emphasis that the latter becomes very critical especially for a power system with a significant share of motor loads. Steady-state voltage stability analysis is often used to analyze slower form of voltage instability making use of power flow simulation as a primary study method. It is commonly tackled by plotting PV and VQ curves. On the other hand, dynamic voltage stability analysis describes the time-dependent voltage performance of the system by revealing the transient and/or the longer-term voltage stability of a power system under study following system disturbances
[14]. The time frame of interest is 10 seconds after the inception of the disturbance, since this period is enough to capture the induction motor dynamics, which are the dominant load type of Dubai Power Grid. Therefore, the use of dynamic voltage stability analysis term denotes short-term or transient voltage stability analysis. Long term voltage stability analysis is not addressed. For steady-state voltage stability analysis, short term transients and dynamics can be ignored; therefore, static load models are exclusively employed as their effects dominate during a voltage disturbance under study. Static load model may consist of a combination of three load components: Constant Impedance (Z), Constant Current (I) and Constant Power loads (P). Constant Power loads maintain a constant power draw from the system regardless the change in voltage. Therefore, constant power loads tend to aggravate a voltage collapse condition. On the other hand, the power drawn by constant current and constant impedance loads decreases with voltage drop providing load relief and hence, better voltage recovery [14]. In this paper, for steady-state voltage stability analysis constant power load is assumed, at all load buses, as a worst case scenario. For dynamic voltage stability analysis, the detailed load model was implemented in order to incorporate the impact of the large proportion of induction motors and reveal if there is any associated transient or quasisteady-state voltage problems that are not captured by steadystate voltage analysis. PSS/E Software Package was used to conduct both steady-state and dynamic voltage stability analysis. Steadystate and dynamic voltage stability analyses were performed for Dubai Power Grid considering various disturbance scenarios. The selected disturbances cover: N-1 contingencies including loss of one circuit or loss of one transformer. N-2 contingencies including loss of two circuits, or loss of one transformer and one circuit (equivalent to the loss of one bus section). N-3 contingencies including loss of two circuits and one transformer. A. Steady-State Voltage Stability Analysis The steady-state voltage stability analysis was performed based on the criteria established by Western Electricity and Coordinating Council (WECC) of North America. The document that was used as a guide for this study was approved by WECC in May of 1998 [15]. The active power transfer limit for each contingency scenario is calculated from the nose of the corresponding PV curve. Then, for each contingency category the active power transfer limit of the worst contingency among this category is considered as the maximum active power transfer margin. According to WECC voltage stability criteria, a safety margin of 5% is taken for N-0 and N-1 contingency conditions and 2.5% for N-2 and N-3 contingency conditions. Fig. 10 illustrates the process of calculating the maximum active power transfer margin.
Figure 10. Calculating Active Power Transfer Margin from PV Curves
The reactive power reserve margin for each contingency scenario is calculated from the knee of the corresponding QV curve. For each contingency scenario, the reactive power margin for the base load and a load increase of 5% is calculated. The change in reactive power margin between base load and base load +5% is calculated, and the contingency having the maximum margin change is considered the worst contingency of its category. This means, under this contingency, the system is very sensitive to load change, i.e. prone to voltage instability. This process is illustrated in Fig 11.
Figure 11. Calculating Reactive Power Reserve Margin from VQ Curves
B. Dynamic Voltage Stability Analysis Transient dynamic voltage analysis was performed considering the worst case contingencies revealed from the steady-state voltage stability analysis. The disturbances were simulated to reflect considering practical scenarios including faults cleared in normal fault clearance time and backup protection clearance time. PSS/E software package was used for simulating these disturbances. The simulation starts with normal system operation for 1 second duration, then the fault is applied at t = 1.0 second. The fault is then cleared after 100 ms (for normal fault clearance time scenarios) or 280 ms (for breaker failure scenarios). After fault clearance the elements outages according to the study scenarios take place. The simulation then is carried out for 20 seconds to monitor system behavior following the simulated disturbance. During the simulation, the voltage trends at the 400 kV and 132 kV buses of the substation under study were plotted to screen any
oscillatory response, overvoltage and or voltage recovery problem resulting from the applied disturbance. IV. RESULTS
A. Steady-State Voltage Stability Analysis Fig. 12 shows the PV curves plotted for one of the main load substations under normal operational conditions (Base Case N-0) and the worst N-1, N-2 and N-3 contingencies. It is obvious that all the plotted PV curves for this substation fall within the acceptable voltage limits. It is also found that this substation can be loaded to its firm capacity for all studied scenarios. Table II presents the calculated maximum active power transfer margin for the base case and all the contingency scenarios. The figures shown are after implementing the 5% safety margin for N-0 and N-1 contingencies, and 2.5% safety margin for N-2 and N-3 contingencies.
Figure 13. QV Curves for a Main Load Substation TABLE III. REACTIVE POWER RESERVE MARGIN FOR A MAIN LOAD SUBSTATION Available Reactive Power Reserve Margin (RRM) (MVAr) Base Case (N-0) Worst N-1 Worst N-2 Worst N-3 1014 796 637 517 Reference RRM > 50% of RRM of N-0 > 50% of RRM of N-0 >0
Contingency Type
WECC Criteria
Figure 12. PV Curves for a Main Load Substation TABLE II. ACTIVE POWER TRANSFER MARGIN FOR A MAIN LOAD SUBSTATION Contingency Type Base Case (N-0) Worst N-1 Worst N-2 Worst N-3 Maximum Power Transfer Margin (MW) 1358 1186 1199 1168
Steady-state voltage stability analysis shows that the maximum power transfer margins and reactive power reserve margins for all load substations of Dubai Power Grid are compliant with WECC voltage stability. B. Dynamic Voltage Stability Analysis It is essential to use the developed dynamic load model for performing dynamic voltage stability analysis in order to assess the impact of the high proportion of AC appliances on voltage stability of Dubai Power Grid. Sample results for one of the main 400/132 kV load substations are presented thereafter. Fig. 14 show the voltage trends during the worst contingencies for normal fault clearance time at the 400 kV bus of the selected load substation.
Fig. 13 shows the QV curves plotted for one of the main load substations under normal operational conditions (Base Case N-0) and the worst N-1, N-2 and N-3 contingencies. The curves are plotted for base load and a load increase of 5% in order to calculate the reactive power reserve margin for base case and all worst contingency scenarios as shown in Table III. It is found that the calculated reactive power reserve margin for each category is compliant with WECC voltage stability criteria.
Figure 14. Voltage Trends at the 400 kV Bus for all Worst Contingencies (Normal Fault Clearance Time)
It was observed that for all simulated disturbances where the triggering fault cleared in normal fault clearance times, there is a fast voltage recovery and a slight transient voltage rise at the 400 kV and the 132 kV buses. However, the voltage returns to the acceptable limit within 10 seconds. Fig. 15 shows the voltage trends under the worst contingencies for backup protection clearance time and at the 400 kV bus of the selected load substation.
disconnection behavior of new load types needs to be incorporated in power system load model. VI. CONCLUSIONS The developed detailed load model revealed two problems related to voltage stability. First, transient and/or quasi steadystate voltage rise was encountered for normal cleared faults. Second, slow voltage recovery after extended fault durations and/or severe contingencies that were also evidenced during real recorded incidents. Both problems could be worsened with higher motor load proportion, more severe contingencies, and more severe fault conditions. Moreover, the dynamic simulation proved that steady-state voltage stability analysis is not sufficient to assess the voltage stability of a system with high proportion of motor loads. REFERENCES
[1] T. V. Cutsem, C. Vournas, Voltage Stability of Electric Power Systems, 2nd. ed., USA: Kluwer Academic Publishers, 2001. [2] Electric Power Research Institute, "Measurement-Based Modeling," EPRI, California, USA, 2006. Load
Figure 15. Voltage Trends at the 400 kV Bus for all Worst Contingencies (Backup Protection Clearance Time)
[3] Pal, M. K., "Voltage Stability Conditions Considering Load Characteristics," IEEE Transactions on Power Systems, vol. 7, no. 1, pp. 243-249, February 1992. [4] A. Borghetti, R. Caldon, A. Mari, C. A. Nucci, "On Dynamic Load Models for Voltage Stability Studies," IEEE Transactions on Power Systems, vol. 12, no. 1, pp. 293-303, February 1997. [5] K.i Tomiyama, J. P. Daniel, S. Ihara, "Modeling Air Conditioner Load for Power System Studies," IEEE Transactions on Power Systems, vol. 13, no. 2, pp. 414-421, May 1998. [6] w. Xu, E.Vaahedi, Y. Mansour, J.Tamby, "Voltage Stability Load Parameter Determination from Field Tests on B.C. Hydro's System," IEEE Transactions on Power Systems, vol. 12, no. 3, pp. 1290-1297, August 1997. [7] K. Morison, H. Hamadani, L. Wang, "Practical Issues in Load Modeling for Voltage Stability Studies," 2003 Power Engineering Society General Meeting, vol. 3, pp. 1392-1397, 2003. [8] K. Morison, H. Hamadani, L. Wang, "Load Modeling for Voltage Stability Studies," in 2006 IEEE Power Systems Conference and Exposition, 2006. [9] Kundur, P., Power System Stability and Control, USA: Mc Grow Hill Inc., 1994. [10] P. Z. V. H.Bai, "A Novel Parameter Identification Approach via Hybrid Learning for Aggregate Load Modeling," IEEE Transactions On Power Systems, vol. 24, no. 3, pp. 1145-1154, 2009. [11] Dubai Electricity and Water Authority, "Dubai Electricity and Water Authority Official Web Page," [Online]. Available: http://www.dewa.gov.ae/aboutus/electstats2012.aspx. [12] P. Z. V. H.Bai, "A Novel Parameter Identification Approach via Hybrid Learning for Aggregate Load Modeling," IEEE Transactions on Power Systems, vol. 24, no. 3, pp. 1145-1154, 2009. [13] S. A. Al Dessi, A. I. Ibrahim, A. H. Osman, "A Comprehensive Methodology for Voltage Stability Assessment of Power Systems Using Modern Analytical Tools," in International Conference on Electric Power and Energy Conversion Systems (EPECS), Sharjah, 2011. [14] (RRWG), Reactive Reserve Working Group, "Guide to WECC/NERC Planning Standards I.D: Voltage Support and Reactive Power," Western Electricity Coordinating Council, 2006. [15] Reactive Power Reserve Work Group (RRWG), "Voltage Stability Criteria, Under Voltage Load Shedding Strategy and Reactive Power Reserve Monitoring Methodolgy," Wester Electricity Coordinating Counci l (WECC), 1998.
For all the simulated disturbances where the triggering fault cleared in backup protection time, it was observed that the voltage recovery at the 400 kV bus is slow. However, the voltage returns to the acceptable limit within maximum time duration of 20 seconds at the 400 kV bus. Although the voltage recovery is slow, it ends up with a voltage rise above acceptable limits for both N-2 and N-3 contingency scenarios. This is probably due to the fault duration which is sufficient to cause motor stalling at the low voltage side causing the slow voltage recovery, and undoubtedly these motors eventually got tripped by their thermal protection causing steady-state voltage rise. V. RECOMMENDATIONS In view of the escalating stress on most of the power systems around the world and the associated growing concern in voltage stability problems, it is necessary to perform accurate voltage stability studies in order to anticipate potential risks and design/operate power systems more appropriately. Therefore, it is recommended for all power system utilities, to develop accurate load model based on real system data and measurements, in order to enhance the accuracy of voltage stability studies. Power systems that are characterized by high proportion of Air Conditioner (AC) load are exposed to shortterm voltage instability, fast voltage collapse, and delayed voltage recovery. Therefore, the dynamic characteristic of AC load and their protection must be adequately represented in voltage stability studies. Furthermore, the new generation of electric appliances that employ power electronics (such as inverter-based appliances) are easily disconnected from the power system following voltage dips causing a significant rise in post-disturbance voltage, and therefore, the self-
Adaptive PID Controller Design by Using Adaptive Interaction Approach Theory

1
1, 2
Tayfun GUNDOGDU
Guven KOMURGOZ
Istanbul Technical University, Faculty of Electrical-Electronics Engineering Department of Electrical Engineering, 34469-Maslak Istanbul, TURKEY 1 2 tgundogdu@itu.edu.tr komurgoz@itu.edu.tr
AbstractA self-tuning algorithm for PID controller based on adaptive interaction approach efficiently used in the Artificial Neural Networks (ANNs) is proposed in this paper. The principle behind the adaptation algorithm is mathematically isometric to the back-propagation algorithm (BPA). By applying Adaptive Interaction (AI), the same adaptation as the well-known BPA can be achieved without the need of a feed-back network. Hereby, by using AI tuning algorithm, the ANN PID controller can be adapted directly without wasting calculation time in order to increase the frequency response of the controller. Speed control of a DC motor under the rapidly changing load condition is simulated to demonstrate the sensitivity of the AI algorithm. PID gains of the ANN controller was tuned directly by using AI tuning algorithm. Simulation results and PID adaptation process have been presented. Keywords- Adaptive Interaction, adaptive neural network, selftuning, PID controller, DC motor control.
difficult to determine the appropriate PID gains when various uncertainties and nonlinearities exist, such as payload variations, hysteresis, friction and others. All these factors could degrade the control system performance. Several tuning methods and discussions were given in [4]. In addition, numerous research papers focused on adaptive PID control [5], self-tuning PID control [6, 7], self-tuning predictive PID control [8] and so on can be found in the existing literature. In adaptive and selftuning PID, the controller parameters were tuned in accordance with changes of the process parameters automatically [5, 7]. ANNs are used to modify the gains of PID controllers [9, 11]. The BPA, which is commonly used in adaptive control systems, cannot be applied directly to NN controllers. Due to BPAs reliance on a dedicated feed-back network to propagate the error back; the system must consist of pure neurons. Nevertheless, instead of BPA, the AI algorithm can be used to the same effect as the BPA [12, 13]. However, AI algorithm is mathematically equivalent to the BPA. ANN controller (ANNC) can be adapted directly without approximating the plant by using AI algorithm. This not only eliminates the error in approximation, but also significantly reduces the complexity of design. Using this theory, the DC motor controlled by self-tuning PID is decomposed into four subsystems consisting DC motor, proportional (P), integral (I), and derivative (D) controller. The parameters of the PID controller , , and are viewed as the interactions between these four subsystems. A simple and effective adaptation algorithm developed in the theory of adaptive interaction is applied to self-tuning these coefficients. The only information required about the plant is its Frchet derivative, which can be easily replaced by a constant that is absorbed into the adaptation coefficient for linear systems, to apply this self-tuning algorithm [12]. In this study, velocity control of a DC motor under a rapidly changing load condition is analyzed by applying this AI algorithm to tune the PID gains on-line. The reference speed, actual speed, error and PID gain curves for the speed control are presented by using MATLAB Simulink. There are many different studies on the self-tuning PID algorithms and control of a DC machine in the existing literature. But, modified AI algorithm, which works very well and gives excellent results, have not been used in any of them to tune the PID gains as the present study does.
I.
INTRODUCTION
ANNs have been one of the most interesting topics in the control community because of their high ability to treat many problems that cannot be handled by traditional analytic approaches. ANNs have found applications in many engineering fields especially in control engineering like linear or nonlinear system identification [1]-[3] and also linear or nonlinear adaptive control [4]-[10]. Conventional PID controller is used extensively in industrial process due to its simple structure, high reliability and easy to implement. However, this controller has many limitations like determination of the PID gains, plant model and situations changes and etc. When the controlled object is changed, the parameters cannot adjust automatically to adapt to the changed condition [3]. With these problems, the limitations of traditional PID controller exposed day by day. PID controllers are classified into two categories according to their tuning algorithm as: initial off-line tuning and continuous online self-tuning. The second class will be focused on in this study, since the objectives and hence requirements of the PID controller often change during the different stages of control and the plant to be controlled often changes from time to time. For DC motor control, the PID controller is one of the most popular methods, since its design is well known for simple systems and does not require detailed knowledge of the system dynamics [3]. However, when using PID structures, it can be
II.
MODEL OF DC MOTOR
DC motors have been widely used in engineering due to their simple structure and ability to be easily integrated in control systems with relatively low cost. The traditional model of a DC motor is a second-order linear one. In this paper, a DC motor is controlled via the input voltage. The control design and theory for controlling a DC motor via current is nearly the same. For simplicity, a constant value as a reference signal is injected to the system to obtain a desired position. The transfer function, including the effect of the rapidly changing load torque , from the input voltage () to the angular velocity, is:
Gv ( s ) w( s ) VT ( s ) ( Ra La s )( Js B ) Kt K e ( Ra La s )TL Kt Ra La s TL
xn t un t c y prec t , n N
c In
(4)
where are the connection weights. With this linear interaction, the dynamics of the system is described by (5).
yn t Fn un t c y prec t , n N
(5)
c In
(1)
The goal of the adaptation algorithm is to adapt the connection weights so the performance index E(y1,....yn , u1,.......un) as a function of the inputs and outputs will be minimized by using the Frchet derivative [14]. The system with dynamics can expressed as in (6).
where is the armature resistance, is the armature inductance, is the moment of inertial of the motor rotor and load, B is the damping ratio of the mechanical system, is the load torque, and and are the torque and back EMF constantans, respectively. III. THEORY AND BACKGROUND OF INTERACTIVE ADAPTATION
yn Fn un c y prec , n N
(6)
c In
If connection weights are adapted according to simplified adaptation algorithm as given in (7) [15].
c c F ' postc x postc y postc
y pre
s s
s O post
c
Depending on the application and configuration of the algorithm, the adjusted coefficients can be ANN weights, PID gains or transfer function coefficients. The theory of IA considers N subsystems called devices. Each device (indexed by n N = {1,2,,N}) has an integrable output signal yn and an integrable input signal xn. The dynamics of each device is described as a causal functional as given in (2).
F ' postc x postc y pre c
E y postc
(7)
Fn X n Yn , n N
(2)
where , and the above equation has a unique solution, then the performance index E will decrease monotonically with time. In fact, (7) is always satisfied, which is the unique solution to (8) where > 0 is the adaptation coefficient [16].
where and are the input and output spaces respectively. Therefore, the relation between input and and output of the nth device is given by (3).
dE d c
,c C
(8)
yn t Fn xn t Fn xn t , n N
where denotes functional composition.
Prec1 Device 1
C4 C2 C1 C3
(3)
Using this algorithm, an ANN can adapt without the need of a feed-back network to back-propagate errors. The algorithm hence provides a biologically plausible mechanism for adaptation in biological neurons. IV. ADAPTIVE NEURAL NETWORK CONTROLLER WITH TUNING ALGORITHM
Postc1 Device 2
C5
Device 3 Device 4 Device 5
An ANN is decomposed into multiple devices, which are taken in the system as neurons, as described in Fig. 1 and shown in Fig. 2(a). The ANN can be described as in (9) by ignoring the dynamics.
Figure 1. Devices and their connections.
Interactions among devices are achieved by connections which are denoted by c as seen in Fig. 1. Devices whose output is conveyed by connection c is denoted by prec n and a device whose input depends on the signal conveyed by c is denoted by postcn. In ={c:prec=n}is the set of input interactions for the nth device , On ={c:postc=n}is the set of output interactions for the nth device. Assuming linear interaction among devices and the external signal (), that is,
vi g hi g wij v j i , i N
N
(9)
where ( ) is the sigmoidal function; is the output of the neuron i; is the input of the neuron i; i is the external input of neuron i; is the weight of the connection from neuron j to neuron i. For output neurons, denoted ti is the desired output of neuron i. The goal is to minimize the error as in (10).
(a)
Self-Tuning PID Controller
(b)
Simplified schema of the controller. Figure 2: NN based control system.
2i
ei , ei
2 N
vi ti , if i is an output neuron (10) 0, otherwise
neuron can be built as a standard unit. In Figure 2(b), are weights which are adjusted by error and logsig is the sigmoidal function and it is equal to (12).
g x x
Applying the adaptation algorithm to ANN, the following substitutions are obtained. [ ] ( )
1 e
x 1
(12)
Mathematically, the NN and adaptation algorithm are described as follows like (9). 1 d 2 pn ws rpre n ws ws ws (13) s 2 dt sDn sDn sDn
Therefore, the adaptation algorithm is given as in (11).
wij g ' hi
vj vi
wki wki g ' hi v j ei

k N
By applying the adaptation law in (7), the weight adaptation becomes as in (14). (11)
ws rpres post s ( p post s ) f post s
(14)
The algorithm (11) is mathematically equivalent to the BPA. However, it does not require a feed-back network to propagate the errors. By eliminating the feed-back network, the algorithm allows a much simpler implementation than that of the BPA. Using the AI algorithm, the adaptation mechanism can be built within each neuron to make the neuron trainable. A trainable
where; n is the label for a particular neuron, s is the label for a particular synapse, and is the set of input synapses of neuron n, and are the presynaptic and postsynaptic neuron corresponding to synapse s, respectively, is the weight of synapse s, is the membrane potential, is the firing rate.
Equation (13) and (14) describe AI algorithm for adaptation in NNs. As shown in [16], it is equivalent to the BPA but requires no feed-back network to back-propagate the error. Fig. 2(b) is obtained through this algorithm. The ANN controller (ANNC) has five inputs: 1 , 2 , 3 , 4 and as shown in Fig. 4(a). 1 is the error between the set point and the output and 2 , 3 , 4 are delayed signals based on e1. is the learning rate. Delayed signals are introduced in order to control the output to depend not only on the current input, but also on past inputs, since the network controller is itself a memory-less device. Mathematically, the input-output relations of neurons can be calculated with the help of Fig. 2(b). The aim is to minimize the system error (e) as presented in (15).
1000_rpm (104.72 rad/s) DC motor having the following parameters are used: Ra = 4.2 , La = 25.2 mH, J = 0.044 kgm2, B = 0.03 Nm/rad/s, K = 2.36 Nm/A [17]. And a random load whose sample time is variable and changes between 5 to 30Nm is applied to the DC motor to observe the adaptation. In order to model the system more realistic, sensor noise which is a white noise with 0.12 noise power, 0.3 sample time and seed : [23341] is taken into account as seen in Fig. 2(a). The modified tuning algorithm is applied to a DC motor with constantly changing load torque. Sample time of load torque is 4 seconds and its magnitude changes 2 N.m to 33 N.m as shown in Fig. 3.
35 30
y where, r is the reference and y is the system output. Weights are calculated as follows by using (13) and (14) as follows:
1 = 1 (3 (3 ) + 0 ) = 1 3 (3 ) 2 = 2 (3 (3 ) + 0 ) = 2 3 (3 ) 3 = 1 (4 (4 ) + 0 ) = 1 4 (4 ) 4 = 2 (4 (4 ) + 0 ) = 2 4 (4 ) 5 = 1 (5 (5 ) + 0 ) = 1 5 (5 ) 6 = 2 (5 (5 ) + 0 ) = 2 5 (5 ) where
E e1 ( r y )
2 2
2 r y 2e1
Load Torque [Nm]
(15)
25 20 15 10 5 0 0 20 60 80 Time [s] Figure 3. Variation of the Load Torque. 40 100 120
3 w7 w7 , 4 w8 w8 , 5 w9 w9
(16)
The adaptation law for 7 , 8 and 9 is more complicated as it is linked to the plant to be controlled. By using (7), since is empty adapted weight is given in (17).
To be able to compare against the Self-Tuning PID, a conventional PID is designed to control the speed of the DC machine. The final stage indicates the control of Self-Tuning PID while the initial stage indicates the control of conventional PID controller. Constant-gains of the conventional PID controller are determined as = 41.8 , = 7.5, = 0.32 by using Ziegler-Nichols method.
w7 F ' postc u r3 2e1
(17)
If the Frchet derivative is approximated by a constant that will be absorbed in , then the above expression is approximated by (18).
w7 r6e1 , w8 r7e1 , w9 r8e1
(18)
The constant , which is varied to analyze the rate adaptation of the ANNC, is considered as the learning rate. Therefore, decomposed subsystems (four devices) are as follows: Device-1 is the proportional part with transfer function 1; Device-2 is the integral part with transfer function s-1; Device-3 is the derivative part with transfer function s; and Device-4 is the DC motors transfer function given in (1). In any case there are three adaptive connections: = , , . , and may be retyped as below by as using (19).
Fig. 4. Speed Error.
K p r3e1 ,
V.
K I r4e1 ,
K D r5e1
(19)
As shown in Fig. 5 the rise time and overshoot are excellent. Furthermore, PID gains which are obtained from the ANNC are given in Fig. 6. As seen in Fig. 6, P gain shows greater change according to the situation of current speed and load. At the initial stage, the maximum control error is about 0.70 5rad/s (6.73 rpm) and average error is 0.074 rad/s (0.706rpm) for 1200 points, and finally max control error is reduced below 0.2 rad/s (1.91 rpm) and average error becomes 0.011 rad/s (0.105 rpm) for 1200 points. The speed error curves for initial and final stages are given in Fig. 4. As seen in Fig. 4, there were sharp and large changes in the error according to speed and load torque change at the initial stage. However, these large errors are reduced and the sharpness of the errors is prevented.
SIMULATION RESULTS
In order to simulate the behavior of the prepared self-tuning PID, MATLAB Simulink model of the proposed method is performed. The reference speed, actual speed and error curves for the speed control of the DC motor, whose transfer function is given in (1), are determined using the Simulink models under the steady state condition. The motor speed is controlled so as to follow the reference speed, which is designed in MATLAB via a signal generator. In this study, 240 V, 12.7 A, 30 Nm,
While the proposed Self-Tuning PID absorbs the load change impact very smoothly, the conventional one cannot, as seen in Fig. 4. Fig. 5 shows the results of the simulation with the voltage input being a velocity reference signal, changing between 0 to 105 rad/s. The adaptation coefficient is taken as 15. PID gains which are obtained from the ANNC are given in Fig. 6.
constantly changing load conditions. This new approach does not require the transformation of the continuous time domain plant into its NN equivalent. Another benefit for applying the proposed algorithm is that it does not require a separate feedback network to back-propagate the error. The adaptation algorithm is mathematically isomorphic to the BPA. Properly designed controller with fixed gains may have sufficient performance in the applications that do not require sensitive speed control. However, some aerospace, biomedical and robotic applications require highly sensitive speed control. Therefore, this method can be used practically in the applications that require sensitive control. REFERENCES
[1] Huang, X., Xu, J., Wang, S., "Nonlinear system identification with continuous piecewise linear neural network", Neurocomputing, vol. 77(1), pp. 167177, 2012 Peng, J., and Dubay, R., "Identification and adaptive neural network control of a DC motor system with dead-zone characteristics", ISA Transactions, Vol. 50(4), pp. 588598, 2011. Bennett, S., Development of the PID controller, IEEE Control Systems Magazine Vol. 13 pp. 2838, 1993. Cominos, P., Munro, N., PID Controllers: Recent Tuning Methods and Design to Specification, IEE Proc. Part D, Contr. Theory Appl., vol. 149(1), pp. 46-53, 2002. Howell, M.N., Gordon, T.J., Best M.C., The application of continuous action reinforcement learning automata to adaptive PID tuning, IEEE Seminar on learning systems for control. pp. 14, 2000 Cameron, F., Seborg, D.E. A self-tuning controller with a PID structure. International Journal of Control, vol. 38(2): pp. 401417, 1983 Kim, J.H, Choi, K.K., Self-turning discrete PID controller, IEEE Transactions on Industrial Electronics, vol. 43,pp. 298300, 1987 Vega, P., Prada, C., Aleixander V., Self-tuning predictive PID controller, IEE Proceedings D - Control Theory and Applications, vol. 138(3), pp. 303 - 311, 1991 Martins, F.G., Man, C., Application of feed-forward artificial neural to improve process control of PID - based control algorithms, Computers and Chemical Engineering, vol. 24, pp. 853858, 2000 Chen, J., Huang T.C., Applying neural networks to on-line updated PID controllers for nonlinear process control, Journal of Process Control, vol. 14(2), pp. 211230, 2004 K. S. Narendra and K. Parthasarathy, Gradient Methods for Optimization of Dynamical Systems Containing Neural Networks, IEEE Transaction on Neural Network, Vol. 2, pp. 252-262, 1991. Lin, F., Brandt, R.D., Saikalis, G., "Self-Tuning of PID Controllers by Adaptive Interaction", Proceedings of American Control Conference, vol. 5, pp. 3676 - 3681, 2000 Brandt, R.D., Lin, F., "Adaptive Interaction and its Application to Neural Networks", Information Sciences, Vol. 121(34), pp. 201215, 1999. Slotine, J.J.E., and Weiping, L., Applied Nonlinear Control, Prentice Hall, 1989. Brandt, R.D., Lin, F., Can supervised learning be achieved without explicit error back-propagation? Proceedings of the International Conference on Neural Networks, pp. 300-305, 1996. Brandt, R.D., and Lin, F., Supervised learning in neural networks without feedback network, IEEE International Symposium on Intelligent Control, pp. 86-90, 1996. http://www.baldor.com/pdf/literature/BR1202-E.pdf, [last access: 25.05.2013
[2]
[3] [4] Fig. 5. Reference and system response. [5]
[6] [7] [8]
[9]
[10]
Fig. 6. Adaptive PID gains determined by ANNC.
[11]
As seen in Fig. 5 Ref signal seems to invisible. Actually, all signals are overlapping. This is because the Ref signal and the response signal of the controller Real are much closed to each other which means the error rate is very low. As seen in Fig. 6, P gain shows greater change according to the situation of the current speed and load. It is also realized that there is no need to use a D controller in such a system, as its average value is almost 0.17. So, the PI controller will be sufficient in terms of economy. VI. CONCLUSIONS
[12]
[13] [14] [15]
[16]
The PID tuning algorithm proposed in this paper has many advantages like a low error rate, low calculation time, and high frequency response in DC motor control applications. The simulation results shows that it performs very well even in
[17]
A Comparison among Reactive Power Compensation Strategies in Wind Farm Power Plant
Vito Calderaro, Vincenzo Galdi
Dept. of Industrial Engineering University of Salerno Fisciano (SA), Italy vcalderaro@unisa.it
Gaspare Conio
Italian Vento Corporation Group Napoli, Italy gaspare.conio@gmail.com
Roberto Fusco
ArcerolMittal Livorno, Italy Robertofusco77@libero.it
AbstractSquirrel cage induction generators are consolidated technologies for wind energy. Nevertheless, they do not perform voltage regulation and absorb reactive power from the utility grid. In this paper a comparison among three different reactive power compensation strategies is presented: centralized, fully decentralized and partially decentralized. The last is the proposal developed in the paper where an optimization problem is solved to individualize optimal sizing and location of reactive power centres. The problem is tested by a case study on a real test grid and encouraging results are presented. Keywords: wind power generation; capacitor compensated distribution lines; genetic algorithms; induction generator; reactive power control
II.
PROBLEM FORMULATION
The goal of the problem is to determine the optimal location and size of shunt capacitors to install in a WFG that minimize energy losses, reactive power absorption from the external distribution network, installation and operational costs within defined limits for bus voltages and line currents. In the analysis we suppose that the voltage magnitude constraint at the point of connection of the wind park to the transmission network is defined by the Transmission System Operator, so that overvoltages in some parts of the distribution network, specially in off peak hours, are avoided [9]. Taking into account that at a WF bus a shunt capacitor system is composed by a basic module, which includes connection, switch and control subsystems, and additional capacitor banks, so the optimization problem can be defined as:
M c L PLossk (t )dt k 1 min N Wi c0 c1 QCi CQTOT i 1
I.
INTRODUCTION
Among technology options, squirrel cage induction generators are a very attractive choice for wind power systems because they are robust, reliable, inexpensive and have low maintenance requirements and costs [1-2]. They are usually installed by a local reactive compensation system that furnishes the reactive power absorbed by the induction generator. Usually, no reactive power compensation of the grid and of the power transformer is available in the standard configuration. In order to control the maximum bus voltages and to minimize the power losses in the Wind Farm Grid (WFG), the WFG owners do not prefer to modify the local reactive compensation system to avoid losing the warranty provided by the seller, choosing between two different approaches for reactive power compensation, widely described in literature: the first is based on a single compensation center (centralized approach), whereas in the second one a total compensation on each bus of the WFG is carried out (distributed approach) [3-8]. In this paper we present a comparison between three different methods for a partial compensation of the reactive power in existing Wind Farms (WFs): the two aforementioned methods proposed in literature and a new one that individualizes the best centers to install capacitor banks, providing a solution based on the characteristics of the plant. The level of compensation is chosen in order to avoid overvoltages in the WFG. The optimization problem for the placement and sizing of the reactive power centers is formulated and solved by Genetic Algorithm.
subject to the following constraints:

Pi (t ) Qi (t )
Pij (t )Vi (t ), V j (t ), i (t ), j (t ) j Qij (t )Vi (t ), V j (t ), i (t ), j (t )

j
Vi min (t ) Vi (t ) Vi max (t )
max I k (t ) I k (t )
where (2) represent the network constraints and (3) the bus voltage and the line current limits, respectively; M is the number of the network lines N is the number of network buses; cL is the medium price of the active energy unsold due to the power losses [/kWh]; PLossk (t ) and
PLoss
( )d are the
power loss at time t [kW] and the energy loss [kWh/year] per
year in the branch k, respectively; c0 is the annual depreciation charges of the total installation and maintenance costs of the basic modules of the capacitor banks [/year], c1 is the per kVAr annual depreciation of the costs of the installation and maintenance of additional capacitor banks [/kVAr/year], QC i is the total reactive power installed at the bus i, CQTOT is the annual cost of the reactive power bought by the grid operator to sustain the reactive power absorbed by asynchronous generators if an imperfect compensation of reactive power in the WFG occurs [/year]; Wi is a binary variable that is 1 if the capacitor bank at bus i exists, 0 otherwise. Pi (t ) ( Qi (t ) ) and Pij (t ) ( Qij (t ) ) are the total active (reactive) power injected or absorbed at bus i at time t and the active (reactive) power flowing to the bus i from the adjacent bus j, respectively; Vi(t) is the nodal voltage at bus i at time t, Ik(t) is the line current at brunch k at time t, Vi min (t ) ( Vi max (t ) ) is the lower (upper)
max voltage limit at bus i at time t, I k (t ) is the current limit on the line k at time t.
With this aim, in this Section the models of the capacitor bank, wind turbine generator and power transformer are introduced. Furthermore, a fast power flow algorithm for radial distribution network is briefly described. A. Capacitor Bank The capacitor banks installed at the bus i at bus voltage Vi according to (6) is modeled by the following admittance:
YC N i YCbase
(7)
Thus, the reactive power is:

QCi N Ci QC _ base N Ci Vi 2 YC _ base
(8)
In order to simplify the notation, in the following we denote all the electric quantities without explicit time dependence. The problem solution is expressed by: where:
S W , QC
B. Wind Turbine In the last two decades there has been a rapid development of technology for wind turbine generators (WTGs). Although many technologies have been proposed, the widespread one is based on asynchronous generators. Typically, it has been modeled with steady-state models without saturation. In literature, many models of induction generators are available [10], some based on a PQ representation, with the constant active and reactive power, others based on a RX model, with active and reactive power calculated as a function of the magnetizing reactance of the generator [10]. In this paper, a P-RX hybrid model is proposed: the injected active power is considered as function of the wind speed, whereas the absorbed reactive power is calculated by using a RX model. The values of R (resistance) and X (reactance) are obtained by the electrical parameters of the WTG.
700 600
W QC
W1,W2 ,,WN
1 2
QC , QC ,, QC
N
Active Power [kW]
The problem (1) is a mixed integer nonlinear programming problem (MINLP). However, assuming that the capacitor banks that guarantees at bus voltage Vi the reactive power QCi consist of a finite number of basic elements of standard sizes YCbase able to furnishes QC _ base reactive power, we can write:
MAX QCi N Ci QC _ base N Ci N C
i
500 400 300 200 100
0 0 5 10 15 20 25
Wind Speed [m/s]

MAX are the numbers of installed basic where N Ci and N C i
elements and the maximum number of basic elements that can be installed at bus i of the WF respectively, is the set of natural numbers and QC _ base is the size of the basic capacitor element. In this way the MINLP problem (1) collapses in a pure integer nonlinear programming problem (INLP).
Figure 1. Wind Generator Power Curve
The active power has been calculated by using the power curve of a commercial WTG [11] depicted in Fig. 1, which is approximated as:
III. DEVICE MODELS FOR POWER FLOW ANALYSIS
Pi wind viwind
Pi wind rated v wind a i 1 e b 0
if 4 viwind 25 otherwise
In order to guarantee a higher accuracy in simulation results and a shorter computational time, a careful modeling of network devices and an efficient power flow algorithm are expected.
where viwind is the wind speed at node i that varies in [0-28 m/s], and where the coefficients a and b have been calculated
minimizing the mean square error between the (9) and the power curve in Fig. 1 (a = 8.8 and b = 1.4). The wind speed dependency by the site orography has been modeled by:
Qicomp
wind 25kW 150kVAr if Pi 200kVAr if 25kW Pi wind 330kW 250kVAr if 330kW P wind 660kW i
Reactive Power [kVar]
viwind
wind v0
zi z0
400 350 300 250 200 150 100 0 100 200 300 400 500 600 700
wind where v0 is the wind speed at the reference altitude z0, zi is the altitude of the site i and is the ground surface friction coefficient that depends on the hour of the day, the season, the wind speed and the weather temperature; has been chosen constant and equal to 1/7, according to [12].
In order to evaluate the reactive power absorbed by the WTG, the relation between the active power generated by the wind turbine and the wind speed has been considered. The absorbed reactive power has been calculated by the equivalent circuit of the asynchronous generator, as shown in Fig. 2.
Active Power [kW]
Figure 3. Reactive Power Curve absorbed by the Wind Generator
The reactive power Qiwind , defined in (12) and absorbed by the WTG, and the reactive power injected by the compensation system Qicomp , defined in (13), furnish the total reactive power injected at bus i and shown in Fig. 4:
Qigen Qicomp Qiwind
400
Figure 2. Induction Generator equivalent circuit
Assuming that the stator resistance Rs is negligible, the functional relations amongst the passive components of the circuit, the produced active power Pi wind and the absorbed reactive power Qiwind at the bus i is:
[kVar]
300
200
100
where the values of the open circuit motor resistance and reactance Rc, Xm, of the stator reactance Xs and of the open circuit rotor resistance and reactance Rr and Xr have been obtained by the datasheet of the WTG [11]. Eq. (11) furnishes the reactive power the WTG:
Qiwind Pi wind
V12 1 2 1 Pi wind jQiwind V1 R c jX m Rr j X X r s s
0 0 -100 200 400 600
gen
-200
-300
Reactive pow er curve of WPG External reactive compensation Total reactive absorbed
Qiwind
absorbed by
-400
Rc s 2 X r X s X m X r X s Rr 2 Xm s 2 X r X s 2 s Rc Rr Rr 2
-500
Pgen [kW]
Figure 4. Reactive Power at Generation Bus
The diagram of (12) is shown in Fig. 3. To compensate the reactive power absorbed by the WTG, the considered wind generator uses an automatic reactive power compensation system with following tree steps:
The power injected at node i is thus:

gen wind Pi Pi gen wind Qicomp Qi Qi
C. Transformer The equivalent circuit of the transformer connecting the WTG to the grid is shown in Fig. 5, where R0 and Xu are the resistance and the reactance of magnetization circuit respectively whereas Re, Xe and Vi are the winding resistance, the leakage reactance and the voltage at secondary side moved to the primary of the transformer, respectively.
used. The filter modifies the number of capacitors coded in the genes of the chromosome according to the following rule:
MAX MAX if N Ci N C then N Ci N Ci N C i i
The fitness function has been pointed out starting from the objective function (1), as described in the next Subsection. The algorithm stops if either the fitness of the better element is close to the average fitness of the population or if a fixed number of generation (1000) has been achieved. B. Solution Algorithm The algorithm to solve the optimal capacitor problem is structured in two parts: in the first one, the optimal capacitor placement problem has been solved for each assigned wind speed and a reference solution is built as a weighted sum of the above mentioned solutions. In the second part, in order to reduce the dispersion of the reactive power compensation centers, a new term is added to the fitness function (1) with the aim to minimize the number of reactive power compensation centers. The new term, which takes into account per each wind speed the distance between the candidate solution and the reference one, is:
K 1 si
l i 1 N j ref
Figure 5. Power Transformer Equivalent Circuit
D. Power Flow The power flow model for the WFG is based on a classical iterative method that here we summarize. The main steps of the power flow procedure are: Step 1: Set the voltage at all nodes at the rated value; Step 2: Calculate the currents injected at the network buses by the WTG and by the capacitors; Step 3: Calculate the branch current; Step 4: Starting from the root (the branch connecting the WF and the HV sub-transmission external network), calculate the voltage at the end bus below each branch; Step 5: Compare the bus voltage calculated at previous step with those calculated at current step; if the error is greater than predetermined margin go to step 2, otherwise stop.
siopt
where K is a dimensional weight assigned to the fitness terms,

l
sij is the i-th gene of the candidate solution j with the wind
ref
speed l, and
ref
siopt is the i-th gene of the reference solution
The slack bus is the point of connection of the WF to the HV sub-transmission network. IV. SOLUTION METHOD AND PROPOSED ALGORITHM
s opt :
ref
with
s opt
ref
opt ref opt opt s1 , s2 ,..., ref s N
A. Solution Method The solution of the INLP problem is obtained by applying a GA because of its ability to represent distribution system design options, and its ability to deal with design constraints. Furthermore, generally, the results give the designer a pool of solutions allowing an engineer to look at various design options with similar features. In particular, the GA starts with a population (generation) of candidate solutions coded by chromosomes randomly generated. Each chromosome codes the capacitors to install in the network buses: in the adopted coding the position of the gene in the chromosome identifies the network bus, whereas its value, coded in binary, specifies the number N Ci of base capacitors installed at the bus, as described in (6). The operators of selection, crossover, mutation and elitism have been applied and a filter after the crossover operator to guarantee the feasibility of the new generated element has been
ref
siopt MAX l siopt

l 1...V
where l siopt the i-th gene of the optimal solution referred to the l-th wind speed among the V wind speeds considered. The algorithm, depicted in Fig. 6, can be synthesized as: for each assigned wind speed, solve the optimal capacitor placement problem; then is set the reference solution and a new term is added. For each assigned wind speed, the optimal capacitor placement problem is solved with the new fitness function and while the new solution is set and the previous are different, the steps 2, 3 and 4 are repeated. V. CASE STUDY The proposed algorithm has been tested on a WFG with 23 branch and 24 buses, depicted in Fig. 7. The power is produced
by 20 asynchronous WTG 660 kW 690 V [12], each one connected at the 20 kV distribution network by a LV/MV transformer.
Start
parameters. The annual depreciations c0 and c1 are estimated by assuming a life expectancy and a payback time of five years. Moreover, the investment cost C0 of the first capacitor includes the costs of protection and switch equipment, too.
TABLE I. SIMULATION PARAMETERS Value 0.11 0.0087 0.90 1250.00 100 Unit [/kWh] [/kVAr] [] [] [kVAr]
Set K=0
Fitness=F1
K=0?
Fitness=F2
Parameter cL c0 c1 C0
QC _ base
Set M=1
The maximum reactive power that may be installed in each bus has been fixed in 700 kVAr.
Solve GA for the wind speed M K=K+1 Store Solution M M=M+1
The size of capacitors considered is 100, 200 and 400 kVAr.

TABLE II. GA PARAMETERS Value 601 600 10 0,02
M<Max
Y
Description Population size Maximum number of generation Maximum number of external iteration Mutation rate [%]
Set reference solution
Satisfied?
Y
In order to solve the capacitor placement problem, a set of 15 runs have been done with an average simulation time of about 6 hours. The optimal capacitors location and sizes obtained for each simulated wind speed are shown in Table III. The optimum number of placement locations suggested by the procedure is equal to 10 and the total reactive power installed is 3.7 MVAr. The results achieved by the proposed capacitor placement strategy has been compared with those obtained applying both a fully distributed and a pure centralized reactive compensation strategy. The comparison of energy losses obtained applying both the proposed method and the traditional reactive power compensation strategy in the [5 m/s, 25 m/s] wind speed range is shown in Fig. 7. As expected, the power losses of the proposed strategy are intermediate between those obtained using the centralized and those obtained using the fully distributed capacitor placement strategy.
TABLE III. BUS 6 8 10 11 13 14 17 19 20 22 14 m/s 300 500 300 300 0 500 300 300 500 0 SIMULATION RESULTS 9 m/s 200 0 0 0 0 100 0 0 0 100
MAX
Stop
Figure 6. Proposed Algorithm
Figure 7. Test network
The simulation parameters, such as the costs, used in the simulations are in Table I whereas in Table II there are the GA
WIND SPEED & CAPACITOR SIZE (kVAr) 13 12 11 10 m/s m/s m/s m/s 300 0 200 0 300 0 0 0 300 300 0 0 0 400 200 0 200 200 0 0 500 300 300 200 300 300 0 100 300 300 300 400 200 300 200 0 300 200 300 0
kVAr 300 500 300 400 400 500 300 400 500 300
However, the proposed strategy allows obtaining very good results in terms of annual cost optimization, as shown in Fig. 8.
50 Fully distribuited Centralized Proposed Method 40
behaviors of the proposed compensation method with the traditional ones described in literature, like the centralized and the fully distributed reactive compensation strategy, has been pointed out and tested on a reference test wind power plant. The results of the simulations carried out have been presented and discussed. The simulations show that the centralized compensation, using an unique compensation center, is characterized by largest losses, whereas the fully distributed compensation strategy, with a large number of centers, is characterized by smallest losses and higher installation and maintenance costs. On the other hand, the proposed method offers a sub-optimal solution that is a tradeoff between the two aforementioned extremes, with a placement and sizing of the compensation centers able to achieve a reduction in power loss and operating costs.
Loss [MW/year]
30
20
10
0 5,00
7,00
9,00
11,00
13,00
15,00
17,00
19,00
21,00
23,00
25,00
Wind Speed [m/s]
ACKNOWLEDGMENT The authors are grateful to Prof. Antonio Piccolo for valuable discussions and Mr. Alfio Al, of TERNA, for the support in data acquisition and result evaluation. REFERENCES
[1] A. Mesemanolis, C. Mademlis, I. Kioskeridis, High-Efficiency Control for a Wind Energy Conversion System With Induction Generator, IEEE Trans. Energy Conversion, vol. 27, pp.958-967, Dec. 2012. [2] S. J. Chapman, Electric Machinery Fundamentals, New York: McGrawHill, 1991. [3] M. Elnashar, M. Kazerani, R. El-Shatshat, M. M A. Salama, Comparative evaluation of reactive power compensation methods for a stand-alone wind energy conversion system, in Proc. 2008 IEEE Power Electronics Specialists Conf., 2008, pp.4539-4544. [4] Zhang Ping, Zeng Xiang Jun, Li Ling, Optimal Allocation of Reactive Power Source in Wind Farms Using Sensitivity Analysis and Tabu Algorithm, in Proc. 2008 IEEE Intelligent Computation Technology and Automation Conf., pp.1230-1234. [5] V. Calderaro, V. Galdi, S. Raiti, A. Vaccaro, ICT in Distribution networks with Wind Power Generation to improve in real time the voltage profiles, in Proc. 2004 39th International Universities Power Engineering Conf. [6] V. Calderaro, V. Galdi, G. Massa, A. Piccolo, Distributed Generation and local voltage regulation: An approach based on sensitivity analysis, in Proc. 2011 IEEE Innovative Smart Grid Technologies Conf., pp.1-8. [7] V. Calderaro, V. Galdi, G. Massa, A. Piccolo, Optimal fuzzy controller for voltage control in distribution systems, in Proc. 2011 IEEE Intelligent Systems Design and Applications Conf. pp.1282-1287. [8] V. Calderaro, V. Galdi, G. Massa, A. Piccolo, Optimal synthesis of a fuzzy controller with PSO for local reactive power support, in Proc. 2012 IEEE Electrical and Power Engineering Conf., pp.201-206. [9] A. Augugliaro, L. Dusonchet, M.G. Ippolito, E. Riva Sanseverino, An Efficient Iterative Method for Load flow Solution in Radial Distribution Networks, in Proc. 2001 IEEE Power Tech Conf. [10] L.H. Hansen, P.H. Madsen, F. Blaabjerg, H.C. Christensen, U. Lindhard, K. Eskildsen, Generators and Power Electronics Technology for Wind Turbines, in Proc. of Annual Conference of the IEEE Industrial Electronics Society, 2001. [11] VESTAS V 47 wind generator data-sheet. [12] G. L. Johnson, Wind Energy System, Electronic Ed., Manhattan, KS, 2006.
Figure 8. Energy losses
The simulation results shown that the proposed strategy allows obtaining a reduction of the total costs, as shown in Fig. 9; this is obtained by a reduction of the number of the power factor compensation centers (PFCC) and, thus, of the related installation costs.
90,0
Installation cost of PFCCs

80,0
Reactive power cost Capacitors cost

37,0 20,0 25,0 12,5 11,0 2,9 3,3
70,0
Power loss cost
60,0
Cost [k/Year]
50,0
40,0
30,0 44,6 44,6 43,5 44,4
20,0
10,0
0,0
No Capacitor
Centralized
Fully distributed
Proposed Method
Compensation Method
Figure 9. Comparison among compensation strategies
The great fall in the investment cost compensates the minimum rise in costs due to the little growth of the power losses. VI. CONCLUSION In the paper the problem of the compensation of reactive power in existing wind farm with induction generators has been analyzed and a semi-decentralized procedure has been presented. The proposed compensation strategy increases the generation efficiency and influences the losses in the internal wind farm grid. The proposed strategy for the optimal sizing and placement of the reactive power center has been formalized and implemented using Genetic Algorithms. A comparison of the
Optimal Substation PMU Placement Method for the Two-Level State Estimator
Heloisa H. Mller
School of Electrical and Computer Engineering University of Campinas, SP, Brazil Email: h.h.muller@ieee.org
Carlos A. Castro
School of Electrical and Computer Engineering University of Campinas, SP, Brazil Email: ccastro@ieee.org
AbstractThis paper presents a new method for the optimal allocation of phasor measurement units in substations with a focus on the two-level state estimation process that was recently proposed in the specialized literature. A mixed heuristic/metaheuristic method is proposed to determine the number and location of those units in such a way to provide robust observability characteristics. Its reliable, robust, and precise results are shown for small and large substation layouts.
I. I NTRODUCTION Energy Management Systems (EMS) have been available for decades and their functionalities have been constantly evolving. Among their functions, the State Estimation (SE) [1], [2] is crucial for determining the power system topology and operating state, as well as allowing other functions to be carried out, as for example security analysis. An important development on this subject was the introduction of the generalized SE [1], [2], [3], [4], [5]. The recent technological advances have been benetting the control and monitoring of power systems in a very signicant way. The increase in information storage and processing capabilities have been allowing the reengineering of several funcionalities related to power systems control, management and planning. In particular one could mention the advent of phasor measurements as one of those important advances. The availability of phasor measurement units (PMUs) has been considered as a pathway to the transition to the smart grid of the future [6]. Meter placement has always been considered as one of the key aspects of the state estimation area. Studies involving conventional meter placement and Remote Terminal Units (RTUs) [7] as well as studies considering the generalized model and the representation of circuit breakers [8] have been presented. Also, studies have been conducted considering generalized node-branch models and the use of PMUs [9], [10], [11], [12], [13], [14], [15]. Recently, a generalized two-level SE model was proposed [6], [16], [17], where each substations topology and operating state are obtained independently through a linear SE formulation. Phasor measurements were assumed to be available. This two-level SE model has a solid foundation and provides high levels of robustness and precision. It seems that the adoption of the two-level SE function by utilities will be the natural way towards a distributed, robust, and fast state estimation
function. Moreover, its principles are perfectly aligned with the ones of a smart grid. Therefore, it was adopted in this paper. Given the two-level SE scenario, it is important to determine the appropriate location of PMUs at the substation to provide the necessary robustness of the SE process. The technological context of the existing power system substations and economical factors require that this allocation be dened in an optimal way, to provide the necessary robustness at a minimum cost. Considering these crucial aspects, the main objective of this paper is to propose an optimal PMU allocation method for substations based on a specialized Genetic Algorithm (GA). Even though the method proposed in this paper was developed to be used for substations, it can also be applied to the whole system, in the conventional SE sense. Special attention has also been paid to the observability issue, as discussed in [6], [16], [17]. In case any measurement is lost, observability must be guaranteed by another measurement strategically located to cover for the former. Actually, those measurements must be allocated to provide coverage for the largest number of measurements in case they fail. Full coverage is achieved when the system maintains its observability after any of the measurements is lost [7], [18]. This principle allows the SE process to be very robust and applicable for real time monitoring and operation using phasor measurements. The following procedures have been implemented in this paper: (a) the allocation of a set of n PMUs in such a way to provide the best possible coverage; (b) the allocation of an extra set of m PMUs to provide a better coverage, considering that the rst n PMUs of (a) had already been installed; and (c) the allocation of the minimum set of PMUs that results in full coverage (in this case neither n or m are specied). II. T WO - LEVEL STATE ESTIMATION AND OBSERVABILITY A. Two-level state estimation State estimation (SE) can be dened as an optimization problem where the power systems state variables are estimated based on a number of available measurements. The quality of such estimation depends on the measurement quality, the location of the measurement devices, and also the capture, storage and treatment of the measurements. The following types of measurements are considered in this paper: bus
voltage magnitudes, the magnitudes of currents through circuit breakers, and nodal current injection magnitudes. The SE model proposed in [6], [16], [17], where the problem is divided into two levels, will be used as a reference in this paper. The rst level, referred to as Substation level state estimator, aims to determining each substation topology, by nding the circuit breaker statuses. Then, busbar voltages are estimated. In the second level, referred to as Control center level state estimator, the information obtained in the rst level is added to the additional system data, as load and generation data, transmission system topology, among others, for the state estimation of the whole system. In particular, this paper is concerned with the rst level, whose formulation leads to a linear programming problem. B. Substation level state estimation The topology of a substation is obtained through a linear model involving complex current injection measurements at busbars and currents through the circuit breakers [6], [16], [17], resulting in z i = z inj z cb = Hi x i + r inj r cb = Hi x i + r i ,
estimated bus voltages voltages. w v(2i1,2i1) and w v(2i,2i) correspond to the elements of the voltage covariance matrix, imag real and z v v (i) and z (i) correspond respectively to the real and imaginary parts of the measured bus voltages.
sb real v(2i1,2i1) z v i=1 w (i) sb v(2i1,2i1) i=1 w sb imag v(2i,2i) z v i=1 w (i) sb v(2i,2i) i=1 w
x real v
(4)
x imag v
(5)
T Eq. (2) has a solution if Hi Hi has an inverse, or, if the system is observable. Therefore, the location of the current measurements is crucial for guaranteeing the systems observability, bad data processing and the overall robustness of the method. The location of voltage measurements is also crucial, and the relevant buses must be determined to complete the process and to determine the operating state of the substation with the linear SE.
(1)
C. Observability analysis The conventional observability analysis is usually carried out by assuming measurement symmetry and the existence of sufcient voltage measurements. The symmetry assumption leads to a simplied problem, where the real part is considered only, that is, the linearized SE is based on the DC model and the P problem is used for observability analysis [1], [2]. In the case of the two-level linear SE, measurements are complex voltage and current quantities and the analysis is divided into two parts. The rst one takes into account complex current measurements and the main goal is to determine the substations topology, or the statuses of the substation circuit breakers. It is assumed that the so-called gain matrix Ci = T Hi Hi has rank r. These conditions result in an observable system, allowing the use of the most important features of the gain matrix Ci , measurement matrix Hi and information on the substation topology to allocate PMUs/measurements. The second part consists of estimating the busbar voltages. It is crucial that all busbar voltages are determined despite the substations topology. Therefore, it is assumed that voltage measurements corresponding to the so-called relevant busbars are available. For instance, substation busbars connected to transmission line receiving ends are network nodes that receive the bulk power ows, therefore, they can be considered as relevant ones. III. G ENETIC ALGORITHM
where z i is an array that contains busbar current injections (z inj ) and circuit breaker currents (z cb ), Hi is the Jacobian matrix, xi is the state variable array of circuit breaker currents, and r i is the array of measurement errors for both current injections (r inj ) and circuit breaker currents (r cb ). The rows of the Jacobian matrix Hi correspond to the current injection Ip and circuit breaker current Ipq , Iqp and measurements. Its columns correspond to the estimated circuit breaker currents. Its elements are either 1, 0 or 1, which is a very important feature [6], [16] that will be explored by the proposed method, as explained in Sec. IV. The state variable array x i is obtained by
T T x i = (Hi Wi Hi )1 Hi Wi z i ,
(2)
i is the covariance matrix related to the current where W measurements. After x i and the substation topology are known, the voltages at each substation busbar can be estimated considering each voltage level using (3), where z v is the array of voltage measurements, Hv is the Jacobian matrix for voltage estimation, x v is the state variable array, e r v is the array of voltage errors. z v = Hv x v +r v (3)
v of existing voltage measurements is Jacobian matrix H diagonal. The state variables are calculated by (4) and (5), where sb is the number of substation buses , x real and x imag v v correspond respectively to the real and imaginary parts of the
Genetic algorithms (GA) are metaheuristics for solving optimization problems inspired in biological evolution processes [19]. GA can be considered as a process with Darwinian characteristics related to an individual and to the population it belongs. Those characteristics are evaluated by performance functions [20].
The GA implemented in this paper has the following features: (a) the initial population can either be generated randomly or be chosen using the memory of previous simulations; (b) a performance function is dened to improve the variety of the descendant population determined by the selection function; (c) a selection function based on roulette chooses parents by simulating a roulette wheel; (d) a reproduction control where the individuals to be preserved in the next generation are dened by an elite count parameter, and the others are created by crossover; (e) a broadcast crossover function that denes a binary random array to determine which genes of each parents chromosome will recombine to generate the children; and (f) a mutation method that makes small random changes in the chromosome. The stopping criteria used in this research work were: (a) number of generations; (b) time limit; (c) threshold for the performance function; and (d) threshold for the performance function variation. IV. F ORMULATION In the proposed formulation three measurement types are considered, namely complex busbar voltages, complex circuit breaker currents and complex busbar current injections. It is assumed that a PMU allocated to a certain busbar will measure its complex bus voltage, complex current injection and complex currents through all branches connected to that busbar. The possibility of allocating PMUs in a substation with or without previously installed PMUs is considered. Also, it is possible to preset PMU allocation to certain busbars, referred to as relevant busbars in this paper, if they are connected to transmission backbones, or connected to generators or synchronous condensers. The main idea is to allocate the minimum number of PMUs while ensuring the systems observability. The different PMU placement alternatives generated by the GA are evaluated through different aspects that reect their qualities and relevances, based on the rank r of matrix Hi from (2), the T characteristics of the gain matrix Ci , computed by (Hi Hi ), and weight factors of measurements p. For solving the PMU allocation problem by a GA, a chromosome (array) s with T positions was dened, where T is the maximum number of busbars that can receive a PMU. Each position k of s may be equal to 0 (PMU is not allocated at busbar k ) or 1 (PMU is allocated at busbar k ). Based on array s, an expanded chromosome sm is obtained. While s indicates the location on the PMUs, sm contains the measurements (either current injections and current ows) available after those PMUs are connected. In case some PMUs are already connected (pre-allocated) in the substation, array spre with T positions is set with 1 in the positions for which there are existing PMUs, and 0 otherwise. Weight factors p are associated to the PMU/measurements according to their importance, quality, etc. The evaluation of the individuals generated by the GA is done by a performance function y , computed as follows, where input n is the desired number of PMUs. Note that penalty
factors ptyrank and ptypmu are also dened. Those factors are related to the rank of matrix Hi and the number of PMUs of each individual, respectively. Performance Function y 01: Inputs: s, n, p, c, ptyrank , ptypmu 02: Obtain sm from s 03: Build matrices Hi and Ci based on sm 04: Let min be the number of columns of Ci 05: Calculate r = rank(Ci ) 06: Compute f1 = p s 07: Compute f2 = |min r| ptyrank 08: If n > 0 T 09: Compute f3 = |( j =1 sj ) n| ptypmu 10: Compute y = f1 + f2 + f3 11: Else 12: Compute y = f1 + f2 13: End if 14: End The GA has been implemented in this paper to achieve either one of the following goals. 1) The allocation of a set of n PMUs in such a way to provide the best possible coverage; 2) The allocation of an extra set of m PMUs to provide a better coverage, considering that the rst n PMUs of (1) had already been installed; 3) The allocation of the minimum set of PMUs that results in full coverage (in this case neither n or m are specied). Assume that for a certain substation a minimum set of n PMUs is necessary to provide observability. All n PMUs are considered critical, in the sense that any missing or discarded PMU would lead to an unobservable system. Missing measurements may occur in case of PMU misfunction, data transmission failure, and so on. Discarded measurements may result from bad data detection procedures. Goal 3) above aims to determining such a set. Of course there may be several critical PMU sets in a substation that meet the requirements. GA then selects the one with the smallest performance function. Goal 2) aims to allocating an extra set of m PMs, considering that a set with n PMUs was previously allocated in 1). Then, the GA looks for the best coverage for bad/missing measurements with the smallest performance function. The so-called Coverage Level Function is carried out for the top ranked solutions determined by the GA. The idea is to obtain the individuals level of coverage. The ideal situation is that the coverage level equals the number of measurements, that is, every measurement is covered by another measurement in case the former is missing, and observability is preserved. This would result in a very robust SE function. By using the current balance equations obtained from the available measurements for each busbar, and the possibility of a measurement
be obtained as a function of other existing measurements, the coverage factor f is dened. This factor is split into two other factors, respectively fInj related to the current injection measurements, and fCB related to the currents through the circuit breakers. The current balance equations are obtained from Hi . The function is shown below. Coverage Level Function 01: For each individual s do 02: Build matrix Hi with s 03: Obtain current balance equation from Hi 04: Compute fInj 05: Compute fCB 06: Compute f = fInj + fCB 07: Compute = c s 08: Show Result 09: End for The coverage level function can also be used to verify whether the substation is still observable in case an entire PMU is lost, that is, all measurements associated to that PMU are missing. Of course, this situation is more critical than the loss of one measurement only. Since the general idea is to allocate the minimum number of PMUs while guaranteeing some degree of robustness regarding observability, it is possible to identify a clear cost/benet problem here. Note that in addition to computing the coverage level of each solution (individual), the coverage level function also determines the installation cost of each solution by
T
V. S IMULATION TESTS Simulations have been carried out for different substation layouts, namely a 4-bus/3-breaker/14-measurements (number of busbars, number of circuit breakers, maximum number of measurements), a 6-bus/6-breaker/24-measurements, a 9-bus/11-breaker/40-measurements, a 14-bus/13-breaker/54measurements, an 18-bus/30-breaker/96-measurements, a 30bus/41-breaker/142-measurements, a 40-bus/79-breaker/238measurements, a 57-bus/80-breaker/274-measurements, an 118-bus/186-breaker/608-measurements, and a 300-bus/411breaker/1422-measurements. The results corresponding to the 9-bus/11-breaker substation layout from Fig. 1 are shown in detail in this paper. Five simulation cases will be described below. A. Case 1 The idea here is to obtain the smallest set of PMUs/measurements such that the system is observable. Therefore, the number of allocated PMUs n will be determined. Busbars 1 and 4 were arbitrarily chosen as relevant buses, in the sense that the allocation of PMUs in those buses is mandatory. Table I shows different solutions that were obtained by the proposed method. The table contains columns s (array that shows busbars that will receive a PMU), sm (array that species the voltage and current measurements), P M U s (number of PMUs), M eas (number of measurements), f (coverage level), y (performance function), Obs (observability Yes/No) and installation cost .
=
j =1
cj sj ,
(6)
where cj is an element of the installation cost array c and indicates the installation cost of a PMU connected at busbar j , and sj is the element j of chromosome s. The GA used in this paper is summarized below. Genetic Algorithm 01: Input data: n, spre 02: Dene GA parameters 03: Read network/substation layout 04: Set p and c for each busbar 05: Dene penalty factor ptyrank based on the number of Circuit Breakers 06: Dene penalty factor ptypmu associated to the number of PMUs to be installed 07: Dene initial population 08: Execute GA with (y , parameters, ptyrank , ptypmu) 09: Call coverage level function
Fig. 1. Example 9-bus/11-breaker substation layout
The best solution corresponds to the allocation of 6 PMUs (leading to 21 measurements) that guarantee the systems observability even if one of them is somehow missing. The top three solutions have a performance function value of y = 977.00. The rst solution is s = 110110110 and sm = 1110111101110011010110, the second is s = 110110101 and sm = 1110111101010011011101, and the third is s = 101110110 and sm = 1101111101100111010110. Array s of the rst solution means that PMUs are located at busbars 1, 2, 4, 5, 7 and 8. Array sm means that injection currents I1 , I2 , I4 , I5 , I7 , I8 , bus voltage V1 , V2 , V4 , V5 , V7 , V8 and circuit breaker currents I1,2 , I1,6 , I2,3 , I4,5 , I5,6 , I5,7 , I5,9 ,
I7,8 , I8,9 , I2,1 , I4,3 , I5,4 , I7,5 , I7,6 , and I8,7 are measured. Note that in this case voltage measurements could be obtained for the relevant busbars (1 and 4) only, however, four additional voltage measurements (busbars 2, 5, 7 and 8) will be available since there will be PMUs at those busbars.
TABLE I C ASE 1 - A LLOCATION OF PMU S WITH RELEVANT BUSBARS 1 E 4
s 110110110 110111010 101110110 110110101 110111110 110110111 111110110 110111111 010111010 010111110 111010110 010111111 111111111 010110110 011110110 100110100 sm 1110111101110011010110 1110111110111011100010 1101111101100111010110 1110111101010011011101 1110111111111011110110 1110111101110011011111 1111111101110111010110 1110111111111011111111 0010111110111011100010 0010111111111011110110 1111011101110101010110 0010111111111011111111 1111111111111111111111 0010111101110011010110 0011111101110111010110 1100111101000011010100 P MUs 6 6 6 6 7 7 7 8 5 6 6 7 9 5 6 4 M eas 21 21 21 21 25 24 24 28 18 22 21 25 31 18 21 15 f 21 21 21 21 25 24 24 28 18 22 21 25 31 18 21 15 y -977.00 -977.00 -977.00 -977.00 -495.00 -494.00 -494.00 -12.00 476.00 958.00 959.00 1441.00 3881.00 3886.00 4369.00 4877.00 Obs Y Y Y Y Y Y Y Y Y Y Y Y Y N N N 33 33 33 33 39 38 38 44 28 34 33 39 49 28 33 23
case. In this simulation there are two good quality solutions. The rst solution presents s = 101011010 and sm = 1101011110101101100010, while the second one presents s = 101010110 and sm = 1101011101100101010110. The top solution of Table III indicates that injection currents I1 , I3 , I5 , I6 , and I8 , bus voltages V1 , V3 , V5 , V6 , and V8 and circuit breaker currents I1,2 , I1,6 , I3,4 , I5,6 , I5,7 , I5,9 , I6,7 , I8,9 , I6,1 , I3,2 , I5,4 , I6,5 , and I8,7 have been allocated. The second solution indicates that injection currents I1 , I3 , I5 , I7 , and I8 , bus voltages V1 , V3 , V5 , V7 , and V8 , and circuit breaker currents I1,2 , I1,6 , I3,4 , I5,6 , I5,7 , I5,9 , I7,8 , I8,9 , I3,2 , I5,4 , I7,5 , I7,6 , and I8,7 have been allocated. Note that Table III contains different solutions with different coverage levels, and some of them result in an unobservable system.
TABLE III C ASE 2 - A LLOCATION OF PMU S WITHOUT RELEVANT BUSES
s 101011010 101010110 011011010 101011110 101011011 101011111 111011110 101111011 101010100 001011110 110011010 001101101 sm 1101011110101101100010 1101011101100101010110 0011011110111101100010 1101011111101101110110 1101011110101101101011 1101011111101101111111 1111011111111101110110 1101111110101111101011 1101011101000101010100 0001011111101101110110 1110011110111001100010 0001100011001110111101 P MUs 5 5 5 6 6 7 7 7 4 5 5 5 M eas 18 18 18 22 21 25 25 24 15 19 18 17 f 18 18 18 22 21 25 25 24 15 19 18 17 y 2412.00 2412.00 2412.00 2894.00 2895.00 3377.00 3377.00 3378.00 5339.00 5821.00 5822.00 5823.00 Obs Y Y Y Y Y Y Y Y N N N N 28 28 28 34 33 39 39 38 23 29 28 27
It is important to point out that a set of good quality measurement sets generated by the GA can be stored in addition to the optimal solution (as shown in Table I). The possibility of considering the adoption of other good quality measurement sets provides an additional exibility to the decision maker, since particular, unpredicted factors may be obstacles to the implementation of the optimal solution. Table II shows the coverage analyses for the cases where an entire PMU (all its measurements) are missing. P M U Out indicates which PMUs are lost (one at a time), and f p indicates the number of PMUs that are covered for in case they are lost. Of course, the situation where PMUs are allocated to all busbars (last row of Table II) is the most expensive and robust situation, since the system maintains its observability no matter what PMU is lost.
TABLE II PMU S C OVERAGE OF C ASE 1
s 110111010 101110110 110111110 110110111 111110110 110111111 010111110 111010110 010111111 111111111 sm 1110111110111011100010 1101111101100111010110 1110111111111011110110 1110111101110011011111 1111111101110111010110 1110111111111011111111 0010111111111011110110 1111011101110101010110 0010111111111011111111 1111111111111111111111 P MUs 6 6 7 7 7 8 6 6 7 9 P M U Out 1 4 1,6,7 8,9 2,3,4 1,5,6,7,8,9 7 2 5,7,8,9 All fp 1 1 3 2 3 6 1 1 4 9 Obs Y Y Y Y Y Y Y Y Y Y
C. Case 3 This simulation case assumes that the PMUs of Case 1 had already been installed, that is, spre = 110110110 and there are two relevant busbars, namely 1 and 4. Table IV shows the results of allocating one additional PMU. The simulation resulted in the allocation of 7 PMUs (s = 110111110) and 25 measurements (sm = 1110111111111011110110) that guarantees the systems observability, with performance function value y = 9688.00. Arrays s and sm mean that injection currents I1 , I2 , I4 , I5 , I6 , I7 , and I8 , bus voltages V1 , V2 , V4 , V5 , V6 , V7 , and V8 , and circuit breaker currents I1,2 , I1,6 , I2,3 , I4,5 , I5,6 , I5,7 , I5,9 , I6,7 , I7,8 , I8,9 , I2,1 , I6,1 , I4,3 , I5,4 , I6,5 , I7,5 , I7,6 and I8,7 have been allocated.
TABLE IV C ASE 3 - A LLOCATION OF ONE ADDITIONAL PMU TO THE SET OF C ASE 1
s 110111110 111110110 110110111 sm 1110111111111011110110 1111111101110111010110 1110111101110011011111 P MUs 7 7 7 M eas 25 24 24 f 25 24 24 y -9688.00 -9687.00 -9687.00 Obs Y Y Y 49 48 48
B. Case 2 The idea here is to obtain the minimum set of PMUs/measurements such that the system is observable as in Case 1, however, no relevant busbars have been dened. The results are shown in Table III. The simulation resulted in the allocation of 5 PMUs and 18 measurements that guarantee the systems observability, with performance function y = 2412.00. The value of y is different from Case 1, since there are no penalty factors related to relevant busbars in this
D. Case 4 In this simulation it was assumed that six PMUs must be allocated, and there are two relevant busbars, namely 1 and 5. Table V shows the results. In this case 6 PMUs (s = 101011110) and 22 measurements (sm = 1101011111101101110110) were allocated, with performance function value y = 321.00.
Arrays s and sm of the top ranked solution mean that injection currents I1 , I3 , I5 , I6 , I7 , and I8 , bus voltages V1 , V3 , V5 , V6 , V7 , and V8 , and circuit breaker currents I1,2 , I1,6 , I3,4 , I5,6 , I5,7 , I5,9 , I6,7 , I7,8 , I8,9 , I6,1 , I3,2 , I5,4 , I6,5 , I7,5 , I7,6 and I8,7 have been allocated.
TABLE V C ASE 4 - A LLOCATION OF 6 PMU S WITH RELEVANT BUSES 1 AND 5
s 101011110 111010110 101110110 101010111 101011011 111011010 101111100 100111110 011011110 sm 1101011111101101110110 1111011101110101010110 1101111101100111010110 1101011101100101011111 1101011110101101101011 1111011110111101100010 1101111111001111110100 1100111111101011110110 0011011111111101110110 P MUs 6 6 6 6 6 6 6 6 6 M eas 22 21 21 21 21 21 22 22 22 f 22 21 21 21 21 21 22 22 22 y -1944.00* -1943.00 -1943.00 -1943.00 -1943.00 -1943.00 -1933.00 -1933.00 -8.00 Obs Y Y Y Y Y Y N N Y 34 33 33 33 33 33 34 34 34
coverage level function. The results show that the proposed method provides very good and robust results. Note that the robustness in the two-level SE is crucial, since it is expected that the results obtained in the substation-level SE are errorfree and used directly in the Control Center level SE, without any further verication. ACKNOWLEDGMENT The authors would like to acknowledge the nancial support provided by Brazilian funding agencies CAPES and CNPq. R EFERENCES
[1] A. Monticelli, State Estimation in Eletric Power Systems. A Generalized Approach. Kluwer, USA, 1999. [2] A. Abur and A. G. Expsito, Power System State Estimation. Marcel Dekker, USA, 2004. [3] Y. Wu, M. Kezunovic, and T. Kostic, The dynamic utilization of substation measurements to maintain power system observability, Power Systems Conf. Expo, vol. 1, pp. 16991704, Nov. 2006. [4] S. Jakovljevic and M. Kezunovic, Advanced substation data collecting and processing for state estimation enhancement, Power Eng. Soc. Summer Meeting, vol. 1, pp. 201216, Jul. 2002. [5] A. G. Expsito and A. de la Villa Jan, Reduced substation models for generalized state estimation, IEEE Trans. Power Syst, vol. 16, no. 4, pp. 839846, Nov. 2001. [6] T. Yang, H. Sun, and A. Bose, Transition to a two-level linear state estimator-part i: Architecture, IEEE Trans. Power Syst, vol. 26, no. 1, pp. 4653, Feb. 2011. [7] J. B. A. London, G. L. U. Brito, and N. G. Bretas, Method for meter and rtu placement for state estimation purposes, IEEE Bologna PowerTech Conference,Bologna, Italy, vol. 1, Jun. 2003. [8] A. S. Costa, E. M. Lourenco, and K. A. Clements, Power system topological observability analysis including switching branches, IEEE Trans. Power Syst, vol. 17, no. 2, pp. 250256 May 2002. [9] J. Chen and A. Abur, Placement of pmus to enable bad data detection in state estimation, IEEE Trans. Power Syst, vol. 21, no. 4, pp. 1608 1615, Nov. 2006. [10] M. Korkali and A. Abur, Placement of pmus to enable bad data detection in state estimation, IEEE PES General Meeting, Calgary, Alberta, Canada, Jul. 2009. [11] J. Chen and A. Abur, Enhanced topology error processing via optimal, IEEE Trans. Power Syst., vol. 23, no. 3, pp. 845852, Aug. 2008. [12] A. A. Fish, S. Chowdhury, and S. P. Chowdhury, Optimal pmu placement in a power network for full system observability, IEEE PES General Meeting, Detroit Mi. USA, Jul. 2011. [13] K. S. Sajan and B. Tyagi, Optimal placement of pmu with optimal branch current phasors for complete and incomplete observability, IEEE PES General Meeting, Detroit Mi. USA, Jul. 2011. [14] M. Zhou, V. A. Centeno, A. G. Phadke, Y. Hu, D. Novosel, and H. A. R. Volskis, A preprocessing method for effective pmu placement studies, 3rd Int. Conf. Electric Utility Deregulation and Restructuring and Power Technologies (DRPT 2008), Nanjing, China, pp. 26822687, 2008. [15] J. B. A. London, S. A. R. Piereti, R. A. S. Benedito, and N. G. Bretas, Redundancy and observability analysis of conventional and pmu measurements, IEEE Trans. Power Syst, vol. 24, no. 3, pp. 16291630, Aug 2009. [16] T. Yang, H. Sun, and A. Bose, Transition to a two-level linear state estimator-part ii: Algorithm, IEEE Trans. Power Syst, vol. 26, no. 1, pp. 5462, Feb. 2011. [17] A. Bose, New computer applications for system operations, IEEE PES General Meeting, pp. 15, July 2012. [18] J. B. A. London, L. F. C. Alberto, and N. G. Bretas, Analysis of measurement-set qualitative characteristics for state-estimation purposes, IET Proc. Gen., Transm., Distrib, vol. 1, pp. 3945, 2007. [19] F. Glover and G. A. Kochenberger, Handbook od Metaheuristics. Kluwer Academic Publishers, USA, Massachusetts, 2003. [20] T. Bck, D. Fogel, and Z. Michalewicz, Handbook of Evolutionary Computation. Institute of Physics Publishing and Oxford University Press, 1997.
E. Case 5 In this case parameter n was set to 6, that is, 6 PMUs will be allocated. No pre-installed PMUs nor relevant busbars were assumed. Table VI shows the simulation results. The simulation resulted in 22 measurements that guarantee the systems observability, with performance function y = 2894.00. In fact, there are three equivalent possibilities marked with . The rst considers the allocation of PMUs to busbars 1, 3, 5, 6, 7, and 8. The second considers the allocation of PMUs to busbars 2, 3, 5, 6, 7, and 8. Finally, the third considers the allocation of PMUs to busbars 1, 3, 5, 6, 7, and 9. Note that the seventh solution of Table VI (marked with ) presents the same number of PMUs and measurements, however, their locations result in an unobservable system.
TABLE VI C ASE 5 - A LLOCATION OF 6 PMU S WITHOUT RELEVANT BUSES OR PRE - ALLOCATION
s 101011110 011011110 101011101 101010111 101111010 111011010 100111110 110011110 sm 1101011111101101110110 0011011111111101110110 1101011111001101111101 1101011101100101011111 1101111110101111100010 1111011110111101100010 1100111111101011110110 1110011111111001110110 P MUs 6 6 6 6 6 6 6 6 M eas 22 22 22 21 21 21 22 22 f 22 22 22 21 21 21 22 22 y 2894.00* 2894.00* 2894.00* 2895.00 2895.00 2895.00 2905.00 2905.00 Obs Y Y Y Y Y Y N N 34 34 34 33 33 33 34 34
VI. C ONCLUSION The main goal of this paper is to propose a method for allocating PMUs/measurements in a substation in order to obtain a robust, efcient and consistent substation-level SE in a two-level SE scheme, recently proposed in the literature. The proposed method is able to provide high quality measurement sets so as to guarantee the systems observability. Moreover, it is capable of allocating additional PMUs/measurements to provide effective coverage, in the sense that the absence of any measurement of the resultant measurement set does not result in observability problems. The method is based on a Genetic Algorithm, with an original, specialized performance function, and on a original
A New Approach for Temperature Rising Test of an Induction Motor Loaded by a Current Controlled Eddy Current Brake
Mehmet Onur GULBAHCE
Department of Electrical Engineering Istanbul Technical University Istanbul, Turkey ogulbahce@itu.edu.tr
Handan NAK
Department of Control Engineering Istanbul Technical University Istanbul, Turkey nak@itu.edu.tr
Derya Ahmet KOCABAS

Department of Electrical Engineering Istanbul Technical University Istanbul, Turkey kocabasde@itu.edu.tr
Abstract Temperature rising test is one of the most important electrical machine tests which used to determine the name plate values of an electrical machine and is a key element in confirming that the machine is designed for use within its designed operating capability. As defined in standards, an induction motor is loaded gradually and winding temperature is measured time steps in temperature rising tests. Generally, dynamo brakes and resistive load banks which needs additional effort to adjust settings during the tests. During the test, winding resistance change depending on the rising temperature resulting in a change in motor torque and extra effort is needed to keep the torque unchanged. In this study, a current controlled eddy current brake is used for loading an induction motor instead of traditional dynamo brakes to perform the temperature rising test with a constant load torque. A closed loop control system is designed with conventional PI controller to obtain constant braking torque. Keywords-component; eddy current brakes; temperature rising tests; excitation current control; conventional PI.
below the limit, then load of tested machine is increased to obtain a newly settled temperature. This procedure is continued until the temperature of the electrical machine is reached to the permitted value. The permitted value of the temperature is related to insulation class [3]. Change in temperature can be measured by resistance method, thermometers and also thermocouples [4]. In resistance method, temperature rise can be determined by rising resistance. The actual temperature to be taken into account is the value around the conductors placed in the slots. Because of the difficulties in measuring the temperature of the windings directly, resistance method is appropriate for determining the actual temperature during the test. In electrical resistance method, temperature rise is calculated by obtaining the change in resistance. For copper conductors, (1) is used for calculation where R1 is the resistance measured at the temperature 1 and R2 is the resistance measured at unknown temperature 2 to be obtained [3].
I.
INTRODUCTION
There are many different methods to determine the operating capability of an induction motor acting under full load. All these methods are given in national and international standards [1]. Some of these methods are direct load method, back to back test, phantom loading test, forward short circuit test, variable inertia test, mixed-frequency test [2]. Here, direct load method is used and change in temperature vs time is determined by obtaining the variation of electrical resistance by temperature. Basically, during a temperature rising test of an electrical machine to be operated as a motor, the machine is coupled with another electrical machine or a mechanical load. Tested electrical machine is loaded with a constant load and the change in winding temperature is obtained. In obtaining the variation of electrical resistance by temperature, the winding resistance is measured periodically and the temperature is then calculated. When the temperature settles to a constant value, it is needed to make a comparison between the actual temperature and the temperature limit of the insulation material. If it is
R2 2 + 235 = R1 1 + 235
(1)
During the test, motor current changes depending on the change in winding resistance resulting in a change in the mechanical torque. In this study, an eddy current brake is used and electromagnetic field is needed to be controlled to keep the torque unchanged. By the virtue of defined temperature rising test procedure, induction motor must be loaded by a constant load torque and changing the mechanical torque is not recommended. The eddy current brake (ECB) is an electromechanical conversion system which converts energy into thermal energy [5]. It consists of a conductive disk which rotates between two static electromagnets. Ferromagnetic or nonmagnetic material can be used for conductive disk which is connected to the shaft of an electrical motor to be tested. When the electromagnet is supplied by direct current, conductive disk falls into a timevarying magnetic flux and eddy currents are induced around the flux lines [6]. These currents have a circular route inside the material [7]. In addition to that, induced eddy currents
consumes the mechanical energy at the rate which is proportional to square of the effective value of the eddy current and the resistance of the conductive material. In other words, mechanical input energy of the rotating system is transformed into heat inside the rotating disk. In this study, the excitation current of an eddy current brake is controlled to keep the load torque of the induction motor constant during a temperature rising test where the change in temperature is obtained by the change in electrical resistance. A simulation is performed by using a model including the relation between temperature and electrical resistance, eddy current brake operational data, standard induction motor d-q model and a PI controller. II. MODELING OF EDDY CURRENT BRAKE
The eddy current brake design having electromagnets has a braking torque versus speed characteristic for different excitation current values as in Fig. 2. Fig. 3 and 4 show the total power dissipation versus mechanical speed and maximum braking torque versus excitation current characteristics of the brake respectively. Hence, the inputs of the eddy current brake model are excitation current and speed, and the output is the produced torque. Fig. 5 shows the 3-D characteristic of the used eddy current brake.
Braking Torque (N.m.) 0 -5 -10 -15 -20 -25 -30 Speed (rpm) 0 1000 2000 3000 4000 5000 6000 7000 Figure 2. Braking torque versus mechanical speed. Total Power Dissipation (W) 0 -1000
Rotational eddy current brake designs can be used to consume the rotationally stored energy which can be used to load a rotating machine which is given in Fig. 2. The rotational design must contain a magnetic field producing source which can be either a permanent magnet or an electromagnet. Conductive disk
Pole and coil
-2000 -3000 -4000 -5000 -6000 Speed (rpm) 7000 0 2000 4000 6000 8000 Figure 3. Total power dissipation versus mechanical speed.
Figure 1. geometry of the eddy current brake.
Because of its complexities, obtaining dynamic equations of an eddy current brake is almost impossible. Instead of using dynamic equations of eddy current brake, experimental or finite element analysis results for modeling. In this study, used eddy current brake is modeled according to the results of finite element analysis which obtained from previous studies [5]. The produced torque of eddy current brake is related to the excitation current and shaft speed, where and the air gap and other design parameters are all kept unchanged. For the brake model, the results obtained from a previously designed eddy current brake [6] is used where excitation current changes between 0-10 A and mechanical speed input between 0-7500 rpm.
Maximum Braking Torque (N.m.) 0 -5 -10 -15 -20 -25 -30 Excitation Current (A) 2 3 4 5 6 7 8 9 10 Figure 4. Maximum braking torque versus excitation current.
Te, ref
0 -5 Torque [Nm] -10 -15 -20 -25 10 8 6 Current [A] 4 2 0 2000 4000 6000 8000
CONTROLLER
Te
ECB
d
DC CHOPPER
ie
DC POWER SUPPLY
Figure 7. Block scheme of the control system.
Shaft Speed [rpm]
Figure 5. 3-D characteristic of the used eddy current brake.
III.
PROPOSED TEMPERATURE RISING TEST MECHANISM
The Direct Load Method (DLM) is the most common method for temperature rising test of three phase induction motors. In this method, a mechanical load is coupled to the machine shaft directly, and the motor is fully loaded. In addition to this, it is possible to obtain exact name plate values of a motor by temperature rising test. The induction motor is loaded gradually until the temperature reaches its permitted value. Since the motor parameters change with the rising temperature, the torque and shaft speed also change. Hence, it is critical to keep motor torque constant during the test, until the winding temperature reaches steady state. Authors used an eddy current brake as the mechanical load coupled with the motor as shown in Fig. 6, where shaft speed is equal to the motor speed. Here Tm is the induction motor torque, Te is the ECB torque and is the shaft rotational speed. When the motor is heated, depending on the change in temperature and the constant load torque, motor speed decreases. Therefore, the eddy current brake must be controlled for obtaining constant torque against the variable shaft speed.
The induction motor is modeled in d-q reference frame [8]. State equations of stator currents referred to dq0 axes can be seen in (2) - (6) where Ls is stator inductance, Lr' is rotor inductance transformed to stator, Lm is mutual inductance, Rs is stator resistance, Rr' is rotor resistance transformed to stator, s is angular frequency of stator currents, r is angular frequency of rotor currents, is angular speed of motor shaft, rd and rq are d and q axes flux-linkages respectively. RE is equivalent resistance, is leakage factor and p is pole number.
disd 1 Lm Rr L rd + p m rq + Vsd = RE isd + Lssisq + 2 dt Ls Lr Lr disq dt = 1 Lm Rr L rd p m rq + Vsq RE isq Lssisq + 2 Ls Lr Lr
(2) (3) (4) (5) (6)
d rd Lm Rr R = isd r rd + r rq dt Lr Lr
d rq dt = Lm Rr R isq r rq r rq Lr Lr Lm ( isq rd isd rq ) Lr Lm Rr 2 Lr 2 Lm 2 Ls Lr
Tm = p
where
RE = Rs +
(7)
= 1
Figure 6. The basic schematic of the proposed temperature rise test system.
(8) (9)
In this study, the torque of eddy current brake is controlled by the excitation current. The block scheme of the control system is given in Fig. 7. The purpose of the control system is to provide the desired eddy current brake torque (Te,ref) by regulating the dc excitation current (ie). Therefore, a conventional PI controller is used to adjust the duty cycle (d) of the dc chopper.
s = r + p
Resistance variation with a change in temperature for conductors expressed as in (10) where RT is the resistance corresponding to the temperature T, R0 is the resistance at the temperature T0, and is the temperature-resistivity coefficient of the conductor for a specific material.
RT = R0 1 + ( T T0 ) (10)
Both stator and rotor winding resistances are modeled using (10).
The mechanical equation of the whole system is given as in (11) where Tm is the induction motor torque, Te is the ECB torque, is the shaft rotational speed, J is the inertia. Damping coefficient of the system B is neglected.
Tm Te = J d dt
IV.
SIMULATION RESULTS
In order to examine the proposed test system behavior, the simulation diagram in Fig. 8 is constructed in MATLAB Simulink R14. The test system is segmented in three different subsystems which are eddy current brake model, induction motor model, and eddy current brake controller. To examine the effect of temperature, a typical temperature profile is defined and the relation between temperature and winding resistance is modeled with (10). System parameters are given in Table 1.
(11)
[w]
e u
30/pi i_eddy
1 J.s Eddy current brake Transfer Fcn
[w]
Te_ref Controller
[w]
Tmotor
f(u) f(u) Clock Temperature Rr f(u) Rs Vs
Rr
isa
Rs
isb
i_motor
Vs isc
Induction motor
Figure 8. Block diagram of the proposed temperature rising test of an induction motor with current controlled eddy current brake.
Temperature [C]
In the simulation model, the motor is supplied by a soft starter in order to control the starting current and time by keeping the ration of voltage and frequency constant, to prevent extra heating created in starting after stopping the motor for measurements. Therefore, the model is suitable for both starting load and no load.
TABLE I. Parameters System inertia (J) Rated eddy current brake power (Pe) Induction motor stator resistance (Rs) Induction motor rotor resistance (Rr) Induction motor stator inductance (Ls) Induction motor rotor inductance (Lr) Induction motor mutual inductance (Ls) Induction motor rated power (Pm) Induction motor pole number (p) SYSTEM PARAMETERS. Value [Unit] 0.061 [kgm2] 6 [kW] 2.5275 [] 3.0018 [] 7.02255 [mH] 6.3283 [mH] 0.21486 [H] 2.2 [kW] 2
the stator winding resistance according to this temperature. It is possible to use different predictions for these quantities.
120 100 80 60 40 20
50 Time [s]
100
150
Figure 9. Predicted change of temperature vs. time.

4
Resistance [ohm]
3.5
2.5
Fig. 9 illustrates the predicted change of temperature for stator and rotor windings while Fig. 10 shows the variation of
50 Time [s]
100
150
Figure 10. Change of stator winding resistance vs. time.
Fig. 11 illustrates the soft starter output and Fig 12-13 shows the simulation results during the soft starting.
350 300 Voltage [V] 250 200 150 100 50 0 5 Time [s] 10 15 Torque [Nm]
20
15
10
50 Time [s]
100
150
Figure 15. Eddy current brake torque change during the temperature rising test.
30 25
Figure 11. Motor voltage during soft starting.

1500 Torque [Nm]
20 15 10 5
Speed [rpm]
1000
500 0 0 50 Time [s] 0 0 5 Time [s] 10 15 100 150
Figure 12. Shaft speed during soft starting.

20
Figure 16. Change of induction motor torque during the temperature rising test.
7.78
7.76 Torque [Nm] Current [A] 15
7.74
10
7.72 5 7.7 0 0 5 Time [s] 10 15 0 50 Time [s] 100 150
Figure 13. Eddy current brake torque during soft starting.
Figure 17. Eddy current brake current change during the temperature rising test.
As seen from figures, during soft starting, shaft speed increases while the induction motor is loaded by 15 Nm constant torque produced by the eddy current brake. Since shaft speed increases, the braking torque of eddy current brake also tends to increase. However, the torque is kept constant by the desired PI controller. Fig. 14-15 show that the torque of eddy current brake is constant while the rotor speed decreases as the motor is heated. Fig. 15 illustrates the speed change between 1375 rpm to 1410 rpm for seeing decrease clearly. Fig. 16 shows the motor torque and Fig. 17 shows the change of excitation current of eddy current brake according to time. Normally, temperature rising test takes a long time but in this study, accelerated testing procedure is applied to the induction motor.
1410 1405 1400 Speed [rpm] 1395 1390 1385 1380 1375 0 50 Time [s] 100 150
V.
CONCLUSION
In this paper, a new approach for temperature rising test of a induction motor is presented. The novelty of the study is the usage of a current controlled eddy current brake as the mechanical load. The constant torque required for the temperature rising test during the variable speed operation is supplied from the eddy current brake which has a conventional PI controller to regulate the excitation current of the brake. By means of excitation current control, manual regulation of load torque against motor parameter change during temperature rising test is not needed. Therefore, the proposed method offers more practical and useful way when it is compared with the usage of dynamo brakes and resistive load banks. REFERENCES
[1] IEEE 112 Test Procedure for Polyphase Induction Motors and Generators and International Standards IEC 60034-29 Equivalent Loading and Superposition Techniques to Determine Temperature Rise. I. olak, G. Bal, . Elmas, Review of the Testing Methods for Full-Load Temperature Rise Testing of Induction Machines, EPE Journal, Vol. 6, no. 1, May 1996. A. F. Mergen, D. A. Kocabas, Elektrik Makinalarnda Sarglar (Windings in Electrical Machines), 1st ed. (in Turkish), Istanbul: Birsen Yaynevi, 2007.
[2]
[3]
Figure 14. Shaft speed during the temperature rising test.
[4] E. R. Summers, Determination of Temperature Rise of Induction Motors, American Institute of Electrical Engineers, Transactions of the, vol. 58, no. 9, pp. 459-467, September 1939. Gulbahce, M. O., Kocabas, D. A., Habir I. Finite Element Analysis of a Small Power Eddy Current Brake, Mechatronica, 2012, IEEE Conference, 5-7 December, 2012. Gulbahce, M. O., Kocabas, D. A, Atalay A. K., A Study to Determine the Act of Excitation Current on Braking Torque for a Low Power Eddy Current Brake, IEMDC-13, IEEE Conference, 12-15 May 2013 Chicago-Illinois. Gulbahce, M. O., Kocabas, D. A, Atalay A. K., Determination of the Effect of Conductive Disk Thickness on Braking Torque for a Low Power Eddy Current Brake, POWERENG-13, IEEE Conference, 13 17 May 2013, Istanbul, Turkey. Kocabas, D.A., Salman, E., Atalay, A.K., Analysis using D-Q transformation of a drive system including load and two identical induction motors, IEMDC-11, IEEE Conference, 15-18 May 2011, Niagara Falls, ON. M. O. Gulbahce, Orta Guclu Bir Girdap Akimi Freni Tasarimi ve Sonlu Elemanlar Yntemi ile Analizi (Design and Finite Element Analysis of a Medium PowerEddy Current Brake), Msc. Thesis, Istanbul Technical University, Graduate School of Science Engineering and Technology, 2012.
[5]
[6]
[7]
[8]
[9]
Harmonic Network Impedance Identification with Pulse Response Analysis

Michael Jordan, Trung Do Thanh, Hauke Langkowski, and Detlef Schulz
Department of Electrical Engineering, Chair of Electrical Power Systems, Helmut-Schmidt-University (HSU) Holstenhofweg 85, 22043 Hamburg, Germany
Abstract-The network impedance at a point of common coupling (PCC) of public mains is an essential parameter for design and control of power electronic systems connected to the PCC. Identifying its spectral characteristic is challenging, because it requires a measureable excitation of the network in the desired frequency range. Further on, the network impedance varies over time. Therefore, the identification process should be fast and repeatable. In this paper a system is presented, which allows a direct excitation of the network impedance on the medium-voltage level with defined current pulses. Short current pulse signals are evaluated and optimized in a simulation model in order to realize a precise and fast determination of the network impedance with this system. Keywords-impedance measurement; R/X-ratio; pulse response;
I.
INTRODUCTION
Driven by the fast growing share of Renewable Energy on the power production with a high number of distributed generators (particularly wind turbines and photovoltaic systems), the number of grid-connected power converters has strongly increased and is likely to continue rising in the next years. An optimized design and failure-free operation of these systems requires a precise knowledge of the respective network connection point. The electrical characteristic of an AC network connection point can be fully described by the open circuit voltage and the internal impedance at the corresponding terminals. In most cases the open circuit voltage can be easily measured and analyzed. Measure the network impedance is more complicated, because it requires an excitation of the network and precise measurement and analyzing equipment. Yet, the network impedance at a PCC is of high interest for a number of applications. The absolute value of the network impedance at the nominal frequency defines the short circuit power of a PCC and therefore its capacity. Absolute value and phase (respectively the R/X-ratio) of the fundamental impedance is also a substantial parameter for filter design [1] as well as for control, stability evaluation, and islanding detection of power converters [2-5]. Further on, the impedance components of higher frequencies above the fundamental one are of special interest for power quality evaluation because harmonic voltage distortions, produced by power converters and other nonlinear loads and generators, are proportional to the corresponding harmonic network impedances [6][7].
In practice the network impedance is often just assessed for the fundamental component with the short circuit power and the nominal voltage. The short circuit power is obtained from rough analytical calculations or grid simulation programs. Higher harmonic components are estimated by multiplying the fundamental component with defined factors [7]. If detailed information of the power system are available, resonance effects at higher frequency can also be analyzed by computer simulations [1]. Yet, the complex and time varying grid structure with a high number of distributed fluctuating loads and generators as well as unknown parameters, especially on the load side, usually prevent a realistic modeling of the network impedance. Due to lag of information and the non-stationary nature of public mains a correct determination of the network impedance can only be achieved through measurement over time. In this contribution short pulse signals are evaluated to determine the harmonic network impedance of a PCC in the range from 50 Hz to 10 kHz. First a general overview of applied impedance identification methods is given. Afterwards a power electronic system is presented to measure the network impedance on the medium-voltage level. A detailed simulation model of the system is combined with a basic network model to assess and optimize the identification process with this system. II. IMPEDANCE IDENTIFICATION METHODS Several methods have been presented to identify the network impedance. In [5] and [8] the transient response of injected current spikes are analyzed to measure the network impedance. Other proposed methods feed a sinusoidal interharmonic [4] or white-noise perturbation [9] current into the grid and measure the corresponding network voltage in steady state. These methods have in common that converters are used to excite the network on the low-voltage level. The impedance identification is carried out online during the normal operation of the power controller and has to be very fast and accurate especially for islanding detection. A drawback of these stated online methods is that a high excitation has to be generated by the converter for the identification of frequency above the fundamental due to the attenuation of the output filter. This has a negative effect of the converter efficiency in normal operation [9]. In further approaches, short transients or pulse width modulated current
test signals elicit by switched capacitors or resistors have been used to identify the network impedance on low-voltage PCC over a wide spectral range [10]. The determination of the network impedance above the low-voltage level is difficult due to the low impedance magnitude and smaller feasible excitation, which requires precise wideband current and voltage probes suitable for these higher voltage levels. The spectral excitation of higher voltage levels has been realized by switched capacitor banks [11] or energisation of a power transformer [12] with good measurement results. However, strong excited spectral components are limited with these methods and cannot be varied. In addition, the switching is done by circuit breakers, which have a maximum cycle of operation. This strongly limits the repetition rate and number of possible measurement cycles. In [13] measurements are presented, which use test signals generated by a power converter and fed by a transformer station to the high-voltage level (110 kV) to identify the network impedance. The spectral range of these measurements was limited from 0.5 to 2.5 kHz, because the traditional voltage transducer used on the high-voltage level have non-linear frequency response with resonances close to the nominal frequency. Investigations have shown that the frequency response of traditional transducer is influenced by several factors, which cannot all correctly be captured by calibration measurements [14][15]. Consequently, special wideband sensors should be used for reproducible harmonic impedance measurement. Further on, current signals with high amplitude have to be generated on the low-voltage level to elicit measurable signals on the higher voltage level due to the transformation ratio and strong attenuation of the transformer above the nominal frequency. For this reason, a direct excitation of the respective voltage level is favorable. III.
NEW IMPEDANCE MEASUREMENT SYSTEM
thyristor bridge
IGBT
fire pulse generators and IGBT driver unit

Figure 1. The power electronic system for impedance measurment on the medium-voltage level.
To identify the network impedance in a three-phase three-wire system on the medium-voltage level, an asymmetrical phase switching method can be applied with the system, presented in [10][17]. Two phases are cyclically linked by the thyristor bridge to the serial circuit of load and IGBT. Depending on the load and the switching of the IGBT different excitation signals can be generated to measure the network impedance with this method. For wideband impedance identification with high spectral resolution frequency-swept pulse width modulated (PWM) or random pulse width modulated (RPWM) signals can be applied to the IGBT-switch in the system. These signals are spectrally analyzed over several fundamental periods and averaged over a few iteration loops, which restricts the time resolution [10]. Another possibility to excite the network with the system is to elicit short pulse signals with high energy. The corresponding transients are analyzed in a short time interval, which allows fast impedance identification, but with lower spectral resolution and accuracy. IV. SIMULATION MODEL FOR NETWORK IMPEDANCE IDENTIFICATION
For a direct and controllable spectral network excitation at medium-voltage level an innovative power electronic system has been developed at the Helmut-Schmidt-University (HSU) in cooperation with ABB Switzerland Ltd. Semiconductors. Fig.1 shows a photo of the system. It consists of six thyristor stacks arranged to a three phase bridge rectifier and an IGBT module with corresponding pulse generators and IGBT driver unit. Every thyristor stack is made of eight serial connected 3.3 kVDC thyristors adding up to a peak off-state voltage of 26.4 kVDC for every stack. The IGBT module consists of 14 serial connected 4.5 kVDC IGBT (divided into two stacks), which theoretically leads to a total blocking voltage of 63 kVDC for the IGBT module. However, due to the overvoltage protection circuitry with metal-oxide varistors (MOV) and active-clamping Zener diodes, the maximum blocking voltage of the IGBT module is reduced to 32 kVDC. The topology and switching-concept of the power electronic system is described in detail under [16].
In order to analyze different load switching configurations a simulation model of the new measurement system has been set up using the PLECS toolbox within the Matlab/Simulink software environment [18]. Fig. 2 shows the main components of the simulation model. The simulation model was originally designed to test and optimize the controller program, which generates the trigger signals and has been expanded with a basic network model. The network voltage VN is modeled with ideal voltage source blocks for the fundamental and additional harmonic distortion. The blocks are set to a phase to phase voltage of 10 kVAC superimposed with a third (2.5 %) and a fifth harmonic (3 %). For the evaluation of different excitation signals a simplified model for the network impedance ZN has been chosen consisting of equal serial ohmic and inductive
network
VN ZN A B C
measurement
power electronic system
IGBT-switch subsystem
Figure 2. Simulation model of the power electronic system.
components for every phase. Coupling impedances, resonances and asymmetries are neglected in this evaluation. The thyristor bridge consists of single standard blocks with high-ohmic parallel resistors considering the voltage balancing circuit of the thyristor stacks. The IGBT module determines the switching characteristic of the system and is modeled in more detail in a subsystem (see below in Fig. 2). It contains 14 serial connected IGBT blocks with parallel balancing and protective circuitry. The current rise and fall characteristics of the IGBT module were measured in a high-voltage laboratory and are included in the IGBT blocks to model the switching characteristic as real as possible. The current rise time is mainly determined by its final value and the inductive reactance of the network impedance. It is set to 2 s. The current fall time of the IGBT module is around 10 to 20 s. It depends also on the switched-off current level, the circuit inductivity and the resulting overvoltage. Critical voltage peaks during the switch-off are suppressed by an active clamping circuit implemented with Zener diodes between collector and gate of every IGBT. At a collector emitter voltage close to 2.3 kVDC the gate of an IGBT is charged and operates in linear range reducing the overvoltage [16]. Unfortunately, this active clamping concept of the IGBT module cannot be realized in the simulation model, because the gates are digitally controlled and the on-resistance of the IGBT models cannot be dynamically changed. Yet, this should introduce only minor errors to the modeled switching characteristics since the network phase to phase voltage is set
to 10 kVAC and is less than half of the 32 kVDC clamping voltage (14 x 2.3 kV). V.
PULSE RESPONSE NETWORK IMPEDANCE IDENTIFICATION
To simulate different excitation signals varied load are switched between phase A and B with defined pulse duration. The corresponding phase to phase voltage and the current characteristics are recorded. To consider the harmonic distortion from the network the open circuit voltage is also recorded before the switching process. The recorded parameters are spectrally analyzed with a fast Fourier transform (FFT) algorithm. The spectral range and resolution depends on the sampling rate and the analyzed time interval. For the following evaluations a time interval with adjusted sampling rate for the FFT has been chosen. Two 50 Hz fundamental periods are sampled and divided into two data sets. The first contains the open circuit voltage over one period. In the second recorded period an excitation signal is triggered. The sampling rate is set 409600 Hz, which results in 8192 data points for every parameter and a spectral resolution of 50 Hz. Since the network is symmetrically designed, the network impedance can be easily calculated in the frequency range: . (1)
Where V ÀB is the open circuit voltage and V `ÀB and I `À the voltage and current characteristics during the load impedance switching process.
A. Measurement Uncertainties To consider the limited precision of the measurement equipment in the simulation model, noise is added to the output signals of the ideal voltage and current measurement blocks. For voltage measurement a precise high-voltage divider is used with a measurement uncertainty of 0.2 % from DC to 10 kHz. The current transducers have a maximum amplitude error of 0.02 % and a phase error of 0.02. Calibration measurements have shown that the total accuracy including measurement and quantization noise is smaller than 25 V for the voltage measurement and 20 mA for the current measurement. The phase error of both sensors is insignificant and can be neglected in the simulation model. A. General Method As stated in section III, the network impedance excitation can be accomplished by transients caused by short energy signals or by longer power signals in steady state. In this section pulse signals with varying energy (respectively amplitude and duration) are evaluated. The corresponding transients are analyzed in a short interval of one fundamental period, which compromises a high time resolution but a low frequency determination. Fig. 3 shows the current and voltage characteristic of a short current pulse generated by switching a resistive 1000 k load at the voltage maximum with a duration of 1 ms. The spectral components of the current pulse and the differential voltage (VAB), are displayed in Fig. 4. The Fourier transform of the approximate rectangular shape of the current pulse is a Sinc-function. Its amplitude and frequency depends on the pulse duration Tp and maximum value Imax. The corresponding voltage components depend on the network impedance, which has been set to 0.5 and
1.59 mH. The network impedance linearly rises with frequency while the current components decrease with the envelope of a Sinc-function. The local minima appear at integral multiples of 1/Tp. At this frequency the impedance determination has high errors, because the current excitation and resulting voltage response is low. Fig. 5 illustrates this effect for different pulse width. The load impedance has not been changed and the energy of the excitation pulses decreases with shorter pulse duration. The spectral components are smaller for lower frequencies, which leads to poor identification results in this range. This is especially the case for the 0.1 ms pulse. The current harmonic components up to 2 kHz are too small to elicit voltage components over the noise level due to the relative small network impedance in this low frequency range. However, the components are spread more evenly over a wider
Figure 3. Current and voltage characteristic of a 1 ms pulse.
Figure 4. Spectral components of the 1 ms pulse signal.
Figure 5. Simulation results with different pulse durations.
spectral range, which improves the impedance determination at higher frequencies. Therefore, the 0.1 ms pulse signal shows good results in the range from 2 kHz. to 9.0 kHz. Above this frequency the excitation is low due to the first minima at 10 kHz. Whereas, the 1 ms pulse signal has the highest voltage excitation in the low frequency range. If the pulse amplitude is fixed, it becomes a trade-off between higher excitation in the lower frequency region with longer pulses and broader spectral identification with short pulses. The right choice depends on the magnitude of the network impedance, which can differ widely in practice. B. Pulse Response Analysis The current sensors used in the system have a range of 60 A, which defines the limit for the maximum pulse amplitude. The load impedance is set to 250 , which leads to a maximum pulse amplitude of 56 A. For the following evaluation two different pulse widths are used to identify the network impedance. For the determination up to 10 kHz a pulse length of 80 s is applied to the IGBT. For a more accurate identification in the low spectral region to 2.5 kHz a pulse duration of 300 s is used. Three different network impedance scenarios for a 10 kV PCC on the medium-voltage level are simulated (see Table I). Magnitude and R/X-ratio are increased for every scenario. The values in these scenarios are typical for network impedances in highly meshed grids in urban areas.
TABLE I. No. 1 2 3 R () 0.05 0.15 0.5
NETWORK IMPEDANCE SCENARIOS
XL(50Hz) () 0.25 0.30 0.5
L (mH) 0.795 0.954 1.59
ZG(50Hz) () .254 0.335 0.707
R/X 0.2 0.5 1
The simulation results are displayed in Fig. 6 and Fig. 7. As expected, the identification with 80 s pulse duration signals show good results at frequency above 2 kHz, but have unacceptable errors at low harmonics. The determination accuracy at low harmonics is increased with the longer pulse width. Yet, the pulse excitation close to the fundamental is too weak, especially for the low network impedances in scenario 1 and 2. C. Fundamental Impedance Determination To identify the fundamental component in low impedance networks the load can be switched in the voltage zero-crossing over a full period. The resulting current wave package is displayed in Fig. 8. The fundamental impedance can be fast identified with high precision by calculating the first Fourier series coefficient of the parameters in (1). The voltage drop caused over the network impedance is small but measurable even for the low impedances in scenario 1 and 2. The results of the impedance determination with this method are stated in Table II.
Figure 6. Simulation results with 80 s pulse signals.
Figure 7. Simulation results with 300 s pulse signals.
REFERENCES
[1] [2] G.J. Wakileh, Power Systems Harmonics - Fundamentals, Analysis and Filter Design. Berlin: Springer, 2001. M. Liserre, R. Theodorescu, and F. Blaabjerg, Stability of photovoltaic and wind turbine grid-parallel inverters for a large set of grid impedance values, IEEE Transactions on Power Delivery, vol. 21, no. 1, pp. 263272, 2006. J. C. Vasquez, J. M. Guerrero, A. Luna, P. Rodriguez, R. Theodorescu, Adaptive Droop Control Applied to Voltage-Source Inverters Operating in Grid-Connected and Islanded Modes, IEEE Transactions on Industrial Electronics, vol. 56, no. 10, pp. 4088 -4096, 2009. L. Asiminoaei, R. Theodorescu, F. Blaabjerg, and U. Borup A digital controlled PV-inverter with grid impedance estimation, IEEE Transactions on Power Electronics, vol. 20, no. 6, pp. 1480-1490, 2005. H. Gu, X. Guo, D. Wang, and W. Wu, Real-time Grid Impedance Estimation Technique for Grid-Connected Power Converters, IEEE International Symposium on Industrial Electronics (ISIE), pp. 16211626, Hangzhou, 28-31 May 2012. M. Cspedes, J. Sun, Online Grid Impedance Identification for Adaptive Control of Grid-Connected Inverters, Energy Conversion Congress and Exposition (ECCE) IEEE, Raleigh, pp. 914-921, 15-20 Sept. 2012. D. Schulz, Power Quality - Theory, Simulation, Measurement and Assessment. (In German) Offenbach: VDE-Verlag, 2004. J. Schlabbach, and W. Mombauer, Power Quality Origin and Assessment of Phase Effects; Grid Connection of Renewable Energies. (In German) Offenbach: VDE-Verlag, 2008. M. Sumner, B. Palethorpe, and D. W. P. Thomas, Impedance measurement for improved power quality-part 1: the measurement technique, IEEE Transactions on Power Delivery, vol. 19, no. 3, pp. 1442-1448, 2004. D. Martin, E. Santi, and A. Barkley, Wide bandwidth system identification of AC system impedances by applying perturbations to an existing converter, Energy Conversion Congress and Exposition (ECCE) IEEE, pp. 2549-2556, Phoenix, 17-22 Sept. 2011. M. Jordan, H. Langkowski, T. Do Thanh, and D. Schulz, Frequency dependent grid-impedance determination with pulse-width-modulation signals, 7th International Conference-Workshop on Compatibility and Power Electronics (CPE) IEEE, pp. 131-136, Tallinn, 1-3 June 2011. M. Nagal, W. Xu, and J. Sawada, Harmonic impedance measurement using three-phase transients, IEEE Transactions on Power Delivery, vol. 13, pp. 272-277, January 1998. C. Xie, S.B. Tennakoon, , R. Langella, D. Gallo, A. Testa, and A. Wixon, Harmonic impedance measurement of a 25 KV sin gle phase AC supply system, Proceedings of Ninth International Conference on Harmonics and Quality of Power, Orlando, vol. 1, pp. 214-219, 1-4 October 2000. J. Xie, Y. X. Feng, and N. Krap, Network impedance measurements for three-phase high-voltage power systems, Asia-Pacific Power and Energy Engineering Conference (APPEEC), Chengdu, 28-31 March 2010. T. Pfajfar, J. Meyer, P. Schegner, and I. Papi, Influence of Instrument Transformers on Harmonic Distortion Assessment IEEE Power and Energy Society General Meeting, San Diego, pp. 1-6, 22-26 July 2012. M. Klatt, J. Meyer, M. Elst, and, P. Schegner, Frequency Responses of MV voltage transformers in the range of 50 Hz to 10kHz 14th International Conference on Harmonics and Quality of Power (ICHQP) IEEE, Bergamo, pp. 1-6, 26-29 Sept. 2010. T. Do Thanh, R. Leutwyler, M. Jordan, H. Langkowski, and D. Schulz, Power Electronic Topology and Switching-Concept for Identification of the Grid Impedance at Medium-Voltage-Level (in German) International ETG-Congress 2011, Wrzburg, 8-9 November 2011. T. Do Thanh, T. Schostan, K.-D. Dettmann, and D. Schulz, Nonsinusoidal power caused by measurements of grid impedances at unbalanced grid voltages, Proceedings of the IEEE conference of the International School on Nonsinusoidal Currents and Compensation ISNCC, Lagow, 10-13 June 2008. Plexim GmbH, PLECS Blocksets The Power Electronic Toolbox for Simulink, [online ] http://www.plexim.com, 2013.
[3]
[4]
Figure 8. Current wave package over one fundamental period. TABLE II. No. 1 2 3
RESULTS OF THE FUNDAMENTAL IMPEDANCE IDENTIFICATION
[5]
ZREF () 0.05+0.25i 0.15 + 0.3i 0.50 + 0.5i
ZM () 0.0482+0. 2732i 0.1476 + 0.3057i 0.5003 + 0.5098i
|Z| () 0.0227 0.0043 0.00751
() 1.3 0.82 0.56
[6]
[6] [7]
D. Discussion The simulated network impedance identification process with short pulse signals or wave packages shows satisfying results. The accuracy of the determination can be further improved by averaging the results of several iterations. However, this will reduce the time resolution. In the simulation model only measurement noise from the sensors and quantization is included. Yet, other noise sources (e.g. electromagnetic interferences) may produce additional errors. Further on, variation of the network voltage amplitude and frequency during the identification process can affect the accuracy of the measurement in practice. The system will soon be tested on a 10 kV laboratory network connection, which is in construction. Measurements will be presented in future publication. VI.
CONCLUSION
[8]
[9]
[10]
[11]
[12]
[13]
In this contribution a power electronic system has been presented to measure the network impedance on the medium-voltage level. The system excites the network impedance with pulsed current signals and measures the corresponding current and voltage characteristics. Different signals have been investigated and optimized in a simulation model for the 10 kV voltage level. The results indicate that the impedance can be determined with adequate precision and a high time resolution with adjusted current pulse signals. ACKNOWLEDGMENT This topic was investigated within the project Development of a measurement device for the determination of the time and frequency dependent grid impedance on the medium-voltagelevel. The project is funded by the German Ministry for the Environment, Nature Conservation and Nuclear Safety under the support code 0325049.
[14]
[15]
[16]
[17]
[18]
Dynamic Analysis of a Telecommunication Station Grounding System, Applying the TLM Method
T. I. A. H. Mustafa, H. D. Almaguer, L. H. Meyer, IEEE Member, S. H. L. Cabral and J. E. Baader
Depart. of Electrical Engineering and Telecommunication University of Blumenau -FURB Blumenau/SC Brazil tim@furb.br, hugo@furb.br, meyer@furb.br, scabral@furb.br, eduardo_baader@hotmail.com
Abstract This work presents a dynamic analysis of a telecommunication station grounding system by applying the TLM- Transmission Line Modeling method to predict the behavior of the grounding system under impulsive condition due lightning discharges. Results obtained from simulations can help in studies and analyses of topologies that are employed in electric power network and associated telecommunication stations. Index Terms Grounding system, TLM, lightning discharges, telecommunication station.
J. Anchieta J. Jr.
Companhia Estadual de Energia Eltrica, CEEE-GT Porto Alegre/RS Brazil anchieta@ceee.com.br adopted by most of electric utilities and telecommunication companies. II. MODELLING OF THE GROUNDING APPLYING THE TLM
I.
INTRODUCTION
Telecommunication towers such as for microwave communication or radio repeaters are very high structures, placed in remote isolated locations and subject to lightning discharges. Among several consequences of the incidence of any lightning discharge, one of the most significant is the distribution of voltage surges along different points of the grounding system. Thus, several studies have been carried out by analyzing transferred or induced voltages due to lightning discharges through grounding systems [1-3]. In this context, numerical method with TLM method [4] appears as an attractive tool to be taken, since it has already been successfully applied in studies of behavior of electric grounding problems under impulsive condition [5]. Thus, application of the TLM method is performed in this work for evaluating the distribution of voltage surges after a lightning discharge and it is based on the use of a transient grounding resistance (TGR), instead of a grounding resistance with constant value, that has been adopted by the electric utilities in studies of topologies that are employed in electric power distribution network. Results of voltage level in a grounding system after the incidence of lightning discharges on a telecommunication tower will be presented, applying the TLM modeling for showing the behavior of the grounding system from practices
The TLM method is a numerical modeling technique based on temporal and spatial sampling of electromagnetic fields [4]. In a typical TLM simulation, a mesh of stubs of transmission lines represents the propagation space, whereas electric and magnetic fields are made equivalent to voltages and currents on the network, respectively. Simulation starts by exciting the mesh at specific points with a voltage impulses and consequent propagation of these impulses over the mesh as they are scattered from the nodes and bounce at boundaries. In the case presented in this work, the 3D mesh that is employed is based on the Symmetrical Condensed Node (SCN). More details about TLM formulations, its applications and computational requirements can be found in [4,5]. Fig. 1 shows that each node has 12 voltage ports, and six total field quantities (Ex, Ey, Ez, Hx, Hy and Hz) at the center of the cell are calculated from these voltages. More details about TLM formulations, its applications and computational requirements can be found in [4,5].
Figure 1. The TLM Symmetrical Condensed Node (SCN). This work was financially supported by CEEE (Companhia Estadual de Energia Eltrica do RGS) and FURB (The University of Blumenau) from Brazil.
Fig.2 shows a grounding topology that was initially studied to demonstrate the efficiency of the TLM method. It consists in a steel copperweld grounding electrode with 13 mm of diameter, 2.4 m of length and vertically placed. Soil parameters are relative dielectric permittivity, r = 10 and soil resistivity, , ranging from 10 to 1500 .m.
Figure 3. Measured and TLM voltage applied to the rod of the grounding topology.
Figure 2. Grounding topology with a vertical copperweld rod;
Thus, Fig. 4 presents results of the TGR for experimental and TLM simulations. It can be observed that wave shape and trend of the TLM results are very similar to the experimentally simulated one. The value obtained for the steady-state stabilizes in about 56 , for both.
The excitation source is an impulsive current surge applied to conductor that takes part of the grounding system and it is connected to the descending conductor. This type of lightning current surge modeling is recommended by the International Electrotechnical Commission IEC 61000 [6]. Thus, the transient grounding resistance (TGR) of this structure is determined through TLM simulation in the time domain. Its evaluation is performed at each iteration by the ratio between the voltage, Vn(t) and the current, In(t), at the rod excitation node:
TGR(t ) = Vn (t ) I n (t )
(1)
Figure 4. Comparison between experimental and TLM for TGR.
The number of iterations in the time domain must be sufficient to allow convergence of value of TGR (the transient component) to the grounding resistance value at steady-state. All conductive structures above the soil surface (such as the descending conductors of the protection system) are not considered in the simulation, as well as any ionization or skin effect in the soil. On the other hand, for the sake of comparison of results, experimental tests were performed at the high voltage laboratory of FURB for a similar condition. Especially, in order to analyzing the experimental behavior of the TGR to compare with TLM simulation. Therefore, Fig. 3 shows experimental and TLM behavior of the voltage to far points applied to the steel copperweld above described in a soil with homogeneous value of 96 .m for the electric resistivity.
After this comparison, the behavior of the transient grounding was then studied for the same grounding topology but taking into account additional items like connecting conductors, changes in the values of electric resistivity and relative dielectric electric permittivity of the soil. Thus, additional simulations of this case were performed with TLM to compare values of the steady state grounding resistance to those calculated by the classical analytic expression, for different values of electric resistivity [3]:
R=
4L ln 1 2L a
(2)
where L is the rod length, a the radius of its circular cross section and is the soil resistivity, taken as being homogeneous. Results are presented on Table I, in which it can be seen that there is a better agreement between TLM and
[3] for lower values of electric resistivity. For higher values of resistivity would be needed experimental measurements to understand the significant difference between the values of resistance.
TABLE I. Resistivity (.m) STEADY STATE GROUNDING RESISTANCE R () from TLM R () from analytic expression (2)
10 200 400 500 1500
4.28 70.23 129.76 158.00 423.35
4.18 83.53 167.05 208.82 626.44
III.
THE DYNAMIC ANALYSIS

Figure 6. Electric field distribution at the surface above the grounding mesh in t=0.05 us.
From the physical and structural characteristics of a grounding system, it is possible to perform simulations by the numerical method TLM to evaluate the performance of the grounding. Based on results presented in previous sections, it is evident the versatility of a TLM simulation, since the input data can be easily changed, thus allowing a complete analysis on a range of values that parameters and conditions, such as the soil resistivity and lightning discharges can take on studies and dynamic analysis of grounding systems. For the grounding system shown in Fig. 5, which was designed for a telecommunications station CEEE-GT (electric utility), a series of simulations were performed using the TLM method, based on measurements of parameters such as soil resistivity and resistance of the mesh, as well as the physical dimensions of the grounding system for the analysis of dynamic behavior of some electrical parameters such as surge impedance and electric field.
Figure 5. Grounding system designed for telecommunications station of CEEE-GT.
The Fig. 6 to 11 show the normalized results obtained from a simulation on the grounding system, considering that an impulsive current surge (1/3s and 3500A of peak [6]) is applied to the tower grounding as shown in Fig. 5. The time interval of the analysis was 15 s, thus allowing to evaluate the dynamic behavior and distribution of the electric field in the geometric plane above the conductors of the mesh, that were placed at 50 cm below the soil surface.
In Fig. 9 (t = 1.7 s) the electric field reaches its maximum value around 984,88 kV/m. Already in Fig.10 the electric field begins to be attenuated, being largely attenuated from Fig. 11 in t= 8 s. It can be seen that high amplitude of electric field is concentrated near the region where the impulsive surge occurred, i.e. a region filled with many conductors. Therefore only a limited area of the grounding system is able to effectively dissipate the currents impulsive in soil. This demonstrates the importance of having a dense region of conductors near to the insertion points of a possible lightning discharge. IV. CONCLUSION
In this work some results of the application of TLM method are presented and allow concluding some important items about behavior of overvoltage level in the grounding system of a Telecommunication Station after the incidence of a lightning discharge. An important aspect is that the concentration of the electric field depends on the point at which impulsive current is injected (mainly due to the geometry of the conductors). As each injection point have different performance characteristics field, it is expected that the surge impedance that is measured at different points also different. It is known from other studies that for soils with low resistivity only a small area of conductors of the grounding system, located around of the injection point of the surge current is sufficient to dissipate the current in the soil. However, for high resistivity soils, it is expected that the surge current is dissipated over a larger area of conductors of the grounding system. For a next step of this research new set of simulations will be performed with addition of other and more realistic configurations of grounding topologies. ACKNOWLEDGMENT The authors would like to thank the Companhia Estadual de Gerao e Transmisso de Energia Eltrica CEEE-GT in Rio Grande do Sul State and University of Blumenau FURB in Santa Catarina State, Brazil for their financial support in the researches. REFERENCES
[1] [2] [3] [4] [5] A. Piantini and J. M. Janiszewski, Induced voltages on distribution lines due to lightning discharges on nearby metallic structures, IEEE Transactions on Magnetics, vol. 34, n. 5, pp. 2799-2802, Sep. 1998 M. Tell, G. Dias, A. Raizer, H. D. Almaguer, T. I. Mustafa and V. Coelho, Aterramento Eltrico Impulsivo em Baixa e Altas Freqncias, EDIPUCRS, 1a Edition, 2008. S. F. Visacro, Aterramentos Eltricos. So Paulo: Ed. Artliber, 2002. C. Christopoulos, The Transmission-Line Modeling Method- TLM, New York: IEEE Press and Oxford University Press, 1995. H. D. Almaguer, Contribuio ao Mtodo da Modelagem por Linhas de Transmisso (TLM) e sua Aplicao aos Estudos em Bioeletromagnetismo, Thesis of PhDs Degree in Electrical Engineering, Federal University of Santa Catarina, Brazil, 2003 International Standard - IEC - International Electrotechnical Commission, Section of IEC 61000 4 5, Second Edition, Nov.2005.
[6] Figure 11. Electric field distribution at the surface above the grounding mesh in t=8.0 us.
Real Time Implementation of H Loop Shaping Robust PSS for Three-Machine Power System using dSPACE
Dr.Jayapal R,
Professor, Dept. of Electrical & Electronics Engg., R.V. College of Engg., Bangalore, India. Email: jayapalr @ rvce.edu.in Abstract This paper presents the experimental evaluation of the designed power system stabilizer for a 3-machine, 9-bus power system. Based on the non-linear robust control theory, a control design is applied to design a Robust Power System Stabilizer (RPSS) to stabilize a linearized uncertain power system using Glover-McFarlane H loop shaping design procedure. The efficiency of the designed controller is tested by implementing on real time environment using dSPACE work stations DS1005 and DS1104. The real time experimental results of RPSS are compared with that of the conventional power system stabilizer (CPSS) for a three phase fault. Also, the real time simulation results of RPSS are compared with the off-line simulation results of RPSS, thus validating the simulation results with the experimental results. Justification of robustness is also presented by considering three different operating points. Keywords dSPACE, H Loop shaping, Multi machine, Real time simulation, Robust power system stabilizer, State space. I. INTRODUCTION
Dr. J K. Mendiratta
Professor, Dept. of Electrical Engg., MVJ College of Engg., Bangalore, India. Email: mendiratta.jk@gmail.com are developed for SMIB systems with synchronous machine models based on Model 1.1[9] and for a series compensated line based on Model 2.2 [10]. Dual-input PSS using GloverMcFarlane H loop-shaping technique was designed to mitigate the Sub Synchronous Resonance [7]. Chuanjiang Zhu et.al, [2] have designed a PSS using Glover-McFarlane H loop-shaping technique for multimachine system based on sequential tuning, while only the authors [11] [12] have used the Glover-McFarlane H loop shaping technique with the normalized coprime factor [13] robust stabilization technique for multimachine system based on decentralized scheme with simultaneous tuning to design the RPSS. Various researchers [14] [17] across the globe have implemented different types of PSS with different real time methodologies by conducting experiments for small laboratory type power system models. The PSS is designed using Fuzzy Logic Control (FLC) and Genetic algorithm [14] and the effectiveness and robustness of the designed stabilizer was investigated by comparing with CPSS by implementing it to the laboratory model. The performance of the PID based PSS designed using H static output feedback (H-SOF) control technique [15] was tested in real time using the Analog Power System Simulator and the results of the proposed full-order H and CPSS are compared to show that the performance of Robust PSS was superior. The performance of the fuzzy logic power system stabilizer (FLPSS) was compared with PID power system stabilizer (PIDPSS) experimentally [16] in the real time on a laboratory set up of SMIB model and showed the effectiveness of FLPSS in vulnerable operating points. A robust controller based on Fuzzy Logic was designed [17] to damp power system oscillations using the SSSC and Real-time simulations are conducted to demonstrate the validity of the proposed damping controller using the RTDS interfaced with an external DSP-based digital control system on SMIB system. Not many attempts are made to implement the PSS in real time for multimachine systems. Majority of the real time works on PSS was implemented using TI based DSP cards, Data acquisition cards, PID controller cards, etc. for small laboratory type power systems. Not many attempts were made to implement the same in real time using dSPACE interfacing cards. In our work, the designed RPSS is implemented in real time using the dSPACE boards [18] for multimachine power system. The paper is organized in the following sequence. A brief introduction about the background literature related to the H control and real time implementation of PSS was presented in the introductory section. In Section II, the power system model description and problem statement are provided. In
Power System Stabilizer (PSS) is installed to achieve desired stability and security by modulating the generator excitation by providing additional damping to electromechanical oscillations of synchronous machine rotors and to enhance the power transfer capability of the power system [1]. Advanced control designs in power systems like adaptive control, H control, synthesis [2], non-linear control, feedback linearization [3], periodic output feedback and sliding mode control [4] are employed recently. The goal of these studies is to achieve power system stability and performance robustness upon the occurrence of faults. Conventional stabilizers are not designed to guarantee the desired level of robustness as their designs are specific for a given operating point; they do not guarantee robustness for a wide range of operating conditions [5]. To include the model uncertainties at the controller design stage, modern robust control methodologies have been used to design PSS [2], [3]. The resulting PSS ensures the stability for a set of perturbed operating points with respect to the nominal system and has good oscillation damping ability. The proposed RPSS based on Glover-McFarlanes H shaping technique [6] is free from common deficiencies of normal power system nonlinear controllers which are network dependent and equilibrium dependent. The H optimal controller design is relatively simpler in terms of the computational burden. Glover-McFarlanes H shaping technique is applied to design RPSS with respect to model uncertainties of the power system for Single Machine connected to Infinite Bus (SMIB) systems [7], [8] only. RPSS using Glover-McFarlanes H loop shaping technique
Section III, the controller design paradigm is given. Real time implementation of the designed RPSS using dSPACE cards is presented in Section IV along with the comparison with real time implementation of CPSS and the simulation results of RPSS. The paper concludes with the final conclusions, followed by the references & the author biographies. II. DEVELOPMENT OF THE POWER SYSTEM MODEL To study the control of power system oscillations, 3machine, 9- bus system taken from [19] is used. The relevant data of the system is available in [11]. In this system, the synchronous machine is modeled using Model 1.1[20] in which case one field winding on d-axis and one equivalent damper on q-axis are considered. Stator and network are modeled by algebraic equations. Each machine is expressed in its own d-q reference frame which rotates with its rotor. For solving interconnecting network equations, all voltages and currents are expressed in a common reference frame rotating at synchronous speed (D-Q or Krons reference frame). Axes transformation equations are used to transform between individual machine (d-q) reference frames and the common (D-Q) reference frame. The system model is created in the Simulink/Matlab using the relevant algebraic and differential equations [11] [20] of Model 1.1. If the PSS design is based on the one machine infinite bus model, after the installations of PSSs on most machines of a large power system, low frequency oscillations may still occur due to lack of coordination of these stabilizers [21]. To achieve the coordination, the state matrix of the entire system is used to design PSS using Glover-McFarlane H loop shaping design procedure. For the system considered this procedure yields three stabilizers one at each machine. Using the method explained in section III, one of the three stabilizers is selected. Using participation factor technique [22], [23] stabilizers are placed only at the machines where PSS is most essential. The Eigen values of this system are obtained and the eigenvalues associated with the two modes at the given operating point without PSS are given in Table I.
TABLE I. EIGEN VALUES OF THE SYSTEM
system considered, using Glover-McFarlane H loop shaping technique is provided. A. Review of the procedure for robust controller design: Among many robust control techniques, H control theory is one of an excellent robust technique for designing RPSS. Based on this theory, three techniques of RPSS design are possible using (i) H optimal controller (ii) H2/ H mixed sensitivity controller and (iii) H loop shaping controller. RPSS design using H optimal controller theory is tedious and so, in practice a suboptimal rather than optimal solution is used. However, this H design procedure produces undesirable controllers whose zeros cancel all the stable plant poles, which is unacceptable when the plant contains lightly damped modes [24]. The design of H PSS via mixed sensitivity approach reveals the high performance and robustness but the selection of weighting functions poses a problem due to the trade off relationship between sensitivity function and complementary sensitivity function [25].The loop shaping is conceptually simple yet powerful design method in the frequency domain. The constraints of the loop phase near crossover frequency (stability requirements) complicate the loop shaping procedure considerably either for systems with RHP poles and zeros or for the multi input and multi out (MIMO) case. In the present work, PSS is designed by modifying the third technique based on the GloverMcFarlane H loop shaping design procedure [6], [13]. This method combines the advantages of loop shaping and H control via normalized coprime factorization (NCF) approach. This design procedure follows the classical loop shaping principles in the choice of the control objectives, while the stability requirement is embedded into a special robust stability framework called NCF robust stability problem. In contrast to the classical loop shaping approach, the loop shaping is done without explicit regard to the nominal plant phase information. The present design is both simple and systematic in terms of design and weighting functions selection and provides some basic guidelines for loop shaping, weighting selection and controller design paradigm formulation. The Glover-McFarlane H loop shaping design procedure [6], [13] consists of three steps [11], [12]. The Eigen values of this system correspond to the inter-area mode. The damping ratio of the system is computed. The detailed design procedure of RPSS, for the example considered in section II, is provided in [11]. The pre compensator W1 and post compensator W2 are obtained as
W1 = 33 * 10 5 * 10s * (1 + 0.33s) (1 + 10s)(1 + 0.5780s )(1 + 1.0406 )
Swing mode M1 M2
Without PSS 0.92893 j11.946 0.2683 j 7.8228
TABLE II. PARTICIPATION FACTORS
M1 M2
Sm1 0.0047 0.1387
Sm2 0.0844 0.3173
Sm3 0.3994 0.0506
Table II gives the participation factors (magnitude) of the system in modes M1 and M2. The speed of that machine with highest participation in a particular mode is the best signal to damp the oscillations due to that mode. In Table II, Smi is the slip of ith machine. Hence, it can be observed from Table II that generators 3 and 2 are the best locations to place PSSs to damp modes M1 and M2 respectively. III. ROBUST CONTROLLER DESIGN To start with, a brief review of the design procedure relating to the design of the robust H controller based on Glover-McFarlane H loop shaping technique is presented. This is followed by the design of the RPSS for the 3-machine
. (s + 1000 ) 3 The final controller K is obtained as detailed in [11]. B. The Controller Selection The above procedure [11] gives three controllers from input to the output such as K (1, 1), K (1, 2) & K (1, 3). To find the best of the three controllers, Bode magnitude plot of each controller is compared with the Bode magnitude plot of general controller K. The controller whose Bode magnitude plot closely matches with the Bode magnitude plot of general controller is selected as the best controller. For the example
W2 =
26 * (s + 10) 3
considered K (1, 3) matches with the general controller K as shown in Fig.1.

200 Singular Values General Controller Controller from input to output3
150 Singular Values (dB)
100
50
-50 -5 10
10 Frequency (rad/sec)
10
Fig.1 The singular value plots of K and K (1, 3)
C. Controller order reduction It is required to conduct nonlinear simulations using Simulink to examine the performance of the designed controller. The resulting controller has a high order. The controller is reduced to a 7th order controller using the Hankel Norm reduction [26]. The transfer function of the reduced order controller is given as
K=
7 7 10 6 13 5 15 4 4.677 *10 s + 7.709*10 s + 4.32*10 s + 5.379*10 s 17 3 18 2 19 19 + 1.726*10 s + 2.384*10 s + 2.179*10 s+ 1.199*10 7 6 7 5 10 4 13 3 s +5625 *s + 1.389*10 s + 1.971*10 s + 1.485*10 s 15 2 17 17 + 5.226*10 s + 4.544*10 s + 9.129*10
Matlab / Simulink models can be implemented and tested in real-time. The Real-Time Interface (RTI) contains a library of blocks which connects the Simulink model to the physical world. The real time dSPACE workstations DS1005 and DS1104 are chosen to simulate the plant and controller respectively. The DS1005 processor board has Motorola Power PC 750, 488Hz processor. This processor is connected with DS2002 multichannel ADC and DS2101 Multichannel DAC along with other digital I/O boards in a chassis Box. The DS1104 has Power PC 603e, 250Hz with built in ADC and DAC along with digital I/Os, which is placed in PCI slot of a desktop CPU. The power system model is loaded on to DS1005 while the power system stabilizer on to DS1104. The overall block diagram of the laboratory setup of implemented PSS using the dSPACE cards is shown in Fig.2. The pictorial representation and connection diagram of the laboratory setup is shown in Fig.3 along with the internal connections. The nonlinear model of the power system is replaced by the I/O components that form the interfaces to real power system. The output of DS1005 (Power System) is taken through DAC (DS2101), available along with DS1005, and connected to ADC of DS1104 through connector panel CLP1104. CLP1104 connector panel serves as an interface between the DS1104 and all external hardware. The ADC of DS1104 in turn is connected to DS1104 (Power System Stabilizer) through digital filter. The low pass digital filter is used to eliminate unwanted harmonics. The output of DS1104 is fedback to DS1005 through DAC of DS1104 and ADC channel of DS2002 I/O board.
The bode plots of the full-order controller and the reducedorder controller are plotted [11] and observed that the characteristics of the reduced order controller match with original controller at the frequencies of interest. After adding the designed controller, the damping of the nominal closedloop system has increased. After obtaining the controller, non-linear simulations are performed and comparisons of the performances are made with the CPSS and the resulting RPSS for three different types of faults [11]. IV. REAL TIME SIMULATION The non-linear simulations are first performed using Matlab / Simulink environment to test the efficiency of the designed RPSS [11] and then real time simulations are carried out using dSPACE real time hardware [18] to validate the designed RPSS.
Fig.3 Pictorial representation & connection diagram of controller and the plant
Fig.2 Block diagram representation of dSPACE setup of controller and the plant
The dSPACE real time hardware is an obvious choice when working within the Matlab/Simulink environment. The
RTI is the link between dSPACE hardware and the Matlab/Simulink software. It acts as a driver to the dSPACE hardware. RTI is integrated into the Simulink model like any other Simulink blocks. Then, in order to download the Matlab model to the dSPACE hardware, a C-code is generated using Real-Time Workshop (RTW). The C-code is then compiled to & downloaded to program memory of dSPACE hardware. Finally, Control Desk, an
experimentation tool, is used to control, tune and monitor the running process. With the software control desk (dSPACE) a virtual instrument panel is built. It enables the operator to change parameters and monitor signals in real-time without regenerating the code. In addition, control desk displays time histories of any variable being used by the application. A. Comparison of RPSS and CPSS in real time environment A three phase to ground fault is created at bus no.7 which is initiated at 1.0 sec and automatically cleared after 0.1sec. The Real-Time simulation is performed with power system model loaded on to dSPACE workstation DS1005 and RPSS and CPSS are alternatively loaded on to DS1104 for the same fault conditions. The time step of simulation for the controller model and the power system model is taken as 0.001 sec. The performance characteristics such as the rotor angles, slip and electric torque at all generators were observed on the control desk of DS1005.
-10 Rotor angle(deg) -15 Generator 1 RPSS CPSS
The various performance curves observed on the control desk during the real time simulation are saved as data files in the Matlab, using which response curves are plotted. Figs 4, 5 and 6 show the variation of rotor angle, relative slip with respect to Centre Of Inertia (COI) and electric torque at generator 1 respectively with RPSS and CPSS. From these plots, it can be clearly seen that the oscillations vanish faster in case of system with RPSS, while oscillations exists up to around 10 sec with CPSS. This indicates that RPSS is more effective in providing damping to the system compared to CPSS, confirming the simulation results with Simulink. B. Comparison of real time simulation with Matlab/Simulink simulation In this case, an increase of 3 times the mechanical torque initiated at 1.0 sec followed by restoring the torque back to initial value after 5 cycles, is considered. For the fault considered, the real time simulation is performed with RPSS loaded on to DS1104 (Real Time) with time step of 0.001sec and the results are compared with the simulation results obtained using Simulink (Off line).
-5 Generator 1 Real Time Off line
-20
Rotor angle(deg)
-10 -15 -20 -25 -30 -35 0 2 4 6 Time(sec)
-25 -30 0
4 6 Time(sec)
10
Fig.4 Plot of rotor angle of generator 1 with RPSS and CPSS on DS1104 with a three phase fault
3 2 Reltive Slip(p.u) 1
x 10
-3
Generator 1 RPSS CPSS
10
Fig.7 Plot of rotor angle of generator 1 with RPSS on DS1104 and simulation using Simulink with an increase in mechanical torque
0.015 Relative Slip(p.u) 0.01 0.005 0 -0.005 -0.01 -0.015 0 2
Generator 2 Real Time Off line
0 -1 -2 -3 0 2 4 6 Time(sec) 8 10
Fig.5 Plot of relative slip of generator 1 with RPSS and CPSS on DS1104 with a three phase fault
1.5 Electric Torque(p.u) 1 0.5 0 -0.5 0
Generator 1 RPSS CPSS
4 6 Time(sec)
10
Fig.8 Plot of relative slip at generator 2 with RPSS on DS1104 and simulation using Simulink with an increase in mechanical torque
4 6 Time(sec)
10
Fig.6 Plot of electric torque of generator 1 with RPSS and CPSS on DS1104 with a three phase fault
Fig.7 shows the variation of rotor angle at generator 1, Fig.8 shows the relative slip with respect to COI at generator 2 and Fig.9 show the variation of elctric torque at generator 3 when subjected to an increase in mechanical torque with RPSS, corresponding to real time and off line simulations. From these plots, it can be concluded that the results of real time simulation deviate slightly from simulation using Simulink. In both the cases the oscillations vanish faster but the real time simulation produces few extra oscillations which are caused by time delay of signal transmission and A/D, D/A converters and also due to noise introduced by the external wiring contacts. Though digital filters are used to eliminate these harmonics, small disturbances still persist in
the system response which can be eliminated by using proper analog filters during actual implementation.
2 Electric Torque(p.u) 1.5 1 0.5 0 0
2.5 Electric Torque(p.u) 2 1.5 1 0.5 0 -0.5 0 2
Generator 1 Case 1 Case 2 Case 3
Generator 3 Real Time Off line
4 6 Time(sec)
10
4 6 Time(sec)
10
Fig.12 Electric torque of generator 1 for 3 operating points for 3 operating points with RPSS on DS1104
Fig.9 Plot of elctric torque at generator 3 with RPSS on DS1104 and simulation using Simulink with an increase in mechanical torque
C. Justification of Robustness with real time simulation For justification of robustness in real time, an increase in Vref is initiated at 1.0 sec followed by restoring it back to the initial value after 5 cycles, is considered. The following 3different Vref errors are considered, viz. Case 1 : Three times Vref Case 2 : Five times Vref Case 3 : Seven times Vref The system is subjected to the above three cases with RPSS on DS1104 for the real time simulation The following plots indicate the responses at different machines, when the system is subjected to the above three faults.
-5 Rotor angle(deg) -10 -15 -20 -25 0 Generator 1 Case 1 Case 2 Case 3
Figs.10, 11 and 12 show the variation of rotor angle, relative slip and electric torque at generator 1. From these plots, it is evident that in case of the system with RPSS the settling time is almost independent of the operating point and the system is subjected to low oscillations and the system consistently reaches steady state within 2 to 3sec after the fault is cleared, even when the system is subjected to wide range of operating conditions. Again the distortion in the responses is caused by time delay of signal transmission and A/D, D/A converters and also due to noise introduced by the external wiring contacts. Though digital filters are used to eliminate these harmonics, small disturbances still persist in the system response which can be eliminated by using proper analog filters during the actual implementation in the field. V. CONCLUSION A systematic approach to design an RPSS using GloverMcFarlane H loop shaping procedure is presented for a 3machine, 9-bus power system. The procedure for feedback loop setting and for selecting the weighting functions W1 and W2 is presented. The performance of the designed RPSS is validated by implementing on real time hardware using dSPACE workstation DS1104. The real time experimental results of RPSS are compared with that of the CPSS for a three phase fault. The experimental results with RPSS and CPSS are observed on the control desk. From the real time results, it could be observed that the system with RPSS reaches the steady state period within 3 to 4sec after the fault is cleared, while with CPSS, the system requires more time to reach steady state, similar to simulation results. Justification of robustness is also provided by taking up three different operating points. The robustness results show that the system with RPSS settles down consistently within 2 to 3sec irrespective of the operating point. Collectively, these real time results show that the designed RPSS provides better damping & robustness, thus validating the effectiveness of designed RPSS The Simulink / Matlab simulation results of the designed RPSS are also compared with the real time experimental results. The Matlab simulation results almost match with the real time results with a small deviation, which is caused by time delay of signal transmission and A/D, D/A converters and also due to noise introduced by the external wiring contacts. Though digital filters are used to eliminate these harmonics, small disturbances still persist in the system response which can be eliminated by using proper analog filters during the actual implementation in the field.
4 6 Time(sec)
10
Fig.10 Rotor angle of generator 2 for 3 operating points for 3 operating points with RPSS on DS1104
4 Relative Slip(p.u) 2 0 -2
x 10
-3
Generator 1 Case 1 Case 2 Case 3
-4 0
4 6 Time(sec)
10
Fig.11 Rotor angle of generator 1 for 3 operating points for 3 operating points with RPSS on DS1104
REFERENCES [1]. E.V.Larsen and D.A.Swann, Applying Power system Stabilizers Part-I: General Concepts, IEEE Transactions on Power apparatus and Systems, Vol. PAS-100, No.6, pp. 3017-3024, June 1981 [2]. Chuanjiang Zhu, Mustafa Khammash, Vijay Vittal, and Wenzheng Qiu, Robust Power System Stabilizer Design Using Loop Shaping Approach, IEEE Transactions on Power Systems, Vol. 18, No. 2, May 2003. [3]. Dequiang Gan, Zhihu Qu and Hongzhu Cai, Multi machine power system excitation control design via theories of feedback linearization control and nonlinear robust control, International Journal of Systems Science 31 (4), pp. 519-527, 2000. [4]. Rajjev Gupta, B. Bandyopadhyay and A.M.Kulkarni, Design of power system stabilizer for single machine power system using periodic output feedback, IEE proceedings on Generation, Transmission and Distribution, Vol. 150, No.2, pp. 211-216, Mar. 2003. [5]. Bikash Pal and Balarko Chaudhuri, Robust Control in Power Systems, Springer,2005. [6]. MacFarlane, D.C. and K. Glover, A loop shaping design procedure using H synthesis, IEEE Transactions on Automatic Control, Vol. AC-37, pp. 759769, 1992. [7]. K.K. Anaparthi, B.C. Pal and H. El-Zobaidi, Coprime factorization approach in designing multi-input stabilizer for damping electromechanical oscillations in power systems, IEE Proceedings- Generation, Transmission and Distribution., Vol. 152, no. 3, May 2005. [8]. Ziad M. M. Ali and Alexander I.Malikov, Robust techniques for designing power system stabilizer, Journal of Theoretical and Applied Information Technology 2005 - 2009 JATIT. [9]. Jayapal.R, J.K.Mendiratta, H controller design for a SMIB based PSS Model 1.1, JATIT, Journal of Theoretical and Applied Information Technology, ISSN: 1992- 8645, Vol.11, no.1, January2010. [10]. Jayapal.R, J.K.Mendiratta, Design and simulation of robust H control based power system stabilizer for SMIB models, IJCSNS, International Journal of Computer Science and Network Security, ISSN: 17387906, Vol.9, no.11, November 2009. [11]. Jayapal.R, J.K.Mendiratta, H Loop Shaping Based Robust Power System Stabilizer for Three Machine Power System, IJCA, International Journal of Computer Applications, ISSN: 0975-8887, Vol. 1, no 7, Article 19, February 2010. [12]. Jayapal.R, J.K.Mendiratta, H Loop Shaping Based Robust Power System Stabilizer for Multimachine Power System, International Conference on IEEE EPEC 2010, Electrical Power and Energy Conference 2010, Halifax, Canada [13]. Duncan McFarlane and Keith Glover, An H Design Procedure Using Robust Stabilization of Normalized Coprime Factors, Proceedings of the 27th Conference on Decision and Control, Austin, Texas, December 1988.. [14]. Talaat, Hossam E. A., Al-Sulaiman, Abdulaziz A., Abdennour, Adel, Experimental investigation of a GA-Optimized neuro-fuzzy power system stabilizer for multimachine system, International Journal of
Electrical Power and Energy Systems, Vol. 32, Issue 7, Sept. 2010, pp. 751-759 2010. [15]. Hassan Bevrani, Takashi Hiyama, Hossein Bevrani, Robust PID based power system stabilizer : Design & Real time implementation, International Journal of Electrical Power and Energy Systems, Sept. 2010. [16]. Saud A. Al-Osaimi, Adel Abdennour, Abdullaziz A. Al-Sulaiman, Hardware implementation of a fuzzy logic stabilizer on a laboratory scale power system, Electric Power System Research., Vol. 74, Issue 1, pp. 9-15, Apr. 2005. [17]. Jianhong Chen, Tjing T. Lie, D. M. Vilathgamuwa, Damping of power system oscillations using SSSC in real time implementation, International Journal of Electrical Power and Energy Systems, Vol. 26, No. 5, pp. 357-364, Jun. 2004. [18]. Control desk experiment guide, RTI and RTI-MP implementation guide, dSPACE release 4.1. [19]. Anderson P.M and Fouad A.A., Power system control and stability, Iowa state university press, Ames, 1977. [20]. Padiyar K.R., Power system dynamics stability and control, Second edition, BS publications, Hyderabad, India, 2002. [21]. Yao-nan Yu, Electric Power system Dynamics, Academic Press, 1983. [22]. Kundur, P., Power System Stability and Control, McGraw-Hill, USA, New York, 1993. [23]. Padiyar, K.R., and H.V. Saikumar, Modal Inertia A New Concept for the Location of PSS in Multimachine Systems, National Systems Conference, NSC 2003, IIT, Kharagpur. [24]. S.Chen and O.P.Malik, H optimisation-based power system stabiliser design, IEE Proceedings Generation, Transmission and Distribution, Vol. 142, No.2, March 1995, PP. 179-184. [25]. Herbert Werner, Petr Korba, And Tai Chen Yang, Robust Tuning Of Power System Stabilizers Using LMI-Techniques, IEEE Transactions On Control Systems Technology, Vol. 11, No. 1, January 2003. Page(s): 147 152. [26]. Andrea Gombani and Michele Pavon, A general hankel-norm approximation scheme for linear recursive filtering, Automatica, Vol. 26, Issue 1, pp. 103-112, Jan. 1990.
3rd International Conference on Electric Power and Energy Conversion Systems (EPECS13)
Analytical Comparison between Two Topologies of a Series Resonant Inverter from Class D
LABIOD Mokhtar*1, ALLAG Hichem *2 and IKHLEF Nabil *3
* 1,2,3
Laboratoire dtudes et de modlisation en lectrotechnique (LAMEL), Universit de Jijel. BP 98 Ouled Aissa, 18000 Jijel, Algrie
*1
mlabiod18@gmail.com, *2 allag_hic@yahoo.fr, *3 ikhlefnabil@yahoo.fr
AbstractThe induction heating systems are the typical field of application of resonant inverters. In this work, we will make an analytical comparison between two topologies of a radio frequency series resonant converter supplies a series load, this comparison is based on a simulation of the induction heating system behaviour is taking into account the variation of the object impedance and was heated above the value of the inductance (L) as a function of temperature, so that it can be concluded that the most appropriate topology for a class D series resonant inverter. Keywords: induction heating inverters series resonant inverter class D.
influences the value of resistance and inductance, which leads to an increase of the resistance and the inductance fall, especially when the object to be heated passes its Curie temperature (magnetic parameter), as shown in figure 1 [1]. So the armature is an unstable load with electrical characteristics (inductance and resistance) variables.
IV.
INTRODUCTION
Class D inverter is generally used to drive the induction coil for generating the magnetic flux density in high frequency induction heating systems. It takes energy from a DC voltage source. The DC voltage is converted into a high frequency AC voltage by the latter. Then, the inverter supplies current to the load (object to be heated). Whose load is purely inductive, and modelled by an inductor in series with a resistor, requires compensation by a capacitor selected according to the property of the resonant frequency, which is utilized to minimize the resonance electric stress on semiconductor switches, and to maximize the power transferred by reducing harmonics at the same order of the resonant frequency( ).During the
Fig. 1 Variation of resistance and inductance depending on the temperature (Experimental Results)
It poses constraints on control level process: the order (we lose the state of resonance) and power transfer to the device (phase shift between the voltage and current) that's why we are interested only to the inductance changes (in a resonant state, the load is purely resistive). Whose objective is to analyse the load behaviour during the heating process and to determine the influence of different variations on the exchange of power between the inverter and load.
heating of object, provided the generator impedance is highly dependent on the load, the generated heat is directly
II.
LOAD ANALYSIS
(quasi-sine) that
is close to
.The Fourier series brings up the appear as (6)
The load is formed by an RLC circuit, the resistance is in series with the resistor and the assembly is in series with a capacitor, so the resonant frequency equals .
decomposition of the input voltage, fundamental, makes to the fundamental ( ) For: The current that it corresponds is: ( ) ( | |) ( ) ( | |) ( )
Particularly thats allow a functioning in natural switching beyond the resonant frequency, so the use of thyristor switches (IGBT) while minimizing the harmonics of the output signal. For example [2], the initial value of inductance is 66 H will be dropped to 62H value when it is heated. By cons is not interested in changes in resistance and is considered 6.4 (circuit resistance included). A 0.32 F capacitor was used to compensate for process a resonant frequency: (1) III. CIRCUIT DESCRIPTION
(7)
Class D inverter comprises two switches S1, S2 with antiparallel diodes D1, D2 using IGBT, a resonant capacitor C r and an inductor which consists of a series combination of a resistor and Req inductance Leq. One of the main advantages of the half-bridge inverter is low voltage across the switch is equal to the voltage supply. Thus, compared to other topologies (E-Class, quasi-resonant converter etc.) for induction heating applications, it is suitable for high voltage applications [2]. The input voltage of 311V continuous line
(Unfortunately this frequency responds for the beginning of the process, the load is still cold; because of the inductor value will vary during heating process). In which the impedance Z is given by: ( Where: ( )) (2)
is directly introduced into inverter [3] and S1, D1 and S2, D2 are alternately used to deliver current to the induction
: The switching pulse.
coil at high frequency. IV. DIAGRAM OF SIMULATED MODEL
: The resonance pulsation. As: : the impedance Z becomes purely resistive. The simulation is done under MATLAB Simulink If one introduces the factor of quality: Then: | | And : ( ) (5) V. The voltage across the load is shaped slots frequency f that can be associated with pulsation .The current In this topology our load is represented by an inductor in series with a resistor and all in series with a capacitor as shown in Figure 2. FIRST TOPOLOGY (4) (3) environment to see the behaviour of the load impedance seen by inverter where switches, power electronics, are insulated gate bipolar transistors (IGBT) ideals control the resonant frequency (f0= 34600Hz) controlled by a full wave control, and the switching is instantaneous.
response to this will be that much closer to the sine wave
For this topology we have four modes of operation as shown in the figure (Fig. 3). Mode 1 (t0-t1) the resonance current flowing in a direct
sense of the interrupter S1. Mode 2 (t1-t2) when S1 is deactivated at the point of (t =
t1), the resonance current through S1 begins to freewheel through the diode D2.
Fig. 2 Schematic of the principle of class D inverter
Mode 3 (t2-t3) after (t = t2), and the resonance current flows freely in a reverse direction through S2 is already
A. Interpretation and discussion of results
turned on. Here, the resonant capacitor Cr, is used as voltage source.
350 300 250 200 150 100 50 0 0 2 4 6 8 Time (seconds) 10
Mode 4 (t3-t4) when S2 is turned off at (t = t3), the

voltage current
resonant current through S2 starts freewheeling through diode D1.
voltage (V) current (A)
12 x 10
-3
a)
General form Fig 4. a) Theoretical waveforms of Class-D inverter.

voltage current
300
250
200
150
100
50
0 3.44 3.45 3.46 3.47 3.48 3.49 Time (seconds) 3.5 3.51 x 10
-3
b)
Real form
Fig 3. Voltage and current forms
For applying a voltage in niche form, we obtain a quasisinusoidal form of current and this is an advantage of the resonant modes of the system are explained and illustrated in figures 4 a) and b).
Fig 4. b) the four modes of operation of the inverter.
Figure 5 shows the active and reactive power consumed by the load, we notice that the reactive power is null when the object is still cold and it is logically because of the load seen
by inverter is purely resistive (resonant mode). Thus, it

350
begins to provide the reactive power when the inductance

300
votage current
starts to vary during the warm-up. And therefore it leads to

active power reactive power 1500
250 200 150 100 50 0
an increase in the value of the voltage and current.

2000
active power (watt) reactive power (VAR)
1000
6 Time (seconds)
10 x 10
12
-3
a)
500
General form
350
0
300 250
votage current
-500
0.002
0.004
0.006 Time (seconds)
0.008
0.01
0.012
200 150 100 50
Fig. 5 Active and reactive power consumed by the load.
VI.
SECOND TOPOLOGY
In this topologie, the inverter comprises two switches S1 and S2 with anti-parallel diodes D1 and D2, two resonant capacitors Cr/2, and an inductor which consists of a series combination of the equivalent resistance Req and inductance Leq, as shown in Figure 6 [3].
3.41
3.42
3.43
3.44 3.45 3.46 Time (seconds)
3.47
3.48
3.49 x 10
-3
a) Real form Fig. 7 Voltage and current forms
For applying a voltage in niche form, we get the same results than the previous topology except that the current amplitude is slightly more amplified. We also too four operating modes of system topology are explained and illustrated in the figures 8 a) and b).
Mode 1 when S2 is turned off, anti-parallel diode D1 is conducted by S1 iLeq negative load current. Then, we obtain
B
Fig. 6 Schematic of the principle of class D inverter second topology
the condition ZVS of S1. Mode 2 when S1 anti-parallel diode D1 is off, and S2 is conducted ZVS is reached. In this mode, current flows, iLeq positive load.
A. Interpretation and discussion of results Mode 3 when S1 is off, similar to mode 1 and anti-parallel The figures below represent respectively the forms of current and voltage. diode D2 S2 is conducted. In this way, we obtain the condition ZVS of S2.
Mode 4 where anti-parallel diode D2 is off S2, S2 is conducted and ZVS is achieved. In this mode, current flow iLeq negative load.
because of the load seen by the inverter is purely resistive (resonant mode). Thus, it begins to provide the reactive power when the inductance starts to vary during the heating at the same time the active power transferred to the load begins to drop with a note that the active power is greater than the previous topology VII. CONCLUSION
Based on this comparison is the fact that the second topology of class D inverters is the most suitable for induction heating systems because it transfers more of active power to the load (to be heated) it improve performance
Fig 8. a) Theoretical waveforms of Class-D inverter.
complete system that is to say more power gives more heat involves a fast heating. VIII. REFERENCES
[1] Liliana Grajales (Analyse and design of 500 kHz Series resonant inverter for induction heating applications) November 6, 1995 - Virginia Polytechnic Institute and state University. [2] K. Kazimierczuk, et al. "Class-D Voltage-switching MOSFET power Amplifier", Proc. Inst. Elect. Eng., Nov, 1991. Vol. 138, No. 6, pp.285-296. [3] S.Arumugam, S.Ramareddy "computer simulation of class d inverter fed induction heated jar".Journal of Theoretical & Applied Information Technology;2009, Vol. 6 Issue 4, p21 [4] M. K. Kazimierczuk, T. Nandakumar, and S.Wang, Analysis of series parallel resonant converter, IEEE Trans. Aerosp. Electron. Syst., vol. 29, no. 1, pp. 8897, Jan. 1992. [5] A. Suresh and S. Rama Reddy Series and Parallel Resonant Inverter Fed Ferromagnetic Load-A Comparative Analysis(ICETCIP'2011) Bangkok Dec., 2011. [6] Young-Sup Kwon, Sang-Bong Yoo, Dong-Seok Hyun Half-Bridge Series Resonant Inverter for Induction Heating Applications with Load-Adaptive PFM Control Strategy. IEEE, Applied Power Electronics Conference and Exposition, 1999. APEC '99. vol.1. 575 581. [7] Won-Suk Choi,Nam-Ju Park, Dong-Yun Leeand DongSeok HyunA New Control Scheme for a Class-D Inverter with Induction Heating Jar Application by Constant Switching Frequency Journal of Power Electronics, Vol. 5, No. 4, October 2005.
Fig. 8. b) the four modes of inverter operations.

2500 active power reactive power
2000
active power (watt) reactive power (VAR)
1500
1000
500
-500
0.002
0.004
0.006 Time (seconds)
0.008
0.01
0.012
Fig. 9 The active and reactive power consumed by load.
Figure 9 shows the active and thereactive power consumed by load, where we note that the reactivepower is null when the object to be heated is still cold and it logically
Performance of a power plant with Fresnel Concentrator that has average power in Algerian sites
H. Beltagy* D. Semmar* and N. Said**
* Department of Mechanics, Saad Dahlab University, W. Blida, Algeria, Email: hani.beltagy@gmail.com, djaffarsemmar@yahoo.fr ** Centre of development of the renewable energies EPST/CDER Bouzarah, W.Algiers, ALGERIA, Email: n.said@cder.dz
Abstract The objective of this work is to characterize a 5MWe Fresnel solar power plant o different sites of Algeria (Sahara). These sites were chosen for comparison by moving the plant in different locations; namely Hassi R'mel, Tamanrasset, Beni-Abbes, and El Oued. Theoretical modeling has allowed the determination of the optimal number of mirrors to avoid shadowing effects, blocking and cosine effect, the optimum number of mirrors was estimated at 40 mirrors. Furthermore, a simulation was performed for the field subsystem and the receiver subsystem. The simulation results are shown on annual basis. Yields of the annual energy production and the cost per kWh are evaluated. Indeed, the calculation of performance differs from one site to another with values of 15.7%, 11.8%, 11.6% and 10.4% respectively for Tamanrasset, El Oued, Beni-Abbes, Hassi R'mel. The thermal performance and the technical and economic results are also evaluated for these sites. Keywords solar energy, solar concentrator mirrors, Fresnel solar thermal power plants, performance, thermoelectric plants.
The figure 1 shows the principal design in linear section of a Fresnel sensor with a reflecting system and an absorber cavity form, that the width (W) is constant for each mirror. The slope of each mirror has been calculated such as the solar radiation is incident on the opening schedule. In other part, an appropriated distance is kept between 2 following mirrors, in order that each one could not make a shadow to adjacent mirror. [3,4].
I. INTRODUCTION The solar systems with concentration give the possibility to produce electricity from solar energy, the temperatures could easily be beyond 500C and conversion production is generally high. We only exploit direct solar radiation that is considered such as the principal resource. In this study, our principal purpose is to make a theoretical design of a solar power plant with a Fresnel concentrator, so that we may simulate into different Algerian sites and adapt it to Algerian climatic conditions. This simulation will allow us to compare the annual production of the power plant into different elected sites and to assess the cost of the KWh produced, (L.E.C.) for each situation; Hassi Rmel for the centre, Beni Abbes for the west, El Oued for the east, and Tamanrasset for the south. II. THEORETICAL MODELING: In order to explain the conception of the theory of Fresnel concentrator, we used the works realized by [1,2] that developed equations considering few essential hypothesis to determine the concentrated solar energy.
Fig.1 : Solar Fresnel Concentrator. [3,4].
Fig.2: Schematic representation of a Linear Fresnel Reflector with a trapezoidal absorber cavity. [4].
solar concentration
A. a mirror parameters: Each mirror is characterized by 3 parameters: The position (Qn); The incline angle (On); The difference (Sn). The following expressions have been used to obtain these parameters by using simple relationships of geometrical optics [3].
15 10 5 0 0 2 4 6 8
number of mirrors
n= tan-1[
( ) ( )
(1) (2) (3) (4) (5)
Fig.3: The relationship between the solar concentration and mirrors number in Fresnel solar reflector for n=7. According to the figure 2, we notice that if we increase the mirrors number (collecting surface), the solar concentration increases, so the solar concentration is relative to the mirrors number, this is the reason of the high quantity of the solar radius captivated by the collectors. In one part, the total solar concentration is the sum of all the elementary solar concentration of each mirror. B. Optimization Calculation: To optimize our power plant and to obtain an optimum number of the mirrors, we propose to increase the mirrors number until observing the stability or the decreasing of the concentration. So, with the equations (4) and (8), we can estimate the Fresnel collector concentration (CR) for different games of reflecting mirrors. The results found are illustrated in the figure 3 that shows the solar concentration variation relative to the mirrors number. We stopped the calculation when we observed the concentration stability or its decreasing. 40
Solar concentration
Sn=W.sin n-1.tan (2 n+0). Qn = Qn-1+W.cos n-1+ Sn CR=2 Cln=
Un Dn ln are values of solar radius reflected on the absorber and calculated as following [3].
Un = Dn=W .cos n.sec2 ln=

W is the mirrors width f is the focal distance
(6) (7) (8)
W is the width of each mirrors schedule, W=100mm. 0 is the half angular distance of the sun in an ordinary point on the earth. (0=16=0.26) [3]. f=1300mm is the focal distance. The initial conditions are the following: 0=0, S0=0, Q0 =-W/2, Q1=W/2. In order to place the mirrors in the Fresnel concentrator structure, we have to solve the equations (1, 2 and 3) and to calculate different parameters such as the position (Q), the incline angle (), and the distance between mirrors (S) from the first central mirror till the last one. In our study we keep 15 reflecting mirrors (the first central mirror + 7 mirrors on its right and 7 on its left). After resolving the equation (1, 2 and 3) and the calculation of the total solar concentration, we solve the 4th equation and obtain the following results.
30 20 10 0 0 20 40 60
number of mirrors
Fig. 4: the relationship between solar concentration and mirrors number in the Fresnel reflector: n=48. We notice that if we increase the mirrors number (collecting surface) the solar concentration increases, but at mirrors number called optimum number that is shown in figure 3, that is evaluated to 40 reflector mirrors, the solar concentration does not change, that means that after this, the rest of the mirrors do not contribute in the concentration.
III. DESCRIPTION OF THE PLANT: The solar power plant chosen for this study is a central with Fresnel concentrators of 5 MWe; its quite similar technically to the Novatec German solar, installed in the calasparra site in Spain. [5] All the characteristics of this power plant are in the following table: [5] Table I: Characteristics of the Power plant Size of the solar field Length of the solar field Net opening area The width of the field Reflector surface La longueur du module The length of the reflector Tube diameter Diameter of the receiver Length between the reflector and absorber Efficiency of the power The power block Inlet temperature Outlet temperature Power output of the generator IV. SIMULATION RESULTS: A .Thermal analysis : With the SAM [6] software, we fixed the different characteristics of the power plant, the different results are presented in graphs with a schedule (hour by hour), for the four considered sites, we first presented the annual profits of the different powers (Total power incident on the field, total field thermal power, and total receiver thermal losses), after, we presented the different efficiencies of the power plant (Collector optical efficiency, solar field thermal efficiency, total solar field collection efficiency, cycle conversion efficiency, and finally the product between total solar field collection efficiency, and cycle conversion efficiency gives total efficiency of the power plant). The profiles presented show the annual mean of the considered parameter. 21571m 806m 18489m 16.56m 513.6m 44.8m 0.75m 0.07m 0.6m 7m 35% 140C 270C 5MWe
Fig. 7: Powers for the site Bni-Abbes Fig. 6: Powers for the site Tamanrasset Fig.5: Powers for the site Hassi R'Mel:
Mirror field
the receiver
Fig. 8: Powers for the site El Oued
Fig. 9: Efficiencies for a site of Hassi Rmel
the purchase price [7]. We also present the profits proportionally to a reference site (the worst site), for the two costs 4 and 12 Algerian Dinars, and at last we will represent the functioning years number in order to make profitable the power plant for each site, that means the number of the years that the power plant will make to recover its investment.
Fig.10: Efficiencies for a site of Tamanrasset
Fig. 13: Average annual direct irradiation for the 4 sites.
Fig.11: Efficiencies for a site of Bni- Abbes
Fig. 14: Annual energy produced by the plant in each sites.
Fig. 12: Efficiencies for a site of El Oued
B. Economic Analysis: We represent the economic analysis results of the simulation, it allows us to obtain: Annual energy produced and the Levelized Electricity Cost for each site. C. Levelized Electricity Cost Estimation: The Levelized Electricity Cost (L.E.C.) is the principal indicator for the economic analysis of the solar concentration systems, it represents the levelized value of the costs converted in fuel annual equal payments and shared by the annual electric production. D. Economic Analysis Results: The financial analysis results groups the annual energy produced in year, the L.E.C in DA/KWh, the direct annual radiation, the turnover for 4 Algerian Dinars the KWh and, the turnover for the KWh sold at 12 Algerian Dinars (three times the cost of the KWh, according to the energy minister decree concerning
Fig. 16: The Turnover for the 4 sites Fig. 15: Levelized electricity cost for the 4 sites.
All the economic analysis results are in the above table number II.
Table II: Results of the economic analysis

The selected sites Annual radiation (KWh/m.year) Net Annual Energy (MWh) The LEC (DA/KWh) Turnover for the KWh 4DA Turnover for the 12DA KWh The gain for the KWh 4DA The gain for the KWh 12 DA The percentage gain Number of years of profitability (4DA) Number of years of profitability (12DA) Hassi Rmel 2008,4 8280 12,42 33.121.300 99.363.900 2.594.900 7.784.700 7.83% 38 years 13 years Tamanrasset 2759,4 11704 8,86 46.816.608 140.449.824 16.290.208 48.870.624 34.7% 27 years 9 years Bni-abbs 2114 7631 13.45 30.526.400 91.579.200 0 0 0 42 years 14 years El Oued 2137 8981 11,47 35.926.780 107.780.340 5.400.380 16.201.140 15% 35 years 12 years
V. CONCLUSION The study we realized concerning the solar power plant with Fresnel linear concentrator allows us to know the functioning of this kind of the power plant, and the parameters that have an influence on its functioning. At a certain number of mirrors, the solar concentration does not change proportionally , there are certainly some mirrors that do not contribute in the concentration, this is principally due to the shadow between mirrors, the cosine effect, and also the pursuit system (the pursuit is not integral). The optimal value of the mirrors that we must not go over is 40 reflector mirrors (20 mirrors in the right half schedule, and 20 in the left half schedule). So our power plant contains an optimal number of 41 mirrors (the central primary mirror + 20 mirrors in the right half schedule and 20 in the left half schedule). We also notice that the blocking effect, the shadow between mirrors and the cosine effect represent a great obstacle to the productivity and the profitability of the power plant. According to this study, we noticed the importance of the good choice of the power plant site, because each site is characterized by its direct clarification, the room temperature, the wind speed, the latitude, the proportional elevation on the sea level, and other factors which have an important part in the productivity and the profitability of the power plant, this is well shown in the obtained results where the energetic production and the KWh cost change while we move the power plant in the different sites. According to the results obtained from the energetic balances, established in annual mean on the hours schedule, for the different characteristics of the power plant and the technical economic study for estimating the KWh cost concerning the four sites elected. We may notice that Algeria has great opportunities to choose the installation of the solar power plant with Fresnel concentrator. At last, for the studied sites, the best site of installing a such kind of power plant to choose will be the site of Tamanrasset, followed by El Oued, Hassi Rmel, and Ben Abbes. VI. REFERENCES
[1] Negi B S, Mathur S S, et Kandpal T C. "Optical and thermal performance evaluation of linear Fresnel reflector solar concentrator". Solar and Wind Technology, Volume 6, page 589-593.1989. [2] Bliss R et Whillier. "The derivation of several, plate efficiency factors, useful in the design of the flat plat solar heat collector". Solar Energy. Vol 4. Pages 55-64.1959. [3] Panna L S, Sarvia R M et Bhagoria J L. "Thermal perforance of linear Fresnel reflecting solar concentrator with trapezoidal cavity absorbers". Applied Energy, Volum 87, September 2009, Pages 541-550. [4] S. Flores Larsen, M. Altamirano, A. Hernndez Heat loss of a trapezoidal cavity absorber for a linear Fresnel reflecting solar concentrator, Volum 39, August 2011, Pages 198 -206. [5] Web site http //www.Novatec solaire.com. Consult en Juin 2012. observed in June 2012 . [6] Web site http //www.Nrel.Gov/Analysis/Sam. observed in June 2012. [7] Executive decree N:04-92 on date of Safar 4th 1425, corresponding to Marsh 25th 2004, concerning costs of Electricity production diversifying. Energy and mines minister. March 2004.
Novel Algorithm for Aggregated Demand Response Strategy for Smart Distribution Network
Muhammad Babar1,2 , T. P. Imthias Ahamed1 , Aqueel Shah2 , Essam A. Al-Ammar1 , Nazar H. Malik1 1 Saudi Aramco Chair in Electrical Power, Department of Electrical Engineering, College of Engineering, King Saud University, Riyadh, KSA. 2 Institute of Industrial and Manufacturing Engineering, PNEC, National University of Science and Technology, Karachi, Pakistan. (email: engr.mbabar@gmail.com, mbabar@ksu.edu.sa)
AbstractAdvancement in demand side management strategies enables smart grid to cope with the ever increasing energy demand and provide economic benet to all of its stakeholders. Moreover, emerging concept of smart pricing and advances in load control can provide new business opportunities for demand side management service provider or aggregator. The aggregator act as a third party between the electricity supply system and the consumers, and facilitate consumers to actively participate in Demand Side Management (DSM) by bidding price against power reduction with some constraints. This work develops a novel algorithm for aggregated demand response for smart distribution network. Simulations are carried out which identify the demand reduction bids and consumer constraints. The simulation results of the proposed algorithm demonstrate the potential impact of an aggregated demand response on the power system.
I. I NTRODUCTION The advancements in information and communication technology and burgeoning challenges in supply and demand of electrical power have led to the concept of Smart Grid [1]. Smart Grid is expected to improve the efciency, quality, reliability, economics and sustainability of complete Supply and Demand chain of the Electricity [2]. In brief, Smart Grid is an energy management system of electrical power grid using advance data communication and networking in order to cope with skyrocketing demand and provide economical benet to all stakeholders. Demand Side Management is one of the most important management strategy that aims to balance electrical supply and demand by reducing the power demand during critical periods instead of increasing the power generation [3]. Efcient demand side management can potentially avoid the construction of an under-utilized electrical infrastructure in terms of generation capacity, transmission lines and distribution networks [4]. Controlling and inuencing the energy usage can reduce the overall peak load demand, reshape the demand curve, and increase the grid sustainability by reducing the overall cost and carbon emission levels [5]. Direct Load Control (DLC) refers to the load management strategy which has been used by the utilities or system operators since 1980s [6]. Conventionally, objective of the DLC was either cost minimization for the peak load [6], [7] or minimization of production cost [8][10]. In 1984, Lee and Breiphol used quasistatic system technique [9] and in
1986, Bhatnagar and Rehman presented quadratic cost curve technique [11] but both of them were specic to only cost minimization. However, Detroit Edison in 1980 [6], Carolina Power and Light Company in 1983 [7], Florida Power & Light [12] and Taiwan Power [13] in late 80s accumulated the experiences regarding DLC system which were specic to only peak minimization. Even till today many researchers are studying several DLC techniques and algorithms for demand side management [5], [13][16]. Most of the DLC scheduling methods use Dynamic Programming [8], [11], [13] or Linear Programming techniques [7], [9], [14]. However DLC, faced the challenge of customers acceptance and people frowned at the idea of relinquishing control over their own loads. Therefore, it is a crucial challenge for any demand side management strategy to guarantee the security of customers privacy and to provide the customers satisfaction [17]. In order to create a balance between the needs and wants of the utility and the customers, load control strategy cannot be implemented without considering the customers satisfaction [18]. In [19], Goran and Kirschen suggested that the active participation of consumer in market pricing process could benet both in minimizing peak demand and maximizing the social welfare. Goel, Aparna and Wang [20] presented the framework for aggregated demand response in which consumers can actively participate in a power reduction program via the market bidding. The emerging concept of aggregated demand response proposes aggregator as a legal third party entity which has a bilateral contract with a utility as a large energy buyer for a negotiated tariff program. At the same time, it contracts a large number of domestic customers and is responsible for designing and implementing the demand side management program including smart monitoring, direct controlling and pricing to satisfy these customers needs and wants [21]. In this paper, a strategy for aggregated demand response is proposed in the smart grid environment. The strategy is based on demand reduction bidding1 and constraints2 identied by
1 This is an incentive amount or price identied by the consumer on the basis of his/her satisfaction and domestic use. 2 Customer constraints are the factors that take care of consumer satisfaction and ergonomics.
the customer to the aggregator at the time of contract. In this paper, dynamic programming algorithm is developed for nding the optimal load schedule and direct controlling of large number of consumers. The rest of paper is organized as follows, section II formulates the demand side strategy in smart grid environment. Section III presents the scheduling algorithm. Section IV presents details of a case study and section V presents the simulation results. Finally, section VI concludes the paper. II. P ROBLEM F ORMULATION The purpose of the aggregator is to facilitate the utility in shaving power demand during the peak hours by using aggregated demand response strategy for load scheduling and DLC for load control of consumers in smart distribution network. For such demand response, the aggregator and the consumers mutually sign the bilateral contracts that facilitate consumers to update their demand reduction bids for a given time period provided some constraints are met. It is assumed min that Tn is minimum duration for which nth consumer must ON max be continuously ON and Tn is maximum duration for OF F th min which n consumer can be continuously OFF. Thus, Tn ON max and TnOF F capture constraints of consumers. If demand reduction bids fn (Pi ) are given by the consumers as a function of power levels i.e. P {0, P1 , P2 , . . . , Pi , . . . , PI }. Let us assume that TnON (k ) and TnOF F (k ) be the consumer constraints during k th interval. Then, Xn (k ) is a binary variable which depends on the previous values of consumer constraints. Mathematically,
min max Xn (k ) = Tn , Tn ON OF F
III. S CHEDULING A LGORITHM Suppose that there are N number of consumers in contract with aggregator. For instance, during control period of [T1 T2 ] in a day, utility calls aggregator for peak demand shaving. Then, rst of all, the aggregator divides the identied period [T1 T2 ] by hstep into |H| intervals, where hstep is a duration of a control interval. Thus, for making optimal decision, aggregator has to nd x(k ) = [X1 (k ), X2 (k ), . . . , Xn (k ), . . . , XN (k )] and p(k ) = [P 1 (k ), P 2 (k ), . . . , P n (k ), . . . , pN (k )] which are used to evaluate the nal solution as expressed in eq. 2. For the rst interval (i.e k = 1), it is assumed that all consumers are available for load reduction means x(1) initializes with ones and customer constraints with zeros. Thus, in order to nd the p(1), the problem is divided into N stages and is solved by using dynamic programming [22]. Next, the problem has to be solved for the remaining intervals i.e. k = 2, 3, .., |H|. The decision variable Xn (k ) for all N consumers is determined based on the consumer constraints at previous interval such that: 1 0 ; ;
max if TnOF F (k 1) < Tn OF F min & TnON (k 1) > Tn ON max if TnOF F (k 1) > TnOF F min & TnON (k 1) < Tn ON
Xn (k ) =
(3)
Once Xn (k + 1) is decided P n (k ) is obtained by solving eq. 2 using dynamic programming [22]. Moreover, during each interval TnOF F (k ) and TnON (k ) are also updated as follows: If Xn (k ) = 1; TnOF F (k ) TnON (k ) If Xn (k ) = 0; TnOF F (k ) TnON (k ) = 0 (5) = TnON (k 1) + hstep = = TnOF F (k 1) + hstep 0 (4)
(1)
Thus, Xn (k ) simply denote whether the nth consumer is participating or not in load reduction and is equal to one if the nth consumer is participating during the k th interval. If utility calls aggregator to reduce P power for N consumers at a given interval. Then, aggregator has to nd the optimal levels of power reductions by considering the constraints of consumers and calculating least aggregated demand reduction bid by all N consumers in order to maximize its own prot and full utility requirement. Hence, the problem could be mathematically expressed as:
N Pn P
Thus, the decision that whether nth consumer participates or not is taken based on the current status of the consumer Xn (k ) and values of TnON (k ) and TnOF F (k ) . The complete algorithm is explained in Fig. 1. IV. C ASE S TUDY The proposed algorithm is investigated considering 15 large consumers who have signed up the contracts of energy management program using DLC with the aggregator. As per the bilateral contract, every consumer provides demand reduction bids for different levels of power reduction suggested by aggregator as shown in table I. In this table, it is also assumed that each consumer proposes the maximum reduction of 10kW at a particular interval with some specied consumer constraints. Thus, the maximum power that can be reduced by the aggregator during any particular interval (i.e each interval
min
fn (P n (k ) Xn (k ))
n=1
(2)
s.t. P 1 (k ) + P 2 (k ) + . . . + P n (k ) + . . . + P N (k ) = P (k ) min TnON (k ) = 0 or TnON (k ) > Tn ON max TnOF F (k ) < TnOF F The procedure to update Tn (k ) , TnON (k ) and TnOF F (k ) is explained in the next section.
Table I D EMAND REDUCTION BIDS PROPOSED BY THE CONSUMERS OVER CORRESPONDING LOAD REDUCTION .
Load Reduction Consumer 11 Consumer 10 Consumer 12 Consumer 13 Consumer 14 Consumer 15 0.02 0.04 0.06 0.08 0.12 0.17 0.18 0.24 0.22 0.25
M AX TOF F minutes 10 20 10 10 10 10 10 10 20 10 10 10 10 20 20
Consumer 5
Consumer 6
Consumer 7
Consumer 8
Consumer 2
kW 1 2 3 4 5 6 7 8 9 10
Consumer 4
0.05 0.09 0.17 0.18 0.2 0.24 0.28 0.26 0.33 0.36
0.01 0.03 0.08 0.08 0.13 0.14 0.2 0.18 0.24 0.28
0.08 0.12 0.1 0.15 0.16 0.22 0.23 0.24 0.28 0.33
0.03 0.08 0.09 0.11 0.11 0.19 0.21 0.21 0.23 0.29
0.07 0.11 0.12 0.13 0.15 0.22 0.23 0.26 0.27 0.3
0.04 0.07 0.09 0.11 0.14 0.19 0.21 0.22 0.24 0.28
Power Reduction Bid /SAR 0.06 0.05 0.04 0.05 0.09 0.09 0.07 0.09 0.11 0.1 0.09 0.1 0.13 0.12 0.12 0.13 0.16 0.15 0.13 0.17 0.2 0.2 0.18 0.21 0.23 0.22 0.2 0.22 0.25 0.24 0.23 0.24 0.27 0.26 0.25 0.27 0.31 0.3 0.29 0.31
Consumer 9
Consumer 1
Consumer 3
0.03 0.06 0.09 0.1 0.15 0.18 0.21 0.23 0.24 0.28
0.06 0.1 0.13 0.13 0.19 0.21 0.23 0.25 0.27 0.33
0.02 0.07 0.1 0.07 0.12 0.18 0.18 0.19 0.24 0.26
0.1 0.14 0.15 0.13 0.17 0.24 0.26 0.26 0.27 0.3
Table II C ONSUMER CONSTRAINTS BY 15 CONSUMERS

Start
Consumer 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Get: Demand Reduction Bids and consumer constraints of all consumers Acquire: load reduction pattern from utility for specied time period [T1 T2 ] Find total number of intervals |H|
Initialize x(1) with ones
Initialize TnOF F (1) and TnON (1) with zeros for all values of n
T M AX minutes 360 360 240 360 300 360 360 240 360 300 360 360 240 290 200
M IN TON minutes 40 40 30 30 30 40 30 40 30 40 40 30 40 30 30
Find out optimal power reduction p(1) by solving eq. 2 using dynamic programming.
k = k + 1
Obtain: x(k) by eq. 3 for all values of n Find out optimal power reduction levels p(k) for all participating consumers by solving eq. 2 using dynamic programming. Obtain: TnOF F (k) , TnON (k) and Tn (k) by using eq.s, 4 & 5 for all values of n
yes
k |H| no Stop
Figure 1.
Flowchart of scheduling program
is of 10 minutes) will always be 150kW for the case under discussion. In order to test the efcacy of this algorithm, it is assumed that utility provide a peak demand curve to the aggregator for a time period of 8 hours ranging from 8:00am to 4:00pm. Thus, 60 the total number of control intervals is equal to 8 10 = 48. The utility wants that the aggregator should execute its aggregated demand response program to shave the peak demand. In Fig. 3, the solid-line shows the demand curve of 15 consumers before the application of any load control by the aggregator. It is observed that collectively all consumers use a maximum power of 127kW during the 25th interval. Although, the utility might be capable to provide this much power to these consumers even during the peak hours, usually it may cost more to the utility, because either it buys power from some other providers or runs some inefcient generators. Thus, it is assumed that the utility is capable of providing 100kW continuously to the 15 consumers at low cost and high power quality during peak hours, as shown in Fig. 3 by dotted-line. So, utility calls the aggregator to shave this peak demand up to its optimal generation capacity of 100kW . Then, the aggregator executes this algorithm for the reduction of power and uses predened demand reduction bids and the consumer
Figure 2.
Power curtailment plots of all consumers for the given 8 hrs
by each consumer during all k intervals. Therefore, it can be observed in Fig. 2 that every consumer has different power curtailment pattern for all k intervals. Moreover, it can also be observed that the aggregator does not curtail the load of 1st consumer because it is among the highest bidding consumers as shown in Table I as well as it is also among the consumers min max with tight constraints i.e. Tn =40 minutes and Tn =10 ON OF F minutes as shown in Table II. On other hand, the aggregator curtails most of the power of 2nd , 9th and 15th consumer because they are among the consumers with least reduction max min bids as well as they offer Tn =20 minutes and Tn =30 OF F ON minutes as shown in Table II. It can also be observed from Fig. 2 that aggregator did not curtail power of consumers continuously through out the control intervals because of their identied consumer constraints. For instance, aggregator controls load of consumer M AX 13 for T13 = 10 minutes at 1st interval of a peak hours OF F M AX and should resume it for T13 = 40 minutes (i.e. for next 4 ON intervals). Now, if aggregator need to curtail more power for 13th consumer, then it has to wait for atleast 4 intervals before the commencement of the next curtailment. So, at 6th interval, the aggregator again curtails the power of the 13th consumer. Although, 13th consumer has identied that the aggregator can curtail its power for total of 240 minutes per day (i.e. 24 intervals per day) as shown in Table II, but even then aggregator can not continuously curtail power of this consumer because of the constraints. Thus, this trend of power reduction during the intervals is observed in Fig. 2 for all consumers because of the predened constraints. It is also observed from Fig. 2 that during the load control of any consumer, the aggregator can change its level of curtailment in next interval. For instance, 15th consumer reduces power of 4kW for 10 minutes at 1st interval. Since, it has M AX identied T15 = 20 minutes, so, aggregator curtails 3kW OF F power of this consumer during the next interval. Similarly, aggregator curtails 2kW power of 2nd consumer at 1st interval,
Figure 3. Power consumed by 15 consumers for a period ranging from 8:00am to 4:00pm
constraints by the 15 consumers as shown in Table I and Table II respectively. V. S IMULATION R ESULTS As mention earlier, the aggregator has discretized the levels of power reduction by 1kW . It should be noted that the aggregator runs the load scheduling algorithm at every interval and curtails power of various consumers. Moreover, at a given interval, it also calculates the prospective optimal load control schedule for the next interval. Fig. 2 shows power curtailment of each consumer for complete 48 intervals as bar graphs, representing the amount of power curtailed per interval. In this gure, y axis shows the levels of power reduction in kW and x axis shows the interval number. Table I shows that each consumer offers demand reduction bidding with different prices. The consumer are shown in Table II. The algorithm has to solve for optimal power reduction using dynamic programming while taking care of the constraints
while during 2nd interval, aggregator increased the power curtailment up to 8kW. However, aggregator always take care max of all N consumers during the change in the level of Tn OF F of curtailment at any interval. Semi-dotted-line in Fig. 3 shows the aggregated consumption of 15 consumers after the optimal load control by aggregator is applied. Moreover, it can be observed from Fig. 3 that the aggregated power after application of load control is lower than the utilitys identied efcient generation capacity. Thus, it is evident from the simulation results that the proposed algorithm successfully achieve the prime objectives of the aggregator i.e. to maximize its prot by paying minimal aggregated demand reduction bid to consumers and to shave the peak demand as per utility requirement thereby providing full satisfaction to consumers by considering consumer constraints in demand side management program. VI. C ONCLUSION Demand side management can facilitate a smarter distribution network of the future. This paper presents a novel aggregated demand response strategy for such distribution network. The paper mathematically formulates a generalize algorithm for the aggregator that solves multi-stage decision problem by using dynamic programming. The proposed algorithm fulls objectives of utility by minimizing demand during the peak hours. It also helps the aggregator in maximizing its prot. Moreover, it satises the consumer by providing high quality and low-cost power and incentives for load reductions with mutual agreements. The simulation results show that the algorithm is able to achieve all the objectives effectively for a large number of consumers. R EFERENCES
[1] S. Amin, For the good of the grid, Power and Energy Magazine, IEEE, vol. 6, no. 6, pp. 4859, 2008. [2] U. D. of Energy, Smart grid. [Online]. Available: http://energy.gov/oe/technology-development/smart-grid [3] B. Kirby, Spinning reserve from responsive loads. United States. Department of Energy, 2003. [4] Assessment of demand response and advanced metering, Federal Energy Regulatory Commission, Department of Energy, 2008. [5] T. Logenthiran, D. Srinivasan, and T. Shun, Demand side management in smart grid using heuristic optimization, Smart Grid, IEEE Transactions on, vol. 3, no. 3, pp. 12441252, 2012. [6] B. Hastings, Ten years of operating experience with a remote controlled water heater load management system at detroit edison, Power Apparatus and Systems, IEEE Transactions on, no. 4, pp. 14371441, 1980. [7] S. Lee and C. Wilkins, A practical approach to appliance load control analysis: a water heater case study, power apparatus and systems, ieee transactions on, no. 4, pp. 10071013, 1983. [8] A. Cohen and C. Wang, An optimization method for load management scheduling, Power Systems, IEEE Transactions on, vol. 3, no. 2, pp. 612618, 1988. [9] F. Lee and A. Breipohl, Operational cost savings of direct load control, Department of Electrical Engineering, The Univ. of Kansas, Lawrence, KS, Tech. Rep., 1984. [10] K. Le, R. Boyle, M. Hunter, and K. Jones, A procedure for coordinating direct-load-control strategies to minimize system production costs, Power Apparatus and Systems, IEEE Transactions on, no. 6, pp. 1843 1849, 1983. [11] R. Bhatnagar and S. Rahman, Dispatch of direct load control for fuel cost minimization, Power Systems, IEEE Transactions on, vol. 1, no. 4, pp. 96102, 1986.
[12] L. J. H. A. A. G. G. G. R. Bhatnagar, J. Latimer and E. Chan, On-line load control dispatch at orida power & light, Power Systems, IEEE Transactions on, vol. 3, no. 3, pp. 12371243, 1988. [13] W. Chu, B. Chen, and C. Fu, Scheduling of direct load control to minimize load reduction for a utility suffering from generation shortage, Power Systems, IEEE Transactions on, vol. 8, no. 4, pp. 15251530, 1993. [14] C. Kurucz, D. Brandt, and S. Sim, A linear programming model for reducing system peak through customer load control programs, Power Systems, IEEE Transactions on, vol. 11, no. 4, pp. 18171824, 1996. [15] S. Lee, S. Kim, and S. Kim, Demand side management with air conditioner loads based on the queuing system model, Power Systems, IEEE Transactions on, vol. 26, no. 2, pp. 661668, 2011. [16] A. Sepulveda, L. Paull, W. Morsi, H. Li, C. Diduch, and L. Chang, A novel demand side management program using water heaters and particle swarm optimization, in Electric Power and Energy Conference (EPEC), 2010 IEEE. IEEE, 2010, pp. 15. [17] D. Callaway, Can smaller loads be protably engaged in power system services? in Power and Energy Society General Meeting, 2011 IEEE. IEEE, 2011, pp. 13. [18] C. Diduch, M. Shaad, R. Errouissi, M. Kaye, J. Meng, and L. Chang, Aggregated domestic electric water heater control-building on smart grid infrastructure, in Power Electronics and Motion Control Conference (IPEMC), 2012 7th International, vol. 1. IEEE, 2012, pp. 128135. [19] G. Strbac and D. Kirschen, Assessing the competitiveness of demandside bidding, Power Systems, IEEE Transactions on, vol. 14, no. 1, pp. 120125, 1999. [20] L. Goel, V. Aparna, and P. Wang, A framework to implement supply and demand side contingency management in reliability assessment of restructured power systems, Power Systems, IEEE Transactions on, vol. 22, no. 1, pp. 205212, 2007. [21] Fast demand response, PARC, A Xerox Company, 2010. [22] M. Babar, I. Ahmed, A. Shah, S. Al Ghannam, E. Al-Ammar, N. Malik, and F. Pazehri, An algorithm for load curtailment in aggregated demand response program, in 2012 IEEE PES Conference on Innovative Smart Grid Technologies-Middle East (ISGT Middle East). IEEE, 2012.
A Novel Microcontroller Based Speed Control Method Employing Current Controller for TWUSM
Altan GENCER
Hurriyet Vocational High School Kayseri, Turkey altangencer@gmail.com
Abstract The relation between rotor speed of the Travelling wave ultrasonic motor (TWUSM) and current drawn by TWUSM is not linear for the TWUSM under different loads. In order to solve this nonlinearities and variation problems of TWUSM, PI controller, current controller, frequency controller and phase difference controller are proposed and implemented. The proposed system evaluates the steady-state performance of TWUSM drive system by using PI controller and current controller. Additionally, in the proposed TWUSM drive system, LC parasitic resonance impedance are added in order to obtain same peak voltage values under different loads by using frequency controller and phase difference controller. Experimental results of the proposed technique are analyzed and presented. Keywords- Travelling wave ultrasonic motor, current controller, pi controller
Flueckiger et al. [7] using neural network. In this system, rotor speed of TWUSM is controlled by using only driving frequency controller. A new driver based on dual-mode frequency and phase control is developed by Chen et al. [8]. This study is realized same peak voltage values in TWUSM speed control [8]. In the aforementioned studies, rotor speed control of TWUSM is realized by using signal of the piezoelectric sensor which cannot represent the accurate rotor speed value or only digital encoder of TWUSM. However, in this paper, accurate rotor speed value of TWUSM is obtained by using both signal of digital encoder of TWUSM and signal of current sensing card. Furthermore, in order to the stator protect from instantaneous high currents under load variations, the current controller is applied in the TWUSM drive circuit system. II. EQUIVALENT ELECTRICAL MODEL OF TWUSM The theory of Equivalent electrical circuits (ECM) is useful in solving complex systems. In its most common form, an equivalent circuit of the TWUSM is composed of capacitors, inductors, and resistors. An accepted equivalent circuit for TWUSM is used in this study and is given in Figure 1 [1]. Two loops of the equivalent circuit of TWUSM represent Aphase and B-phase which is made of rotor component and stator component. The stators mass is equivalent to the capacitance Cd, represents the piezoelectric ceramic. The rotors mass is equivalent to the inductor Lm, its capacitance is given by Cm, and an overall loss is modelled by a resistor denoted by ro.
I.
INTRODUCTION
The development of travelling wave ultrasonic motors has led to a new actuator type that has many attractive properties for precision positioning tasks. TWUSMs use mechanical vibrations to produce a cyclic friction based driving force between one stationary and one moving component. The first working prototype of a TWUSM was built and tested by T. Sashida in 1982 [1]. TWUSM is recently used in different industrial applications, especially for medical or automotive applications due to their low-power, high torque [2, 3].However, due to the complexity in the operating principle, TWUSMs characteristic is difficult to model. In order to overcome with these problems a lot of successful speed control systems were proposed in the literature. Some of them: Conventional control strategies are fixed structure and fixed parameter design [4, 5]. Hence the tuning and optimization of these controllers are a challenging, particularly under varying load conditions, parameter changes, and abnormal modes of operation. A highly effective load adaptive servo system of speed control of TWUSM is realized by Bal et al.[6] using digital signal processor. The variations in the speed characteristics of the TWUSM are obtained in [6]. Speed sensorless control for TWUSM is designed by
Figure 1. Loaded equivalent electrical circuits of two phase TWUSM
Figure 2. The proposed TWUSM driving circuit
The loaded equivalent electrical circuit of TWUSM is given in Figure 1. The single phase total equivalent impedance of TWUSM is given by Eq. (1). Z=(Zm Zd)/( Zm + Zd) (1)
signal which is sent to microcontroller will be low. To amplify this signal, it is passed a non-inverting amplifier circuit and then it is applied to microcontroller ADC input.
Impedance of Zm (stator) is written in Eq. (2) and impedance of Zd (rotor) is written in Eq. (3). Zm=r0 + j(XLm + XCm) Zd=XCd III. (2) (3)
THE PROPOSED DRIVING AND CURRENT CONTROLLER SYSTEM FOR TWUSM

Figure 3. Block scheme of the current sensing card
In Figure 2, rotor speed of TWUSM is controlled by driving frequency (f), phase difference () between applied voltages, amplitude of the applied voltage (V). In order to simply the driving circuit both driving frequency and phase difference of applied voltage are set as control inputs. The amplitude of applied voltages is fixed as V=140Vrms. The operating current sensing card using HAS50-S type current transformer is shown in Figure 3. The transformation ratio of this transformer is 1/1000. Due to nominal operating current value of TWUSM is too small, the value of current
For TWUSM under different load, the relation between rotor speed of TWUSM and current drawn by TWUSM is not linear. To achieve a good relation between rotor speed of TWUSM and current drawn by TWUSM, the block diagram of rotor speed/current control TWUSM drive system is implemented as shown in Figure 4.
Figure 4. Block diagram of rotor speed/current control TWUSM drive system
The rotor speed/current controller of TWUSM is regulated by the driving frequency and phase difference of applied two AC voltages of TWUSM. The two variation techniques are used in the close-loop rotor speed/current control diagram is shown in Figure 4. For the external loop, A ProportionalIntegral (PI) speed controller is implemented. The rotor speed of the TWUSM is measured by digital encoder signal and is calculated by PIC18F452. Then, measured rotor speed signal is compared to reference rotor speed and the rotor speed error es is generated. us is calculated in the output of PI control in Figure 4. PI controller is given by Eq. (4). us = kpe + ki edt (4)
Figure 5. Experimental setup of TWUSM
Where, us is the computed output of the microcontroller. kp is the proportional gain and ki is the integral gain of the controller. For the interior loop, current controller is implemented. The current drawn by the TWUSM is measured by a current sensing card and is computed by PIC18F452. Then, measured current signal is compared reference current and current error ec is generated. uc is calculated output of current control in Figure 4. Furthermore, to protect from instantaneous high current under variation load which could detriment the stator, current controller is applied in the TWUSM drive circuit system. The proposed system evaluates the steady-state performance of TWUSM drive system by using PI controller and current controller For dual-mode control, there are two rotor speed control techniques, which are frequency control and phase difference control. Frequency control is noted that there is rotor speed drop when the driving frequency is increased above the resonant frequency fr (40 kHz). Phase difference control can be varied from /2 to /2 according to the sinusoidal relationship between speed and phase difference. Similarly, us can also be used to decide the direction of rotation by controlling the phase difference between the two AC voltages. When us is positive, varies from 0 to /2, and the TWUSM rotates clockwise; when u is negative, varies from /2 to 0, and the TWUSM rotates anticlockwise. IV. EXPERIMENTAL STUDIES A prototype of new driving circuit and control circuit for a TWUSM is built and setup which composed from inverter, keypad, 4x20LCD, electromagnetic brake and then it is tested. The experimental setup of TWUSM is given in Figure 5. The amplitudes, frequencies and phase differences of two AC voltages of TWUSM are observed by digital oscilloscope. The rotor speed of TWUSM is obtained by signal of the digital encoder. An electromagnetic brake is attached to TWUSM shaft for load tests. The applied load is sensed by a torque sensor. In this study, the rotor speed of TWUSM is controlled both driving frequency and phase difference.
In this study, the rotor speed of TWUSM is controlled by changing driving frequency under different load values as well. In Figure 6; the variation in rotor speed of TWUSM is given by changing the driving frequency between 40,65 kHz and 42,55 kHz for different load values (0,02 Nm 1 Nm). Figure 8 illustrates that actual motor speed and applied voltage frequency are changed with constant rate. From the Figure 8, as load torque increases, the speed of motor decreases. TWUSM rotor speed is measured 91 rpm in frequency 40,65kHz in 0,02 Nm load condition. In the same experiment conditions, motor speed is measured 88 rpm in 0,04 Nm load condition. Motor speed is measured 86 rpm in 0,06 Nm load condition. Motor speed is measured 84 rpm in 0,08 Nm load condition. Motor speed is measured 81 rpm in 0,1 Nm load condition.
100 90 80 70 60 50 40 30 20 10 0 40,65 41,32 42,01 Driving frequency (kHz)
Rotor speed (rpm)
load (0,02Nm) load (0,04Nm) load (0,06Nm) load (0,08Nm) load(0,1Nm)
42,55
Figure 6. Variation of rotor speed for different frequency and load values
In this study, the rotor speed of TWUSM is controlled by changing In this study, the rotor speed of TWUSM is controlled by changing phase difference under different load values as well. In Figure 7; the variation in rotor speed of TWUSM is given by changing the phase difference between /2 and /2 for different load values (0,02 Nm 1 Nm). TWUSM rotor speed is measured 46 rpm in frequency 42,37kHz in 0,02 Nm load condition. In the same experiment conditions, motor speed is measured 44 rpm in 0,04 Nm load condition. Motor speed is measured 42 rpm in 0,06 Nm load condition. Motor speed is measured 41 rpm in 0,08 Nm load
condition. Motor speed is measured 39 rpm in 0,1 Nm load condition.

50 40 30
Rotor speed (rpm)
20 10 0 -10 -20 -30 -40 -50

Phase difference (degree)
-90 -75 -60 -45 -30 -25
25 30 45 60 75 90
load (0,02Nm) load (0,04Nm) load (0,06Nm) load (0,08Nm) load (0,1Nm)
Figure 8. Variation of rotor speed for different phase difference and load values
In Figure 8 phase difference of phase voltages applied to the TWUSM is observed. In these figures the phase difference of voltages is changed between 0 and 2/3 and amplitude of applied two voltages is fixed at 140V. In this phase difference values of two voltage, actual speeds of TWUSM is in Figure 8a phase difference voltage=0 for 0rpm, in Figure 8b phase difference voltage=/3 for 27rpm, in Figure 8c phase difference voltage=/2 for 36rpm. It can be easily seen from the results that the phase difference of the TWUSM has changed as sinusoidal with varied rotor speed of TWUSM.
(b)
(c)
Figure 8. The curves of two-phase output voltages with variation phase difference voltage a) 0 b) /3 c) /2 The rotor speed of TWUSM is controlled by the changing applied voltage frequency and phase difference of two voltages. When the rotor speed of TWUSM is 80rpm, open loop and closed loop are examined experimentally as can be seen in Figure 9. In the Figure 9.a, open loop is observed, the ripple of rotor speed is big and rotor speed drops. From the Figure 9.b and the Figure 9.c, the loaded work TWUSM is observed stable rotor speed.
(a)
Figure 10 shows the measured output voltage and current waveforms of two voltages. It is observed 280mA current value of phase-A and 140V voltage value of phase-A. In this case, the output current of phase-A is front of voltage of phase-A.
(a)
Figure 10. Measured voltage and current of two phase
V.
CONCLUSION
(b)
(c) Figure 9. The experimental measurements of 80 rpm speed curves of loaded TWUSM a)open loop b) two loop no load c) 0,8 Nm load
In this paper, a novel speed control for TWUSM is proposed and implemented in order to investigate the speed response of PI controller, current controller, frequency controller and phase difference controller under variation load condition. While the driver system has a simple construction and small size it provides access to enter speed and direction commands digitally as well as tracking the actual speed of TWUSM. The proposed system evaluates the constant state performance of TWUSM drive system by using PI controller and current controller. The proposed system is tested in Research and Development Laboratory in Gazi University for different speed and load cases under dual-mode operation. The proposed driver and control system have been applied to the speed control loop and current control loop of TWUSM. The experimental results show that rotor speed of TWUSM can be controlled more widely by the proposed technique with frequency; while the phase difference of the phase voltages has less effect on the speed interval. The proposed PIC microcontroller based digital control scheme of TWUSM is simple, cost-effective, reliable and very suitable for practical applications.
REFERENCES
[1] [2] [3] T. Sashida, T. Kenjo, An Introduction to Ultrasonic Motors, Oxford University Press, New York, 1993. K. Uchino, Piezoelectric ultrasonic motors: Overview, Smart Material Structure, vol. 7, p.p. 273-285, 1998. K. Nakamura, S. Ueha, Potential ability of ultrasonic motors: A discussion focused on the friction control mechanism, Electronics and Communications in Japan, vol. 2, 81(4) , pp. 57-67, 1998. B.A. White, R.T. Lipczynski and A.R. Daniels, A simple digital control scheme for a dc motor, Proc. IEE, Pt. B,vol. 130, pp. 143-147, 1983 J. Zhang and T.H. Baston, Robustness enhancement of dc drives with a smooth optimal sliding mode control, IEEE Trans. Ind. Appl., vol. 27, pp. 686-693, 1990. G. Bal, E. Bekiroglu, A highly effective load adaptive servo drive system for travelling wave ultrasonic motor, IEEE Trans. Power Electron. 20 (5) p.p. 11431149, 2005. M. Flueckiger, M. Bullo, Y. Perriard, Sensorless Speed Control of Traveling Wave Ultrasonic Motor, IEEE Trans. Industry Applications Conference, pp. 2488-2495,2006. T.C. Chen, C.H. Yu, C.J. Chen and M.C. Tsai, Neuro- Fuzzy Speed Control of Travelling Wave Type Ultrasonic Motor Drive Using Frequency and Phase Modulation, ISA Transactions, vol. 47, pp. 325338, 2008.
APPENDIX
The name plate of USR60 TWUSM Drive frequency 40-45 kHz. Drive voltage 100-140 Vrms. Rated torque 0,4Nm. Rated power 4 W. Rated speed 10 rad/s. Holding torque<=800mNm. Rotor inertia 7,2x106 kg/m2. Weight 0,23 kg. The equivalent circuit parameters of USR60 TWUSM Cd= 9 nF. Lm= 0,1 H. r0 = 150 . Cm = 168 pF.
[4] [5]
[6]
[7]
[8]
Voltage Control Using Coordinated Control of Heat Pump Water Heaters with Large Penetration of Photovoltaic Systems
A.L.M. Mufaris , S. Kawachi, J. Baba
Department of Electrical Engineering and Information Systems, The University of Tokyo, Tokyo, Japan
Abstract It is expected that large amount of photovoltaic (PV) systems will be installed in future power systems. However, introduction of large PV systems create significant problems specially in the distribution system. One of main problems is voltage rise at the end of distribution line due to reverse power flow by the presence of PV systems. In order to mitigate this problem, we propose load control at the customer end. Heat Pump Water Heater (HPWH) has been selected as customer controllable load due its large thermal energy consumption and flexibility in operating time. In this paper, coordinated control of HPWHs is attained based on the node voltages and customer hot water demands. When one of nodes voltage goes closer to maximum allowable voltage or hot water in the tank becomes its minimum level, HPWH is operated in order to avoid voltage rise or hot water shortage. In the voltage control method, primary control priority for operating HPWH is set to maintain reasonable hot water and then set to avoid the voltage rise with intention to keep the participating customers convenience. It is found that proposed voltage control method not only avoids voltage rise but also maintain customer convenience. Keywords-Heat Pump Water Heater; PV systems; voltage control; input pattern
customer ends is able to regulate customer voltage within acceptable range by compensating reactive power. Proper coordination of shunt capacitors and OLTC (On Load Tap Changers) is used to control voltage and reactive power in the system [8]. Coordinated control of distributed energy storage system (ESS) with traditional voltage regulators including the OLTC and step voltage regulator is proposed to solve voltage rise problem caused by PV systems [9]. Though all researchers come up with solution for distribution voltage control with presence of distribution generations, voltage rise is alleviated up to certain extent. On the other hand, there are considerable problems combined with their proposed methods such as distribution line loss as a result of reactive power flow, high cost due to energy storage installation and complexity in controlling distribution installations. In this paper, HPWH is selected as controllable load at customer end in order to minimize voltage violation. Input pattern of HPWH is major concern of this research so as to control all node voltages within acceptable range by reducing reverse power flow. The author proposes coordinated control of HPWHs from a central controller based on all node voltages and customer hot water demands. Coordinated control of HPWHs means, HPWHs are operated simultaneously with same operating pattern. II. RESIDENTIAL DISTRIBUTION SYSTEM
I.
INTRODUCTION
It is expected that large amount of photovoltaic (PV) systems will be installed in future power systems. In Japan, target capacity for PV systems in 2020 would be 28GW, 20 times the capacity in 2005. The large installation of PV systems creates significant problems in the power system. One of the main problems is voltage rise at distribution line end due to the reverse power flow created during the daytime by PV systems. Reverse power flow is created during off-peak period as supply power and PV generation is larger than demand. Voltage rise problem at the line end becomes first in comparison with other significant problems such as surplus power creation, insufficient frequency control. Several methods for voltage control in distribution systems with distribution generations have been proposed. Optimal control of distribution voltage is attained by proper coordination of distribution installations such as load ratio control transformer, step voltage regulator, shunt capacitor, shunt reactor and static var compensator [1-4]. Application of electrical energy storage with demand side response has been proposed in [5] to control voltage. Centralized control of DSTATCOM with controllable dead band to control all node voltages dynamically is proposed in [6]. In [7], they proposed coordinated control of STATCOM which is installed at
A. Model of distribution System It makes sense that large amount of PV systems will be installed in residential distribution system. The main reason is that number of customers residing in residential distribution system is comparatively large and many of them have their own PV systems. The model of residential distribution system considered for this study and its corresponding features are given as follows in Fig. 1 and Table 1[10]. In order to detail Fig. 1, arrow head showing with color black represents node which has three single phase pole transformers with capacity of 50kVA, each pole transformer has 15 single phase customers whereas, white arrow head shows the 3-phase customers. Hereafter the single and three phase customers are called as typical and commercial customers respectively. Dotted line in the distribution model divides the area into two sections with intention of representing the tap arrangement of pole transformers. First section ends at node 23 with voltage ratio of 6750/105V whereas, second section is with voltage ratio of 6600/105V.
10
22 0.34 0.25 7 0.07 8
34 0.39 33 0.06 32 0.21 0.535

23 24
40 0.34
55
0.34
0.34
54
9 0.02 6
21 0.02 20
0.18 19
39 0.02 38
0.18 0.535
35 36
0.02
51 50 37
0.25
52
0.07
53
5 Distribution 0.18 Substation 4 0.535 0.47 (km) 1 2 3 11 0.18 AL 120 AL 58 AL 25

14
31
0.18
49
0.535
12 17 18
0.535
41
0.535
42
0.03 0.14 0.04

56 57
58
47 48
25 13
analyse voltage with and without PV systems, residential distribution model given in Fig.1 is used. It is apparent that when PV output becomes higher, voltage increases and eventually voltage violation occur at line end. In order to understand worst voltage violation, voltage at 12pm is calculated for all nodes in residential distribution system as PV output becomes large at 12pm due to irradiance. The voltage of the residential distribution system is found by using Newton-Raphson method in which base values for voltage and apparent power are taken as 6.6kV and 10MVA respectively. Power consumption of both typical and commercial customers at 12pm is found from Fig. 2 and maximum power output of PV system (at 12pm) is assumed as 4kW. The residential distribution line model has been implemented in C language and load at each node is applied. Voltage of all nodes is calculated with PV installation of 20%, 40%, 60%, 80% and 100% and without PV systems. Fig. 3 shows the calculated voltage of all nodes in the distribution system with and without PV systems at 12pm. It is considered that PV systems are distributed uniformly. It is clear from Fig. 3 that voltage violation occurs at most of the nodes for 60% and above of PV systems installation. In Japan, distribution system voltage should be 101 6 V at end customer. However, it is assumed that voltage drop between customer end and nearest node (pole transformer) is 0.5V in this paper. Therefore, node voltage (at pole transformer) should be 101 5.5 V. B. Continuous Voltage Analysis It is found from Fig.3 that voltage at node 23, 55 and 58 rapidly increase with PV installation. The reason is that those nodes are at line end of tap sections. In previous section A, voltage is found only at 12pm with intention of finding worst voltage violation condition. In order to analyze voltage throughout daytime, voltage at node 55 is only found for distribution system with PV systems placed at some of nodes as voltage at node 55 is higher in comparison with all other node voltages with PV distribution from Fig.3. Simulation is carried out from early morning 5am to evening 8pm. The voltage analysis is carried out for PV output on sunny, cloudy and rainy days. Fig. 4 shows the voltage variation at node 55 for different PV outputs. The PV systems distribution considered for this study is illustrated in Fig. 5. The nodes selected for PV systems distribution represent that all customers in the node has capacity of 4kW PV systems.
0.21
26 27
0.18 43 0.26 0.12

28 29
0.02
0.34
15
16
TAP 6750/105
0.39
30
0.02
45
44
TAP 6600/105
0.34
46
Figure 1. Model of residential distribution system TABLE I. RESIDENTIAL DISTRIBUTION SYSTEM DATA Pole transformers Typical customers Commercial customers Total length AL 120 Impedance AL 58 AL 25 117 (39 x 3) 1755 (39x3x15) 8 3.89km 0.24+j0.41 /km 0.48+j0.43 /km 1.15+j0.46 /km
B. Customer Loading Pattern It is obvious that average load pattern of customers is different for different season. Typically, customer load demand is higher in summer and winter in comparison with other seasons. In this study, load pattern of both customers, typical and commercial, on holiday of spring season is considered so as to find the maximum voltage violation with PV systems. The load pattern of typical and commercial customers is illustrated in Fig. 2 for spring season [11-12]. Load pattern for commercial customer is obtained by multiplying maximum load for one phase (40kW) with percentage load given in [12].
Figure 2.
Load pattern of typical and commercial customers [11-12]
Voltage allowable limit
III.
VOLTAGE ANALYSIS WITH PV SYSTEMS
A. Voltage at 12pm In conventional power distribution system, voltage decreases from substation to end customer. Conversely, distribution system installed with PV systems show higher voltage at line end in comparison with voltage at line end in conventional system for certain period in daytime. In order to
Figure 3. Voltage of all nodes at 12pm with and without PV systems
increase the water temperature from initial value to set value. In demand unit, total amount of used hot water, , is calculated. In order to make more practical HPWH model, certain amount of heat loss rate, L, is considered [13]. Thus, amount of remaining hot water in the tank is calculated from (2).
= + -
(2)
Where, is the initial hot water, is created hot water, is heat loss and is amount of hot water in the tank. All symbols are denoted to express the quantity in the form of energy.
Figure 4. Voltage at node 55 with PV systems
10
22 0.34 0.25 0.07 7 8
34 0.39 33 0.06 32 0.21 0.535

18 23 24
40 0.34
55
0.34
0.34
54
9 0.02 6
21 0.02 20
0.18 19
39 0.02 38
0.18 0.535
35 36
0.02
51 50 37
0.25
52
0.07
53
5 Distribution 0.18 Substation 4 0.535 0.47 (km) 1 2 3 11 0.18
31
0.18
49
0.535
12 17
0.535
41
0.535
42
0.03 0.14 0.04

56 57
58
47 48
25 13
0.21
26 27
0.18 43 0.26 0.12

28 29
PV Systems Location
14
0.02
0.34
15
0.39
30
0.02
45
44
0.34
46
16
Figure 5. PV systems distribution
Figure 6. Heat pump water heater model TABLE II. HPWH MODEL PARAMETERS Rated power consumption [kW] Thermal capacity of hot water tank [kWh] Start delay [s] Conversion delay [s] COP Normalization coefficient , K Heat loss rate per minute , L 1 20 30 300 3 1/60 0.01%
IV.
HEAT PUMP WATER HEATER (HPWH)
A. HPWH model Use of HPWH would be a possible controllable customer load. Daytime operation of HPWH is possible as large amount of hot water is used in night time. Therefore, suitable operation of HPWH at daytime would minimize the reverse power flow and eventually mitigate the voltage violation when large amount of PV systems are installed. It is considered that participating customers in control method should set the amount of next day hot water demand on the present day and there should be certain amount of hot water initially in order to cover their early morning hot water need without shortage so that HPWH can be operated only during the voltage violation period. The amount of hot water per second is considered as power (kW) and amount of hot water stored in the tank as energy (kWh) in this paper. The operating time of HPWH is found based on the energy demand, power consumption of HPWH and coefficient of performance (COP) from (1).
Where, is power consumption of an HPWH, is created hot water in terms of thermal energy and T is total operating time. The HPWH model consists of heating unit and demand unit. Fig. 6 illustrates proposed HPWH model and its parameters are given in Table 2. Heating unit represents amount of hot water created. During hot water creation, power consumption of HPWH is represented by first order delay of 30s (electrical starting delay) and 300s (conversion delay). These delays are introduced as certain time takes in order to
B. Hot water demand The energy consumption for hot water is comparatively large for typical household customer. Fig. 7 shows the average measured hot water demand pattern in terms of electrical power for a typical customer on holiday. It is assumed that there is no hot water consumption for commercial customer. The total daily hot water demand per typical customer on holiday of spring season is about 11kWh from Fig 7. It is obvious that hot water consumption during a holiday would be less than hot water consumption on any other day as people go out on holidays. Hence, holiday hot water consumption presented in Fig. 7 is obtained by multiplying a factor (= 0.9 assumption) with actual hot water consumption given in [11].
Where, represents power consumption of an HPWH. When hot water tank level is less than its minimum allowable limit (10% of = 2kWh), controller applies rated input power (=1kW) for HPWH to avoid hot water shortage (3). Whereas, controller stops the HPWH operation, in case hot water in the tank exceeds its full capacity (4). Step (5) if the amount of hot water is sufficient, controller checks voltage of all nodes. In other words, controller decides power consumption of HPWHs based on node voltages, if hot water in the tank is within 10% to 100% of tank capacity. From all node voltages ( ), central controller finds the maximum node voltage ( ). Then central controller decides power consumption of all HPWHs based on maximum node voltage. It is represented from (5)-(8). =0 = 0.5kW = 0.75kW = 1kW for for 106.0 < = for 106.5 < = for 107.0 < = < 106.0 < 106.5 < 107.0 (5) (6) (7) (8)
Figure 7. Hot water demand pattern of typical household customer [11]
V.
COORDINATED CONTROL OF HPWHS FOR VOLTAGE CONTROL
The purpose of this study is to mitigate voltage violation and keep the voltage at any node within safe limit (95.5V 106.5V). In the proposed control method, the central controller placed at substation finds the operating pattern for all HPWHs based on the node voltages and amount of hot water in the tank. Central controller operates all HPWHs simultaneously with same operating pattern when one of nodes voltage becomes near maximum acceptable value or amount of hot water in the tank becomes less than the minimum level. In order to make the proposed method clear, Fig. 8 shows proposed coordinated control method in which central controller obtains necessary information from all customers HPWH and all node voltages in order to activate the operation of HPWHs coordinately to avoid voltage violation. It is considered that participating customers in this method should not be affected with hot water shortage. A. Detail procedure of proposed method Fig. 9 shows a flowchart of the proposed voltage control method. The following is a summary of the procedure: Step (1) initially central controller receives next day hot water demand ( ) from all customers before the day starts. Step (2) for every minute, central controller acquires all node voltages ( ), all customers hot water tank level ( ) and created hot water ( ). Step (3) central controller checks whether hot water request from all customers is completed by comparing their
with
The reason for applying many steps in the power consumption is to increase operating time of HPWH as to control voltage for long time. When node voltage is less than 106V, controller does not initiate operation of HPWHs. Conversely controller initiates all HPWHs operation for voltage higher than 106V. Based on the power consumption, HPWH creates hot water until it completes hot water request from customer.
Figure 8. Coordinated control of HPWHs
If If
< =>
, continue
control operation
, stop control operation
Step (4) in order to keep reasonable amount of hot water in the tank, controller checks hot water tank energy level all the time. Whenever controller finds shortage or excess amount of hot water in the tank, following operation takes place as not to affect participated customer. It is represented from (3) (4). = 1kW =0 for for < 10% of > (3) (4)
B. Central Controller functions The central controller is responsible to maintain all node voltages within allowable limit and reasonable hot water in all participating customers tanks. In addition, it should complete all customers hot water requests. The following are the main responsibilities of the central controller. 95.5 < 10% of < 106.5 V, for all nodes i, (i=2,3,..,58)
< =
<=
for whole day
If HPWH operations are initiated, then central controller completes all customers hot water demands before HPWH stops. Therefore, frequent ON and OFF operation and rapid fluctuation of power consumption of HPWHs are avoided. In order to ensure the customer convenience in the proposed method, central controller provides primary control priority to maintain reasonable hot water in all customers tanks. Then it provides control priority for controlling the voltage. In the control method, controller checks previous minute power consumption of HPWH before deciding the present value in step 5 to avoid OFF operation for a case when node voltages and hot water tank energy level are within safe limit. In other words, if the previous minute input, is greater than zero then the next minute input, , is also should be same as previous value in order to avoid many ON and OFF operation.
Start
outputs. Proposed coordinated control of HPWHs method is applied in order to avoid voltage violation. B. Results Results are given only for node 55 as it is found that most critical node in residential distribution system is node 55 with PV systems. If voltage at most critical node (55) is controlled within safe limit, then all other node voltages are within safe limit. Fig. 10 and 11 show voltage at node 55 for sunny and cloudy days respectively with and without proposed method. Without HPWH control, voltage violation occurs for certain period during daytime. However, proposed method controls the voltage and avoids voltage violation. When node voltage becomes 106V, proposed method initiates power consumption of HPWH so as to increase the load in customer premises. As a result, voltage suddenly drops and is within allowable voltage limit. Result on rainy day is not presented as there is no voltage violation in the most critical node (55) from Fig. 4. Fig. 12 shows input pattern of an HPWH and its corresponding hot water energy level in the tank for both sunny and cloudy days. Proposed control method activates operation of HPWH based on either minimum hot water energy level or node voltage. In this analysis, HPWHs are triggered by node voltage that becomes higher than 106.0V before HPWH hot water energy level becomes its minimum level on both sunny and cloudy days. Power consumption of HPWH is at 0.5kW throughout whole operating period in sunny day. Conversely, it changes from 0.5kW to 0.75kW for few minutes in cloudy day case as node 55 voltage goes above 106.5V. Hot water energy level in tank reduces with hot water consumption (illustrated in Fig. 7). When HPWH is operated, tank energy level gradually increases. Though hot water demand and created amount of hot water is almost same (= 11kWh), initial amount of hot water at day starts is not equal to the amount of hot water at day ends from Fig. 12. The reason is that there is an energy loss from the tank. Since Fig. 12 shows characteristics of one customer HPWH, however it is similar for all customers as all of them have same initial and requested hot water, and hot water demand. Hence, it is confirmed from Fig.12 that no customer has shortage of hot water. Initially, customers have sufficient amount of hot water to cover their needs until early morning. During control operation, central controller ensures reasonable amount of hot water in all customers tanks.
t=0
Obtain Energy Request (Erequest) Obtain Data (Etank, Vi & Ecreated) Finish
No
Ecreated < Erequest
Yes
Check Etank
Etank < 2
2< Etank < 20
Etank > 20
Check Vm
Vm < 106
106< Vm < 106.5
106.5 < Vm < 107
Vm > 107
No
It-1 > 0
Yes
It = 1 It = 0
0
It = It-1
It = 0.5
It = 0.75
It = 1
It = 0
HPWH Model
t = t+1
Figure 9. Flow diagram of proposed voltage control method
VI.
VOLTAGE CONTROL BY PROPOSED METHOD
A. Simulation conditions In order to analyse the proposed control method, Japanese residential distribution model (Fig.1) with PV distribution illustrated in Fig. 5 is considered. It is assumed that all typical household customers residing in the distribution model have HPWH. On the other hand, commercial customers do not have HPWH. In the simulation, it is assumed that all customers have similar hot water demand pattern, though they have their own hot water demand pattern in practical. Initial and requested amount of hot water for all customers are assumed to be same and are 6kWh and 11kWh respectively. Loading pattern of both customers illustrated in Fig.2 and PV output on sunny and cloudy days is used for continuous voltage calculation. Simulations are conducted separately for all considered PV
Figure 10. Voltage at node 55 on sunny day
proposed coordinated control of HPWHs method completely removes voltage violation for PV output on sunny and cloudy days for the case presented. However, there is a possibility that proposed control method cannot completely remove voltage violation in all nodes for larger PV distribution than presented in Fig. 5. Customers are not affected with hot water shortage as they had sufficient initial amount of hot water to cover their morning demand until HPWH starts its operation. However, during voltage control period, hot water demand is satisfied by operation from central controller using proposed method. In future works, author proposes local operation of HPWHs based on the nearest node voltage. ACKNOWLEDGEMENT The authors would like to thank Prof. Taniguchi Haruhito from the University of Tokyo, for providing the typical household customer measured load pattern and hot water demand pattern, and his comments. REFERENCES
[1] T. Senjyu etc., Optimal Distribution Voltage Control and Coordination with Distributed Generation IEEE transactions on power delivery, vol. 23, no. 2, April 2008 [2] S.Toma, T. Senjyu, A. Yona, H. Sekine, T. Funabashi and Chul-Hwan Kim, Optimal Control of Voltage in Distribution Systems by Voltage Reference Management, 2nd IEEE International Conference on Power and Energy (PECon 08), December 1-3, 2008, Johor Baharu, Malaysia [3] S. Yoshizawa, Y. Hayashi, M. Tsuji, and E. Kamiya, Centralized Voltage Control Method of Load Ratio Control Transformer and Step Voltage Regulator for Bank Fault Restoration, 2012 3rd IEEE PES Innovative Smart Grid Technologies Europe (ISGT Europe), Berlin [4] P. Esslinger and R. Witzmann, Regulated Distribution Transformers in Low-Voltage Networks with a high Degree of Distributed Generation, 2012 3rd IEEE PES Innovative Smart Grid Technologies Europe (ISGT Europe), Berlin [5] J. Yi, P. Wang etc., Distribution Network Voltage Control Using Energy Storage and Demand Side Response, 2012 3rd IEEE PES, ISGT Europe, Berlin [6] N.Takahashi, and Y.Hayashi, Centralized Voltage Control Method Using Plural D-STATCOM with Controllable Dead Band in Distribution System with Renewable Energy, 2012 3rd IEEE PES Innovative Smart Grid Technologies Europe (ISGT Europe), Berlin [7] Y. Kabasawa, T. Noda, K. Fukushima and K. Nemoto, Consumer voltage regulation using coordinated control of distributed static synchronous compensators STATCOMs, Innovative Smart Grid Technologies (ISGT Europe), 14-17 Oct. 2012, pp. 1- 7 [8] Ferry A. Viawan and Daniel Karlsson, Coordinated Voltage and Reactive Power Control in the Presence of Distributed Generation 2008 IEEE Power and Energy Society General Meeting [9] X. Liu, A. Aichhorn, L. Liu and H. Li, "Coordinated Control of Distributed Energy Storage System With Tap Changer Transformers for Voltage Rise Mitigation Under High Photovoltaic Penetration," IEEE Trans. Smart Grid, vol.3, no.2, pp.897-906, June 2012 [10] "Countermeasure for Harmonics Disturbance in Distribution Network", Electric Technology Research Vol.37, No.3 (1981) [11] "Distributed Autonomous Urban Systems for Mitigating Environmental Impact", The Japan Society for the Promotion of Science [12] Online: http://www2.iee.or.jp/~pes/model/chiiki/tables/index.3.5.a.html (in Japanese) [13] Kei Inoue and Yumiko Iwafune, "Operation of Heat Pump Water Heaters for Restriction of Photovoltaic Reverse Power Flow", 2010 International Conference on Power System Technology
Figure 11. Voltage at node 55 on cloudy day
Figure 12. Characteristics of an HPWH
In order to find maximum voltage violation duration, results obtained for voltage at node 55 is considered. Voltage higher than 106.5V is assumed as voltage violation. Table 3 shows total voltage violation durations for node 55 which are obtained from Fig.10 and 11. Voltage violation on sunny day is higher than voltage violation on cloudy day. However, it is completely eliminated by proposed coordinated control method.
TABLE III. DURATION OF VOLTAGE VIOLATION AT NODE 55 Duration of voltage violation (min) Sunny day PV only Proposed Method 285 0 Cloudy day 173 0
VII.
CONCLUSION
Voltage of the Japanese residential distribution system model is analysed with PV systems and proposed coordinated control method. It is found that 60% of uniformly distributed PV systems cause voltage violation at line end section in distribution system. Application of customer controllable load, HPWH, is suggested in order to control the voltage when PV systems output becomes large. The coordinated operating pattern of HPWHs is found in order to mitigate voltage violation so that reverse power flow can be limited. The

EPECs 2013 Papers

Uploaded by

Copyright:

Available Formats

EPECs 2013 Papers

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

EPECs 2013 Papers

Uploaded by

Copyright:

Available Formats

CONFERENCE COMMITTEES

400 300 200 100

Solar Photovoltaic Power

2/3 Power Electronics

Renewable Energies (PV, Wind Turbines, etc.) References (Local/Centralized)

Power Grid Intelligent Control Communication

Fig. 4. 2L-VSC Back-to-Back (2L-VSC BTB) converter.

Generator Filter Filter

Fig. 5. 3L-NPC Back-to-Back (3L-NPC BTB) converter.

1 or 3 phase 30 kW~10 kW Commercial/ Residential Type B

1 phase 10 kW~1 kW Residential Type C

1 phase 300 W~ Small System Type D

~30 kW Commercial/ PV Plants Type A

Fig. 3. Two dominant power electronics based WTS configurations.

(b) HERIC inverter

Fig. 8. Two transformerless PV inverters (H6 and HERIC) [20], [22].

Fig. 9. Neutral point clamped transformerless topology for PV application.

O Ground Current: iCMV CP vCMV

Fig. 7. Single-phase full-bridge PV inverter with an LCL-filter.

Vdc Current/Voltage Grid Control Synchronization Control

Power Maximization & Limitation Inertia Emulation

Fault ride through & Grid support Power Quality

WTS Specific Functions

Mission profiles Communication

Monitoring and Control

Supervisory command from DSO/TSO

PV Panels/ Strings PPV

Vdc Control Anti-Islanding Protection

Basic Control Functions

PV System Specific Functions

Fault Ride Through

Harmonic Compensation Constant Power Generation Control

Supervisory command from DSO/TSO

Monitoring and Control

1980 1985 1990 0% Power Rating: Electronics Role: Soft starter

2000 2005 10% 30% Rotor Rotor resistance power

2011 2018 (Estimated) 100% Full generator power

Yearly Solar Irrandiance

20 % Reduction of Feed-in Power 6.23 % Reduction of Energy Yield

Yearly Ambient Temperature

CDev  CCap  CO & M E Annual

Mission Profile to Component Stress

Design Tools for Reliability

Thermal Thermal Picture Picture

Six sigma Devices variation Production robustness ...

Cost of fossil fuel generation

[16] [17] [18] [19]

EPECS13 Conference Agenda Wednesday, October 2, 2013

Registration and Reception (Registration is open during the conference)

Transportation to Gala Dinner Gala Dinner Transportation to Hotels

Thursday, October 3, 2013

Parallel Session-Hall A Electric Machines&Drives

Parallel Session-Hall B High Voltage Coffee Break

Parallel Session-Hall C Power Distribution

Parallel Session-Hall B Power Systems Lunch Break

Parallel Session-Hall C High Voltage/ Power Systems

Parallel Session-Hall A Power Distribution/Smart Grid

Parallel Session-Hall C Electric Machines&Drives

Friday, October 4, 2013

Parallel Session-Main Hall Smart Grid

CDev CCap CO & M E Annual