Issue No. 2

International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 745-750 ISSN 2078-2365 Power Quality Enhancement using Custom Power Devices Pudi Sekhar#1, K. Venkateswara rao#2, T.Devaraju#3 #1 M.Tech Student, EEE Department, GITAM University #2 Lecturer, EEE Department, GITAM University #3 HOD, EEE Department, Sri Vidyanikethan Engineering college Abstract— A Power quality problem is an occurrence manifested as a nonstandard voltage, current or frequency that results in a failure or mis-operation of end use Equipments. Utility distribution networks, sensitive industrial loads, and critical commercial operations all suffer from various types of outages and service interruptions which can cost significant financial 1oss per incident based on process down-time, lost production, idle work forces, and other factors. With the restructuring of Power Systems and with shifting trend towards distributed and dispersed Generation, the issue of Power Quality is going to take newer dimensions. The aim therefore, in this work, is to identify the prominent concerns in the area and thereby to recommend measures that can enhance the quality of the power, keeping in mind their economic viability and technical repercussions. In this paper three custom power controllers: DVR, DSTATCOM (modelled in MATLAB/Simulink), PWM Switched Autotransformer (modelled in PSCAD/EMTDC & MATLAB/SIMULINK) are presented. Comprehensive results are presented to assess the performance of each device to mitigate the Voltage sag and economical device is proposed. Index terms— Power Quality Problems, Voltage sag, DVR, DSTATCOM, PWM Switched Autotransformer I. INTRODUCTION Power quality is certainly a major concern in the present era; it becomes especially important with the introduction of sophisticated devices, whose performance is very sensitive to the quality of power supply. Modern industrial processes are based a large amount of electronic devices such as programmable logic controllers and adjustable speed drives. The electronic devices are very sensitive to disturbances [1] and thus industrial loads become less tolerant to power quality problems such as voltage dips, voltage swells, and harmonics. Voltage dips are considered one of the most severe disturbances to the industrial equipment. Swells and over voltages can cause over heating tripping or even destruction of industrial equipment such as motor drives. Electronic equipments are very sensitive loads against harmonics because their control depends on either the peak value or the zero crossing of the supplied voltage, which are all influenced by the harmonic distortion. This paper analyzes the key issues in the Power Quality problems. As one of the prominent power quality problems, the origin, consequences and mitigation techniques of voltage sag problem has been discussed in detail. The study describes the techniques of mitigating voltage sag in a distribution system by two power electronics based devices called Dynamic Voltage Restorer (DVR) and Distribution STATCOM (D-STATCOM) and a new mitigation technique called PWM Switched Autotransformer. II. SOURCES AND EFFECTS OF POWER QUALITY PROBLEMS The distortion in the quality of supply power can be introduced /enhanced at various stages; however, some of the primary sources of distortion [2] can be identified as below: A. Power Electronic Devices B. IT and Office Equipments C. Arcing Devices D. Load Switching E. Large Motor Starting F. Embedded Generation G. Electromagnetic Radiations and Cables H. Storm and Environment Related Causes etc. Some of the common power quality issues and their prominent impact are summarized in the table below: Problem TABLE I Effects Page 745 Pudi et. al., Power Quality Enhancement using Custom Power Devices International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 745-750 ISSN 2078-2365 Voltage sags Transients Harmonics Flicker Devices /Process down time, effect on product quality, failure/malfunction of customer equipments and associated scrap cost, clean up costs, maintenance and repair costs etc. Tripping, components failures, flashover of instrument insulation hardware re booting, software glitches, poor product quality etc. Excessive losses and heating in motors, capacitors and transformers connected to the system Visual irritation, introduction of many harmonic components in the supply power and their associated equipment. Dynamic Voltage Restorer is a series connected device designed to maintain a constant RMS voltage value across a sensitive load. The DVR considered consists of:  an injection / series transformer  a harmonic filter  a Voltage Source Converter (VSC)  an energy storage and  a control system as shown in Figure 1 III. USE OF CUSTOM POWER DEVICES TO IMPROVE POWER QUALITY In order to overcome the problems mentioned above conventional devices such as: • • • • • • • • Line- voltage regulators: Tap changers, buck-boost regulators, CVT (Constant- voltage transformer). M-G Sets (Motor-generator Sets) Magnetic Synthesizers SVC (Static VAR Compensators) UPS (Uninterruptible Power Supplies) SMES (Superconducting magnetic energy storage) Static Transfer Switch Fuel Cell Based Inverter System can be used. But Present day modern equipments are very sensitive to voltage sags and they need the mitigating device to be very fast in acting, which cannot possible by the above conventional devices. So in order to overcome the above disadvantages, a new category of devices called custom power devices are developed. Custom power devices are the new generation of power electronics-based equipment aimed at enhancing the reliability and quality of power flows in lowvoltage distribution networks. There are various custom power devices available such as DVR (dynamic voltage restorer), Dstatcom, UPQC and PWM switching auto transformer. IV. MODELING OF CUSTOM POWER DEVICES AND SIMULATION RESULTS Fig. 1 Schematic diagram of DVR The main function of a DVR is the protection of sensitive loads from voltage sags/swells coming from the network. Therefore as shown in Figure 1, the DVR is located on approach of sensitive loads. If a fault occurs on other lines, DVR inserts series voltage VDVR and compensates load voltage to pre fault value. The momentary amplitudes of the three injected phase voltages are controlled such as to eliminate any detrimental effects of a bus fault to the load voltage VL. This means that any differential voltages caused by transient disturbances in the ac feeder will be compensated by an equivalent voltage generated by the converter and injected on the medium voltage level through the booster transformer. The series injected voltage of the DVR can be written as [3] Vinj = Vload + Vs (1) Where; Vload is the desired load voltage magnitude Vs is the source voltage during sags/swells condition The test system employed to carry out the simulations concerning the DVR actuation is shown in Figure 2 A. DVR Page 746 Pudi et. al., Power Quality Enhancement using Custom Power Devices International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 745-750 ISSN 2078-2365 Fig. 3 Simulation results showing: (a) Supply Voltage, (b) Injected Voltage (c) Load Voltage Fig. 2 Proposed System Configuration (DVR) To verify the working of a DVR employed to avoid voltage sags, simulation studies are carried out as follows: A three–phase short-circuit fault is applied during the period of 200 to 500 ms with a fault resistance of 0.001. The first simulation shows three phase voltage sag is simulated. The simulation started with the supply voltage 30% sagging as shown in Figure 3(a).Figure 3(a) also shows a 30% voltage sag initiated at 0.2s and it is kept until 0.4s, with total voltage sag duration of 0.2s and Figure 3(b) shows voltage injected by the DVR and the corresponding load voltage in Figure 3(c) with compensation. As a result of DVR, the load voltage is kept at 1pu. When the DVR is in operation the voltage sag is mitigated almost completely, and the rms voltage at the sensitive load point is maintained at 98%, as shown in Figure 3. B. D-STATCOM The STATCOM consists mainly of a PWM inverter connected to the network through a transformer. The dc link voltage is provided by capacitor C which is charged with power taken from the network [4].The control system ensures the regulation of the bus voltage and the dc link voltage. The D-STATCOM function is to regulate the bus voltage by absorbing or generating reactive power to the network, like a thyristor static compensator. Page 747 Pudi et. al., Power Quality Enhancement using Custom Power Devices International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 745-750 ISSN 2078-2365 Fig. 6 Voltage sag generated due to sudden increase of load Fig. 4 Block diagram of the proposed model The block diagram of the proposed model is shown in Fig. 4. The load 1 shown is the normal running load. At any specified time interval, a sudden load is included. The introduction of this sudden load introduces voltage sag. Hence if a DSTATCOM is connected in parallel to the distribution, it will correct the voltage sag in that interval. The SIMULINK model for generating a three phase sag in a power system is shown in Fig. 5. A three phase linear transformer is provided for isolation. Two feeders originate from this transformer. In one of the feeders, a sudden load is included from the period of 0.2 to 0.3s. This introduction of sudden load produces a reduction in voltage causing a voltage dip is shown in Fig.6. Fig. 7 Restored Voltage Waveform The error signal is obtained by comparing the per unit value of the voltage with a constant 1. The reference sinusoidal signals required for the generation of PWM pulses is obtained by using a PI controller in conjunction with the error signal obtained. The PWM techniques have several advantages compared to other techniques. The restored voltage waveform is shown in Fig. 7. C. VOLTAGE SAG MITIGATION USING PWM SWITCHED AUTOTRANSFORMER Fig. 5 Simulink model for Voltage Sag mitigation using D-STATCOM Fig. 8 Block diagram of the single phase voltage sag mitigation scheme Page 748 Pudi et. al., Power Quality Enhancement using Custom Power Devices International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 745-750 ISSN 2078-2365 (i) Principle of operation An IGBT is used as power electronic device to inject the error voltage into the line so as to maintain the load voltage constant. Four power diodes (D1 to D4) connected to IGBT switch (SW) controls the direction of power flow and connected in ac voltage controller configuration. This combination with a suitable control circuit maintains constant rms load voltage. In this scheme sinusoidal PWM pulse technique is used. RMS value of the load voltage V L is calculated and compared with the reference rms voltage Vref Under normal condition when there is no voltage disturbance the power flow is through the anti parallel thyristors used as the ac bypass switch. Output filters containing a main capacitor filter and a notch filter are used at the output side to filter out the switching noise and reduce harmonics. During this normal condition VL=Vref, and the error voltage V err is zero. The gate pulses are blocked to IGBT. A sag or swell occurs in the system may be due to sudden increase or decrease in the load, or due to faults. The supply voltage Vs and hence VL decreases during disturbance. When the sensing circuit detects an error voltage greater than of the normal voltage the voltage controller acts immediately to switch off the thyristors. Voltage Verr applied to the pi controller gives the phase angle  . The control voltage given in (2) is constructed at power frequency f = 50 Hz. Vcontrol = sin (wt+  ) (2) Where ma is the modulation index. The phase angle delta is dependent on the percentage of disturbance and hence controls the magnitude of Vcontrol.This control voltage is then compared with the triangular voltage Vtri to generate the PWM pulses VG which are applied to the IGBT to regulate the output voltage. Hence the IGBT switch operates only during voltage sag condition and regulates the output voltage according to the PWM duty-cycle. Single phase voltage sag supporter using a PWM Switched autotransformer with R.M.S voltage as a reference Fig. 9 PSCAD/EMTDC model of single phase PWM switched auto transformer with RMS voltage as reference The Fig. 9 shows the PSCAD/EMTDC model for single phase PWM switched auto transformer for mitigation of voltage sag with peak voltage as a reference. It is tested on a simple a.c system [5] [6] at which a single phase to ground fault is created at 0.2 sec for a duration of 0.2 sec with a fault resistance of 0.1 ohm, which creates a voltage sag of 50% at load .At 0.2 sec, the PWM switched auto transformer switches ON and it takes one cycle to detect the sag and compensate the voltage sag. The waveforms for source and load voltages, load currents before, during and after the fault are as shown in Fig. 10. RMS detection method: In this method the instantaneous load voltage is converted into rms voltage and compared with the reference rms voltage, which gives an error voltage and this voltage given PI controller, which calculates the required angle delta in proportion to error voltage. Page 749 Pudi et. al., Power Quality Enhancement using Custom Power Devices International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 745-750 ISSN 2078-2365 Fig. 13 Waveform of load voltage showing mitigated sag using PWM Switched autotransformer Fig. 10 Simulation results (a) Source voltage with sag (b) Load Voltage without sag Fig. 11 Simulink model PWM Switched Autotransformer of a 3-phase system used for voltage sag studies Figure 11 shows Simulink model PWM Switched Autotransformer [7] of a 3-phase system for voltage sag mitigation. A sudden loading is applied at 0.1 sec to 0.2 sec and a sag is observed as shown in Figure 12 and sag is mitigated using PWM Switched Autotransformer and the corresponding waveform is shown in 13. V. CONCLUSIONS Power quality measures can be applied both at the user end and also at the utility level. This paper has presented models of custom power equipment, namely DVR, D-STATCOM and PWM switched autotransformer, applied them to mitigate voltage dip which is very prominent as per utilities are concerned. Though conventional techniques are available, the proposed devices are very fast acting and efficient. A new Voltage sag mitigation topology called PWM switched auto transformer is modelled and simulated. This topology requires only one PWM switch per phase as compared to DVR or DSTATCOM requires two switches per phase. The PWM switched auto transformer does not require energy storage device for mitigation of voltage sag as compared to DVR and DSTATCOM requires energy storage elements. so, PWM switched autotransformer is efficient and economical among the custom power devices. REFERENCES [1] [2] [3] [4] [5] Fig. 12Waveform of the load voltage with sag during 0.1 sec to 0.2 sec [6] H. Hingorani “Introducing Custom Power” IEEE Spectrum, vol.32 no.6 June 1995 p 41-48. N.G. Hingorani and L. Gyugyi, “Understanding FACTS: Concepts and Technology of Flexible AC Transmission Systems”. Mitigation Of Voltage Sags/Swells Using Dynamic Voltage Restorer (DVR) Rosli Omar and Nasrudin Abd Rahim. 2008 Australasian Universities Power Engineering Conference (AUPEC'08) Paper P-027. Mitigation of voltage sag using Distribution Static Compensator (DSTATCOM) S. Elango Dr. E.Chandra Sekaran, 2011 IEEE. Mitigation of Voltage Sags/Swells using PWM Switched Autotransformer C. Venkatesh, Student Member, IEEE, V. Prasad Reddy, Student Member, IEEE, and Dr. D.V.S.S. Siva Sarma, Senior Member, IEEE. IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 22, NO. 2, MARCH 2007 A Voltage Sag Supporter Utilizing a PWM-Switched Autotransformer Dong-Myung Lee, Member, IEEE, Thomas G. Habetler, Fellow, IEEE, Ronald G. Harley, Fellow, IEEE, Thomas L. Keister, Member, IEEE, and Joseph R. Rostron, Member, IEEE. Page 750 Pudi et. al., Power Quality Enhancement using Custom Power Devices International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 745-750 ISSN 2078-2365 [7] Modeling and simulation of PWM Switched Autotransformer for voltage sag mitigation using MATLAB, International Journal of electrical and power engineering 4(3), 164-168, 2010. Page 751 Pudi et. al., Power Quality Enhancement using Custom Power Devices 

International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 632-641 ISSN 2078-2365 Dynamic analysis of Single Machine Infinite Bus system using Single input and Dual input PSS P. PAVAN KUMAR EEE Department, Gitam University, Visakhapatnam, Andhra Pradesh, India-533045 Email: pavan.kumar270489@gmail.com M. RAVINDRA BABU EEE Department, Gitam University, Visakhapatnam, Andhra Pradesh, India-533045 Email: raviravi1983@gmail.com Abstract – This paper deals with the design of both single input and dual input conventional PSS which is used to damp the low frequency rotor oscillations taking place in power systems. The single input PSS used here are power based derivative type and speed based lead-lag type stabilizer, the dual input stabilizer, PSS3B has two inputs namely from change in speed and deviation of electrical power and has two frequency bands, lower and higher unlike the single input PSS. The PSS parameters are tuned, considering the machine data and operating point of the system used. The optimal parameters of the PSS are obtained using pole placement and genetic algorithm technique and the respective results are compared graphically .The system used is Single Machine Infinite Bus (SMIB) system which is modelled using state space analysis and its dynamic response is analyzed both for system without PSS and with PSS (both single and dual input) using Simulink/Matlab. Keywords: temperature microchange, wireless sensor network, global change, integral equation, WSN. I. INTRODUCTION Power systems experience low-frequency oscillations due to disturbances. These low frequency oscillations are related to the small signal stability of a power system. The phenomenon of stability of synchronous machine under small perturbations is explored by examining the case of a single machine connected to an infinite bus system (SMIB). The analysis of SMIB gives physical insight into the problem of low frequency oscillations. These low frequency oscillations are classified into local mode, inter area mode and torsional mode of oscillations. The SMIB system is predominant in local mode low frequency oscillations [7]. These oscillations may SARASWATHI EEE Department, Gitam University, Visakhapatnam, Andhra Pradesh, India-533045, Email: g_saraswathi@gitam.edu sustain and grow to cause system separation if no adequate damping is available. Small signal disturbances observed on the power system are caused by many factors such as heavy power transmitted over weak tie line and the effect of fast acting, high gain automatic voltage regulator (AVRs) [6]. The main function of the AVR is to improve the transient stability during faults conditions. However, its high gain and fast acting effect have an adverse effect on the system damping which is reduced to a negative value. The under damped system exhibits low frequency oscillations also known as electromechanical oscillations. These oscillations limit the power transfer over the network and if not properly damped, they can grow in magnitude to cause system separation. To counteract the adverse effects of the AVRS, Power system stabilizer (PSS) is used in the auxiliary feedback to provide supplementary damping [6] to the system to damp these low frequency oscillations on the rotor. To overcome this problem, several approaches based on modern control theory, such as Optimal control, Variable control and intelligent control were simulated and tested with satisfactory results. But these stabilizers have been proved to be difficult to implement in real systems. Thus, CPSS remains widely used by power utilities for its simple structure and reliability. Over the past 15 years, interests have been focused on the optimization of the PSS parameters to provide adequate performance for all operating conditions. Hence, many optimizations techniques based on artificial intelligence have been used to find the optimum set of parameters to effectively tune the PSS. In this paper both single input (speed & power based) and dual input stabilizers (PSS3B) are used to damp the low frequency oscillations associated with the system. PSS3B is used with combination of shaft speed deviation (∆ω) and Page 632 Pavan et. al., Dynamic analysis of Single Machine Infinite Bus system using Single input and Dual input PSS International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 632-641 ISSN 2078-2365 change in electrical power (∆Pe) which has its own advantages when compared to single input PSS which is described below in section 3. The parameters of both the types of PSS are tuned using Pole Placement technique and Genetic Algorithm and results are thus analyzed. II. METHODOLOGY A single machine-infinite bus (SMIB) system is considered for the present investigation. A machine connected to a large system through a transmission line may be reduced to a SMIB system, by using Thevenin’s equivalent of the transmission network external to the machine. The synchronous machine is described as the fourth order model. The two-axis synchronous machine representation with a field circuit in the direct axis but without damper windings is considered for the analysis. The system dynamics of the synchronous machine can be expressed as a set of four first order linear differential equations given in equations below [6]. These equations represent a fourth order generator model. The Heffron-Phillips constants are dependent on the machine parameters and the operating condition considered for the system. Here K1, K2, K3 and K6 are positive [6]. K4 is mostly positive except for cases where Re is high. K5 can be either positive or negative and K5 is positive for low to medium external impedances (Re+ jXe) and low to medium loadings. K5 is usually negative for moderate to high external impedances and heavy loadings [6]. The overall linearized block diagram of the SMIB system is shown in Fig.1 below. For the system considered four state variables are considered and linearized differential equations can be written in the state space form as, (9) Where, (10) (1) (2) (11) (3) (4) (12) The constants (K1-K6) are called Heffron-Phillips constants and are computed using the equations given in Appendix. The system data considered is: xd = 0.973 = 0.19 xq = 0.55 = 7.765s D=0 H=5 f=60Hz (5) Transmission line (p.u): Re = 0 Xe = 0.4 (6) Exciter: KE = 200 TE = 0.05s (7) Operating point: Vto = 1.0 P0 =1.0 Q0 = 0.2 δ0 = 28.26 o (8) Page 633 Pavan et. al., Dynamic analysis of Single Machine Infinite Bus system using Single input and Dual input PSS International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 632-641 ISSN 2078-2365 Since PSSs are tuned at the nominal operating point, the damping is only adequate in the vicinity of those operating points. But power systems are highly nonlinear systems, therefore, the machine parameters change with loading and time. The dynamic characteristics also vary at different points. IV. CONVENTIONAL POWER SYSTEM STABILIZER The basic function of CPSS is to damp electromechanical oscillations. To achieve the damping, the CPSS proceeds by controlling the AVR excitation using auxiliary stabilizing signal. The CPSS’s structure is illustrated in Figure 2. Fig. 1. Linearized block diagram of SMIB In the above state space equation system state matrix A is a function of the system parameters, which depend on operating conditions, control matrix B depends on system parameters only and control signal U is the PSS output. Using these state equations and state matrices the overall transfer function of the system is computed, since here no controller is used, it is considered as open loop system whose transfer function is G(s). III. Fig. 2. Structure of CPSS The CPSS classically uses the following inputs [5]: · The shaft speed deviation ∆ω · Active power output, ∆Pa (Change in accelerating power) · ∆Pe (change in electric power), · Bus frequency ∆f POWER SYSTEM STABILIZER One problem that faces power systems nowadays is the low frequency oscillations arising from interconnected systems. Sometimes, these oscillations sustain for minutes and grow to cause system separation. The separation occurs if no adequate damping is available to compensate for the insufficiency of the damping torque in the synchronous generator unit. This insufficiency of damping is mainly due to the AVR exciter’s high speed and gain and the system’s loading. In order to overcome the problem, PSSs have been successfully tested and implemented to damp low frequency oscillations. The PSS provides supplementary feedback stabilizing signal in the excitation system. The feedback is implemented in such a way that electrical torque on the rotor is in phase with speed variations [7]. PSS parameters are normally fixed for certain values that are determined under particular operating conditions. Once the system operating conditions are changed, PSS may not produce adequate damping into an unstable system. 1. Gain The gain determines the amount of damping introduced by the stabilizer. Therefore, increasing the gain can move unstable oscillatory modes into the left – hand complex plane. Ideally, the gain should be set to a value corresponding to a maximum damping. However, in practice the gain Kpss is set to a value satisfactory to damp the critical mode without compromising the stability of other modes. 2. Washout The washout stage is a High Pass Filter (HPF) with purpose to respond only to oscillations in speed and block the Page 634 Pavan et. al., Dynamic analysis of Single Machine Infinite Bus system using Single input and Dual input PSS International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 632-641 ISSN 2078-2365 dc offsets. The Washout filter prevents the terminal voltage of the generator to drift away due to any steady change in speed. 3. Phase compensation This stage consists of two lead – lag compensators as shown in Figure 2 (lead – lag compensation stage). The lead stage is used to compensate for the phase lag introduced by the AVR and the field circuit of the generator. The lead – lag parameters T1-T4 are tuned in such as way that speed oscillations give a damping torque on the rotor. When the terminal voltage is varied, the PSS affects the power flow from the generator, which efficiently damps the local modes. 1) Speed based lead-lag PSS: These stabilizers employ the direct measurement of shaft speed (∆ω) and employ it as input signal for it. The stabilizer, while damping the rotor oscillations, could reduce the damping of the lower-frequency torsional modes if adequate filtering measures were not taken [1 & 5]. In addition to careful pickup placement at a location along the shaft where low-frequency shaft torsionals were at a minimum electronic filters called torsional filters should be used for adequate damping of low frequency oscillations. The structure of this PSS is in the form as shown below [1], for which the parameter such as stabilizer gain Kc, lead lag time constants T1 and T2 are to be computed such that the overall closed loop system will be stable when the PSS is included in the feedback loop. 4. Torsional Filter This stage is added to reduce the impact on the torsional dynamics of the generator while preventing the voltage errors due to the frequency offset. 5. Limiter The PSS output requires limits in order to prevent conflicts with AVR actions during load rejection. The AVR acts to reduce the terminal voltage while it increases the rotor speed and the bus frequency. Thus, the PSS is compelled to counteract and produce more positive output. As described in by P. Kundur in [8], the positive and negative limit should be around the AVR set point to avoid any counteraction. The positive limit of the PSS output voltage contributes to improve the transient stability in the first swing during a fault. The negative limit appears to be very important during the back swing of the rotor. V. (13) 2) Power based derivative PSS: Due to the simplicity of measuring electrical power and its relationship to shaft speed, it was considered to be a natural candidate as an input signal to early stabilizers. The equation of motion for the rotor can be written as follows [1 & 5]: (14) Where, H = inertia constant ΔPm= change in mechanical power input ΔPe= change in electric power output Δω = speed deviation As previously mentioned this type of stabilizer uses electrical power (∆Pe) as input and is of derivative type whose structure is as shown below [1], and the optimal stabilizer parameter K and T are to be computed which ensure closed loop stability of the system. (15) SINGLE INPUT PSS The input signals include deviations in the rotor speed (∆ω=ωmech – ωo), the frequency (∆f), the electrical power (∆P e) and the accelerating power (∆Pa) [5]. As mentioned above in this paper two types of PSS are considered to damp the low frequency oscillations they are, 1.1 Dual input CPSS (PSS3B) In this paper a dual input PSS is used, the two inputs to dualinput PSS are Δω and ΔPe, with two frequency bands, lower frequency and higher frequency bands, unlike the conventional single-input (Δω) PSS [2]. The performance of IEEE type Page 635 Pavan et. al., Dynamic analysis of Single Machine Infinite Bus system using Single input and Dual input PSS International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 632-641 ISSN 2078-2365 PSS3B is found to be the best one within the periphery of the studied system model. This dual input PSS configuration is considered for the present work and its block diagram representation is shown in Figure 3 for the control of low frequency speed oscillations, whose transfer function is taken as H(s) [4]. The simple block diagram considered for pole placement technique is shown below. Fig. 4. Closed loop system including PSS .Fig. 3. IEEE type PSS3B structure In the above PSS structure used [2], the unknown parameters are computed using pole placement and genetic algorithm techniques, in case of pole placement technique the transfer function of the pss is computed and is used in feedback to form a closed loop system, for which characteristic equation is formed to compute the unknown parameters of PSS by placing dominant eigen values in place of ‘s’ in the characteristic equation.. Let the linearized equations of single machine, infinite bus system be expressed in the form, sX(s) = AX(s)+BU(s) (17) Y(s) = CX(s) (18) The PSS with the following structure is used [4], (19) (16) The transfer function of the PSS3B used is shown above, and the pole placement technique is explained in detail in section VI. VI. Where the PSS parameter are to determined such that system dominant eigen values are equal to desired eigen values. Using equations (17),(18) and (19), it can be readily shown that the closed loop system characteristic equation is given by, POLE PLACEMENT TECHNIQUE (20) Pole placement is a method employed in feedback control system theory to place the closed-loop poles of a plant in predetermined locations in the s-plane. This method is also known as Full State Feedback (FSF) technique. Placing poles is desirable because the location of the poles corresponds directly to the eigen values of the system, which control the characteristics of the response of the system. Based on the system data considered and the operating condition, the Heffron-Phillips constants for the system are computed. The state equations are then considered using these constants to compute the state matrices and then the transfer function of the open loop system is computed in matlab using these state matrices. The open loop system transfer function is taken as G(s). Now in the feedback loop, the stabilizer is used From eqn.(20) the required stabilizer parameters can be computed by replacing ‘s’ by the desired eigen value λ and equating the real and imaginary terms on both sides of the equation [4]. Using the state equations and state matrices mentioned in section 2, the open loop transfer function G(s) of the system is obtained, and the PSS of structure shown in eqns.(13),(15) and (16) is used as feedback H(s) for the open loop system and thus forming the closed loop system with unknown parameters, which are computed as mentioned above by replacing ‘s’ by dominant eigen values. VII. GENETIC ALGORITHM (GA) Page 636 Pavan et. al., Dynamic analysis of Single Machine Infinite Bus system using Single input and Dual input PSS International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 632-641 ISSN 2078-2365 Genetic Algorithms (GAs) are heuristic search procedures inspired by the mechanism of evolution and natural genetic. They combine the survival of the fittest principle with information exchange among individuals. GA’s are simple yet powerful tools for system optimization and other applications [11]. This technique has been pioneered few decades ago by Holland, basing the approach on the Darwin’s survival of the fittest hypothesis. In GA’s candidates solutions to a problem are similar to individuals in a population. A population of individuals is maintained within the search space of GAs, each representing a possible solution to a given problem. The individuals are randomly collected to form the initial population from which improvement is sought. The individuals are then selected according to their level of fitness within the problem domain and breed together. The breeding is done by using the operators borrowed from the natural genetic, to form future generations (offsprings) [11]. The population is successively improved with respect to the search objective. The least fit individuals are replaced with new and fitter offspring from previous generation. trapped in a local minimum. Mutation plays the role of recovering the lost genetic materials as well as for randomly disturbing genetic information. Mutation has traditionally considered as a simple search operator [11]. If crossover is supposed to exploit the current solution to find better ones, mutation is supposed to help for the exploration of the whole search space. 4) Replacement: Replacement is the last stage of any breeding cycle. It is in this process that children populate the next generation by replacing parents, if fitter. Reinsertion can be made partially or completely, uniformly (offspring replace parents uniformly at random) or fitness-based. The most common operators handled in genetic algorithm are described in detail below, which in whole called as breeding cycle. 1) Selection (Reproduction): In this stage, individuals are selected from the current population according to their fitness value, obtained from the objective function previously described. The purpose of the selection is to choose individuals to be mated. The selection can be performed in several ways. But many selection techniques employ a “roulette wheel” [11]. It is a mechanism to probabilistically select individuals based on some measure of their performances. 2) Crossover (Recombination): In this stage, the individuals retained (in pairs), from the above stage, exchange genetic information to form new individuals (offsprings). This process helps the optimization search to escape from possible local optima and search different zones of the search space [11]. The combination or crossover is done by randomly choosing a cutting point where both parents are divided in two. Then the parents exchange information to form two offsprings that may replace them if the children are fitter. 3) Mutation: After crossover, the strings are subjected to mutation. Mutation prevents the algorithm to be Fig. 5. General Scheme of Genetic Algorithm All these operation are carried out in Genetic Algorithm toolbox in which the following fitness function has to be defined. The problem of computing optimal parameters of a single power system stabilizer for different operating points implies that power system stabilizer must stabilize the family of N plants [1]: , k= 1,2,3….N (21) Where X(t) is the state vector and U(t) is the input stabilizing signal. A necessary and sufficient condition for the set of plants in the system to be simultaneously stabilizable with stabilizing signal is that Eigen values of the closed-loop system lie in the left- hand side of the complex s-plane [1]. This condition motivates the following approach for determining parameters Ks1, Ks2, T1 and T2 of the power system stabilizer. Selection of Ks1, Ks2, T1 and T2 to minimize the following fitness function, Page 637 Pavan et. al., Dynamic analysis of Single Machine Infinite Bus system using Single input and Dual input PSS International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 632-641 ISSN 2078-2365 i=1,2,…N, k=1,2,..N (22) consideration which are further used to compute the HeffronPhillips constants (A.2). Where λi,k is the kth closed-loop eigen value of the ith plant [1]. If a solution is found such that J<0, then the resulting Ks1, Ks2, T1 and T2 stabilize the collection of plants. For running the GA toolbox the command gatool [10], is to be given in command window of MATLAB and in the tool the fitness function is to be defined in which the state matrix A including PSS is used and the unknown PSS parameters are taken as unknown variables which are to be optimized such that the eigen values of the matrix lie on the left half of s-plane i.e., in the stability region. This method of finding the parameter is applied for the type of PSS described in section 3. The state matrices ‘A’ and the specifications used for running GA toolbox are mentioned in Appendix. VIII. APPENDIX 1.2 Calculation of Heffron-Phillips constants All the variables with subscript ‘0’ are values of variables evaluated at their pre-disturbance steady-state operating point from the known values of P0 , Q0 and Vt0. (A.2) 1.3 Modelling of System including Speed based PSS (∆ω) When PSS of structure described in equation (13) is used as feedback of open loop system, it forms a closed loop system. The state equations involved are, (A.1) The above equations indicated in (A.1) are used to calculate the initial conditions of the system under Page 638 Pavan et. al., Dynamic analysis of Single Machine Infinite Bus system using Single input and Dual input PSS International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 632-641 ISSN 2078-2365 (A.3) 1.4 Modelling of System including Power based PSS (∆Pe) When PSS of structure described in equation (16) is used as feedback of open loop system, it forms a closed loop system. The state equations involved are, (A.4) The wash out time constant for the both speed and power based PSS is taken as Tw= 2sec 1.5 Modelling of System including PSS3B The state equations of the system when PSS of structure shown in section 3 is used in the feedback loop are derived as below (A.3). Page 639 Pavan et. al., Dynamic analysis of Single Machine Infinite Bus system using Single input and Dual input PSS International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 632-641 ISSN 2078-2365 Crossover fraction 0.7 Mutation function Use constraint dependent default Crossover function Single point Migration direction Forward Number of generations 300 The application of GA tool box for optimization of PSS parameters, the following constraints on the parameters has to be considered, (A.5) The state matrix ‘A’ of the system including PSS3B is shown below (A.6) which is used in the objective function to evaluate the fitness using GA tool box. For speed based PSS, 10 ≤ Kc ≤ 50; 0.01 ≤ T1 ≤ 1; 0.01 ≤ T2 ≤ 0.1 For power based PSS, 0.1 ≤ K ≤ 10; 0.01 ≤ T ≤ 1 -3 ≤ Ks1 ≤ 0 ; 20 ≤ Ks2 ≤ 60 ; 0 ≤ T1 ≤ 0.3 ; 0 ≤ T2 ≤ 0.1 IX. RESULT AND DISCUSSION The parameters of the PSS obtained using pole placement and Genetic Algorithm techniques are shown below. Single input parameters: 1) Speed based PSS using Pole placement technique are, Kc=9.6763, T1=0.285sec, T2=0.05sec (A.6) The washout time constants is taken as T w1=Tw2=10sec. 1.6 Specifications of Genetic Algorithm For using GA toolbox to optimize the PSS parameters the following specifications are used, Parameterd obtained using GeneticAlgorithm is, Kc=10.541, T1=0.498sec, T2=0.1sec 2) Power based PSS using Pole placement technique are, K=0.8954, T=0.3104sec Parameters obtained using Genetic Algorithm is, K=3.4, T=0.498sec Table 1. Genetic Algorithm Specifications for Toolbox Table 2. PSS3B parameters Population size 75 Creation function Use constraint dependent default Scaling function Rank Selection function Roulette PSS3B Parameters Ks1 Ks2 T1 T2 Pole Placement -0.5 48.259 0.05sec 0.25sec Genetic algorithm -0.354 20.003 0.15sec 0.1sec Page 640 Pavan et. al., Dynamic analysis of Single Machine Infinite Bus system using Single input and Dual input PSS International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 632-641 ISSN 2078-2365 technique and are simulated to analyse the dynamic response in both the cases. The technique of computing parameters becomes complex with the increase in number of machines in case of pole placement technique, whereas the technique of Genetic Algorithm can be used to compute optimal parameters of PSS for wide range of operating conditions in power system and also can be implemented for multi-machine system. The settling time of the PSS is less in case of Genetic Algorithm technique when compared to Pole Placement Technique. XI. Fig. 6. Simulation output of SMIB with GA-PSS REFERENCES [1] “Tuning of Power System Stabilizers via Genetic Algorithm for Stabilization of power Systems” by MehranRashidi, FarzanRashidi, Hamid Moaavar, 0-7803-7952-7/03/$17.00 0 2003 IEEE [2]. M. Sreedevi and P. Jeno Paul, “Comparison of Two Power system stabilizers for Power system stability” International Journal of Signal System Control and Engineering Applications, 3(4): 70-76, 2010 ISSN: 1997-5422 [3]. Joe H. Chow, George E. Boukarim, and Alexander Murdoch, “Power System Stabilizers as Undergraduate Control Design Projects”, IEEE transactions on power systems, vol. 19, no. 1, February 2004. pp. 144-151 [4]. “Efficient pole-assignment method for designing stabilisers in multimachine power systems” by S. Elangovan and CM. Lim, IEE PROCEEDINGS, Vol. 134, Pt. C, No. 6, NOVEMBER 1987 [5]. G.R. Bérubé, L.M. Hajagos, Members Kestrel Power Engineering Ltd. Accelerating-Power Based Power System Stabilizers Fig. 7. Simulation output of SMIB with Pole Placement-PSS The settling time of the simulation response for PSS3B are compared in table shown below, Settling Time Without PSS Pole placement PSS GA PSS Table 3. Settling time comparison Single input PSS Dual input PSS Speed based Power PSS3B PSS based PSS 56.43sec 56.43sec 56.43sec 4.14sec 5.79sec 3.66sec 3.29sec X. 1.93sec 1.74sec [6].K.R.Padiyar, “Power system Dynamics Stability and Control”. John Wiley; Interline Publishing, 1996 [7].P. M. Anderson and A.A Fouad “Power system control and stability”,Iowa state university Press 1977 [8].P.Kundur, “Power system stability and control”. McGraw-Hill, New York 1994. [9]. E.V Larsen and D.A. Swann, “Applying Power System Stabilizers, Parts I, I1 and III”, IEEE Trans., Vol. PAS-100, June 1981, pp. 3017- 3046 [10]. A. J. Chipperfield and P. J. Fleming “The MATLAB Genetic Algorithm Toolbox” From IEE Colloquium on Applied Control Techniques Using MATLAB, Digest No. 1995/014, 26/01/95 [11]. S.N.Sivanandam, S.N .Deepa “Introduction Algorithms”.Springer-Verlag Berlin Heidelberg 2008. to Genetic CONCLUSION The optimal parameters of dual input conventional pss, PSS3B is obtained using pole placement and genetic algorithm Page 641 Pavan et. al., Dynamic analysis of Single Machine Infinite Bus system using Single input and Dual input PSS 

International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 738-744 ISSN 2078-2365 Capacitors Natural Voltage Balancing Mechanism Investigation in Flying Capacitor Multicell Converters Vahid Dargahi and Abbas Shoulaie Department of Electrical Engineering, Iran University of Science and Technology, Tehran 16846, Iran vdargahi@elec.iust.ac.ir, shoulaie@iust.ac.ir Abstract—This paper presents an analytical analysis of flying capacitors voltage balancing process in flying capacitor multicell converters. The analysis is based upon renowned theory of Double Fourier series which leads to modeling and state-space representation of converters. State-space representation of converter can be utilized to investigate the transient and steady states of internal flying capacitors voltages. To provide verification, experimental results acquired from a laboratory prototype are compared against numerical solution of differential equations of converter state-space representation and simulation results. Keywords Flying Capacitor Multicell Converter, Double Fourier Series, Self balancing. I. INTRODUCTION As a consequence of reaching higher power and lack of its suitable ranged switches, multilevel converters popped up in 1975 and have been continuously developed in recent years due to the necessity of increase in power level of industrial applications especially high power applications such as high power AC motor drives, active power filters, reactive power compensation and FACTS devices. The main reason is the capability of these topologies to handle voltage/power in the range of kilovolts/megawatts as a result of recent developments in the area of high power semiconductors [1]-[5]. The concept of multilevel arises from acquiring a staircase output voltage waveform as voltage levels from input dc voltages by means of converter appropriate configuration and its proper switching pattern. This staircase voltage by its resemblance to sinusoidal voltage waveform leads to primitive advantages of utilizing switches with low-voltage ratings, higher power quality, lower total harmonic distortion, etc [1][5]. The term multilevel starts with the three-level converter introduced by Nabae et al. The Neutral Point Clamped (NPC) converter, presented in the early 80’s, is a standard topology in industry on its 3-level version. However, for a higher number of levels, this topology has some drawbacks such as: voltage balance of the dc-link capacitors and the number of clamping diodes [2]-[4]. Alternatives for the NPC converters are the multicell topologies. Different cells and approaches to interconnect them lead to many topologies which the most important ones are the Cascaded Multicell (CM) and the Flying Capacitor Multicell (FCM) accompanied by its sub-topology Stacked Multicell (SM) converters [2][3][5]. The FCM converter, and its derivative, the SM converter, have many advantageous properties for medium voltage applications, particularly the transformer-less operation and the ability to naturally maintain the flying capacitors voltages at their target operating levels. This substantial property is called natural balancing and allows the construction of such converters with a large number of voltage levels. Natural selfbalancing of the flying capacitors voltages occurs without any feedback control. A necessary condition for this phenomenon is that average currents of the flying capacitors must be zero. As a result, each cell must be controlled with the same duty cycle and a regular phase shifted progression along the cells. Generally, an output RLC filter (balance booster circuit), tuned to the switching frequency or multiple of that, is suggested to be connected across the load in order to accelerate this self balancing process in the transient states [2][3][5]. The FCM converter uses a series connection of “cells” comprising a flying capacitor and its associated complimentary switch pair and produces a switched voltage that is the sum of the individual cell states [2][3][5]. Despite of mentioned appreciable advantages, multilevel converters possess some following main drawbacks: increased number of isolated dc voltages, clamping diodes, capacitors and of power semiconductor switches accompanied by their related gating and protection circuits which result in a sophisticated overall system [2][3][5]. As mentioned, voltage natural balancing mechanism is a fundamental principle in flying capacitor and stacked multicell converters which consents to construction of voltage levels at the converter output [8]-[11]. The main objective of this paper is to provide a mathematical model for flying capacitor multicell converters intended for investigation of transient and steady state of flying capacitors voltages accompanied by taking into account the effect of balance booster circuit. Page 738 Vahid and Shoulaie, Capacitors Natural Voltage Balancing Mechanism Investigation in Flying Capacitor Multicell Converters International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 738-744 ISSN 2078-2365 II. INSTANTANEOUS MODELING OF FLYING CAPACITOR MULTICELL CONVERTERS IN STATE-SPACE REPRESENTATION A. Fundamental Concepts of Flying Capacitors Multicell Converters Flying capacitor multicell converters (FCMCs) which have been proposed by T.A. Meynard are relatively new breed of multilevel converters in comparison with conventional neutral point clamped (NPC) and cascade H-bridge (CHB) ones. A typical configuration of FCMC is depicted in Fig 1. As illustrated, R cells in a FCMC are overlapped to form a required converter’s leg. Each cell consists of one voltage source (a dc voltage source equal to E in Rth cell and capacitors possessing specific voltages in remaining cells) and two power semiconductor switches which are in complementary state to each other to avoid short-circuiting of voltage sources. Phase shifted carrier sinusoidal pulse width modulation (PSCSPWM) technique is the most common control scheme which is applied to switching strategy of FCMCs to guaranty both best harmonic performance and voltage balancing mechanism of flying capacitors. It should be noted that in a R-cell FC converter each switch sustains just a fraction of DC link voltage, i.e. E/R. This R-cell configuration leads to R+1 levels of voltage with peak to peak voltage value of E at the converter output. Flying capacitor multicell converters are in preference to the NPC and CHB ones as considering advantages such as: modularity, noninterdependency of cells as fault occurs and ease of reaching higher voltage levels just by introducing new cells [2][3][5]. Switches states of a 4-cell-5-level FCMC and output voltage, using PSCSPWM control method, are illustrated in Table-1. 0.5E SR 0.5E SR S3 + EC( R-1)= ( R-1)E/ R S3 S2 + EC2= 2E/R S2 S1 + EC1 = E/R S1 Cell- R Cell-2 + l r Cell-1 _ Fig. 1. R-cell flying capacitor multicell converter with maximum output voltage value of E. Table-1: States of switches in a 4-cell-5-level conventional FCMC. Output Number of State of Switches {(S4, S3, S2, S1)} Voltage States Level +E {(1,1,1,1)} 1 + 0.5E {(1,1,1,0)(1,1,0,1)(1,0,1,1) (0,1,1,1)} 4 0 {(1,1,0,0)(1,0,0,1)(0,0,1,1)(1,0,1,0)(0,0,1,1)(0,1,0,1)} 6 -0.5E {(1,0,0,0)(0,1,0,0)(0,0,1,0)(0,0,0,1)} 4 -E {(0,0,0,0)} 1 B. Instantaneous Model of the Converter in the State-Space Representation By utilizing proper switching pattern in FCMCs, their capacitor voltages would reach to the specific values which allow to constructing the desired output voltage levels. This property is known as natural voltage balancing mechanism and is achieved using phase shifted carrier pulse width modulation (PSCPWM) switching technique in the converter [8]-[11]. Natural self-balancing process of the flying capacitors voltages, as one of the advantages of FCM converters occurs without any feedback control. A necessary condition for this phenomenon is that average currents of the flying capacitors must be zero. As a result, each cell must be controlled with the same duty cycle and a regular phase shifted progression along the cells. Generally, an output RLC filter (balance booster circuit), tuned to the switching frequency or multiple of that, is suggested to be connected across the load in order to accelerate this self balancing process in the transient states. In this case, the dynamic of the self-balancing process depends on the impedance of load at the switching frequency. If the impedance at the switching frequency is high then the natural balancing is very slow and vice versa. The output RLC filter should be tuned to the switching frequency as follow [8]-[11]: Lb  Cb  1 2    fSW (1) Where, fSW is the switching frequency, Lb and Cb are inductance and capacitance of the output RLC booster circuit, respectively. iout( t ) Vout The output voltage of a R-cell FC converter has R+1-levels and its frequency spectrum has the harmonics around the  R  k  fSW  th harmonic where k and fSW are the integer number and the switching frequency, respectively [2][3][5]. However in this section a mathematical model of the converter will be presented to verify this balancing property. This model describes time domain differential equations of the flying capacitors voltages and state-space representation of the converter. Utilization a numerical solution of differential equations leads to acquire transient and steady state response of flying capacitors voltages. According to Fig. 1, a switching function of a cell  is defined as follows [8]-[11]:   1 H  t      1 if S is on if S is off   1: R (2) And its double Fourier series expansion can be expressed as follows [6]-[7]: Page 739 Vahid and Shoulaie, Capacitors Natural Voltage Balancing Mechanism Investigation in Flying Capacitor Multicell Converters International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 738-744 ISSN 2078-2365 H  (t )  M cos r t  r       4   sin   m  n   J n  m M       2  2    m          1 2   m1 n   cos  m c t  c    n r t  r      R      (3)  Where M , r , r , J n . , c , c are modulation index, angular frequency of reference sinusoidal waveform, phase angle of reference sinusoidal waveform, nth order Bessel function of first kind, triangular carrier waveform angular frequency and triangular carrier waveform phase angle, respectively. Output voltage and current of the converter and associated differential equations of the flying capacitors voltages based on mentioned switching functions can be obtained as follows [8][11]: E 1 R1 H R (t )    H   t   H  1  t   Ec 2 2  1 Vout  iout  t   Vout  t  (5) ZL 1   H  1  t   H   t  iout  t  dt 2 dEc C (4) (6) Where, ZL is a series connection of resistance  r  and inductance  l  and C is capacitance of flying capacitors. Flying capacitor multicell converter’s state-space representation can be written as follows:  Ec  A Ec  B  E 1  Hi 1  t   H i  t   j 1  t    j t  4Ci  Bi1  R  t   H i 1  t   H i  t  4Ci  ZL r   r l  M 2 2      l  cos  r t  r  tan 1  r    r       4  sin   m  n      2 2  2   m r    mc  nr  l           (   1)2      m  t  c         c R   m1 n      J  m  M  cos   n  t       r r  n  2           m n l r    tan 1  c           r      (11)   1: R To consider the balance booster circuit effect on the dynamic of the self-balancing process following modification should be applied to the acquired state-space equations: Aij  1  H i 1  t   H i  t    j 1  t    j 1  t      j  t    j  t   4Ci  (12) 1 R  t   R  t    H i 1  t   H i  t   4Ci  (13) i, j  1: R  1 Bi1  i  1: R  1 And: (7) Ec   Ec1 Ec 2 Ec3 ...... Ec ( R1)  Aij  H  t    t   T i, j  1: R  1 i  1: R  1 (8) (9) (10) And: Page 740 Vahid and Shoulaie, Capacitors Natural Voltage Balancing Mechanism Investigation in Flying Capacitor Multicell Converters International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 738-744 ISSN 2078-2365   t    H  t  Zb  M 1  2  Rb  r Lb  r Cb      r t  r     r Lb  r Cb 1   cos  1 2    tan     Rb                     4    2  1    m R 2  m  n L  m  n C        b c r b c r b                      sin   m  n   J n  m M    2  2   m1 n              m  t    (   1)2    c    c R         cos   n r t  r       1    m  n  L    m  n  C       r b c r b c 1       tan     Rb          1: R  III.  (14) Fig. 2. Transient and steady state of internal flying capacitors voltages of a 6cell-7-level FCMC acquired from state-space numerical solution and simulation for resistive-inductive load in parallel with RLC booster circuit. Table-3: Parameters used in simulation and numerical solution of state-space representation of 6-cell-7-level FCMC. System Parameters Values DC voltage (E) 600 V Internal flying capacitors (C) 2200 uF PS-SPWM carrier frequency (fSW) 5 kHz Fundamental output voltage frequency 50 Hz Modulation index 0.8 Resistive-inductive load (r-l) 10 Ω – 50 mH Booster circuit(RLC) 10 Ω – 10 uH-101.32 uF NUMERICAL AND SIMULATION RESULTS To provide verification to the elaborated state-space representation of the FCMC, numerical solution is utilized to solve differential equations of 6-cell-7-level and 4-cell-5-level converters. Transient and steady state of internal flying capacitors voltages of mentioned converters acquired from state-space numerical solution are shown in Figs. 2-6. System parameters used in numerical solution are given in Tables 2-5. Table-2: Parameters used in simulation and numerical solution of state-space representation of 6-cell-7-level FCMC. System Parameters Values DC voltage (E) 600 V Internal flying capacitors (C) 560 uF PS-SPWM carrier frequency (fSW) 5 kHz Fundamental output voltage frequency 50 Hz Modulation index 0.8 Resistive-inductive load (r-l) 10 Ω – 50 mH Booster circuit(RLC) 20 Ω – 10 uH-101.32 uF Fig. 3. Transient and steady state of internal flying capacitors voltages of a 6cell-7-level FCMC acquired from state-space numerical solution and simulation for resistive-inductive load in parallel with RLC booster circuit. Table-4: Parameters used in simulation and numerical solution of state-space representation of 4-cell-5-level FCMC. System Parameters Values DC voltage (E) 600 V Internal flying capacitors (C) 2200 uF PS-SPWM carrier frequency (fSW) 5 kHz Fundamental output voltage frequency 50 Hz Modulation index 0.8 Resistive-inductive load (r-l) 10 Ω – 50 mH Booster circuit(RLC) 10 Ω – 10 uH-101.32 uF Page 741 Vahid and Shoulaie, Capacitors Natural Voltage Balancing Mechanism Investigation in Flying Capacitor Multicell Converters International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 738-744 ISSN 2078-2365 Numerical solutions are in accordance with simulation results and admit the model of the converter. It is worthwhile noting that the value of resistor in the balance booster circuit plays an important role in transient state of flying capacitors voltages and its reduction accelerates the self balancing process in flying capacitors and vice versa. IV. Fig. 4. Transient and steady state of internal flying capacitors voltages of a 4cell-5-level FCMC acquired from state-space numerical solution and simulation for resistive-inductive load in parallel with RLC booster circuit. Table-5: Parameters used in simulation and numerical solution of state-space representation of 4-cell-5-level FCMC. System Parameters Values DC voltage (E) 600 V Internal flying capacitors (C) 2200 uF PS-SPWM carrier frequency (fSW) 5 kHz Fundamental output voltage frequency 50 Hz Modulation index 0.9 Resistive-inductive load (r-l) 10 Ω – 50 mH Booster circuit(RLC) 10 Ω – 10 uH-101.32 uF EXPERIMENTAL RESULTS To verify the mathematical model of FCMC, the measured transient and steady state voltages of flying capacitors of a 3cell-4-level converter acquired from the prototype system, illustrated in Fig. 7 and Fig. 9, are compared against simulation results and numerical solution of the converter time-domain differential equations, presented in Fig. 8 and Fig.10. Also main parameters of the converter are given in Tables 6 and 7. The match between simulation, numerical solution and experimental results confirms the modified mathematical model of FCMC. Table-6: Parameters used in simulation, numerical solution, and experimental converter. System Parameters Values DC voltage (E) 75 V Internal flying capacitors (C) 560 uF PS-SPWM carrier frequency (fSW) 5 kHz Fundamental output voltage frequency 50 Hz Modulation index 0.8 Resistive-inductive load (RL- LL) 25 Ω – 30 mH Booster circuit 24 Ω – 10 uH- 101.32 uF Fig. 5. Transient and steady state of internal flying capacitors voltages of a 4-cell-5-level FCMC acquired from state-space numerical solution and simulation for resistive-inductive load in parallel with RLC booster circuit. Fig. 7. Transient and steady state of internal flying capacitors voltages of a 3-cell-4-level FCMC acquired from state-space numerical solution and simulation for resistive-inductive load in parallel with RLC booster circuit. Fig. 6. Transient and steady state of internal flying capacitors voltages of a 4-cell-5-level FCMC acquired from state-space numerical solution and simulation for resistive-inductive load in parallel with RLC booster circuit for two different modulation indexes. Page 742 Vahid and Shoulaie, Capacitors Natural Voltage Balancing Mechanism Investigation in Flying Capacitor Multicell Converters International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 738-744 ISSN 2078-2365 circuit. Fig. 8. Transient and steady state of internal flying capacitors voltages of a 3-cell-4-level FCMC acquired from experimental prototype for resistive-inductive load in parallel with RLC booster circuit. Table-7: Parameters used in simulation, numerical solution, and experimental converter. System Parameters Values DC voltage (E) 75 V Internal flying capacitors (C) 560 uF PS-SPWM carrier frequency (fSW) 5 kHz Fundamental output voltage frequency 50 Hz Modulation index 0.8 Resistive-inductive load (RL- LL) 25 Ω – 30 mH Booster circuit 24 Ω – 10 uH- 101.32 uF Fig. 10. Transient and steady state of internal flying capacitors voltages of a 3-cell-4-level FCMC acquired from experimental prototype for resistive-inductive load in parallel with RLC booster circuit. V. CONCLUSION Multicell converters are very interesting for highpower/medium-voltage applications, for considerably improvement of the output voltage frequency spectrum and reduction of the conduction losses, switching ripple and value of dV/dt. This paper presents a modified mathematical model of flying capacitor multicell converters. In the proposed model the effect of balance booster circuit which is usually connected in parallel with load to accelerate the self balancing process of flying capacitors, is also considered. Numerical solutions, simulation and experimental results are in accordance with each other and confirm the validity of proposed model for FCMCs. REFERENCES [1] Fig. 9. Transient and steady state of internal flying capacitors voltages of a 3-cell-4-level FCMC acquired from state-space numerical solution and simulation for resistive-inductive load in parallel with RLC booster R. H. Baker and L. H. Bannister, “Electric power converter,” U.S. Patent 3 867 643, Feb. 1975. Page 743 Vahid and Shoulaie, Capacitors Natural Voltage Balancing Mechanism Investigation in Flying Capacitor Multicell Converters International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 738-744 ISSN 2078-2365 [2] [3] [4] [5] [6] [7] J. S. Lai and F. Z. Peng, “Multilevel converters—A new breed of power converters,” in Proc. IEEE Ind. Applicat. Soc. Annu. Meeting, 1995, pp. 2348–2356. J. Rodriguez, J. Lai, and F. Z. Peng, “Multilevel inverters: A survey of topologies, controls and applications,” IEEE Trans. Ind. Electron., vol. 49, no. 4, pp. 724–738, Aug. 2002. A. Nabae, I. Takahashi, and H. Akagi, “A new neutral-point clamped PWM inverter,” IEEE Trans. Ind. Applicat., vol. IA-17, pp. 518–523, Sept./Oct. 1981. T. A. Meynard and H. Foch, “Multi-level choppers for high voltage applications,” Eur. Power Electron. Drives J., vol. 2, no. 1, p. 41, Mar. 1992. H. S. Black, Modulation Theory. New York: Van Nostrand, 1953. D. G. Holmes and T. A. Lipo, Pulse Width Modulation for Power Converters: Principles and Practice. Piscataway, NJ: IEEE Press, 2003. X. Yuan, H. Stemmler, and I. Barbi, “Self-balancing of the clampingcapacitor- voltages in the multilevel capacitor-clampinginverter under sub-harmonic PWM modulation,” IEEE Trans. Power Electron., vol. 16, no. 2, pp. 256–263, Mar. 2001. [9] R.Wilkinson, H. du Mouton, and T. Meynard, “Natural balance of multicell converters : The two-cell case,” IEEE Trans. Power Electron., vol. 21, no. 6, pp. 1649–1657, Nov. 2006. [10] R.Wilkinson, H. du Mouton, and T. Meynard, “Natural balance of multicell converters: The general case,” IEEE Trans. Power Electron., vol. 21, no. 6, pp. 1658–1666, Nov. 2006. [11] B. P. McGrath and D. G. Holmes, “Analytical modelling of voltage balance dynamics for a flying capacitor multilevel converter,” in Conf. Rec. IEEE Power Electron. Specialists Conf. (PESC), 2007, pp. 968– 974. [8] Page 744 Vahid and Shoulaie, Capacitors Natural Voltage Balancing Mechanism Investigation in Flying Capacitor Multicell Converters 

International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 731-737 ISSN 2078-2365 Capacitors Voltage Balancing Modeling in Three Phase Flying Capacitor Converters with Booster Vahid Dargahi and Abbas Shoulaie Department of Electrical Engineering, Iran University of Science and Technology, Tehran 16846, Iran vdargahi@elec.iust.ac.ir, shoulaie@iust.ac.ir Abstract— This paper provides a mathematical model for voltage natural balancing process in three phase capacitor-clamped multicell converters. The analysis leads to state-space model of the converters. State-space representation of converter can be utilized to investigate the start-up and steady states of internal flying capacitors voltages. To provide verification, numerical solutions for three phase capacitor-clamped multicell converter’s analytic model are presented. Keywords Three Phase Flying Capacitor Multicell Converter, Self-balancing , Voltage Balance Booster. I. INTRODUCTION As a consequence of reaching higher power and lack of its suitable ranged switches, multilevel converters were introduced in 1975 and have been continuously developed in recent years due to the necessity of increase in power level of industrial applications especially high power applications such as high power AC motor drives, active power filters, reactive power compensation and FACTS devices. The main reason is the capability of these topologies to handle voltage/power in the range of kilovolts/megawatts as a result of serial connection of power switches in these converters [1]. The concept of multilevel arises from acquiring a staircase output voltage waveform as voltage levels from input dc voltages by means of converter appropriate configuration and its proper switching pattern [2]-[3]. In comparison with the conventional two-level converters, multilevel ones excel at producing an output voltage comprising several steps with considerable enhancements to power quality, harmonic content, and efficiency. Multilevel converters have other appreciable advantages such as lower switching losses, lower voltage ratings of used semiconductor switches, reduction of output dv/dt stress and filter inductance, etc. [4]-[6]. The term multilevel starts with the three-level converter introduced by Nabae et al. The Neutral Point Clamped (NPC) converter, presented in the early 80’s, is a standard topology in industry on its 3-level version. However, for a higher number of levels, this topology has some drawbacks such as: voltage imbalance issue of the dc-link capacitors and the excessive use of clamping diodes [7]. Cascade multicell (CM) converters use a series connection of H-bridge modules. Dc link voltage of each H-bridge module must be an isolated one. Modularity, ease of extension the number of output voltage levels via adding new modules, reliability, and fault tolerant feature are the most notable advantages of these topologies [8]-[9]. Flying capacitor (FC) based converters use ladder connection of units called as ‘cells’. Each cell is composed of one flying capacitor and two complimentary power switches. Redundant switching states in flying capacitor based converters can be implemented to stabilize the voltage across flying capacitors at their requisite values. The difference between voltages across two adjacent flying capacitors determines the step value of staircase-form output voltage [10]-[12]. The FCM converter, and its derivative, the SM converter, have many advantageous properties for medium voltage applications, particularly the transformer-less operation and the ability to naturally maintain the clamping capacitors voltages at their target operating levels. This substantial property is called natural balancing and allows the construction of such converters with a large number of voltage levels. Natural selfbalancing of the flying capacitors voltages occurs without any feedback control. A necessary condition for this phenomenon is that average current of the clamping capacitors must be zero. As a result, each cell must be controlled with the same duty cycle and a regular phase shifted progression along the cells. Generally, an output RLC filter (balance booster circuit), tuned to the switching frequency or multiple of that, is suggested to be connected across the load in order to accelerate this selfbalancing process in the transient states. The FCM converter uses a series connection of “cells” comprising a flying capacitor and its associated complimentary switch pair and produces a switched voltage that is the sum of the individual cell states [13]. Despite of mentioned appreciable advantages, multilevel converters possess some following main drawbacks: increased number of isolated dc voltage sources, clamping diodes, capacitors and of power semiconductor switches accompanied Page 731 Vahid and Shoulaie, Capacitors Voltage Balancing Modeling in Three Phase Flying Capacitor Converters with Booster International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 731-737 ISSN 2078-2365 by their related gating and protection circuits which result in a sophisticated overall system [14]. As mentioned, voltage natural balancing mechanism is a fundamental principle in flying capacitor and stacked multicell converters which consents to construction of voltage levels at the converter output [11]-[21]. The main objective of this paper is to provide a mathematical model for three phase flying capacitor multicell converters with intention for investigation of start-up and steady state of clamping capacitors voltages by taking into account the effect of balance booster circuit. 0.5E SR 0.5E SR S3 + EC( R-1)= ( R-1)E/ R S3 S2 + EC2= 2E/R S2 S1 + EC1 = E/R S1 iout( t ) + l Vout Cell- R Cell-2 r Cell-1 _ Fig. 1. R-cell flying capacitor multicell converter with maximum output voltage value of E. II. INSTANTANEOUS MODELING OF THREE PHASE FLYING CAPACITOR MULTICELL CONVERTERS IN STATE-SPACE REPRESENTATION A. Fundamental Concepts of Flying Capacitors Multicell Converters Flying capacitor multicell converters (FCMCs) which have been proposed by T.A. Meynard are relatively new breed of multilevel converters in comparison with conventional neutral point clamped (NPC) and cascade H-bridge (CHB) ones. A typical single phase configuration of FCMC is depicted in Fig 1. As illustrated, R cells in a FCMC are overlapped to form a required converter’s leg. Each cell consists of one voltage source (a dc voltage source equal to E in Rth cell and capacitors possessing specific voltages in remaining cells) and two power semiconductor switches which are in complementary state to each other to avoid short-circuiting of voltage sources. Phase Shifted Carrier Sinusoidal Pulse Width Modulation (PSCSPWM) technique is the most common control scheme which is applied to switching strategy of FCMCs to guaranty both best harmonic performance and voltage balancing mechanism in clamping capacitors. It should be noted that in a R-cell FCMC each switch sustains just a fraction of DC link voltage, i.e. E/R. This R-cell configuration leads to R+1 levels of voltage with peak to peak value of E at the converter output. FCMCs are in preference to the NPC and CHB ones as considering appreciable advantages such as: modularity, noninterdependency of cells as fault occurs and ease of reaching higher voltage levels just by introducing new cells [1]-[10]. Control strategy, switches states and output voltage of a 4cell-5-level FCMC are illustrated in Fig. 2 and Table-1, respectively. Fig. 2. Switching pattern of a typical 4-cell 5-level FCMC, switches states and output voltage. Table-1: States of switches in a 4-cell-5-level conventional FCMC. Output Number State of Switches {(S4, S3, S2, S1)} Voltage of States Level +0.5E {(1,1,1,1)} 1 + 0.25E {(1,1,1,0)(1,1,0,1)(1,0,1,1) (0,1,1,1)} 4 0 {(1,1,0,0)(1,0,0,1)(0,0,1,1)(1,0,1,0)(0,0,1,1)(0,1,0,1)} 6 -0.25E {(1,0,0,0)(0,1,0,0)(0,0,1,0)(0,0,0,1)} 4 -0.5E {(0,0,0,0)} 1 The output voltage of a R-cell FCM converter has R+1levels and its frequency spectrum has the harmonics around the  R  k  fSW  th harmonic where k and fSW are the integer number and the switching frequency, respectively [11]-[14]. A typical three phase clamping capacitor multicell converter.is shown in Fig 3. Page 732 Vahid and Shoulaie, Capacitors Voltage Balancing Modeling in Three Phase Flying Capacitor Converters with Booster International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 731-737 ISSN 2078-2365 0.5E SR,a O 0.5E + EC(R-1),a= (R-1)E/R - SR,a S3,a S3,a Cell-R,a SR,b + EC(R-1),b= (R-1)E/R - SR,b S3,b S3,b Cell-R,b SR,c + EC(R-1),c= (R-1)E/R - SR,c S3,c S3,c S2,a + E C2,a = 2E/R S2,a S1,a + E C1,a = E /R S1,a Cell-2,a Cell-1,a S2,b + E C2,b= 2E/R S2,b S1,b + E C1,b= E /R S1,b Cell-2,b Cell-1,b S2,c + E C2,c= 2E/R S2,c S1,c + E C1,c = E /R S1,c ia (t) r l + Va (t) _ ib (t) r l + Vb(t) _ ic (t) r l + Vc (t) _ balancing property. In the derived model, capacitors voltages are state variables. Utilization a numerical solution for proposed model differential equations leads to acquire transient and steady state response of flying capacitors voltages. According to Fig. 3, switching function of cell  in each phase of the converter is defined as follows [15]-[21]: n   1 H  , x t      1 if S , x is on if S , x is off   1: R, x  a , b, c (2) And its Double Fourier Series expansion can be expressed as follows [15]: H  , x (t )  M cos     4        Fig. 3. Three phase R-cell flying capacitor multicell converter with maximum output sin   m  n   J n  m M  cos  m  n    2  2  voltage value of E in each phase.  m 1 n   m  Cell-2,c Cell-R,c Cell-1,c  B. Instantaneous Model of the Three Phase Converter in the State-Space Representation By utilizing proper switching pattern in FCMCs, their capacitor voltages would reach to the specific values which allow to constructing the desired output voltage levels. This property is known as voltage natural balancing mechanism and is achieved using PSC-SPWM switching technique in these converters [11]-[21]. Natural self-balancing process of the clamping capacitors voltages, as one of the advantages of FCMCs occurs without any feedback control. A necessary condition for this phenomenon is that average current of the flying capacitors must be zero. As a result, each cell must be controlled with the same duty cycle and a regular phase shifted progression along the cells. Generally, an output RLC filter (balance booster circuit), tuned to the switching frequency or multiple of that, is suggested to be connected across the load in order to accelerate this self-balancing process in the start-up states. In this case, the dynamic of the self-balancing process depends on the impedance of balance booster circuit at the switching frequency. If the impedance at the switching frequency is high then the natural balancing is very slow and vice versa. The output RLC filter should be tuned to the switching frequency as follow [1]: Lb  Cb  1 2    fSW (1) Where, fSW is the switching frequency, Lb and Cb are inductance and capacitance of the output RLC booster circuit, respectively. In this section a mathematical model for three phase capacitor-clamped converter will be presented to verify natural    c t   c     1 2 ,   t     r r x  0 xa   2 x   xb  3  4 xc   3 R (3) where M , r , r , J n . , c , c are modulation index, angular frequency of reference sinusoidal waveform, phase angle of reference sinusoidal waveform, nth order Bessel function of first kind, triangular carrier waveform angular frequency and triangular carrier waveform phase angle, respectively. According to Fig. 3, output voltage and current of the converter based on mentioned switching functions can be obtained as follows: Vao  t   Van  t   Vno  t  Vbo  t   Vbn  t   Vno  t  Vco  t   Vcn  t   Vno  t  Vao  t   Vbo  t   Vco  t   (4) Van  t   Vbn  t   Vcn  t   3Vno  t  0 Van  t    2 1 1 Vao  t     1   Vbn  t     1 2 1 Vbo  t     3  1 1 2      Vco  t   Vcn  t   (5)  Page 733 Vahid and Shoulaie, Capacitors Voltage Balancing Modeling in Three Phase Flying Capacitor Converters with Booster International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 731-737 ISSN 2078-2365 ia  t   T    Vao  t  Vbo  t  Vco  t   i t   . , ,  b     ZL ZL     ZL ic  t   Vxo  t   ix  t   C , x C , a (6)  E 1 R1 H R, x (t )    H  , x  t   H  1, x  t   Ec , x  t  2 2  1 Vxn  t  ZL dEc , x  t  dt dEc  , a  t     1  H  1, x t   H  , x t  ix t  2 (8) 1  H  1,a  t   H  ,a  t    2 ZL   2 E 2 1 R1   H R, a (t )     H j , a  t   H j 1, a  t   Ecj , a  t  3 2 j 1 3 2  1 E 1 1 R1   H R,b (t )     H j ,b  t   H j 1,b  t   Ecj ,b  t  3 2 j 1  3 2  1 E 1 1 R1   H R,c (t )     H j ,c  t   H j 1,c  t   Ecj ,c  t  3 2 j 1  3 2 C , a dt dEc  , a  t  R, a  t  H dt (7)              (9)  H  1, a t   H  ,a t   E  R,c  t  H  1, a t   H  ,a t    E   R 1 1 H  1, a  t   H  , a  t      j ,c  t    j 1,c  t   Ecj , c  t   12 j 1 12 x  y :  1   H i 1, x  t   H i, x  t    j , x t    j 1, x t   6Ci , x Aij , xy  t    x  y :  1   H i 1, x  t   H i, x  t    j , y t    j 1, y t  12Ci , x   x, y  a , b, c i, j  1: R  1  , x  t     (11)  Eci , a  t    Bi1, a  t        Eci ,b  t     Bi1,b  t   E      Eci , c  t    Bi1, c  t   and: R 1 1 H  1, a  t   H  , a  t      j ,b  t    j 1,b  t   Ecj ,b  t    12 j 1 12 Aij , ac  t    Aij ,bc  t    Aij , cc  t      6 R 1 1 H  1, a  t   H  , a  t      j , a  t    j 1, a  t   Ecj , a  t    6 j 1 R,b  t  Aij , ab  t  Aij ,bb  t  Aij , cb  t   2  R, x  t     1 Bi1, x  t    H i 1, x t   H i, x t     R, y  t   12Ci , x  y|   x y  i, j  1: R  1 x, y  a , b, c   1, a  t   H  , a  t    E   dEci , a  t      dt   A  t   dEci ,b  t    ij , aa     Aij ,ba  t   dt    dE  t    Aij , ca  t   ci , c   dt  (10)  Where, ZL is a series connection of resistance  r  and inductance  l  and C is capacitance of flying capacitors. Three phase clamping capacitor multicell converter’s statespace representation can be written as follows: H  ,x t  ZL  (13) M cos   tan 1  0,1  r  0,1       4  sin   m  n   J n  m M      2   2  m r  m, n    m1 n    1  cos  m  n  tan  m, n      (12)   m, n   m, n  1 2   m, n 2  (14)  mc  nr  l r To consider the balance booster circuit effect on the dynamic of the self-balancing process following modification should be applied to the acquired state-space equations: Page 734 Vahid and Shoulaie, Capacitors Voltage Balancing Modeling in Three Phase Flying Capacitor Converters with Booster International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 731-737 ISSN 2078-2365 x  y :    j , x  t    j , x  t     1  H i 1, x  t   H i , x  t       6Ci , x t    j 1, x  t     j 1, x    Aij , xy  t     x y :      t    j , y  t     1  j, y  12C H i 1, x  t   H i , x  t    j 1, y  t    j 1, y  t   i, x                   2 R, x  t   R , x  t      1 H i 1, x  t   H i , x  t   R, y  t   R , y  t   12Ci , x   y|   x y  i, j  1: R  1 x, y  a , b, c  Bi1, x  t    , x  t    H  ,x t  Zb    M cos   tan 1  0,1 Rb 0,1   m, n   m, n  1    m, n 2    mc  nr  Lb    mc  nr  Cb  2  (16) (17) 1 Rb III. Fig. 4. Transient and steady state of internal flying capacitors voltages of a 2cell-3-level three phase FCMC acquired from state-space numerical solution for resistive-inductive load in parallel with RLC booster circuit.   4       sin   m  n   J n  m M    2   2       m Rb  m, n m1 n    1  cos m  n  tan  m, n    (15) NUMERICAL SOLUTION RESULTS To provide verification to the elaborated state-space representation of the three phase FCMCs, numerical solution is utilized to solve differential equations of 2-cell-3-level and 3cell-4-level three phase converters. Transient and steady state of internal flying capacitors voltages of mentioned converters acquired from state-space numerical solution are shown in Figs. 4-9. System parameters used in numerical solution are given in Tables 2-4. Fig. 5. Start-up state of internal flying capacitors voltages of a 2-cell-3-level three phase FCMC acquired from state-space numerical solution for resistiveinductive load in parallel with RLC booster circuit. Table-3: Parameters used in numerical solution of state-space representation of 2-cell-3-level three phase FCMC. System Parameters Values DC voltage (E) 600 V Internal flying capacitors (C) 560 uF PSCSPWM carrier frequency (fSW) 5 kHz Fundamental output voltage frequency 50 Hz Modulation index 0.8 Resistive-inductive load (r-l) 10 Ω – 80 mH Booster circuit(RLC) 200 Ω – 6.97 uH-144 uF Table-2: Parameters used in numerical solution of state-space representation of 2-cell-3-level three phase FCMC. System Parameters Values DC voltage (E) 600 V Internal flying capacitors (C) 560 uF PSCSPWM carrier frequency (fSW) 5 kHz Fundamental output voltage frequency 50 Hz Modulation index 0.8 Resistive-inductive load (r-l) 10 Ω – 80 mH Booster circuit(RLC) 10 Ω – 6.97 uH-144 uF Page 735 Vahid and Shoulaie, Capacitors Voltage Balancing Modeling in Three Phase Flying Capacitor Converters with Booster International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 731-737 ISSN 2078-2365 Fig. 6. Transient and steady state of internal flying capacitors voltages of a 2cell-3-level three phase FCMC acquired from state-space numerical solution for resistive-inductive load in parallel with RLC booster circuit. Fig. 8. Transient and steady state of internal flying capacitors voltages of a 3cell-4-level three phase FCMC acquired from state-space numerical solution for resistive-inductive load in parallel with RLC booster circuit. Fig. 7. Start-up state of internal flying capacitors voltages of a 2-cell-3-level three phase FCMC acquired from state-space numerical solution for resistiveinductive load in parallel with RLC booster circuit. Fig. 9. Start-up state of internal flying capacitors voltages of a 3-cell-4-level three phase FCMC acquired from state-space numerical solution for resistiveinductive load in parallel with RLC booster circuit. Table-4: Parameters used in numerical solution of state-space representation of 3-cell-4-level three phase FCMC. System Parameters Values DC voltage (E) 600 V Internal flying capacitors (C) 560 uF PSCSPWM carrier frequency (fSW) 5 kHz Fundamental output voltage frequency 50 Hz Modulation index 0.8 Resistive-inductive load (r-l) 10 Ω – 50 mH Booster circuit(RLC) 20 Ω – 6.97 uH-144 uF Numerical solutions predict the start-up and steady state of clamping capacitors voltages precisely and admit the proposed model for three phase clamping capacitor multicell converters. It is worth noting that the value of resistor in the balance booster circuit plays an important role in start-up state of flying capacitors voltages and its reduction accelerates the selfbalancing process and vice versa. IV. CONCLUSION Clamping capacitor multicell converters are very interesting for high-power/medium-voltage applications, for considerably improvement of the output voltage frequency spectrum, voltage natural balancing of clamping capacitors and fault tolerance. This paper presents a mathematical model for three phase clamping capacitor multicell converters. In the proposed model the effect of balance booster circuit which is usually connected in parallel with load to accelerate the self-balancing process of flying capacitors, is considered which has not been reported in Page 736 Vahid and Shoulaie, Capacitors Voltage Balancing Modeling in Three Phase Flying Capacitor Converters with Booster International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 731-737 ISSN 2078-2365 literature. Numerical solutions confirm the validity of proposed model for three phase flying capacitor multicell converters. REFERENCES A. K. Sadigh, V. Dargahi, and A. Shoulaie, “Elimination of one DC voltage source in stacked multicell converters,” IET. Power. Electron, in press. [2] J. S. Lai and F. Z. Peng, “Multilevel converters—A new breed of power converters,” in Proc. IEEE Ind. Applicat. Soc. Annu. Meeting, 1995, pp. 2348–2356. [3] J. Rodriguez, J. Lai, and F. Z. Peng, “Multilevel inverters: A survey of topologies, controls and applications,” IEEE Trans. Ind. Electron., vol. 49, no. 4, pp. 724–738, Aug. 2002. [4] J. Rodriguez, S. Bernet, B. Wu, J. Pontt, and S. Kouro, “Multi-level voltage-source-converter topologies for industrial medium-voltage drives,” IEEE Trans. Ind. Electron., vol. 54, no. 6, pp. 2930–2945, Dec. 2007. [5] V. Dargahi, A. Shoulaie, “Capacitors natural voltage balancing mechanism investigation in flying capacitor multicell converters,” in Proc. 19th Iranian Conference on Electrical Engineering (ICEE), pp. 1– 6, Tehran, 2011. [6] A. Nabae, I. Takahashi, and H. Akagi, “A new neutral-point clamped PWM inverter,” IEEE Trans. Ind. Applicat., vol. IA-17, pp. 518–523, Sept./Oct. 1981. [7] S. Kouro, M. Malinowski, K. Gopakumar, J. Pou, L. G. Franquelo, B.Wu, J. Rodriguez, M. A. Perez, and J. I. Leon, “Recent advances and industrial applications of multilevel converters,” IEEE Trans. Ind. Electron., vol. 57, no. 8, pp. 2553–2580, Aug. 2010. [8] T. A. Meynard and H. Foch, “Multi-level choppers for high voltage applications,” Eur. Power Electron. Drives J., vol. 2, no. 1, p. 41, Mar. 1992. [9] L. G. Franquelo, J. Rodríguez, J. I. Leon, S. Kouro, R. Portillo, and M. A. M. Prats, “The age of multilevel converters arrives,” IEEE Ind. Electron. Mag., vol. 2, no. 2, pp. 28–39, Jun. 2008. [10] A. Shukla, A. Ghosh, and A. Joshi, “Natural balancing of flying capacitor voltages in multicell inverter under PD carrier-based pwm,” IEEE Trans. Power Electron., vol. 26, no. 6, pp. 1682– 1693, Jun. 2011. [1] [11] M. Salehifar, V. Dargahi, and A. Shoulaie, “A series active power filter using a flying capacitor multicell inverter,” in Proc. 1st Power Electronic & Drive Systems & Technologies Conference (PEDSTC), pp. 191–195, Tehran, 2010. [12] V. Dargahi, M. Salehifar, M. Abarzadeh, and A. Shoulaie, “Grid interaction of DG units with a modified mixed cascade flying capacitor multicell inverter,” in Proc. 2nd Power Electronic & Drive Systems & Technologies Conference (PEDSTC), pp. 1–6, Tehran, 2011. [13] V. Dargahi, M. Abarzadeh, M. Salehifar, and A. Shoulaie, “Voltage compensation by a modified mixed cascade flying capacitor multicell inverter,” in Proc. 1st Power Quality Conference (PQC), pp. 1–5, Tehran, 2010. [14] V. Dargahi, M. J. Zandzadeh, M. Salehifar, and A. Shoulaie, “Utilization a flying capacitor multicell converter based adaptive shunt active power filter to enhance power quality,” in Proc. 25th International power System Conference (PSC), pp. 1–9, Tehran, 2010. [15] D. G. Holmes and T. A. Lipo, Pulse Width Modulation for Power Converters: Principles and Practice. Piscataway, NJ: IEEE Press, 2003. [16] X. Yuan, H. Stemmler, and I. Barbi, “Self-balancing of the clampingcapacitor- voltages in the multilevel capacitor-clampinginverter under sub-harmonic PWM modulation,” IEEE Trans. Power Electron., vol. 16, no. 2, pp. 256–263, Mar. 2001. [17] R.Wilkinson, H. du Mouton, and T. Meynard, “Natural balance of multicell converters : The two-cell case,” IEEE Trans. Power Electron., vol. 21, no. 6, pp. 1649–1657, Nov. 2006. [18] R.Wilkinson, H. du Mouton, and T. Meynard, “Natural balance of multicell converters: The general case,” IEEE Trans. Power Electron., vol. 21, no. 6, pp. 1658–1666, Nov. 2006. [19] B. P. McGrath and D. G. Holmes, “Analytical modelling of voltage balance dynamics for a flying capacitor multilevel converter,” IEEE Trans. Power Electron., vol. 23, no. 2, pp. 543–550, Mar. 2008. [20] B. P. McGrath and D. G. Holmes, “Enhanced voltage balancing of a flying capacitor multilevel converter using phase disposition (PD) modulation,” IEEE Trans. Power Electron., vol. 26, no. 7, pp. 1933– 1942, Jul. 2011. [21] B. McGrath and D. Holmes, “Analytical determination of the capacitor voltage balancing dynamics for three phase flying capacitor converters,’’ in Conf. Rec. 42nd IEEE IAS Annu. Meeting, Sep. 23–27, 2007, pp. 1974–1981. Page 737 Vahid and Shoulaie, Capacitors Voltage Balancing Modeling in Three Phase Flying Capacitor Converters with Booster 

International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 757-763 ISSN 2078-2365 Estimation of Re-striking Transient Over voltages in a 132KV Gas insulated Substation M. Kondalu1, P.S. Subramanyam2 Electrical & Electronics Engineering, JNT University. Hyderabad. 1 Kondalu_m@yahoo.com 2 subramanyamps@gmail.com Joginpally B.R. Engineering College, Moinabad, Hyderabad Abstract: --This paper presents the most significant results of a Very fast transient Overvoltages generated due to switching operations have been analyzed and presented. Since the contact speed of dis-connector switches is low, re-striking occurs many times before the current interruption is completed .Each re-striking generates transient overvoltage with level of magnitude. These transient have travelling wave behaviour, they travel to the external systems through enclosures, bushings, cable joints etc.. and cause damage to the outside equipment. They can lead secondary break downs in GIS and may give rise to electromagnetic interference. The Earth faults give rise to TEV which can interfere with the operation and control of secondary equipment in a 3-phase 132kv GIS. Thus it is important to develop a suitable MATLAB7.8 models for estimation of these overvoltages. Keywords—Gas Insulated Substation (GIS), very fast Transient overvoltages, 3phase faults, MATLAB 7.8 software and Control circuitry I. INTRODUCTION For accurate analysis of transients, it is essential to find the VFTO’s and circuit parameters. Due to the travelling nature of the transients the modelling of GIS makes use of electrical equivalent circuits composed by lumped elements and especially by distributed parameter lines, surge impedances and travelling times. The simulation depends on the quality of the model of each individual GIS component. In order to achieve reasonable results in GIS structures highly accurate models for each internal equipment and also for components connected to the GIS are necessary. The dis-connector spark itself has to be taken into account by transient resistance according to the Toepler’s equation and subsequent arc resistance of a few ohms. The wave shape of the over voltage surge due to dis-connector switch is affected by all GIS elements. Accordingly, the simulation of transients in GIS assumes an establishment of the models for the Bus, Bushing, Elbow, Transformers, Surge Arresters, Breakers, Spacers, dis-connectors, and Enclosures and so on. A GIS system comprising of an Input Cable, Spacer, Dis-connector Switch, Bus bar of 10mts length and load has been considered for modelling into electrical network and analysis. The Fast Transient Over voltage waveform generated during Closing and Opening operation of Dis-connector Switch and 3-phase faults has been considered for calculations. Spacers are simulated by lumped Capacitance. The Inductance of the bus duct is calculated from the diameters of Conductor and Enclosure. Capacitances are calculated on the basis of actual diameters of inner and outer cylinders of central conductor and outer enclosure. Cone Insulators used for supporting inner conductor against outer enclosure are assumed to be disk type for approximate calculation of spacer capacitance. The busduct can be modelled as a series of Pi-network or as sequence parameters. However in this model, it is considered as distributed Pi-network. The Schematic Diagram of a Typical Gas Insulated System (GIS) is shown in below figure 3. Frequency nature, the VFTO imposed on the transformers connected directly to the GIS would not be distributed evenly on all transformer windings. Some windings, e.g. the first few turns connecting to the 132kV GIS, would be subject to a higher magnitude of overvoltage, posing a potential risk of insulation breakdown of the transformers[9][10]. Page 757 Kondalu and Subramanyam, Estimation of Re-striking Transient Overvoltages in a 132KV Gas insulated Substation International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 757-763 ISSN 2078-2365 II. MODELLING OF 132KV GAS INSULATED SUBSTATION During the current operation of dis-connector switch in a GIS, re-strikes(pre-strikes) occur because of low speed of the dis-connector switch moving contact, hence Very fast Transient Over voltage are developed. These VFTO’s are caused by switching operations and 3-phase fault When a dis-connector switch is opened on a floating section of switchgear, trapped charge may be left on the floating section. In the opening operation of dis-connector switch, transients are produced and the magnitude of these transients and rise times depends on the circuits parameters. When there is a fault occurs, there is a short circuit in the system. Transients are also produced due to the faults in the system. Due to this VFTO’s are caused by switching operation can also lead to secondary breakdown with in GIS. Re-striking surges generated by the dis-connector switches at GIS generally possess extremely high frequencies ranging from several hundred KHz to several MHz . For the development of equivalent circuit low voltage step response measurements of the main GIS components have been made. Using MATLAB 7.8 of the equivalent models is developed. During opening of Dis- connector switch (DS), transients are produced due to internal oscillations. The magnitude of transients and rise times depends on the circuit parameters like inductance, Capacitance and connected Load. Assuming that some trapped charge is left is left during operating operation; transients can be calculated during closing operation of DS. Fast transient over voltages generated during Dis-connector Switch operation are a sequence of voltage of voltage steps created by voltage collapse across the gas at re-striking specific over voltage shape is formed by multiple reflections and refractions. Operation of Dis-connector Switch (DS) can be shown by using the fig 1 Where L1 = Inductance of Source C1 = Capacitance of Source C2 = Capacitance of GIS Open part U1 = Power Frequency Voltage U2 = Power GIS Voltage The more frequent service situation of the isolator is its use to connect or dis-connect unloaded parts of the installation as is shown in fig 1 for example apart of the of the GIS is disconnected by an isolator from an overhead supply line. Where by the self-capacitance C2 of this part of circuit can be upto several nF, depending on its length. First re-strike across the gap occurs when voltage across the gap exceeds the breakdown voltage. The occurrence of re-strikes is described with the following Fig 2 The voltage across the gap is the difference between U1 and U2, if it is assumed that the breakdown voltage UB of the gap increases with increasing separation and therefore with time as shown in fig 2.Then the curve U2 can be constructed as follows. At the instant of mechanical contact separation, U1 and U2 have the same value, the voltage U2 continues to retain this value, while U1 changes with power frequency, the voltage (U2-U1) Across the gap of the isolator also changes. As soon as, (U2-U1) exceeds the dielectric strength UB of the gap, a breakdown and thus first re-strike occurs. Both electrodes are there by electrically connected by conducting spark, whereby GIS section with initial voltage U2 is very rapidly charged to instantaneous value of U1. The transient current flowing through the spark then interrupts as soon as the GIS have been charged to U1 and spark extinguishes. The voltage U2 now remains constant with time, while the voltage U1.on the side of supply keeps changing. This continues until the second re- strike occurs with an increased breakdown voltage UB as a consequence of larger separation. Hence U2 follows U1, until finally at the end of the switching process the gap no longer can be broken down. Transients are also produced due to faults in the system. When there is a fault, there will be short circuit in the system. Due to this, oscillations occur due to presence of inductance and capacitance on both sides of the fault section causing transients. Dis-connector Switch (DS) operation typically involves slow moving contacts which results in numerous discharges during operation .For example, a floating section of switchgear between a disconnect switch and an open breaker (load side may be disconnected from an energized Gas insulated system (supply side). For capacitive currents below—1 amp, are-strike occurs every time the voltage between the connects exceeds the dielectric strength of the gaseous medium between them. Each re-strike generates a spark, which equalizes the potentials between the switch contacts. Following spark extinction, the supply and load side potentials will deviate according to the AC supply voltage variation and the discharge characteristics of the load side respectively. Another spark will result when the voltage across the electrode gap dependent breakdown voltage UB and the potential difference of the load and supply side U. Page 758 Kondalu and Subramanyam, Estimation of Re-striking Transient Overvoltages in a 132KV Gas insulated Substation International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 757-763 ISSN 2078-2365 Each Dis-connector Switch (DS) operation generates a large number of ignitions between the moving contacts. The number of ignitions depends on the speed of the contacts. The largest and steepest surge voltages are generated only by those breakdowns at the largest contact gap. Therefore, only by those breakdowns (10-50) need be considered for dielectric purpose. A. Calculation of variable arc resistance The Variable arc resistance is calculated using the formula: ∫ Where, = Toepler’s constant = 0.005 volt.sec/mt for SF6 under uniform field conditions L = spark length in meters = Initial charge or charge at the instant of breakdown T = spark collapse time in sec. Fig. 1 Electric Circuit for explaining re-strikes Fig 2 Voltage of the open –ended GIS side of the Isolator Page 759 Kondalu and Subramanyam, Estimation of Re-striking Transient Overvoltages in a 132KV Gas insulated Substation International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 757-763 ISSN 2078-2365 Fig. 3 Schematic diagram of a typical Gas Insulated Substation Fig. 4. 132kv GIS MATLAB implemented circuit when CB in Re-Striking Condition The value of time varying spark resistance R(t) is calculated three post type spacers and a 132kv gas bushing containing until it reaches a value of 1 to 3 ohms. The integral in the stress capacitor. denominator sums up the absolute value of current ‘i’ through . the resistance R(t) over the time beginning at breakdown inception. Thus, it corresponds to the charge conducted through the spark channel up to time‘t’. III. 3-PHASE EQUIVALENT CIRCUIT FOR 132KV The busduct can be modelled as a series of Pi-network GIS SYSTEM FOR 10MTS LENGTH or as sequence parameters. However in this model, it is considered as distributed Pi-network. The Schematic Diagram The bus duct is dividing into three sections of length of a Typical Gas Insulated System (GIS) is shown in below 1mts, 4mts, 5mts from load side. The GIS bushing is figure 3. represented by a capacitance of 125pf. The resistance of 1 ohm Assuming that some trapped charge is left on the floating spark channel is connected in series with circuit breaker. section of switchgear during opening operation of disMATLAB Circuit for 10 mts. length in a 3-phase 132kv GIS connector switch, a voltage of certain value is considered shown in the fig. 4. during MATLAB Due to trapped charge some voltage remains on the floating The apparatus as disconnected with an earthing switch, three section which can create severe conditions because the first redisc type Spencer’s , a load bus bar above to 10mts long width strike can occur at the peak of power frequency voltage giving a voltage of 2.0 p.u. On re-strike the voltage on each side will Page 760 Kondalu and Subramanyam, Estimation of Re-striking Transient Overvoltages in a 132KV Gas insulated Substation International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 757-763 ISSN 2078-2365 collapse initially zero and hence creating two 1.0pu voltage steps of opposite polarities. In this, it is assumed that restriking is created at 1.0 p.u. respectively on either side of disconnector switch (DS). The transients due to closing of the circuit breaker are calculated and maximum voltage obtained with a rise time. This method implemented on MATLAB 7.8. the voltage before and after circuit breaker is taken to be 1.0 pu and -1.0pu as the most enormous condition but depending on the time of closing of circuit breaker the magnitude of the voltage on the load side changes. For different values of voltages on the load side the magnitudes and rise time of the voltage wave are calculated keeping source side voltages as constant as 1.0p.u the values are tabulated in table I. Similarly by changing the magnitudes of the voltage on the source side, keeping voltage on load side constant at 1.0p.u. Then the transient due to variation of voltage on source side obtained. The values are tabulated in Table II. TABLE I 7 8 9 10 0.4 0.3 0.2 0.1 1.79 1.71 1.42 1.39 9 12 11 9 IV. RESULTS AND DISCUSSION The phenomenon that occurs during the DS closing into a capacitive load is very nearly the reserve of processes that occur during its opening. Here, the first restrike occurs due to the residual voltage left behind by a previous opening on the load side. Circuit breaker or load break switch closing or openings also generate VFTO in the case of re-strikes but the number of such VFTO is much lower than those generated by DS operation. The various transient voltage and current at different positions in a 3 phase 132kv GIS for the first switching operation presented in results. Assumed that there is a second re-strike another switch is connected in parallel to the circuit breaker for simulation in MATLAB modeling. Transients are calculated by closing this switch when voltage difference across the contacts of the circuit breaker reaches maximum value. TRANSIENT DUE TO VARIATION OF RE-STRIKE VOLTAGE ON LODE SIDE S.no 1 2 3 4 5 6 7 8 9 10 Load side Voltage (p.u) -1.0 -0.9 -0.8 -0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 Magnitude of the voltage (p.u) 2.45 2.39 2.19 2.15 2.03 1.96 1.82 1.77 1.53 1.45 Rise Time (Nanos) 10 13 12 10 12 11 10 13 12 9 During Re-strike operation (source & Load) the voltage through the resistance of the circuit breaker is shown in fig.5 and fig.6. From the graph it was found the maximum current is 25A at a rise time of 13ns. TABLE II TRANSIENTS DUE TO VARIATION OF RE-STRIKE VOLTAGE ON SOURCE SIDE S.no 1 2 3 4 5 6 Source side Voltage (p.u) 1.0 0.9 0.8 0.7 0.6 0.5 Magnitude of the voltage (p.u) 2.43 2.37 2.17 2.11 2.01 1.91 Rise Time (nanos) 9 10 11 10 9 10 Fig. 5 Transient voltage waveform during Re-strike for 10mts from source side in a 3-phase 132kv GIS Page 761 Kondalu and Subramanyam, Estimation of Re-striking Transient Overvoltages in a 132KV Gas insulated Substation International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 757-763 ISSN 2078-2365 Fig 6Transient voltage waveform during Re-strike for 10mts from Load side in a 3-phase 132kv GIS Fig.8 Current waveform during Re-strike for 10mts from Load side in a 3-phase 132kv GIS TABLE III THE ANALYSIS VALUES ARE TABULATED AS FOLLOWS: Mode of operation Magnitude of voltage(p.u) Rise time (Nano sec) During closing operation 2.46 69 During opening operation 1.22 56 During second restrike 2.45 112 Fig 7 Current waveform during Re-strike for 10mts from source side in a 3-phase 132kv GIS V. CONCLUSION To introduce the current chopping, the circuit breaker is opened remains. Hence to calculate transients due to opening operation the CB is opened at 12ns. The transients are obtained and show in fig.7. The transient calculated due to re-strike gives the peak voltage of 2.45p.u at a rise time of 112ns show in fig.8. A model is Developed for the prediction of the VFTO phenomena in the circuit of voltage and current transformers in GIS. The main advantage of such model is to enable the transient analysis of GIS.A spark collapse time was correctly simulated by the variable resistor. By this spark collapse time, resistance of the VFTO is extended, and the component caused by short surge impedance discontinuities such as spacers, disconnectors and short bus branches were damped. A GIS system comprising of spacers, bus bar and disconnectors has been considered for modeling into electric network. The inductance of the bus bar is calculated from Page 762 Kondalu and Subramanyam, Estimation of Re-striking Transient Overvoltages in a 132KV Gas insulated Substation International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 757-763 ISSN 2078-2365 diameters of conductors and enclosure using standard formulae. Cone insulators used for supporting inner conductor against outer enclosure are assumed to be disk type for approximate calculation of spacer capacitance is calculated using formulae for concentric cylinders. The entire bus length is modeled as distributed pi-network. The peak magnitude of fast transient currents generated during switching event changes from one position to another in a 132kv GIS for a particular switching operation. These transients over voltages are reduced by connecting suitable resistor in an equivalent circuit during closing and opening operation. VI. REFERENCES [1] Boggs SA., Chu F.Y. and Pujimotor N. (IYXZ), 'Disconnect Switch Induced Transients and Trapped Charge in GIs', EEE Trans. PAS, Vol. PAS-101, No.IO, pp3593-3601. [2] MohanaRao M., Naidu M.S. (199% 'Estimation of Fast Transient Overvoltages in the case of Disconnnector operation in a GIS', 3d workshop & conference on EHV Technology, IISC Bangalore. [3] J.B. Kim,M.S. Kim,K.S.Park, W.P.Son.,.D.S. Kim, G.S. Kil. Development of monitoring and diagnostic system for SF6 gas insulated switchgear. IEEE Conference Record of the 2002 IEEE International Symposium on Electrical Insulation. Boston, Massachusetts, UnitedStates, pp.453456, 2002. [4] M.kondalu, G.Sreekanthreddy, Dr. P.S. subramanyam,” Estimation Transientover voltages in gas insulated bus duct from 220kv gas insulated substation”, International journal of Computer applications, (0975-8887) volume 20-no.8 april 2011. [5] M.kondalu, G.Sreekanthreddy, Dr. P.S. subramanyam,” Analysis and Calculation of very fast transient over voltages in 220kv gas insulated substationIinternational Journal of Engineering &techsciencevol 2(4) 2011 [6] Li Liu-ling, Hu Pan-feng, Qiu Yu-chang, Analysis of Very Fast Transient Overvoltage Calculation Affected by Different Transformer WindingModels, Journal of Xi’an Jiaotong University, 2005 ′ 10 ″ Vol.39 No.10:1160-1164. [7] Yang Linghui, Zhang Jiamin, Research on transient overvoltage during Operation of 500kV GIS disconnecting switch, East China Electric Power, Jan.2004, Vol.32 No.1:37-41 [8] Shibuya Y, Fujita S, Shimomura T. Effects of very fast transient overVoltages on Transformer [J].IEE Proceedings, Part C,1999,146(4): 459 464. [9] G. Ecklin, D. Schlicht, and A. Plessel. Over voltages in GIS caused by the Operation of isolators in Surges in High Voltage Networks. K.Ragaller, Ed. New York City: Plenum Press, 1980. [10] H. Hiesinger,RWitzmann. Very fast Transient Breakdown at a needle Shaped Protrusion, IX Int. Conf. on Gas Dis. and Their Appli. Sep 1988. Page 763 Kondalu and Subramanyam, Estimation of Re-striking Transient Overvoltages in a 132KV Gas insulated Substation 

International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 784-790 ISSN 2078-2365 Different Braking Techniques Employed to a Brushless DC Motor Drive used in Locomotives M.Rakesh, P.V.R.L. Narasimham Department of EEE., Gudlavalleru Engineering College, Gudlavalleru-521356, A.P, India Abstract— Brushless Direct current (BLDC) motors are gaining attention these days in many applications because of their simplicity in its control and high power density. Due to their usage many advances are taking place in field of automobiles in general and locomotives in particular. To have an effective control over the locomotive it is desirable to have control over starting, running and braking of a bldc drive. This paper mainly describes about different types of braking that can be applied to a bldc drive used in locomotives. Various braking methods are introduced and described. The braking of bldc motor is simulated in MATLAB/Simulink. The simulation results are presented and comparative study is made. Key words: BLDC drive, braking methods, locomotives, Matlab/Simulink. I. INTRODUCTION Brushless Direct current (BLDC) motors are one of the motor types rapidly gaining popularity in industry such as appliances, automotives, aerospace, consumer, medical, industrial automation equipment and instrumentation. Recent trend in automobile industry is using these BLDC motors as electric vehicles as these are energy efficient and pollutant free. Simulation studies indicate that a 15% longer driving range is possible for an electric vehicle with PM brushless motor drive systems compared with induction types. As the name implies, BLDC motors do not use brushes for commutation; instead they are electronically commutated. In BLDC motor since the back emf is non sinusoidal, the inductance do not vary sinusoidally in the abc frame and it does not seem advantageous to transform the equations to d-q frame since inductances will not be constant after transformation [5]. The braking of BLDC motors is quite easier as these machines employ a permanent magnet as its rotor. The braking methods of a BLDC motor are similar to that of a direct current machine. This paper deals with different types of braking applicable to a BLDC drive. The performance of locomotive is examined for dynamic braking, plugging and regenerative braking and simulation results are presented. II. MATHEMATICAL MODELING OF BLDC MOTOR In modeling a BLDC motor, abc phase variable model is preferred to d-q axis model as the mutual inductance between stator and rotor is non-sinusoidal[1]. The mathematical modeling is done in abc phase variable model and is expressed in state-space form. Following assumptions are made in modeling the BLDC motor[10].  The motor is not saturated.  Stator resistances of all the windings are equal and self and mutual inductances are constant.  The power semiconductor devices are ideal. The voltage equations of the three phase stator windings are va R 0 0 ia L M M ia ea vb = 0 R 0 ib + p M L M ib + eb vc ic ic 0 0 R M M L (1) ec The generated electro-magnetic torque equation is Te = (eaia+ebib+ecic)/ωm (2) The equation of motion is pωm =(Te - Tl – Bωm)/J (3) These voltage equations are transformed to state-space form and are arranged as follows: ia p ib ic -R/L 0 0 = 0 -R/L 0 0 0 -R/L ia 1/L 0 0 va ib + 0 1/L 0 vb ic 0 0 1/L vc (4) Where va, vb, vc are the voltages of the three phases a,b and c in volts R is resistance of each phase of motor in ohms ia, ib, ic are the currents of the three phases a,b and c in amperes. p is the derivative Page 784 Rakesh and Narasimham, Different Braking Techniques Employed to a Brushless DC Motor Drive used in Locomotives International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 784-790 ISSN 2078-2365 L is self inductance of each phase of motor in henrys M is mutual inductance between respective phases in henrys ea, eb, ec are the back emfs of the three phases a,b and c in volts Te is the electromagnetic torque in Newton meters ωm is the mechanical speed of the motor in radians per second Tl is load torque in Newton meters B is damping constant in newtons per radian per second J is inertia of rotor in kg – m2 L is difference of self and mutual inductances in henrys The modeling of the machine during motoring operation is presented above and is modified for different braking operations. III. BRAKING AND TYPES OF BRAKING which opposes the motion. If the input current is in phase with back emf, motoring torque is developed otherwise if the input current is in out of phase with back emf then braking torque is developed. The electro magnetic torque developed in phase ‘a’ is Tea = (ea*ia)/ωr (5) Braking during forward motoring is called forward braking while it is in reverse rotation is called reverse braking. In Speed-Torque plane, forward braking will result in second quadrant operation where as reverse braking results in fourth quadrant operation. Figure.1. is a basic bldc motor drive used in locomotives. a three phase inverter is used for exciting the three phase bldc motor. For the motor control, Commutation logic and control block takes the rotor position, torque command and current feed back as inputs for switching the gate drives of the switches. In locomotives, precise control over stopping of machine is important along with start. In such a case to stop the machine quickly and accurately, braking methods are useful. Braking is nothing but stopping the machine at a desired position. Ideal braking is bringing the machine to rest in no time. Braking of locomotive can be done as electric braking or mechanical braking. In mechanical braking the motion is restricted by the friction applied by mechanical brakes which is preferred during low speeds. In electric braking the motor works as a generator developing a negative torque which restricts the motion. The purpose of electrical braking is to restrict the motion of the machine as quick as possible. Electric braking cannot replace the ordinary mechanical brakes, as the vehicle cannot be held stationary by it. In locomotives, for the braking to be done perfectly and smoothly, electric braking in conjugation with mechanical braking is used. This is done by applying electrical braking to slow down the locomotive to a lower speed and then mechanical brakes are applied. Electric braking to above drive can be implemented in three ways namely Dynamic Braking, Plugging and Regenerative Braking. During electric braking the motor torque will reverse and the machine will work as a generator, absorbing mechanical energy from load and converting it into electrical energy. The mechanical energy is obtained from the load either from the energy stored in the inertia of the motor load system or from the active load torque when the locomotive is moving down gradient. Electrical braking reduces the wear of the brake shoes and gives higher rate of braking retardation, thus brings the vehicle quickly to rest and shortens the running time to a considerable extent. This braking can be implemented by disconnecting the power supply to windings and short circuiting them. The short circuited windings carrying higher magnitudes of current will damage the windings. To limit the short circuit current flowing in these windings, an external resistance R is connected in series with the windings. This resistance is used to limit current and is called braking resistance. Thus, the power generated in the three stator windings during Figure.1 Basic BLDC Motor drive during normal operation A. DYNAMIC BRAKING Dynamic braking is bringing the machine to rest position by dissipating the kinetic energy possessed by the rotor of motor in the form of heat energy through some external resistance. Braking action can be achieved by generating a torque of opposite polarity (braking torque) to that of motoring torque Page 785 Rakesh and Narasimham, Different Braking Techniques Employed to a Brushless DC Motor Drive used in Locomotives International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 784-790 ISSN 2078-2365 braking is wasted in the external resistance. Figure.2 Circuit configuration of bldc motor drive during dynamic braking Fig.2 is an implementation of dynamic braking to the drive in fig.1. The generated power is dissipated in the resistor R through the fed back diodes. When fast braking is desired, that is in order to keep the braking torque at a fixed value the resistance has to be decreased with time. This is done by controlling switch So [4]. The equation of speed during normal operation is ω= (V/K) - (R*T)/K2 (6) For dynamic braking V=0, then the speed equation becomes ω= - (R*T)/K2 (7) Here R is the resistance of winding after insertion of braking resistance. So during dynamic braking the speed-torque relation is in the form of a straight line which passes through the origin with a negative slope of –R/K2. B. PLUGGING Plugging is a method of braking obtained by reversing the applied supply voltage, so that the input voltage assists the back emf in forcing armature current in reverse direction. This reversed current will have impact on torque, thus producing deceleration. Plugging provides faster braking response because braking torque is high as the magnitude of current during this braking is high. Even though plugging provides faster braking response it is highly in-efficient because in addition to generated power, the power supplied by the source is also wasted in resistances and plugging increases the inverter rating also. Plugging can be implemented to the drive in figure.1 by reversing the voltage (by high speed switches) and a braking resistance is connected just as in dynamic braking. The speed equation now becomes ω = (-V/K) - (R*T)/K2 (8) So plugging provides torque at zero speed. when reverse voltage is applied for stopping the locomotive the supply must be disconnected at the instant where speed is close to zero. Otherwise it will rotate in reverse direction (reverse motoring takes place). The speed-torque relation is of the form of a straight line with a negative intercept. Thus the speed-torque plot during plugging doesn’t pass through origin, it is a straight line having a slope of –R/K2 and having a negative intercept of –V/K. C. REGENERATIVE BRAKING In regenerative braking, instead of wasting the power in external resistance the power generated during retardation is fed back towards the source i.e., the motor works as a generator developing a negative torque which opposes the motion and the generated energy is supplied to the source. For the generated energy to be supplied to the source two conditions should be satisfied i) back emf should be greater than supply voltage (E > V) for all speeds ii) Current has to reverse its direction For the above two conditions to be satisfied, increase the back emf so that it is greater than the supply voltage. In order to increase the back emf, increase the speed. The speed increases when the locomotive is moving down the gradient or by increasing the field flux. But increasing the field flux beyond rated is not possible as the permanent magnets are used in field system. So, for a source of fixed voltage of rated value regenerative braking is possible only for speeds higher than rated value and for a variable voltage source it is possible for below rated speeds also. During regeneration if the generated power is not absorbed by the load, it will be supplied to the line and the line voltage will rise to dangerous values leading to insulation break down. Hence regenerative braking should be used only when there are loads connected to absorb regenerated power. Figure.3 basic four-quadrant bldc motor drive for regeneration Figure.3 shows a basic four-quadrant electronically commutated motor drive which provides regenerative braking. During regeneration the capacitor C stores the energy recovered from the load through the feed back diodes Page 786 Rakesh and Narasimham, Different Braking Techniques Employed to a Brushless DC Motor Drive used in Locomotives International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 784-790 ISSN 2078-2365 across the switches. To limit the capacitor C voltage to a safer value Switch S0 is used to dissipate the excess energy through the resistor. The speed equation during regeneration becomes ω = (V/K) - (R*T)/K2 (9) The above equation is of the form of a straight line with a positive intercept. As we keep on decreasing the voltage with reference to back emf, the voltage becomes zero finally. Thus the speed equation becomes ω = (V/K) + (R*T)/K2 (10) So during regenerative braking the speed-torque relation is in the form of a straight line which passes through the origin with a slope of R/K2. IV. reversal in current is observed (at 0.05 sec) but not in back emf, this produces negative torque called braking torque. The magnitude of current during braking will decide the braking time and it exists till the kinetic energy possessed by rotor is dissipated completely. This current has to be restricted in order to protect the circuitry from damage. To limit the current we can use fixed or variable resistance, but variable resistance is employed in order to decrease the braking time. SIMULATION RESULTS The locomotive system is developed using the motor parameters listed in below table: Voltage 200 Torque constant 0.66 Vdc,(volts) Kt(Nm/A) No of Poles, 2P 4 Rotor inertia 0.79 J(kg-m ) Winding 1.4 Noload Speed 5400 resistance Rs(Ω) Nnl(rpm) Winding 8.90 Stalling Torque 8 inductance Tstall, (Nm) Ls(mH) 2 In my model, initially the machine is started as a motor and is subjected to braking after the locomotive reaches a steady speed. Three different types of brakings are performed in MATLAB / SIMULINK. A. DYNAMIC BRAKING During simulation, machine is allowed to attain a steady speed initially and after that at 0.05sec dynamic braking is applied. The simulation results are obtained. Figure.5 a) Torque and b) Speed waveforms during dynamic braking Initially the vehicle attained a steady speed in 0.03sec during motoring. At time t =0.05sec, the dynamic brake is applied which developed a negative torque that opposed the motion and tended the locomotive to rest. The time taken by the locomotive to reach zero speed is 37.5 milliseconds. Figure.4 Current and Back emf waveforms during dynamic braking Figure.4 shows the variation of current and back emf waveforms from motoring to dynamic braking. Here a Page 787 Rakesh and Narasimham, Different Braking Techniques Employed to a Brushless DC Motor Drive used in Locomotives International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 784-790 ISSN 2078-2365 Figure.6 Speed-Torque plot of locomotive from motoring to dynamic braking When Dynamic brake is applied, the operating point has shifted to second quadrant with same magnitude of speed. Speed can’t be changed abruptly but the torque maintains the same magnitude but with opposite in sign (negative torque) as shown in figure.6. This negative torque means the negative power; i.e. power generated which is dissipated in the braking resistance. The locomotive stops when the operating point is origin. B. PLUGGING During simulation, machine is allowed to attain a steady speed initially and after that at 0.05sec reverse voltage is applied to stop the vehicle. The simulation results are obtained. Figure.7 shows the variations in current and back emf waveforms from motoring to braking Figure.8 a) Torque b) Speed waveforms during plugging In plugging, time taken by the locomotive to reach zero speed is 31.2 milliseconds (figure.8), which is 37.5 milliseconds in case of dynamic braking for the same machine. So plugging gives quick response compared to dynamic braking. Figure.7 Current and Back emf waveforms during plugging At a time of 0.05sec, when plugging is applied current has changed its direction. In this case of braking, the magnitude of current has increased to 2.67 times to that of steady value with out braking resistance. Figure.8 shows the waveforms of torque and speed during plugging. With out braking resistance, the braking torque has increased with respect to current, and it also increased by 2.7 times to that of steady value in magnitude but with a negative sign. So to incorporate plugging the inverter has to be redesigned and proper care is to be taken for the windings to with stand this much of current. Figure.9 Speed-Torque plot of locomotive from motoring to braking for plugging When plugging is applied for the machine operating in first quadrant the operating point has shifted to second quadrant with same magnitude of speed but with a change in the magnitude of torque with an opposite sign (negative torque). The path of the curve is a straight line as shown in figure.9. When speed is zero there does torque exists. So mechanical Page 788 Rakesh and Narasimham, Different Braking Techniques Employed to a Brushless DC Motor Drive used in Locomotives International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 784-790 ISSN 2078-2365 brakes are applied at that instant to prevent reverse motoring. C. REGENERATIVE BRAKING During simulation, at 0.05sec to stop the locomotive regenerative braking is applied. The simulation results are shown. Figure.10 shows the variations in current and back emf waveforms from motoring to braking during regeneration. Figure.11 a)Torque b) Speed waveforms during regenerative braking The time taken by the machine to reach zero speed is 38.3 milliseconds (figure.11), which is almost same during dynamic braking. This time interval is the time for which regeneration took place. Till 0.05seconds, the three windings of the machine had consumed the power and after that instant the power has become negative because the three windings are generating power which is fed to an external load. Figure.10 Current and Back emf waveforms during regenerative braking In this braking, in order to keep the magnitude of current with in the bounds proper care has to be taken in varying voltage. Figure.11 shows the torque and speed waveforms during regenerative braking. Figure.12 Speed-Torque plot of locomotive from motoring to regenerative braking The speed-torque plot of the machine during its operation from motoring to regeneration is shown in figure.12. When Regenerative braking is applied, the power generated is delivered to an external load and the operating point has shifted to second quadrant, this quadrant is called Forward regeneration. The locomotive stops regeneration when the operating point tends to origin. V. CONCLUSION From the simulation results analysis of the three brakings, i) Regenerative braking is more useful as no power is wasted but this process is costlier as this requires some external circuitry for regeneration. Page 789 Rakesh and Narasimham, Different Braking Techniques Employed to a Brushless DC Motor Drive used in Locomotives International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 784-790 ISSN 2078-2365 ii) Dynamic braking can be used where stopping the machine is important not wasted power i.e., where economy is a factor. iii) Plugging is the most in-efficient method as this will damage the windings, though this gives faster braking to incorporate this the inverter has to be redesigned. Power supplied by the source is wasted along with the power generated. Hence regenerative braking is preferred in locomotives. For consumers using bldc motor drive regenerative braking is suggested, if they can afford the extra cost for the regenerative circuitry else dynamic braking is preferred. REFERENCES [1] P.Pillay and R.Krishnan, “Modeling, simulation and analysis of permanent-magnet motor drives, part-II: the brushless DC motor drives,” IEEE Trans. on Industry Applications, vol. 25, pp.274-279, March/April1989. [2] T. Jahns, R. C. Bccerra, and M. Ehsani, “lntcgrated current regulation for a brushless ECM drive,” IEEE Trans. Pother( -Iron., vol. 6, no. I, pp. 118-126, Jan. 1991. [3] Roger C.Becerra, “Four-Quadrant Brushless ECM Drive with Integrated Current Regulation,” IEEE Trans. on Industry Applications, vol. 28, No.4, July/August 1992. [4] Gopal K Dubey “Fundamentals of Electrical Drives”, Narosa Publishing House, New Delhi, Second Edn, 2001, Chapter 7, pp271-277 [5] Pragasan Pillay and R.Krishnan, ‘‘Modelling of permanent magnet motor drives’’ IEEE Transactions on industrial electronics, Vol.35, No.4, November 1988. [6] Krishnan R “motor Drives Modeling, Analysis and Control”, Prentice Hall of India, First Edn, 2002. [7] Pragasan Pillay and R.Krishnan, ‘‘Control Characteristics and Speed Controller Design for a High Performance Permanent Magnet Synchronous Motor Drive’’ IEEE Transactions on Power Electronics Vol.5 No.2 April 1990. [8] P.Pillay and R.Krishnan, “An investigation into the torque behaviour of a brushledd dc motor drive”,IEEE Transactions 1988 [9] Pragasan Pillay and R.Krishnan, ‘‘Application characteristics of permanent magnet synchronous and brushless dc motors for servo drives’’ IEEE Trans. on Industry Applications, vol. 27, No.5, september/october 1991. [10] Vinatha U, Swetha Pola, Dr K.P.Vittal, ‘‘Simulation of Four Quadrant Operation & Speed Control of BLDC Motor on MATLAB / SIMULINK’’. [11] Padmaraja yedmale,‘‘Brushless dc motor fundamentals’’ Microchip AN885- Microchip technology Inc Page 790 Rakesh and Narasimham, Different Braking Techniques Employed to a Brushless DC Motor Drive used in Locomotives 

International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 751-756 ISSN 2078-2365 Step-Up Dc/Dc Converter for Distributed Power Generation Systems T. Karthikeyan, B.Gowdhami and. Sathishkumar M.E. 1 PG Student, 2PG Student and 3Assitant professor EEE Mailam Engineering College, Villupuram, India E-mail :karthimec2011@gmail.com,gowdhamiped@gmail.com and mecposk@yahoo.co.in can be easily transported and converted to other forms for the benefit of society. In terms of power they come in various sizes from mW to MW ranges. Abstract: This paper presents new step-up dc/dc converter topologies intended for distributed power generation systems. The topologies contain a voltage-fed quasi-Zsource inverter with continuous input current on the primary side, a single-phase isolation transformer, and a voltage doublers rectifier (VDR). To increase the power density of the converter, a three-phase auxiliary ac link (a three-phase inverter and a three-phase isolation transformer) and a three-phase VDR are proposed to be implemented. This paper describes the operation principles of the proposed topologies and analyzes the theoretical and experimental results Keywords: QZSI, voltage doublers rectifier, DC/DC converter, PV panel. 1. INTRODUCTION To interconnect a low-dc-voltage-producing PV (typically 40–80 Vdc) to residential loads (typically 230Vac single phase or 3 × 400 Vac), a special voltage matching converter is required. Due to safety and dynamic performance requirements, the interface converter should be realized within the dc/dc/ac concept. This means that low voltage from the PV first passes through the front-end stepup dc/dc converter with the galvanic isolation; subsequently, the output dc voltage is inverted in the three-phase inverter and filtered to comply with the imposed standards and requirements (second dc/ac stage). Solar energy is the most abundant renewable resource. The electromagnetic waves emitted by the sun are referred to as solar radiation. The amount of sunlight received by any surface on earth will depend on several factors including; geographical location, time of the day, season, local landscape and local weather. The light's angle of incidence on a given surface will depend on the orientation since the Earth's surface is round and the intensity will depend on the distance that the light has to travel to reach the respective surface. The radiation received by a surface will have two components one which is direct and will depend on the distance the rays travel (air mass). The other component is called diffuse radiation and is illustrated in figure 2.1. The range of wavelengths of light that reach the earth varies for 300nm to 400nm approximately. The spectrum outside the atmosphere, which closely resembles 'black body' radiation, since the atmosphere selectively, absorbs certain wavelengths. They can directly convert the sun's energy into electricity which The design of the front-end isolated dc/dc converter is most challenging because this stage is the main contributor of interface converter efficiency, weight, and overall dimensions. The low voltage provided by the PV is always associated with high currents in the primary part of the dc/dc converter (switching transistors and primary winding of the isolation transformer). These high currents lead to high conduction and switching losses in the semiconductors and therefore reduce the efficiency. Moreover, the large voltage boost factor requirement presents a unique challenge to the dc/dc converter design . This specific requirement could be fulfilled in different ways: by use of an auxiliary boost converter before the isolated dc/dc converter or by use of an isolation transformer with a large turns ratio for effective voltage step-up. A direct step-up dc/dc converter without input voltage preregulation is simpler in control and protection. Due to the reduced number of switching devices, the converter tends to have better efficiency and reliability. The varying voltage from the PV passes through the high- Page 751 Karthikeyan et. Al., Step-Up Dc/Dc Converter for Distributed Power Generation Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 751-756 ISSN 2078-2365 frequency inverter to the step-up isolation transformer. The magnitude of the primary winding voltage is controlled by the duty cycle variation of inverter switches in accordance with the PV output voltage and converter load conditions. The isolation transformer should have an increased turns ratio (approximately 1 : 17) to provide effective voltage step-up in the whole range of input voltage and load variations. The choice of dc/dc converter topology in that case can be broadly categorized as a push–pull or a singlephase full-bridge topology. Because of the symmetrical transformer flux and minimized stress of primary inverter switches, the fullbridge topology has been found to be most useful in terms of cost and efficiency, particularly when implemented for power levels higher than 3 kW. This paper is devoted to a new power circuit topology to be implemented in the front-end dc/dc converter for distributed power generation. The topology proposed contains a voltage-fed quasi-Z-source inverter (qZSI) with continuous input current at the converter input side, a high-frequency stepup isolation transformer, and a voltage doubler rectifier (VDR). In contrast to earlier presented topologies, the novel converter provides such advantages as increased reliability, isolation transformer with reduced turns ratio, and reduced impact on the PV due to continuous input current. To improve the power density of the converter, the topology with a three phase intermediate ac link is discussed in the final section of this paper. 2. DESCRIPTION OF PROPOSED TOPOLOGY The voltage-fed qZSI with continuous input current implemented at the converter input side has a unique feature: It can boost the input voltage by utilizing extra switching state—the shoot-through state. The shoot-though state here is the simultaneous conduction of both switches of the same phase leg of the inverter. This operation state is forbidden for the traditional voltage source inverter (VSI) because it causes the short circuit of the dc-link capacitors. In the discussed qZSI, the shoot-through state is used to boost the magnetic energy stored in the dc-side inductors (L1 and L2) without short-circuiting the dc capacitors. This increase in inductive energy, in turn, presented in as a modification of a currently popular voltage-fed Z-source inverter (ZSI). The drawback associated with the conventional ZSI is substantial— discontinuous input current during the boost mode that could have a negative influence on the PV. The discussed qZSI features continuous current drawn from the FC as well as lower operating voltage of the capacitor C2, as compared to the ZSI topology. Uc1= 1-Ds/1-2Ds.UIN…………1 Uc2=Ds/1-2Ds.UIN…………….2 Fig.1 Proposed power circuit diagram of Quasi Z-source converter provides the boost of voltage seen on the transformer primary winding during the traditional operating states (active states) of the inverter. Thus, the varying output voltage of the PV is first preregulated by adjusting the shoot-through duty cycle; afterward, the isolation transformer is being supplied with a voltage of constant amplitude value. Although the control principle of the qZSI is more complicated than that of a traditional VSI, it provides a potentially cheaper, more powerful, reliable, and efficient approach to be used for FC powered systems. The voltage-fed qZSI with continuous input current was first where Ds is the duty cycle of the shoot-through state Ds= ts/T…………….3 where tS is the duration of the shoot-through state and T is the operation period. When the input voltage is high enough, the shoot-through states are eliminated, and the qZSI starts to operate as a traditional VSI, thus performing only the buck function of the input voltage. Thus, the qZSI could realize both the voltage boost and the buck functions without any additional switches using a special control algorithm only. Page 752 Karthikeyan et. Al., Step-Up Dc/Dc Converter for Distributed Power Generation Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 751-756 ISSN 2078-2365 3.Voltage Boost Control Method of qZSI-Based SinglePhase DC/DC Converter Fig. 2 shows the control principle of the singlephase qZSI in the shoot-through (voltage boost) operating mode. Fig. 2(a) shows the switching pattern of the traditional single-phase VSI. These switching states are known as active states when one and only one switch in each phase leg conducts. To generate the shoot-through states, two reference signals (Up and Un) were introduced [Fig. 2(b)]. If the triangle waveform is greater than Up or lower than Un, the inverter switches turn into the shootthrough state [Fig. 2(b)]. During this operating mode, the current through the inverter switches reaches its maximum. Depending on the control algorithm, the shoot through current could be distributed between one or both inverter legs. The dc-link voltage and the primary winding voltage waveforms of the isolation transformer during shoot-through are shown in Fig. 2(c) and (d), respectively. According to the presented control methodology (Fig. 2), the shootthrough states are created during the zero states of the fullbridge inverter, where the primary winding of the isolation transformer is shorted through either the top (T1 and T3) or bottom (T2 and T4) inverter switches. To provide a sufficient regulation margin, the zero-state time tZ should always exceed the maximum duration of the shoot-through states tS,max per one switching period. Thus, each operating period of the qZSI during the shoot-through always consists of an active state tA, shootthrough state tS, and zero state tZ. T = tA + tS + tZ………4 DA+ DS+ DZ=1…………..5 where DA is the duty cycle of an active state, DS is the duty cycle of a shoot-through state, and DZ is the duty cycle of a zero state. It should be noted that the duty cycle of the shoot-through state must never exceed 0.5. It should be noted here that, in the presented control scheme, the shootthrough time interval is evenly split into two intervals of half the duration. Fig.2. Proposed operating principle and resulting voltages of the single-phase qZSI in the shoot-through (voltage boost) mode. In that case, the operating frequency of the quasiZ-source (qZS) network will be two times higher, and the resulting switching frequency of the power transistors will be up to three times higher than the fundamental harmonic frequency of the isolation transformer. That fact is very relevant for proper component and operating frequency selection. In the operating points, when the input voltage is high enough, the shoot-through states are eliminated, and the qZSI operates as a traditional VSI. Thus, the qZSI discussed could provide both the voltage boost and buck functions by the single stage energy conversion. 4.Power Circuit Design Considerations This section provides an overview of the design process of the proposed dc/dc converter. In the given application, the desired value selected for the dclink voltage UDC was 80 V. It is assumed that the converter is always operating with the rated load and between two boundary operating points, which correspond to the minimal UIN,min and maximal UIN,max input voltages. In the first case, the shoot-through states should be used to boost the input Page 753 Karthikeyan et. Al., Step-Up Dc/Dc Converter for Distributed Power Generation Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 751-756 ISSN 2078-2365 voltage to the predefined dc-link voltage level. In the second case, when the input voltage is equal to the desired dc-link voltage, no shoot-through is applied, and the qZSI operates as a traditional VSI. The design of the power converter should be performed for the operating point with a minimal possible input voltage and at rated power, when the shootthrough duty cycle reaches its maximum. As a consequence, the boost ratio of the input voltage is also maximal Bmax= UDC/UIN,min= 80/40= 2 To achieve proper efficiency of the converter and better transformer utilization, in real designs, proper balance between the boost ratio and the transformer turns ratio should be found. In the current application, the maximal duty cycle of the shootthrough state is Fig.3 Proposed Quasi Z-source Series-parallel resonant converter Ds,max=0.25 During the active states, the transformer primary winding is being supplied from the inverter by a voltage with an amplitude value UTR,pr = UDC = 80 V. To reduce the turns ratio n of the isolation transformer, a VDR was implemented on the secondary side of the converter. In contrast to the traditional full-bridge rectifier, two diodes of one leg in the VDR topology are replaced by the capacitors. Since each capacitor charges to the peak secondary voltage UTR,sec, the output voltage from this circuit will be the sum of the two capacitor voltages or twice the peak voltage of the secondary winding. This circuit then produces an output voltage that is twice the transformer secondary voltage. Due to the voltage doubling effect, the VDR enables the use of the isolation transformer with a reduced secondary turns ratio, i.e., 1 : 3.75 for the application discussed. Furthermore, the VDR improves the rectification efficiency due to minimized voltage drops in the components (twice reduced number of rectifying diodes and full elimination of a smoothing inductor). Fig.4 Simulation diagram Page 754 Karthikeyan et. Al., Step-Up Dc/Dc Converter for Distributed Power Generation Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 751-756 ISSN 2078-2365 Fig5 dc link voltage Fig.5 Output voltage current waveform 5.CONCLUSION This paper has presented two new isolated step-up dc/dc converter topologies with qZSIs. The topologies are intended for applications with widely varying input voltage and stabilized output voltage and when the galvanic separation of the input and output sides is required. The high-frequency transformer stack is responsible for providing the input/output galvanic isolation demanded in many applications. This paper has focused on an example of the step-up dc/dc converter with high-frequency isolation for the distributed power generation systems. The proposed converters have the following key features in comparison to traditional topologies. 1) The qZSI implemented on the primary side of the converter could provide both the voltage boost and buck functions with no additional switches, only by use of a special control algorithm. 2) The qZSI has an excellent immunity against the cross conduction of the top- and bottom-side inverter switches. Moreover, the qZSI implemented can boost the input voltage by introducing a shoot-through operation mode, which is forbidden in traditional VSIs. 3) The qZSI implemented has the continuous input current (input current never drops to zero) during the shoot-through (voltage boost) mode. 4) The high-frequency step-up isolation transformer provides the required voltage gain as well as input–output galvanic isolation demanded in several applications. 5) The VDR implemented on the converter secondary side has the improved rectification efficiency due to the reduced voltage drop (twice reduced number of rectifying diodes and full elimination of the smoothing inductor). 6) The turns number of the secondary winding of the isolation transformer could be reduced by 62% (turns ratio of 1 : 3.75 in the case of VDR instead of 1 : 10 of traditional full-bridge rectifiers) due to the voltage doubling effect available with the VDR. 6.REFERENCE . [1] A. F. Zobaa and C. Cecati, “A omprehensive review on distributed power generation,” in Proc. SPEEDAM, 2006, pp. 514–518. [2] J. Padulles, G.W. Ault, and J. R.McDonald, “An approach to the dynamic modelling of fuel cell characteristics for distributed generation operation,”in Proc. IEEE Power Eng. Soc.Winter Meeting, 2000, vol. 1, pp. 134–138. [3] W. Choi, P. Enjeti, and J. W. Howze, “Fuel cell powered UPS systems:Design considerations,” in Proc. IEEE 34th PESC, Jun. 15–19, 2003,vol. 1, pp. 385–390. [4] M. H. Todorovic, L. Palma, and P. N. Enjeti, “Design of a wide inputrange DC–DC converter with a Robust power control scheme suitable forfuel cell power conversion,” IEEE Trans. Ind. Electron., vol. 55, no. 3,pp. 1247–1255, Mar. 2008. [5] S. K. Mazumder, R. K. Burra, and K. Acharya, “A ripple-mitigating andenergy-efficient fuel cell powerconditioning system,” IEEE Trans. PowerElectron., vol. 22, no. 4, pp. 1437–1452, Jul. 2007. Page 755 Karthikeyan et. Al., Step-Up Dc/Dc Converter for Distributed Power Generation Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 751-756 ISSN 2078-2365 [6] J. S. Yu and P. N. Enjeti, “A high frequency link direct dc-ac converterfor residential fuel cell power systems,” in Proc. IEEE 35th PESC,Jun. 20–25, 2004, vol. 6, pp. 4755– 4761. [7] J. C. Han and P. N. Enjeti, “A new soft switching direct converter for residential fuel cell power system,” in Conf. Rec. 39th IEEE IAS Annu.Meeting, Oct. 3–7, 2004, vol. 2, pp. 1172–1177. [8] S. K. Mazumder, R. Burra, R. Huang, M. Tahir, K. Acharya,G. Garcia, S. Pro, O. dodrigues, and E. Duheric, “A high-efficiency universalgrid-connected fuel-cell inverter for residential application,” IEEETrans. Power Electron., 2009, to be published. Page 756 Karthikeyan et. Al., Step-Up Dc/Dc Converter for Distributed Power Generation Systems 

International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 625-631 ISSN 2078-2365 Study of Reverberation Time Series and Echo Detection Algorithm in Reverberation Limited Scenarios Sowmya S.T.V Department of ECE, GITAM Institute of Technology, GITAM University, Visakhapatnam,India Email: sowmy482@gmail.com Abstract— An innovative approach to the generation of reverberation time series and echo detection algorithms is presented The time series approach utilizes recent developments in linear spectral prediction research in which the spectra of stochastic process are modelled as rational functions and algorithms are used to efficiently compute optimal estimates of coefficients which specify the spectra. The approach taken in this paper is to detect echo signal in two steps . In the first part the reverberation time series is generated using autoregressive formulation and in the second part echo is detected using order partition prewhiten algorithm. Keywords- Active sonar, Reverberation, Autoregressive model, Reverberation spectrum, pre-whiten I. INTRODUCTION S ONAR is an acronym for sound navigation and ranging .sonar is a system that uses transmitted and reflected underwater sound waves to detect and locate submerged objects or measure the distance of underwater target. Sonar Reverberation (and radar clutter) has been modeled in a variety of ways and for a diversity of applications. Expected reverberation power (intensity) level models are perhaps the most common. In these, the expected reverberation power level at the input, output, or some intermediate point in the sonar (radar) system is estimated as a function of the environmental and system parameters. The models are useful in evaluating system performance for signal processing approaches which depend primarily upon power level, such as single beam energy detectors and matched filters. The time series simulation models can be developed to generate P.Chandra sekhar Department of ECE, GITAM Institute of Technology, GITAM University, Visakhapatnam,India Email: chandrasekhar.au@gmail.com complex (in phase and quadrature) reverberation voltage levels in time series form, as they would occur at some point in the sonar system. The time series data can be run through emulations of alternative signal processing algorithms to evaluate relative performance of the various processes. Reverberation is caused by seabed, sea surface and the In homogeneity of the granule in the seawater. As a noise,reverberation can influence the detection performance of target echo and cause some serious problems to active sonar . Due to the fact that reverberation and target echo are correlative and their spectrums are close, how to restrain reverberation is a problem necessary to be solved for active sonar. Inorder to restrain the reverberation signal we are using pre whiten method here. The principle of this method is that an AR model is established from reverberation and then a whiten filter is designed using the power spectrum of this model. II. TIME SERIES MODELING Recent developments in linear spectral prediction (UP) techniques allow stochastic processes to be modeled in a straightforward manner. Common time series models of sampled stochastic processes, which are basic to many LSP techniques, include autoregressive (AR), moving average (MA), and autoregressive moving average (ARMA) processes. These processes can be realized as the outputs of linear digital filters driven by white noise processes, where the filter transfer functions have all-pole, all-zero, and pole-zero realizations for AR, MA, and ARMA processes, respectively. The digital filters can be implemented as recursive infinite impulse response (IIR) filters for AR and ARMA processes, and as a transversal finite impulse response (FIR) filter for an MA process. If the stochastic process to be modelled is non Page 625 Sowmya and Chandra, Study of Reverberation Time Series and Echo Detection Algorithm in Reverberation Limited Scenarios International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 625-631 ISSN 2078-2365 stationary: but quasi-stationary, the digital filter realization is time varying, with time-varying poles and/or zeros. in   bk i k ,n p k 1  on   bk  kn   2 i=1,2,…….p (6a) p A. Autoregressive Process In an auto regressive model of order p, the value xn of the complex stochastic process at time n given as a linear combination of past values and a random input  n , such that xn   bk xn  k   n p (6b) k 1 where i k ,n is the time-varying autocorrelation function of the non-stationary process. i k,n E{xnk xn*i } (7) is a complex white noise process with zero mean If the spectrum (which in general is time-varying) of the process is known, then the auto-correlation function can be found by inverse Fourier transformation and the parameters { bk ,  } evaluated. If the spectrum is not known, but one has and unity standard deviation. The system transfer function H(Z),between input and output is represented in terms of model parameters as an all pole function. the past p values of the process, as in the above Wiener filtering formulation, then the auto covariance function can be used as a local estimate of the autocorrelation function. k 1 Where n H ( z)  (1)  1   bk z k p k j (2) The poles of H(Z) are the zeros of the polynomial in the denominator and the number of poles p is referred to as the model order. The discrete power density spectrum is pm   2 t 1   bk e p k 1 Where  j 2 mk / M 2 t is the sampling interval and pm is the power at (3) radial frequency.The problem of modeling an arbitrary stochastic process {x,, n = e-, -1, 0, 1, .-} as an AR process reduces to the selection of the model parameters. The manner in which they are selected will depend on a priori information about {x,;. A standard formulation is to select the model parameters such that the linear estimate of the process {x,} at the present time n, given the past p values of the process { X n - k , k = 1: 2, --, p } , is best in a least squares sense. That is, defining the linear estimate of order p as  n 1 k *   x j  k x j i ,  j  n  p i i<k i  k ,n   n 1i *  x j  k x j i ,  j  n p j i>k (8) The auto covariance function can then replace the autocorrelation function in the normal equations. With this replacement, are known as the Yule-Walker equations and will be referred to in this way in this paper, regardless of whether the auto covariance or autocorrelation functions are used in them. A variety of ways to solve the Yule-Walker equations are available. We will use the Levinson-Durbin approach . Having solved for estimates of the model parameters in this way, one can evaluate the spectral estimate or the linear estimation filter transfer function &z) by using the parameter estimates in expressions (3) and (2), respectively. The filter will produce a statistical realization of the process {x} when driven by white noise. The spectral estimate is sometimes referred to as the maximum entropy spectral estimate. (4) the parameters are selected such that the estimation error is minimized in a mean square sense min E{ }  min E{( xn   bk xn k ) 2 } p {a k } 2 n {a k } k 1 (5) This is the digital Wiener optimal one-step prediction filtering problem, and it leads to the specification of the { bk ,  } as the solution of the normal equations Page 626 Sowmya and Chandra, Study of Reverberation Time Series and Echo Detection Algorithm in Reverberation Limited Scenarios International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 625-631 ISSN 2078-2365 . Fig-1. Linear Spectral Prediction Approach To Modeling Stochastic Quasi-Stationary Coherent Time Series. The Levinson-Durbin algorithm is utilized to solve the YuleWalker equations to obtain the optimal set of non stationary poles, or more precisely, the optimal set of non stationary filter coefficients for each beam. Fig-2. Processing Flow Of Reverberation Simulation Model. The generation of coherent reverberation which is correlated between beams and overlapping spectra is illustrated in Fig.3, where Ha, Hb, .-, Hn,, represent transfer functions of the linear filters associated with beams, a, b, ...,n, respectively. evaluation scheme based on a spatial grid approach of Ackerman for a general formulation of reverberation developed by Faure, Ol'shevskii and Middleton. This leads to the computation of surface, volume, and/or bottom reverberation spectra and Dower levels as functions of: a) transmit signal waveform and power level, b) spreading and absorption propagation losses, c) backscattering strength, d) transmit and receive beam patterns, e) sonar platform-ocean geometry, and f) sonar and scatter motion. The formulation represents reverberation spectra at the input of a receiver after beam forming, but prior to signal processing operations. The reverberation is non stationary, in that the spectra vary with time (and range), both in their spectral shape and their power levels. The reverberation is quasi stationary,in that the rate of change with time is assumed to be small relative to the transmit pulse duration (or the reverberation correlation time).Although the formulation can be extended to relax some of the following , intrinsic assumptions include: a) primary scattering only, b) iso-sound-speed ocean, c) direct propagation path, d) narrow-band transmit signals, e) back scattering is from a large number of randomly distributed weak discrete scatters, f) radial velocity distribution of scatters is spatially uniform. A. Faure, Olshevskii and Middleton Formulation: The power density spectrum Pf ( f , r ) of the reverberation envelope at the receiver input from scatters at range r and frequency f, letting * designate the convolution operations given by Pf ( f , r )   2 (r )Yf ( f , r ) * STf ( f ) * D f ( f ) 2  2 (r ) (9) Where = Total reverberation power from scatters at range r, Yf ( f , r ) = Sonar motion envelope (power density) spectrum Fig-3. Parallel Structure of Coherent Multiple-Beam Time Series Filter resulting from sonar motion and stationary (non moving) scatters at range r,  Y ( f , r )df  1  III. REVERBERATION SPECTRUM MODELING The generation of a time series based on the existance of a spectrum was discussed. Here, an approach to generating the expected spectrum for sonar reverberation as a function of pertinent sonar system parameters and environmental conditions is presented. The approach is a numerical  STf ( f ) f (10) = Transmit signal envelope (energy density) Spectrum, Page 627 Sowmya and Chandra, Study of Reverberation Time Series and Echo Detection Algorithm in Reverberation Limited Scenarios International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 625-631 ISSN 2078-2365   STf ( f ) df  1  Df ( f ) (11) = Scatter motion spectrum resulting from random motion of the scatters. This is equal to the probability density function of Doppler frequencies of backscattered signals from a large number of scatters with random motion and received by a stationary (non moving) sonar which transmitted a pure tone signal. Since it is a probability function.  D f ( f )df  1  (12)  The terms  , Yf and D f 2 depends on whether the p02cTsv102r  (r )  2r 2  S2 (r )   b S  B 0 TR ( , )dd (13) p02cTsS 102r 0 bTR ( S , ) cos  S d 2r 3   B2 (r )  same as  S2 (r ) , except s B  S respectively. and B replace (14) s S and (15) Where p02 = transmit source mean square pressure, C = speed of sound in water (m/s) T = time duration of transmit signal (s), r = range from mono static sonar to center of set of discrete scatterers (m), = absorption loss ,  sv , s B , s S = = transmit power beam pattern = receive power beam pattern The surface backscattering strength can be represented as a function of surface grazing angle w i nd speed, and frequency (f) through models such as the Chapman-Harris model SS  10 log sS  3.3 log( S 30)  42.4 log   2.6   158(f 0.33 ) 0.58 (18) SB  10 log sB  10 log(sin 2  B )  27 . (19) The bottom backscattering strength can be represented as reverberation is from volume, surface or bottom scatters. The reverberation power levels are given by 2 v bT ( , ) bR ( , ) Where 2 volume (m-3), surface (m-') bottom (m-2) bTR ( , ) = transmit-receive product power beampattern. = azimuthal angle (rad),  = elevation angle (rad),  S = sin^-1 (z/r) = elevation angle to surface at The sonar motion envelope spectrum for volume, surface, and bottom reverberation, for a sonar moving at constant speed v0 and constant direction are given by j 4 T  S  0 )dd     b ( , ) exp( TR    Y ( f , r )  F B 0  vf S      bTR ( , )dd   B 0   j 4v T   0 cos  cos ) cos  dd    b ( , ) exp( TR S b    Y ( f ,r)  F  0  sf     ( , ) cos    b d    TR S S 0   (20) (21) YBf ( f , r ) =same as Ysf ( f , r ) except  S replace  b . (22) backscattering strength, respectively, B range r and sonar depth z (rad), Fig- 4. Spatial Division Of Reverberation Field. =sin^-1(zb-z/r) = elevation angle to bottom, at range r and sonar vertical distance above bottom of (zB - Z ) (rad). bTR ( , )  [bT ( , )  bT ( , )].[bR ( , )  bR ( , )] The transmit-receive beam pattern is further defined by (16) The power density spectrum Pf(f, r) of the reverberation envelope given by (9) is evaluated numerically dividing space into a set of cells, as illustrated in Fig. 6. The ocean is divided into spherical shells which represent the portion of the ocean that is ensonified by the signal wavefront at particular instants of time after transmission (and corresponding ranges). The Page 628 Sowmya and Chandra, Study of Reverberation Time Series and Echo Detection Algorithm in Reverberation Limited Scenarios International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 625-631 ISSN 2078-2365 spherical shells are subdivided into a grid with three types of cells corresponding to those which contribute to surface, volume, and bottom reverberation. The location of each cell is defined by the range and the azimuth and elevation angles to the center of the cell relative to the platform velocity vector. When the parameters ai of this AR model have been estimated, the system function of pre-whiten filter is as follows: H ( z)  1   ai z 1 p i 1 i (25) Output data y(n) is obtained by data x(n) passing the above system function y(n)  x(n) * z1[ H ( z)] (26) The parameters of this AR model can be estimated through autocorrelation method, Burg method, Lattice recursive algorithm and so on. Fig- 5. Reverberation Spectrum Model IV. DETECTION OF ECHO SIGNAL IN REVERBERATION BACKGROUND Reverberation is caused by seabed, sea surface and the in homogeneity of the granule in the seawater. As a noise, reverberation can influence the detection performance of target echo and cause some serious problems to active sonar . Due to the fact that reverberation and target echo are correlative and their spectrums are close, how to restrain reverberation is a problem necessary to be solved for active sonar. Inorder to restrain the reverberation signal we are using pre whiten method here. B. Order Partition Pre-Whiten Algorithm Order partition pre-whitens algorithm means that the sampled reverberation data is processed according to time order. Firstly, reverberation data is partitioned into several segments. Supposing that the echo signal s(n) is included in the signal x(n) which is received by hydrophone during the observation time T. The pulse width of s(n) is TP . Data x(n) is partitioned into several segments and the width of each segment is A. Pre-whiten method The principle of this method is that an AR model is established from reverberation and then a whiten filter is designed using the power spectrum of this model. Reverberation data is considered as Gaussian color noise. And the spectrum transformations between adjacent data segments are not obvious. The rationality of this hypothesis has been verified. The performance of this method is nicer even echotoreverberation ratio is comparative low. On the premise of this hypothesis, a method using all-pole pre-whiten filters is proposed based on AR model at first. Then ,based on this model ,order partition algorithim is brought forward in detail. The AR model of reverberation data is shown in equation r (n)   a i r (n  i)  w(n) p i 1 (23) where w(n) is a Gaussian white noise with an average of 0, {a1, a2, …, ap} are estimated using the adjacent segment data. And its power spectrum is shown in equation (2). s( w)   2 1   a i e  jwt (24) Fig 6. Sketch Map Of Data Partition Firstly, reverberation data is partitioned into several segments. Supposing that the echo signal s(n) is included in the signal x(n) which is received by hydrophone during the observation time T. The pulse width of s(n) is TP . Data x(n) is partitioned into several segments and the width of each segment is TD as manifested in Fig.1, where dT denotes offset between the adjacent kth and k+1th data segments. The following conditions should be satisfied when partitioning the data into several segments: (1) The width of each data segment should be equivalent to the usable signal s(n) in order to satisfy the requirement of local stationary. p i 1 Page 629 Sowmya and Chandra, Study of Reverberation Time Series and Echo Detection Algorithm in Reverberation Limited Scenarios International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 625-631 ISSN 2078-2365 (3) To ensure that the echo signal s(n) is situated in the next adjacent data segment when s(n) is not completely included in a certain segment, following in equation should be satisfied: (28) dT TD  TP power spectrum of transmitted signal 50 power spectrum of transmitted signal 45 40 power spectrum values in db (2) To ensure that the echo signal s(n) is fully included in one segment, the width of s(n) should be smaller than each data segment. The following in equation should be satisfied: (27) TD  TP 30 25 20 15 10 5 According to the actual situation, the width of each data segment is double of the usable signal and the data overlapping rate is ½. (29) TD  2TP dT  1 / 2(TD ) 35 0 50 100 150 sample values 200 250 300 Fig-8 Transmitted Signal Power Spectrum scatter motion spectrum, 0.08 scatter motion spectrum 0.07 0.06 (30) power spectrum 0.05 0.04 0.03 0.02 0.01 Secondly, the AR model of the kth data segment is established. In the method, the system function of the prewhiten filter which is based on the kth data segment for the k+1th data segment is as follows: 1   a k ,i z  i pk 0 50 100 150 no of samples 200 250 300 Fig-9. Scattering Motion Spectrum 23 18 power density spectrum of reverberation x 10 power density spectrum of reverberation 16 (31) 14 12 power density values H k ( z)  1 0 i 1 10 8 6 Lastly, the output data yk1 (n) is obtained by passing data 4 2 xk1 (n) through above system function: 0 0 (32) The flow chart of order partition pre-whiten algorithm is shown in below figure. 100 200 300 no.of samples 400 500 600 Fig-10. Power Density Spectrum of Reverberation 80 AUTOCORRELATION FUNCTION yk 1 (n)  xk 1 (n) * z1[ H k ( z)] 60 40 20 0 -20 0 10 20 30 40 50 FREQUENCY IN HZ 60 70 80 90 100 0 10 20 30 40 50 TIME SAMPLE 60 70 80 90 100 NORMAL TIME SERIES 80 60 40 20 0 -20 -40 Fig-11. Time Series from autocorrelation function Fig 7:Flow Chart Of Order Partition Pre-Whiten Algorithm 7 4 time series signal x 10 time series signal 3 2 V. SIMULATIONS AND RESULTS The reverberation power spectrum using faure,olshevskii, and middeleton formulation is obtained and according to the block diagram and the time series using autoregressive model is obtained. normal time series 1 0 -1 -2 -3 -4 0 100 200 300 time sample 400 500 600 Fig-12. Time Series from the reverberation spectra Page 630 Sowmya and Chandra, Study of Reverberation Time Series and Echo Detection Algorithm in Reverberation Limited Scenarios International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 625-631 ISSN 2078-2365 partition algorithm is the better approach for detecting the echo signal in reverberation background. echo signal 1 0.8 0.6 REFERENCES 0.4 amplitude 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 0 200 400 600 time samples 800 1000 1200 Fig 13: Echo Signal reverberation signal 1 0.5 amplitude 0 -0.5 -1 -1.5 0 200 400 600 time samples 800 1000 1200 Fig 14:ReverberationSignal echo+revereberation signal 1 0.5 amplitude 0 [1]. S. G. Chamberlain and J. C. Galli, “A Model for Numerical Simulation of Non stationary Sonar Reverberation Using Linear Spectral Prediction,” IEEE journal of oceanic engineering VOL. OE-8, NO. 1, Jan. 1983, pp 21-36. [2]. H. Weinbeg, "Navy interim surface ship model(NISSM) 11,"Naval Underwater Svstems Ctr., New London, CT. NUSC Tech. Publ. 372 and NUSC Tech. Rep. 4527, 1973. [3]. P. C. Etter and R. S. Plum, "A survey of underwater acoustic models and environmental acoustic data banks." Naval Anti- Submarine Warfare Syst.Proj. Office, Dept. of Navy.Washington DC, ASWR Tech. Rep.80-1 15,1980. [4] . C. L. Ackerman and R. L. Kesser. "Reverberation spectrum model for matched filter homing systems." Pennsylvania State Univ. Applied Research Lab, Tech. Memo. File TM73-285, Dec. 4, 1973. [5]. P. Faure, "Theoretical model of reverberation noise,'' J. Acoust.Soc.Amer.,vol. 36. no. 2. pp. 259-266.Feb. 1964. [6]. V. V. Ol'shevskii, Characteristics of Sea Reverberation. English translation by V. M. Albers. New York: Consultants Bureau,1967. [7]. A. D. Waite, Sonar for Practising Engineers, 3rd ed., John Wiley & Sons, Ltd., 2002. [8]. Robert J. Urick, Principles of Underwater Sound, 3rd ed., New York McGraw-Hill, 1983. [9]. Manuel Aineto, Stuart Lawson, “Narrowband signal detection in a reverberation-limited environment”, Oceans 97, MTS/IEEE Proceedings, Nova Scotia, 1997, pp.27-32. [10]. Liang Hong, Li Zhishun, “An Adaptive Method to Detect Moving Target in a Reverberation Background”, Applied Acoustics, Vol.22, No.2, 2003, pp.26-29. -0.5 -1 -1.5 0 200 400 600 time samples 800 1000 1200 Fig 15 : Echo+Reverberation Signal restraining the reveraberation 1 0.8 0.6 0.4 amplitude 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 0 200 400 600 time samples 800 1000 1200 Fig 16: Restraining The Reverberation Signal VI. CONCLUSION This paper describes the approach to the numerical generation of reverberation time series and echo detection algorithm. on the premise of local stationary of reverberation, a method using all pole partition pre whiten filters which is based on AR model is proposed. The simulation results indicate that this method is effective even in the background of low echo to reverberation ratio of input. Moreover, the order Page 631 Sowmya and Chandra, Study of Reverberation Time Series and Echo Detection Algorithm in Reverberation Limited Scenarios 

International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 620-624 ISSN 2078-2365 Climate Change Prediction by Wireless Sensor Technology Chin-Yuan Hsieh Graduate School of Information Technology and Applications, Kao Yuan University, Kaohsiung 821, Taiwan Email: cyhsieh@cc.kyu.edu.tw Abstract – The prediction of the temperature microchange in the forest becomes critically important due to the study of global change. In this paper we develop a model to predict and modify the measured temperature by the Zigbee and wireless sensor network technology. The modification model is developed by a pair of integral equations. The tangential surface electric and magnetic fields are induced by the surface current flowing in the tree skin. The surface current is generated by the electromagnetic wave hinging on the rough tree surfaces. The scatter field including the Kirchhoff and complementary fields from the tangential surface electric and magnetic fields is measured by the Zigbee receiver with the wireless sensor network. The wireless sensor network (WSN) measurement system includes temperature sensor, zigbee and Tmote Sky chip. From the measured temperature in the tree skin we found the temperature data of 24 hours in a day show an exponential function distribution. We also found the temperature reaches peak at different time from the measured data. The time shift is in the range of 1 to 2 hours. The time shift could be from the different dielectric constant of tree skin and the tree density in the forest. After comparing the model prediction with the measured data, the correlation coefficient is 0.94. The excellent prediction reaches in this model prediction. Keywords: temperature microchange, wireless sensor network, global change, integral equation, WSN. I. INTRODUCTION The study of wireless sensor networks is challenging in that it requires an enormous breadth of knowledge from an enormous variety of disciplines. The temperature microchange of forest tree cannot be measured by the thermometer. The microchange of forest temperature is a critical factor to the growth and the type of trees, therefore the forest density, distribution and the temperature will affect the global temperature change. In general, forests are sensitive to the variability and change of climate. The Climatic factors influence forest health-temperature, rainfall, atmospheric levels of carbon dioxide, other greenhouse gases and extreme weather. The microchange of tree temperature cannot be Chen-Yu Hsieh Dept. of Mechatronics, School of Engineering Science, Simon Fraser University, BC, Canada Email: chen-yu_hsieh@sfu.ca measured by the traditional thermometer and need to be measures and studied by the wireless sensor network technology [1-4]. In this paper a wireless sensor network based on a ZigBee technique for the temperature microchange was proposed. The wireless sensor network consists of spatially distributed autonomous sensors to monitor the environmental or physical conditions, such as temperature, humidity, light, pollutants, or vibration and to pass their data through the network to a main location of coordinator, notebook, smart phone or ipad cooperatively. The ZigBee device is a specification for a suite of high level communication protocols using small, low-power digital radios based on an IEEE 802 standard for personal area networks. The technology defined by the ZigBee specification is intended to be simpler and less expensive than other wireless personal area network, such as Bluetooth. ZigBee is targeted at radio-frequency applications that require a low data rate, long battery life, and secure networking. ZigBee has a defined rate of 250 kbps best suited for periodic or intermittent data or a single signal transmission from a sensor or input device. The proposed modern measurement technique is able to offers the change rate of forest temperature by the wireless communication in real time. The architecture of a wireless sensor network for measuring the forest temperature microchange includes the sensor, electromagnetic wave transmitter and receiver, hardware circuitry and software system. The hardware circuitry of the network node is designed based on a Tmote Sky chip. The The Tmote Sky platform is a wireless sensor module IEEE 802.15.4 compliant and has the characteristics of Ultra low power based on a TI MSP430 and Chipcon CC2420 radio. From the data collected by the system, the model is developed to predict the temperature microchange. In last section we compare the model prediction and the measured data to evaluate the prediction of model. Page 620 Chin and Chen, Climate Change Prediction by Wireless Sensor Technology International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 620-624 ISSN 2078-2365 II. METHODOLOGY The aim of the paper is to apply the wireless sensor technologies and wireless communications technology for the temperature microchange measurement of trees in the forest. A wireless sensor communication network is composed of nodes, each of which can compute power, transmit and receive messages over communication links, wireless or cabled. A single wireless sensor communication network may consist of several interconnected subnets of different topologies. In this measurement the mesh network is applied. Mesh nets can be good models for large-scale networks of wireless sensors that are distributed over a geographic region. Mesh networks are regularly distributed networks that generally allow transmission only to a node’s nearest neighbors. The nodes in these networks are generally identical, so that mesh nets are also referred to as peer-to-peer nets. An advantage of mesh nets is that, although all nodes may be identical and have the same computing and transmission capabilities, certain nodes can be designated as ‘group leaders’ that take on additional functions. If a group leader is disabled, another node can then take over these duties. The required transmission power increases as the square of the distance between source and destination. Therefore, multiple short message transmission hops require less power than one long hop. In fact, if the distance between source and destination is R, the power required for single-hop transmission is proportional to R2. If nodes between source and destination are taken advantage of to transmit n short hops instead, the power required by each node is proportional to R2/n2. In this paper the measurement system focuses on ZigBee devices based on the Wireless Sensor Networks. A general Wireless Sensor Network protocol consists of the application layer, transport layer, network layer, data link layer, physical layer, power management plane, mobility management plane and the task management plane [1-4]. In the paper we applied the standard WSN with zigbee technology operating in the Industrial Scientific and Medical (ISM) frequency band of 2.4 GHz. The ISM frequency band provides license free operations, huge spectrum allocation and worldwide compatibility. For monitoring the long distance data the architecture of the WSN technology with Wi-Fi (IEEE 802.11) and PC-based systems is developed in Fig. 1. Fig. 1 The layout structure of multi-hop in wireless sensor network technology system For measuring the tree temperature we install the zigbee sensor node in the fir and gingko to collect the temperature by the wireless sensor network technology. The data are retrieved, processed and transmitted for the real time monitoring. For avoiding the human interference the height of sensor node is installed in 3 meters far from the ground. Meantime we consider the sunlight will affect the data transmission status, the directions of sensor nodes faces north to make the measurement precisely. For the tree routing of Zigbee, a Zigbee node transmitting a packet to the destination is set up to follow the tree mesh topology. The direction of zigbee antenna is directional radiation pattern to increase the received signal strength and the communication distance. All the nodes are distributed and installed along the zigzag path to enhance the successful transmission rate. The TmoteSky wireless system can transmit the power consumption status by wireless at contant time period. Tmote Sky is the wireless sensor module for high data-rate sensor network applications with ultra low-power, high-reliability and ease of development. Tmote Sky also offers a number of integrated peripherals including a 12-bit ADC and DAC, Timer, and UART bus protocols, and a performance boosting DMA controller. The TmoteSky wireless sensor with USB connector is developed by the TinyOS operation system in Fig. 2. Fig. 2 The TmoteSky wireless sensor network chip with USB connector The operating voltage is in the range of 2.1V to 3.6V. The time interval can be adjusted as requirement. To save the consumption of power the transmission period is set up 5 minutes for each transmission. The memory for the data storage is 1MB. If the system is busy, the data to be transmitted will be stored in a short time. The Zigbee is a wireless network standard based on IEEE 802.15.4. The transmission range of a wireless sensor network is usually low, less than 100 meters, therefore the characteristics of WSN are low data rate, short distance, low price and low power consumption. The compute capability is limited based on the low complexity, and limited resources. Page 621 Chin and Chen, Climate Change Prediction by Wireless Sensor Technology International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 620-624 ISSN 2078-2365 III. MODEL DEVELOPMENT For adjusting the measured temperature the electromagnetic wave scattering from the rough surface of a tree will be predicted. The prediction model is developed based on a pair of integral equations. The integral equation is to solve the scatter strength of electromagnetic wave. The scatter strength is induced by the surface current along the tree skin, and the surface current is generated by electric and magnetic fields from the transmitter. In medium 1 (incident plane) the governing integral equations for the tangential surface fields on a dielectric surface are [5]    2 nˆ  E  2nˆ  E i  nˆ   E ds  4 [1]  i  2 nˆ  H  2nˆ  H  nˆ   H ds  4   2 nˆ t  Et   nˆ t   Et ds  4   2 nˆ t  H t  nˆ t   H t ds  4  2Re Eqp c Eqp k  Eqp c Eqpc * * * (7) where Re is the real part operator and * is the symbol for complex conjugate. (b) Fig. 3 The scatter and rescatter fields from random rough surfaces of a tree To obtain the coherent and incoherent power, we have to subtract the mean-squared power from the total power. That is, [2] Eqp s Eqp s  Eqp s Eqp s * (3)     H t  jkt (nˆ t  Et )Gt /  t  (nˆt  H t )  Gt  (nˆt  H t )Gt (4) The spectral representation for the Green's function and its gradient are 1 j ) exp[ ju(x  x)  iv(y  y)  jq z  z]dudv 2  q 1 g G   ( ) exp[ ju(x  x )  iv(y  y)  jqz  z]dudv 2 q G  ( 2 2 2 where q = k - u - v and g  xˆu  yˆv  zˆq . The tangential surface field includes two components, the Kirchhoff component (scatter field) and its complementary component (rescatter field), and is described in Fig. 3. Two corresponding components for the scattered fields are [6]  [1] E qp  E qp  E qp (6) Where s means the tangential surface field, k means the Kirchhoff field and c means the complimentary field in c  Eqp k Eqpk (a)     Et  jkt t (nˆ   H t )Gt  (nˆt  Et )  Gt  (nˆt  Et )Gt k * (2) where     E   jk (nˆ   H )G  (nˆ   E )  G  (nˆ   E )G     H   jk (nˆ   H )G /   (nˆ   H )  G  (nˆ   H )G The fields in the lower medium (medium 2) can be written in terms of the fields in the upper medium (medium 1) by applying the boundary conditions on the continuity of the tangential fields as follows: s Eqp s Eqp s (1) In medium 2, we have (5) equation (6). In terms of the surface tangential field for the dielectric surfaces, the far-zone scattered fields can be derived. Consequently the average scattered power and the scattering coefficients can be found in terms of the far-zone scattered field. With the field expression the average scattered power is given by  Eqp Eqp k k*  Eqp k * Eqp k *  Eqp Eqp c 2 Re[ Eqp c Eqp k  Eqp c Eqp k ] * c*  E qp c c * Eqp * (8) To carry out the ensemble average operation we make an assumption the surface with Gaussian height distribution. After the operation of ensemble average and integrating, the scattered power can be obtained. The bistatic scattering coefficient, Pqp  qp0 , is related to the average power expression ,  qp  (4R2 Pqp ) /( E0 2 A0 ) , as 0 (9) Where Eo is the incident field, R is the distance, Ao is the illuminated area of a tree and P qp represents the polarized and depolarized scatter power. The subscript represents the polarization of incident electromagnetic wave and subscript represents the polarization of scatter electromagnetic wave. From the complex calculation we summarized the scatter coefficient of the summation of Kirchhoff and complementary scatter strength from the rough tree skin and becomes  qp s  exp{ 2 (kL) 2 ( k )2 (cos s 2 cos 2 )    (k ) 2 n I qp e 4 n 1 (kL) 2  [(sin  s cos s  sin  cos  ) 2  (sin  s sin s  sin  sin  ) 2 ]} 4n n  n! (10) Page 622 Chin and Chen, Climate Change Prediction by Wireless Sensor Technology International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 620-624 ISSN 2078-2365 I qp  (cos  s  cos  ) 2 n fqp exp[ 2(k ) 2 cos  s cos  2 2  {[cos  s (cos  s  cos  )]  f Fqp (kx ,k y ) n * qp  [cos  (cos  s  cos  )]n  fqp Fqp (ksx ,ksy )]} exp[ (k ) 2 (cos  cos  s )] * 2 1 2n *  [(cos  s ) Fqp (kx ,k y )  (cos  s cos  ) n Fqp (kx ,k y ) Fqp (ksx ,ksy ) 4  (cos  cos  s ) n Fqp (kx ,k y ) Fqp (ksx ,ksy )  (cos  2 n ) Fqp (ksx ,ksy ) ] 2 * where The f Kirchhoff and the complementary field coefficients qp and Fqp are given in appendix of [6]. The measured is modified by the prediction of a pair of integral equations. To evaluate the prediction model we further apply the correlation coefficient in this paper. The correlation coefficient is a measure of how well trends in the predicted values. It is a measure of how well the predicted values from a forecast model "fit" with the real measured data. The correlation coefficient, , is a quantity that gives the quality of a least squares fitting to the original data. 2 The coefficient of determination, r , is the square of the correlation coefficient. The coefficient of determination is useful because it gives the proportion of the variance of one variable that is predictable from the other variable. The coefficient of determination is the ratio of the variation of the measured average temperature to the total variation and represents the percent of the measured average temperature that is the closest to the line of best fit. The coefficient of determination is computed as: r 2  1  (SSE / SST ) SSE   (Ti  Tˆi ) 2 (13) The SST (Total Sum of Square) measures the deviations of the measured data from their mean and is the sum of squared deviations of individual measurements from the mean: SST   (Ti  T ) 2 Ti 20 Temperature 15 10 5 0 10 20 30 Time (Hour) means the temperature measured. In (14) T means the average value of the measured temperature. The smaller sum of square error, SSE, the more reliable the predictions obtained from the model. The coefficient of determination takes on values between 0 and 1. The higher the coefficient of determination, the more available the model prediction is. IV. RESULT AND DISCUSSION Temperature 25 0 (14) In (13) T̂ means the model prediction value and Table 1 the average temperature during 24 hours among 25 days o o o time Temp.( C tim Temp.( C tim Temp.( C ) e ) e ) 1 11 9 11.82 17 15.06 2 12 10 14.01 18 14.01 3 11 11 15.06 19 11.82 4 12 12 18.7 20 12.27 5 11 13 20.04 21 11.07 6 12.01 14 22 22 12.01 7 11.07 15 20.04 23 11 8 12.27 16 18.7 24 12 Temperature (degree c) (12) Where SSE (Error Sum of Square) measures the deviations of measured data from their predicted values. In general the SSE is the sum of the squared differences between each observation and its group's mean. It can be used as a measure of variation within a cluster: From the measured data the change rate of temperature is proportional to the time. During the time closing the noon the change of temperature is large each day. The data are measured 25 days in a month. The average temperatures of 24 hours are listed in table 1. From the measured data in Fig. 3 the change rate of temperature is related with time and show the exponential relation. From table 1 the peak value of average temperature is at 14:00 instead of 12:00. The time shift may be from the forest density and the dielectric permittivity of tree skin. Fig.4 The measured average temperature v.s. hours From Fig. 4 the minimum average temperature of trees in the forest is 11oC, but the maximum average one is 22oC. The average temperature in a month is 13.83 oC and the variance of average temperature is 11.7oC. It is interesting to find the time of temperature peak is 22oC at 14:00 instead of at noon. The time shift may be caused by the dielectric coefficient of the tree skin and the tree density in the forest. From the measured data we also found the change rate of temperature is proportional to its second-order differentiation. In other Page 623 Chin and Chen, Climate Change Prediction by Wireless Sensor Technology International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 620-624 ISSN 2078-2365 words the second-order differentiation of average temperature collected is proportional to the measured average temperature. It means the average temperature has the characteristics of the exponential function distribution: (10) T  k  k1e It is interesting to find the peak value of average temperature in not at noon in a whole day, but time shift happens in Fig. 5. The measurement is taken 25 days in a month. The mean value of temperature measured is 13.83. The model of exponential function shows the temperature change rate is faster in the daytime, but slower at night. The predicted exponential equation (10) with both of k1= 10 and k2= 50 and modified by constant k= 3. The value of m is the hour on maximum average temperature. In this study the value of m is 14 in Fig. 5. After comparing the measured average temperature of the ginkgo tree with that of the model prediction, the correlation coefficient is 0.935736. The model prediction is excellent.  ( t  m) 2 / k2 V. CONCLUSION The microchange of forest temperature is a critical factor to the growth of trees. Due that the forest temperature and density will affect the global change the forest temperature microchange becomes critically important for studying the global climate change. In the paper we develop a novel model to study the forest temperature microchange during a month. We also construct a novel measurement technology by wireless sensor network. From the measurement we found the microchange of tree average temperature has an exponential function distribution 24 hours in a day. It is attractive that the peal value of tree temperature reaches at the different time interval instead of noon in a day. From the comparisons of the measured temperature with the model predictions, we found the correlation coefficient between the prediction and the measurement is 0.94. From the model analysis, the dielectric constant of tree skin and density of forest trees maybe the main factors to cause the time shift of the temperature peak. In the future we will investigate it. Fig. 5 the comparisons of model prediction with measured temperature of ginkgo tree VI. REFERENCES [1] Yoo, S.; Kim, J.; Kim, T.; Ahn, S.; Sung, J.; Kim, D. A2S: Automated agriculture system based on WSN. In ISCE 2007. IEEE International Symposium on Consumer Electronics, 2007, Irving, TX, USA, 2007. [2] Goense, D.; Thelen, J. Wireless sensor networks for precise phytophthora decision support, ASAE Annual International Meeting; Tampa, FL, USA, 2005. [3] Lea-Cox, J.D.; Kantor, G.; Anhalt, J.; Ristvey, A.; Ross, D.S. A wireless sensor network for the nursery and greenhouse industry. In Southern Nursery Association Research Conference, Vol. 52, 2007. [4] Ruiz-Garcia, L.; Barreiro, P.; Rodríguez-Bermejo, J.; Robla, J.I. Monitoring intermodal refrigerated fruit transport using sensor networks: a review. Span. J. Agric. Res., 5, 142-156, 2007. [5] Poggio, A.J., and E.K. Miller, “Integral Equation Solution of Three Dimensional Scattering Problems,” Computer Techniques for Electromagnetics, Pergamon, New York, Chapter 4, 1973. [6] Chin-Yuan Hsieh, Adrian K. Fung, Giuseppe Nesti, A. Sieber and Peter Coppe, A Further Study of the IEM Surface Scattering Model, IEEE Transactions On Geoscience and Remote Sensing Society, PP. 901-909, Vol. 35, N0. 4, July 1997 [7] Chin-Yuan Hsieh, Polarimetric Bistatic Scattering From Randomly very Rough Surfaces, Microwave and Optical Technology Letters, Vol.25, No. 4, May 20, 2000 25 20 15 10 T(Ginkgo) Data(Ginkgo) 5 0 1 3 5 7 9 11 13 15 17 19 21 23 Page 624 Chin and Chen, Climate Change Prediction by Wireless Sensor Technology 

International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 Robust Neuronal Adaptive Control for a Class of Uncertain Nonlinear Complex Dynamical Multivariable Systems Farouk Zouari LAboratoire de Recherche en Automatique (LARA) École Nationale d’Ingénieurs de Tunis (ENIT), B.P. 37, 1002 Tunis Email:zouari.farouk@gmail.com Abstract: In this paper, we proposed the development of neural adaptive controls to ensure the robustness of uncertain nonlinear multivariable systems. We used two techniques: Robust neural adaptive control and neural indirect adaptive control. The study of the stability and robustness of both techniques was performed by Lyapunov theory. To validate these techniques and discover their effectiveness, a simulation example was considered. The simulation results obtained by these two control techniques have shown the effects of disturbance compensation, good performance tracking data paths and stability control systems. Comparative studies between these two techniques show that the neural indirect adaptive control cannot mitigate the effect of disturbances compared to the robust neural adaptive control. Keywords: Neural adaptive controls; Robust neural adaptive control; Neural indirect adaptive control; Lyapunov theory; uncertain nonlinear multivariable systems. I. INTRODUCTION In recent decades, the robust neuronal adaptive control of complex nonlinear dynamic systems has been studied in several research works which we quote [1-6]. It is used in several industrial applications and particularly in the cases where we are confronted with complex nonlinear dynamics and inaccuracies due to Page 671 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 uncertainties attached to the system to be controlled. The use of the method of robustification is essential to improve the tracking performance and ensure the robustness of the closed loop system in front of the structural uncertainties and external disturbances. This control technology adds to the main control signal a supervisory signal by sliding mode or of type. Several studies of robustification of sliding mode adaptive neural control are based on the use of adaptive neural networks for modeling the process or to calculate the desired control law [7-11]. Generally, the control laws were derived from the examination of stability. Overall, the constructed command ensures stability and good tracking performance. The disadvantage of sliding mode adaptive neural control is the existence of the sign function that causes sudden and rapid changes of the control signal, which can excite the high frequency of the process and cause damage it. Many solutions have been proposed in literature in particular Lie Slotine and added a transition band around the sliding surface to transform the sign function in saturation and thus remove the abrupt changes [12]. Several research works have studied the technique like [13-16]. This technique aims at determining the tracking performances based on a criterion connecting on the one hand the norms of prosecution errors and on the other hand the desired level of disturbance attenuation. This criterion can be interpreted in the state space by obtaining a positive definite matrix unique solution of the Riccati equation. The contribution of this paper is to propose two adaptive controls neuronal structures of nonlinear dynamical multivariable systems rested on the theory of Lyapunov. The architecture and learning algorithm of these two neural adaptive control structures require the modeling of system to be controlled, that is to say the determination of its state equations using the concepts of neural networks. In this sense, we proposed linearization technique inputs- outputs of the system to be controlled based on neural networks. This technique consists in finding linear relationships between inputs and outputs of the system. The neural model of the system obtained online by this linearization technique is used to calculate the commands laws. In fact, the first proposed control structure which is the neural indirect adaptive control uses the Jacobian matrix of the neural model during the calculation of the parameters of neural controller. by cons the second proposed control structure which is the robust neural adaptive control, use the state equations derived from the neural model to determine the neuronal controller and add to main control signals supervision signals by the technique . This document is organized as follows: In Section 2 we present the proposed architecture of recurrent neural network and its learning algorithm based on Lyapunov theory, used in the calculations of the model parameters of a complex nonlinear dynamic multivariable system. In Section 3 we show the architecture of the neuronal controller and its proposed learning algorithm in the structure of neuronal indirect adaptive control. The proposed structure of robust neural adaptive control by the technique is described in Section 4. The numerical results and discussions of these two commands mentioned previously are presented in Section 5. Finally, the conclusion is given in Section 6. II. NEURAL NETWORK MODELING APPROACH Neural network modeling of a system from samples affected by noise usually requires four steps: Page 672 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 - The first step is the choice of the architecture of the neural network, that is to say the number of neurons in the input layer which is a function of past values of inputs and outputs, the number of hidden layers, the number of neurons in each hidden layer, the number of neurons in the output layer, the activation functions of each neuron and organization of these neurons between themselves [17-28]. -The second step is the normalization or the transformation performed on the data inputs-outputs to distribute them uniformly and adapt them to an acceptable level for the neural network [29-31]. All data values must be between or . a. Proposed architecture neural network: of recurrent The figure 1 shows the architecture of the neural network used during the identification phase of an uncertain and perturbed nonlinear complex dynamic multivariable system (with inputs and outputs). - The third step is learning or in other words the calculation of network parameters from samples inputsoutputs system to be identified [32-34]. - The fourth step is the validation of the neural network obtained by using the tests measurements performances criteria. The structure of this neural network is composed of three parts: two linear parts to model the behavior linear of the system and a nonlinear part to approximate the nonlinear dynamics. Page 673 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 u1  k  1 u p  k  1 u1  k  2  u1  k  nb1  u p  k  2 u p  k  nbp  f1 y1  k  1 y1  k  na1  f1 yp  k  1 f2 l1  k  1 f2 yp  k  nap  l k  n  f1 l1  k  lp k z f2 f2 y1  k  yp k Page 674 1 1 Farouk Zouar, 1 Robust cNeuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems z1 l p  k  1 l p  k  ncp  z1 z1 International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 Figure 1- Proposed architecture of the neural network y  k    y1  k  ,  with: , y p  k   is the vector of the neural network outputs at instant k .  T u  k  1  u1  k  1 , y  k    y1  k  , , u p  k  1 is the vector of the system inputs. T , yp  k  is the vector of the system outputs. U  k  2   u1  k  2  , T , u1  k  nb1  , , u j  k  2 ,  , u j  k  nbj  , (1) Y  k  1   y1  k  1 , , y1  k  na1  ,  , y j  k  naj  , , u p  k  2 , , y j  k  1 , , yp  k  1 , , u p  k  nbp  , yp  k  nap   T T such as 2  nbj , 1  j  p such as 1  naj , 1  j  p (2) Page 675 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365   k   l1  k  , , l p  k  is the vector of the second hidden layer outputs of neural model T Yr  k  1  l1  k  1 , , l1  k  nc1  ,  , l j  k  ncj  , , l j  k  1 , , l p  k  1 , , l p  k  ncp   T such as 1  ncj , 1  j  p (3) f1  x  e x  1 and f2  x  x are the activation functions of neurons. e x  1 nh the number of neurons in the first hidden layer. The coefficients of the vector of the neural model parameters w are decomposed into eight groups, formed respectively by:  w111  w1    1  w nh 1 w11nr    the weights between neurons of the input layer and neurons of the first hidden layer,  w1nh nr   w311  w3    3  w p1 w31nh     3 w pnn  the weights between neurons of the first hidden layer and neurons of the second hidden layer,  w2   11  w2    the bias of neurons in the first hidden layer,  2   w nh 1   w4   11  w4     4   w p1  the bias of neurons in the second hidden layer, Page 676 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365  w511  w5    5  w p1 w51nr     5 w pnr   w611  w6    6  w nh 1 the weights between neurons of the input layer and neurons of the second hidden layer, w61nh    the weights between neurons of the first hidden layer,  6 w nh nh   w711  w7    7  w p1 w71 p    the weights between neurons of the second hidden layer, 7  w pp   w811  w8    8  w p1 w81 p    the weights between neurons of the input layer and output layer. 8  w pp   xh  k    11  x k    the first hidden layer outputs of neural model,  h   x nh 1  k   h nr   naj   ncj   nbj  p p p p j 1 j 1 j 1 (4) The vector of the neural model parameters is as follows: w   w111 , ,w1nh nr ,w211 , w311 , ,w2 nh 1 , ,w3 pnh ,w411 , ,w5 pnr ,w611 , ,w4 p1 ,w511 , ,w6 nh nh ,w711 , ,w7 pp ,w811 , ,w8 pp  T (5) T T T T  (k)  U  k  2   , Y  k  1  , Yr  k  1     1  k  , , n  k    T r (6) The vector of the first hidden layer outputs is in the following form: Page 677 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365  xh1  k    f1  s1  k        xh  k          h  x n  k   h   f1 snh  k     (7) Such as: 1  s1  k    w 11  k     S k       1  snh  k    w nh 1  k  w11nr  k     1  k    w611  k       1  w nh nr  k   nr  k    w6 nh 1  k    w61nh  k    xh1  k  1   w211  k          w6 nh nh  k    xh nh  k  1   w2 nh 1  k    (8) The vector of neural model outputs y  k  is given by:  y  k    w7  k   1   11      7  y p  k    w p1  k  3 w71 p  k    l1  k  1   w 11  k       7 w pp  k   l p  k  1   w3 p1  k    w511  k      w5  k   p1 w31nh  k    xh1  k       3 h   w pnn  k   x nh  k    w51nr  k     1  k    w411  k    w811  k          w5 pnr  k    nr  k    w4 p1  k    w8 p1  k   w81 p  k    u1  k  1      w8 pp  k   u p  k  1  (9) Assuming that: Page 678 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365  h1  (k)    w711  k        hp  (k)    w7 p1  k      w511  k      w5  k   p1 3 w71 p  k    x1  k  1   w 11  k       7 w pp  k    xp  k  1   w3 p1  k   w51nr  k     1  k    w411  k          w5 pnr  k    nr  k    w4 p1  k    w31nh  k    xh1  k       w3 pnn  k    xh nh  k    (10) Equation (9) then becomes:  y  k    h  (k)    w8  k  11  1   1         h  (k)    w8  k    p1  y p  k    p w81 p  k    u1  k  1      w8 pp  k   u p  k  1  (11) , p,qs  max  nb1 , Equation (11) can be rewritten in the following state representation:  xi  k  1  xi 1  k  i  n j ,       (12) , mj  1, j  1, , nbp , na1 , , nap , nc1 , , ncp  , n j   j  1 * qs  1 mj   j * qs  xmj  k  1  h j  (k  qs )    w8 ji  k  qs  ui  k  qs  1 p i 1 y j  k   xn j  k  Where: n  p * qs  T   x  k    x1  k  , , xn  k    n (13) We can write equation (12) as follows: Page 679 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365   x  k  1  Ax  k   B  H  k  qs   g  k  qs  u  k  qs  1   T   y k   C x k  (14) With:   M1 0    A  0       0  A   j  1,  0 1  0 0    M j     0  0     Mj   0 n n 0    0   M p  ,p 0 1 0 0 0 0   0  1 0  (15) qs  qs Page 680 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365  0 b1 0    B   0    0    0 bp    0  B  n p  j  1, p   0       b   j 0     1   b j  qs  (16)  0 c1 0  0  C      0    0 c p    0  C  n p  j  1, , p   1   0      c  j       0   c j  qs  (17) Page 681 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365   h1  (k  qs )       H ( k  qs )      hp  (k  qs )      p H ( k  qs )    j  1, , p  h j  (k  qs )    (18)   w811  k  qs  w81 p  k  qs       g  k  qs     8   w8 p1  k  qs    w k q   pp s    g ( k  qs )  p  p   i  1, , p j  1, , p  w8ij  k  qs    (19) Define the modeling errors by: ei  k   yi  k   yi  k  i  p (20) e  k   yy  k   x  k  and the modeling errors vector of all states is defined by: (21) with: yy  k    y1  k  , , y1  k  qs  1 , , y j  k  , , y j  k  qs 1 , yp  k  , , yp  k  qs 1  T n , 1 j  p (22) Combining (14) and (21), the dynamic of modeling errors is then given by: Page 682 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365  e  k  1  Ae  k   B   H  k  qs   g  k  qs  u  k  qs  1  y  k  qs    emoo  k   C T e  k    (23) where : emoo  k   e1  k  , , e p  k  T (24) III. NORMALIZATION TECHNIQUES OF DATA There are two techniques of normalization: -Min-Max normalization: This technique performs a linear transformation on the original data so that all values are in the interval  a , b . The formula of the normalization min-max is the following:    min '  b  a   a  max   min '  with :  is the data value to normalize.  ' is the new data value after the normalization,  '   a , b .  10 (26)  is the smallest integer such as max   '   1 . where : IV. (25) LEARNING ALGORITHM OF NEURAL NETWORK In this section, we propose a theorem 1 that can be used during the learning phase of a neural network.  min and  max are respectively the minimum and the maximum of data value to normalize .     0 w(k  1)  1    p  2     yi  k   i   w k  i 1   - Normalization by decimal scaling: The data are normalized by the following formula: Theorem 1: The learning procedure of a neural network may be given by the following equation:  p   y k  0  i i ei  k   m w k  i 1  w k    i w k  i    2  2    p i 1  yi  k    2     i    w k   i 1    (27) m1 such as : (28) Page 683 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365  i  0i  0, m  (29) 1 m i 0 i  0 (30)  0 (31) 0  i i  1, p  (32) i  i  1, p  (33)   (34)  (35) p i 1 i  (36) the Euclidean norm. T designates the transpose operator. Proof: V k    e  k     2  e  k   2 i i Using the following Lyapunov function: p i 1 2 i p i 1 i 2   2 w k   2  2 w k  2 (37) ei  k   ei  k   ei  k  1 with : w k   w k   w k  1 (38) (39) Page 684 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 V  k    i  ei  k    ei  k     i  ei  k      w k    w k     w k   0 The learning procedure of the neural network is stable if: p p i 1 i 1 T 2 2 (40) V  k    i  ei  k    ei  k     i  ei  k      w k   Using the previous equation, we can write: p p i 1 i 1 2 T  w k     w k   1 2 Such as: 1  0 . (41) therefore: w k  2 p  ei  k      i  w k  i 1  2 p   e  k    T     w k     w k    i  ei  k    i  w k     1  0   i 1     (42)  e  k    w k    i  ei  k    i  i 1  w k   1  2 p  ei  k     4    i    w k   i 1   For equation (42) has a unique solution, it is necessary that: p 2 The term w k  can be written as follows: (43) p   ei  k      w k    i  ei  k      w k     1 i   w k     2 p  ei  k    2     i  w k   i 1   (44) Like: lim  i zi 1  1 m z1 i 0 (45) Page 685 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 We can write: w  k  1  m  i 1    i z  w  k  1  i 0    0 w k    i w  k  i  m i 1     0    p   ei  k   2     i  w  k   i 1       0     p  yi  k   2   i    w  k  i 1   (46) therefore :  p  e  k     i  i   ei  k   m  0  w  k   i 1   w k     i w  k  i  2 2 p   i 1 ei  k     2     i   w  k   i 1     p   y k    0  i  i  e  k    w  k    i m i 1    w k     i w  k  i  2 2   p i 1  yi  k    2     i    w  k   i 1        0 w k  11    p  2     yi  k   i  w k   i 1    p   yi  k       e  k  0 i    w k    i m i 1    w k     i w k  i   2  2    p i 1  yi  k    2     i    w k   i 1    (47) The choice of the initial synaptic weights and the bias can influence the convergence speed of the learning algorithm of neural network [35-46]. According to [47], the weights can be initialized by a random number generator with a   uniform distribution between  and  or a normal distribution N 0, 2 . - If the weights are initialized by a random number generator with a uniform distribution:  s  3 nr  2  nr  1 1   m  0    m1   (48) Page 686 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 - If the weights are initialized by a random number generator with a normal distribution:  s  1 nr  2  nr  1 1   m  0    m1   (49) where : V. s  2.29 VALIDATION TESTS OF THE NEURAL MODEL Most validation tests use a set of samples inputsoutputs which have not been used in learning. Such a test set or validation should if possible cover the same range operating as the set of training samples. The residues ei  k i  1, , p obtained from the estimated model parameters represent non-measurable disturbance presented within the system. The residues must constitute independent random sequences thus assimilating the prediction errors to white noise.    Qi  100% 1       N  k 1 yi (k )  yi (k )   1  yi (k )   N  k 1   N  N k 1  2  yi (k )     2    i  1,     Various tests called whiteness tests residues were developed to validate this property. These validation tests of a model are based on analysis of prediction errors, on the Nash-Sutcliffe criterion, on the autocorrelation of residues, on the cross-correlation function between the residues and other inputs in the system [31, 48-52]. The Nash-Sutcliffe criterion relating to on each output is given by the following relation: ,p (50) The correlation functions are: - Autocorrelation functions of the residues:   e (k)   N  e (k)   e (k   )   N  e (k)   N   Rei ei ( )  k 1  1 i    N k 1 i   1  e (k)   N  i k 1   N 1   ei (k)    k 1  N 2  N k 1 i i  i  1, , p (51) - Cross-correlation function between the residues and previous inputs: Page 687 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365   u (k)   N  u (k)   e (k   )   N  e (k)   N  Rui e j ( )  k 1    1 i  1  ui (k)   N  k 1   N   N k 1  i  ui (k)    k 1  N  2 1  N k 1 j   1  e j (k)   N  k 1   N  j   e j (k)    k 1  N 2 i  1, , p j  1, ,p (52) N is the number of samples. 1,  0 R ( )  0 , ui e j  Rei ei (  )   0,  0 (53) Ideally, if the model is validated, the results of these correlation tests and criterion Nash lead to the following results: and Qi  100% , i  1, ,p, j  1, ,p Typically, it is verified that Qi  100% and the functions R are null for the interval    20, 20 with a confidence interval 95% , that is to say:  1.96 N  R 1.96 N (54) - Average error on each output is defined as follows: AREi  1 N  ei  k i  1, N k 1 , p (55) - Mean absolute error on each output is: AAREi  1 N  ei  k  i  1, N k 1 , p (56) - Root mean square error on each output: RMSi  2 1 N ei  k   i  1,   N k 1 , p (57) - MAE (Mean Absolute Error): MEA  p N 1 ei  k   p * N   i 1 k 1 (58) The desired value of AREi , AAREi , RMSi and MEA is zero. Page 688 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 VI. NEURONAL INDIRECT ADAPTIVE CONTROL In this section, we propose a neuronal indirect adaptive control structure of a complex dynamic multivariable system (with p inputs and p outputs) and a learning algorithm of a neural controller. VII. PROPOSED STRUCTURE OF THE NEURONAL INDIRECT ADAPTIVE CONTROL In this work, the structure of the neural indirect adaptive control is given in Figure 2. The architecture On-line update of ANN by Lyapunov-Base algorithm y  k  1 rm  k  y  k  na  r k Referenc e model r  k  nc  u  k  2 u  k  nb  z1 Recurrent Neural Network u  k  1 z1 of connections of the neurons between them in the neural controller is shown in Figure 3. The real-time adjustment of the controller parameters is performed in two steps. The first step is the estimation of the model neuronal parameters of the system, using Theorem 1 and from the knowledge of several inputs-outputs couples. The second step is the online calculation of controller parameters based on Theorem 2 and the   yk  .  u  k  1    Jacobian matrix of the neural model  + - z1 System yk z1 Online calculation of the Jacobian matrix of the neural model   yk     u  k  1     w k  , H  k  , g  k  Online estimation of neural model parameters of system Page 689 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 Figure 2. Structure of the neuronal indirect adaptive control y1  k  1 y1  k  na  yp  k  1 yp  k  na  f1 r1  k  r1  k  nc  f2 rp  k  rp  k  nc  u1  k  2  u1  k  nb  u1  k  1 f1 f1 u p  k  1 f2 z1 Page 690 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable z1 u p  k  2 Systems u p  k  nb  z1 z1 International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 Figure 3- Proposed architecture of the neural controller VIII. LEARNING ALGORITHM OF NEURAL CONTROLLER The learning algorithm of neural controller of a nonlinear complex uncertain and perturbed multivariate system (with p inputs and p outputs) can use Theorem 2. Theorem 2: The learning of the neural controller can be made by the following equation: Page 691 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365     c0 c wc(k  1)  1  2   p p  ui  k  1 y k   i  2  c   c   ci   i   wc  k  wc  k  i 1 i 1    c0   ci p  i 1  yi  k  eci  k  w k   p  yi  k  2  c    ci  wc  k  i 1  2 2 ui  k  1   ci  wc  k   i 1  p     wc  k  2         ci wc  k  i  mc i 1 (59) Such as: mc  1  ci  0i  0, mc (60)  c (61) mc i 0 i 1 c  0 (62) c  0 (63) 0  ci i  1, p  (64)  ci  i  1, p  (65)  c  c (66) (67) Page 692 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365  c p i 1 i  c i  1, p   ci  0 (68) i  1, p  eci  k   ri  k   yi  k  (69) nrc  p  na  nb  nc  (70) (71) wc   wc111 , ,wc1nhc nrc ,wc 211 , ,wc 2nhc 1 , wc311 , ,wc3 pnhc ,wc 411 , ,wc 4 p1 ,wc511 , ,wc5 pnrc ,wc611 , ,wc6nhc nhc ,wc711 , wc7 pp  is the weights vector of neural controller. T nhc is the number of neurons in the hidden layer of neural controller. Proof: V k   ci  ec  k      ci  ec  k   From the following Lyapunov function: ns i 1 2 ns 2 i 1 i 2 i 2  c 2 wc  k   2 c 2 wc  k    2 nu i 1 ci 2  u  k  1  2 i (72) The adjustment parameters procedure of the neural controller is stable if: V  k     ci  eci  k    ei  k      ci  eci  k   ns ns i 1 i 1   c  wc  k   T 2  0 (73) 2  wc  k    c wc  k  2    ci  ui  k  1    2 nu 2 i 1 with : Equation (73) then becomes: Page 693 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 wc  k  2 2 ns nu  eci  k  ui  k  1  c    ci    ci  wc  k  wc  k  i 1 i 1  2     ns  eci  k    T    wc  k     c  wc  k      ci  eci  k        2  0  i 1  wc  k      If the previous equation has a unique solution, the term  2 is as follows: (74)  ec  k    ci  eci  k    i    c  wc  k    i 1  wc  k   ns 2  2 2 ns nu  eci  k  ui  k  1    ci 4  c    ci  wc  k  wc  k  i 1 i 1  (75) 2     ns   eci  k      c  wc  k      ci  eci  k      i 1  wc  k     The adjustment parameters equation of the controller neural can be written: wc  k    2 ns nu  eci  k  ui  k  1    ci 2  c    ci  wc  k  wc  k  i 1 i 1  2     (76) Like: lim  ci zi 1  1 m z1 i 0 (77) We can write: Page 694 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365  m  wc  k  1   ci zi 1  wc  k  1  i 0   eci  k      m i 1  wc  k     ci wc  k  i     2 2 ns nu    1 i eci  k  ui  k  1  2  c    ci    ci  wc  k  wc  k   i 1 i 1     c0   c  wc  k      ci  eci  k       ns (78) therefore : wc  k  1  wc  k   wc  k  1     c0 c  1  2 n nu   s eci  k  ui  k  1  2  c    ci    ci  wc  k  wc  k   i 1 i 1     ns  ec  k      c0    ci  eci  k    i    i 1   wc  k      wc  k   2 2 ns nu   eci  k  ui  k  1     ci 2  c    ci   wc  k  wc  k  i 1 i 1   2       ci wc  k  i  m i 1 If all conditions are met proper identification, the neural model outputs y  k  are good approximations of the system (79) outputs y  k  , which allow writing: yi  k  yi  k  i  1, ,p (80) Equation (79) then becomes:     c0 c wc  k  1 1  2   ns nu  2  c   c  yi  k    c ui  k  1   i i   wc  k  wc  k  i 1 i 1   (81) The calculation of the term   ns   yi  k      c c ec k          i i 0   wc  k     i 1     wc  k   2 2    ns nu  yi  k  u  k  1     ci i 2  c    ci    wc  k  wc  k  i 1 i 1    yi  k  can be determined through the neural model as follows: wc  k  2       ci wc  k  i  m i 1 Page 695 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365   y1  k     y1  k      wc  k    u1  k  1       y k    y k p  p     wc k u     1  k  1   y1  k    u1  k  1    u p  k  1   wc  k        y p  k    u p  k  1    u p  k  1   wc  k   (82)  yi  k  u j  k  1 We can write: yi  k   yi  k  1 , i  1, u j  k  1  u j  k  2  , p , j  1, ,p (83) IX. ROBUST NEURONAL ADAPTIVE CONTROL The structure of the proposed robust neural adaptive control is given in Figure 4. Control additive R  k  qs    r k ec  k  qs  K Neural controller (Recurrent Neural Network) ur  k  1   uc  k  1 u  k  1 Nonlinear multivariable system TDL Neural model (Recurrent Neural Network) H (k) yy  k  qs  y  k  qs  CT + emoo  k  qs  - g (k) Online Learning TDL Page 696 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 Figure 4. Proposed structure of robust neuronal adaptive control With: TDL : designates delays. The vectors of reference signals are: r  k    r1  k  , (84) R  k   r1  k  , , rp  k  T , r1  k  qs  1 , , rj k, , r j  k  qs  1 , , rp  k  , , rp  k  qs 1 , 1  j  p T (85) The command applied to the system is given by: uc  k  1  u  k  1  ur  k  1 (86) The architecture of the neuronal controller (Figure 5) is deduced from the architecture of the neural model (Figure 1). From equation (23), we can write: u  k  qs  1   g  k  qs   1  H  k  q   Ke  k   y  k  q  s s (87) For equation (87) is realizable, the matrix g  k  must be invertible. Such as: The matrix K   K1 , , K p   T pn is calculated so that the matrix  A BK  has all its eigenvalues strictly less than 1. The neural controller equation is then: Page 697 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 u  k  1   g  k   1  r  k   H  k   Ke  k  q  c s (88) It is also assumed:  g  k  1  G1  k  , , G p  k  T (89) r1  k  u1  k  2  rp  k  e  k  q  u1  k  nb1  u p  k  2 u p  k  nbp  y1  k  1 y1  k  na1  yp  k  1 yp  k  nap  l1  k  1 l1  k  nc1  T c f1 z1 K1 f1 f2 f1 s f2 z1 l1  k  lp k + - G1  k  + Gp  k  + - + u p  k  1 z1 Page 698 - z1 u1  k  1 Kp Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable z1 1 z Systems l  k  1 l p  k  ncp  p z1 z1 e  k  q  T c s International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 Figure 5- Proposed Architecture of the neural controller In reality, there are always modeling errors during the identification phase and disturbance which may affect the system. The equation of the dynamics of tracking errors without the additive component of control may be as follows: ec  k  1  qs    A BK  ec  k  qs   B  l  k  qs  (90) With  l  k  represents the set of disturbances and uncertainty estimates  l  k  emoo  k  (91) If we put: A0  A BK (92) Page 699 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 er  k   C T ec  k  (93) The transfer function between the term  l  z  and prosecution errors er  z  is as follows: H er ,l  z   C T  zI  A0  B 1 (94) Theorem 3: The additive component ur can be calculated using techniques based on optimization H  . It can be given by: ur  k  1    g  k   1 3 1 BT P  ec  k  qc   (95) where : H er ,l  z    2 such as  2  0 (96)  3 is a positive scalar and P a symmetric positive definite matrix verifying the following Riccati equation:  1 2  T P  A0  I    A0  I  P  CC T     PBBT P  Q  0 2    3   2  (97) With: Q  0 and  2  1 2  3 2 0 Proof: The dynamic of the system prosecution errors with the additive command ur is in the following form: Page 700 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 ec  k  1  qs    A BK  ec  k  qs   B  l  k  qs   g  k  ur  k  1   A0 ec  k  qs  B l  k  qs Bg  k  ur  k  1 (98) To verify the stability of the system, define the following candidate Lyapunov function: V k  T 1 ec  k  qs   P  ec  k  qs    2 (99) The term V  k  is given by: T T 1 1 ec  k  qs   P  ec  k  qs     ec  k  qs   P  ec  k  qs    2 2 T T 1 T   ec  k  qs    A0  I  P  P  A0  I    ec  k  qs     ec  k  qs   PB   l  k  qs g  k  ur  k  1    2  (100) V  k   Using the Riccati equation (97), we obtain: T T  1 1 1 2    PBBT P  ec  k  qs   ec  k  qs   Q  ec  k  qs     ec  k  qs     2    3  2 2  2  T T 1   ec  k  qs   CC T  ec  k  qs     ec  k  qs   PBg  k  ur  k  1 2 V  k      ec  k  qs   PB   l  k  qs   T (101) Replacing the value of ur  k  1 in (95), equation (101) becomes: V  k    T T 1 1 ec  k  qs   Q  ec  k  qs    e k  qs   PBBT P  ec  k  qs    2  c 2 2  2   T T 1 ec  k  qs   CC T  ec  k  qs     ec  k  qs   PB   l  k  qs    2 (102) Page 701 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 If we assume: a  T T 1 1 ec  k  qs   Q  ec  k  qs    e k  qs   PBBT P  ec  k  qs    2  c 2 2  2  T T 1 ec  k  qs   CC T  ec  k  qs     ec  k  qs   PB   l  k  qs    2 2 T T 1 1 1 2    ec  k  qs   Q  ec  k  qs    ec  k  qs   CC T  ec  k  qs      2   l  k  qs  2 2 2 T  ec  k  qs   PB   k  q   2  k  q 2  T 1 1 T 2      e k q PBB P e k q         l s 2 l c s c s s   2   2    2 2 2    (103) V  k   a   T 1 1 2 ec  k  qs   CC T  ec  k  qs     2   l  k  qs   2 2 According to the previous equation, we can write: 2 (104) then: V  k   V  0   k T 1 k 1 2 T    e i q CC e i q   c  s   2  2    l i  qs    c  s  2 i 0 i 0 2 As V  k   0 , in this case inequality (105) can be written as follows: (105) k 2 T 1 k 1 2 T    e i q CC e i q   l  i  qs   V  0              c s c s 2 2 i 0 2 i 0 (106) If V  0   0 , we can write:  e i  q  k i 0 k r s i 0 l s   i  q  2 2   2  2 (107) Finally, we can write: Page 702 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 H er ,l  z    2 (108) X. NUMERICAL RESULTS AND DISCUSSION: Either the nonlinear system described by the equations system:  x1  x2  2  x2   6 x2  3 x1  1 sin  u1    2 x1   1   x3  x4 2   x4  5  25  5  53 x1  5 1 sin  u1    2 x1   25  25  5 6 x2   x3  3x4   4  2 4   4 u2  2 4  2 4 u2   4u2   2    y1  x1 y  x 3  2         (109) with : u1 and u2 are the system inputs. y1 and y2 are the system outputs.  1 and  2 are noises such as: 1  max  y1   2  10 max  y1  (110) (111) Figure 6 shows the evolution of the system parameters. Page 703 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 0.155 9.5 0.15 0.145 9 Amplitude Amplitude 0.14 8.5 8 0.135 0.13 0.125 7.5 0.12 7 0.115 0 1 2 3 Time(s) 4 5 0 1 2 5 x 10 (a) 4 5 5 x 10 (b) 0.7 10.5 0.65 Amplitude 10 Amplitude 3 Time(s) 9.5 0.6 9 0.55 8.5 0.5 0 1 2 3 Time(s) (c) 4 5 0 5 x 10 1 2 3 Time(s) 4 5 5 x 10 (d) Page 704 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 16 9.8 9.6 14 9.4 12 9 Amplitude Amplitude 9.2 8.8 8.6 8.4 10 8 6 8.2 4 8 7.8 2 0 1 2 3 Time(s) 4 5 0 1 2 5 x 10 (e) 3 Time(s) 4 5 5 x 10 (f) Figure 6 – Evolution of system parameters: (a) parameter 1 ; (b) parameter  2 ; (c) parameter  3 ; (d) parameter  4 ; (e) parameter  5 ; (f) parameter  6 studied system is given in Figure 2. For online learning of this model we used Theorem 1. The maximum number of iterations is 1000 during this learning phase. 1 1 0.8 0.8 0.6 0.6 0.4 0.4 input u2 sequences input u1 sequences The figures 7 and 8 represent respectively the training sequences and assessment performance sequences (or test sequences) which are normalized by the technical of MinMax normalization. The neural model structure of the 0.2 0 -0.2 0.2 0 -0.2 -0.4 -0.4 -0.6 -0.6 -0.8 -0.8 -1 0 1 2 3 Time(s) (a) 4 -1 5 5 x 10 0 1 2 3 Time(s) 4 5 5 x 10 (b) Page 705 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems 1 1 0.8 0.8 0.6 0.6 0.4 0.4 output y2 sequences output y1 sequences International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 0.2 0 -0.2 0.2 0 -0.2 -0.4 -0.4 -0.6 -0.6 -0.8 -0.8 -1 0 1 2 3 Time(s) 4 -1 5 0 1 2 5 x 10 (c) 3 Time(s) 4 5 5 x 10 (d) Figure 7 – Training sequences: (a) control input u1 ; (b) control input u2 ; (c) desired output y1 ; (d) desired output 1 1 0.8 0.8 0.6 0.6 0.4 0.4 input u2 sequences input u1 sequences y2 0.2 0 -0.2 0.2 0 -0.2 -0.4 -0.4 -0.6 -0.6 -0.8 -0.8 -1 0 1 2 3 Time(s) (a) 4 -1 5 5 x 10 0 1 2 3 Time(s) 4 5 5 x 10 (b) Page 706 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems 1 1 0.8 0.8 0.6 0.6 0.4 0.4 output y2 sequences output y1 sequences International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 0.2 0 -0.2 0.2 0 -0.2 -0.4 -0.4 -0.6 -0.6 -0.8 -0.8 -1 0 1 2 3 Time(s) 4 -1 5 0 1 2 5 x 10 3 Time(s) (c) 4 5 5 x 10 (d) Figure 8 – Test sequences: (a) input u1 ; (b) input u2 ; (c) output y1 ; (d) output y2 Tables 1 and 2 show the obtained test results from different candidate neural models. Note that to obtain a neural model of the system studied of a satisfactory accuracy, it requires that: m  2 , nb1  2 , nb 2  2 , na1  2 , na 2  2 , nc1  1, nc 2  1 , nh  8 , 1  0.7 , 2  0.8 , 1  2 , 2  2.4 ,   1.8 ,  0  0.96 ,  1  0.025 , p  2 ,   0.4 and  2  0.015 . Tableau 1. Evolution of MEA of different candidate neural models in the case m  1 nb1 nb 2 na 1 na 2 nc1 nc 2 nh 1 2  1 2  0 1 1 1 1 1 1 1 1 0.2 0.3 0.05 0.1 0.2 0.6 0.1 0.9 0.84 1 2 2 1 1 2 2 0.6 0.5 0.1 0.7 0.9 1.2 0.2 0.8 0.76 MEA Page 707 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 1 3 3 2 2 1 3 0.3 0.4 0.2 1 0.98 0.8 0.3 0.7 0.67 1 1 2 1 2 2 4 0.15 0.7 0.3 1.1 1 0.9 0.4 0.6 0.43 2 2 2 1 3 1 5 0.3 0.3 0.4 1.3 1.5 0.7 0.5 0.5 0.38 2 3 3 2 2 1 6 0.5 0.6 0.5 1.4 1.7 1.2 0.6 0.4 0.27 2 1 2 1 2 2 7 0.6 0.7 0.6 1.5 2 1.4 0.7 0.3 0.14 2 2 2 2 1 1 8 0.7 0.8 0.4 2 2.4 1.8 0.96 0.04 0.06 3 3 3 1 2 2 9 0.7 0.9 0.5 2.1 2.5 1.8 0.9 0.1 0.12 Tableau 2. Evolution of MEA of different candidate neural models in the case m  2 nb1 nb 2 na 1 na 2 nc1 nc 2 nh 1  2 1 2  0 1 2 1 1 1 1 1 1 1 0.2 0.05 0.3 0.1 0.2 0.6 0.1 0.8 0.1 0.772 1 2 2 1 1 2 2 0.6 0.1 0.5 0.7 0.9 1.2 0.2 0.4 0.4 0.093 1 3 3 2 2 1 3 0.3 0.2 0.4 1 0.98 0.8 0.3 0.3 0.4 0.086 1 1 2 1 2 2 4 0.15 0.3 0.7 1.1 1 0.9 0.4 0.4 0.2 0.063 2 2 2 1 3 1 5 0.3 0.4 0.3 1.3 1.5 0.7 0.5 0.3 0.2 0.045 2 3 3 2 2 1 6 0.5 0.5 0.6 1.4 1.7 1.2 0.6 0.2 0.2 0.037 2 1 2 1 2 2 7 0.6 0.6 0.7 1.5 2 1.4 0.7 0.15 0.15 0.022 2 2 2 2 1 1 8 0.7 0.4 0.8 2 2.4 1.8 0.96 0.025 0.015 0.001 3 3 3 1 2 2 9 0.7 0.5 0.9 2.1 2.5 1.8 0.9 0.05 0.0.5 0.081 The autocorrelation functions of residuals and crosscorrelation functions between inputs and residues (figure 9) are within the confidence intervals, MEA validating the use of neural network of characteristics ( nb1  2 , nb 2  2 , na1  2 , na 2  2 , nc1  1 , nc 2  1 , nh  8 ) as a model of the studied system. Page 708 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems 1.2 1.2 1 1 Auto-correlation of residuals e2 Auto-correlation of residuals e1 International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 0.8 0.6 0.4 0.2 0 0.6 0.4 0.2 0 0 5 10 15 20 -0.2 25 0 10 15 lag (a) (b) 0.08 0.08 0.06 0.06 0.04 0.02 0 -0.02 -0.04 -0.06 -0.08 -25 5 lag Cross correlation function between input u1 and output residues e2 Cross correlation function between input u1 and output residues e1 -0.2 0.8 20 25 0.04 0.02 0 -0.02 -0.04 -0.06 -20 -15 -10 (c) -5 0 lag 5 10 15 20 25 -0.08 -25 -20 -15 -10 -5 0 lag 5 10 15 20 25 (d) Page 709 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems 0.08 0.08 0.06 0.06 Cross correlation function between input u2 and output residues e2 Cross correlation function between input u2 and output residues e1 International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 0.04 0.02 0 -0.02 -0.04 -0.06 -0.08 -25 0.04 0.02 0 -0.02 -0.04 -0.06 -20 -15 -10 -5 0 lag 5 10 15 20 25 (e) -0.08 -25 -20 -15 -10 -5 0 lag 5 10 15 20 25 (f) Figure 9 – Validation tests of the chosen neural model: (a) Autocorrelation function of the prediction error e1 ; (b) Autocorrelation function of the prediction error e2 ; (c) Cross-correlation function between the input u1 and the residues e1 ; (d) Cross-correlation function between the input u1 and the residues e2 ; (e) Cross-correlation function between the input u2 and the residues e1 ; (f) Cross-correlation function between the input u2 and the residues e2 . After the determination phase of a neural network capable to best approximate the desired relationships of inputs-outputs of the studied system, the proposed structures of neural adaptive control (indirect adaptive control and robust neural adaptive control by the technique H  ) are applied to this system. First, we will control the system by neural indirect adaptive control. The structure of this command is that presented in Figure 2. The architecture of the neural controller is that given by Figure 3. The online calculation procedure of the controller parameters uses the theorem 2. The maximum number of iterations is 1000 during the phase of the calculation of the parameters. The evolution of mean values of the absolute differences between reference signals and outputs of the system as a function to the parameters ( na , nb , nc and nhc ) are presented in Tables 3 and 4. Page 710 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 From the results of these evolutions, the chosen neural controller has the characteristics:  na  2 n  2  b   nc  1  nhc  7 and during learning of this controller, the chosen parameters (  c1 ,  c2 ,  c1 ,  c2 ,  c ,  c0 ,  c1 ,  c2 , m ) are as (112)   c1  0.8   c  0.9 2    c1  2.1    c2  2.7   c  1.8    c  1.7  m 2    c0  0.96  c  0.021  1  c2  0.019 follows: (113) The average value of absolute differences between reference signals and the outputs of the system is given by the following equation: VME  1 P 1 N  eci  k      p i 1  N k 1   1 P 1 N  ri  k   yi  k      p i 1  N k 1  (114) Page 711 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 Tableau 3. Evolution of the average values of absolute differences between the reference signals and the system outputs of different candidate neuronal controllers in the case m  1 na nb nc nhc  c1  c2 c  c1  c2 c  c0  c1 VME 1 1 1 2 0.4 0.2 0.4 0.1 0.2 0.7 0.1 0.9 0.0943 1 2 1 3 0.5 0.3 0.7 0.5 0.8 0.9 0.2 0.8 0.0741 2 1 2 4 0.6 0.4 1.2 1.3 0.9 1.3 0.2 0.8 0.0526 2 2 1 5 0.7 0.5 1.3 1.4 1.3 1.4 0.3 0.8 0.0431 2 1 1 6 0.8 0.7 1.4 1.5 2.1 1.6 0.8 0.2 0.0291 2 2 1 7 0.8 0.9 1.7 2.1 2.7 1.8 0.96 0.04 0.0048 3 2 1 8 0.9 1 1.8 2.5 2.9 2.1 0.82 0.18 0.0069 2 2 2 9 1.1 1.4 2.05 2.7 2.8 2.7 0.8 0.2 0.0087 2 2 1 10 1.2 1.5 2 2.8 2.9 2.9 0.7 0.3 0.0091 Tableau 4. Evolution of the average values of absolute differences between the reference signals and the system outputs of different candidate neuronal controllers in the case m  2 na nb nc nhc  c1  c2 c  c1  c2 c  c0  c1  c2 VME 1 1 1 2 0.4 0.2 0.4 0.1 0.2 0.7 0.1 0.2 0.7 0.0874 1 2 1 3 0.5 0.3 0.7 0.5 0.8 0.9 0.2 0.2 0.6 0.0532 Page 712 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 2 1 2 4 0.6 0.4 1.2 1.3 0.9 1.3 0.2 0.3 0.5 0.0214 2 2 1 5 0.7 0.5 1.3 1.4 1.3 1.4 0.3 0.4 0.4 0.0172 2 1 1 6 0.8 0.7 1.4 1.5 2.1 1.6 0.8 0.1 0.1 0.0091 2 2 1 7 0.8 0.9 1.7 2.1 2.7 1.8 0.96 0.021 0.019 0.0016 3 2 1 8 0.9 1 1.8 2.5 2.9 2.1 0.82 0.015 0.165 0.0054 2 2 2 9 1.1 1.4 2.05 2.7 2.8 2.7 0.8 0.1 0.1 0.0076 2 2 1 10 1.2 1.5 2 2.8 2.9 2.9 0.7 0.15 0.15 0.0083 The results obtained by the proposed neural indirect adaptive control applied to the system are defined in Figures 10, 11 and 12. 2 80 60 Control signal u2 applied to the system Control signal u1 applied to the system 1.5 1 0.5 40 20 0 -20 -40 0 -60 0 1 2 3 Time(s) (a) 4 5 0 5 x 10 1 2 3 Time(s) 4 5 5 x 10 (b) Page 713 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 6 Signal reference r1 System output y1 70 5 50 Signal reference r2 and System output y2 Signal reference r1 and System output y1 Signal reference r2 System output y2 60 4 3 2 40 30 20 10 0 -10 1 -20 -30 0 0 1 2 3 Time(s) 4 5 0 1 2 5 x 10 3 2 1 0 -1 -2 -3 0 1 2 3 Time(s) (e) 4 5 5 x 10 (d) Evolution of the difference between the reference signal r2 and the system output y2 Evolution of the difference between the reference signal r1 and the system output y1 (c) 3 Time(s) 4 5 30 20 10 0 -10 -20 -30 0 5 x 10 1 2 3 Time(s) 4 5 5 x 10 (f) Figure 10 – Results of neuronal indirect adaptive control where r1 and r2 are respectively a signal with random uniformly distributed amplitudes and a triangular signal: (a) control signal u1 applied to the system ; (b) control signal u2 applied to the system; (c) reference signal r1 and the system output y1 ; (d) reference signal r2 and the system output y2 ; (e) Evolution of the difference between the reference signal r1 and the system output y1 ; (f) Evolution of the difference between the reference signal r2 and the system output y2 . Page 714 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 14 0.7 12 0.6 10 Control signal u2 applied to the system Control signal u1 applied to the system 0.5 0.4 0.3 0.2 8 6 0.1 4 0 2 -0.1 0 0 1 2 3 Time(s) 4 5 0 1 2 5 x 10 (a) 3 Time(s) 4 5 5 x 10 (b) Signal reference r1 System output y1 Signal reference r2 System output y2 30 2.5 2 Signal reference r2 and System output y2 Signal reference r1 and System output y1 25 1.5 1 0.5 15 10 5 0 0 20 0 1 2 3 Time(s) (c) 4 0 5 5 x 10 1 2 3 Time(s) 4 5 5 x 10 (d) Page 715 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems 1 Evolution of the difference between the reference signal r2 and the system output y2 Evolution of the difference between the reference signal r1 and the system output y1 International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 0 1 2 3 Time(s) 4 5 6 4 2 0 -2 -4 -6 -8 0 1 2 5 x 10 (e) 3 Time(s) 4 5 5 x 10 (f) Figure 11 – Results of neuronal indirect adaptive control where r1 and r2 are respectively a sinusoidal signal and a sinusoidal signal : (a) Control signal u1 applied to the system ; (b) Control signal u2 applied to the system ; (c) reference signal r1 and the system output y1 ; (d) reference signal r2 and the system output y2 ; (e) Evolution of the difference between the reference signal r1 and the system output y1 ; (f) Evolution of the difference between the reference signal r2 and the system output y2 . 1.4 1.2 30 25 0.8 Control signal u2 applied to the system Control signal u1 applied to the system 1 0.6 0.4 0.2 20 15 10 0 5 -0.2 0 -0.4 0 1 2 3 Time(s) 4 5 0 5 x 10 1 2 3 Time(s) 4 5 5 x 10 Page 716 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 (a) (b) 60 Signal reference r2 System output y2 Signal reference r1 System output y1 5 4.5 50 Signal reference r2 and System output y2 Signal reference r1 and System output y1 4 3.5 3 2.5 2 40 30 20 1.5 10 1 0.5 0 0 0 1 2 3 Time(s) 4 0 5 2000 4000 5 x 10 6000 8000 Samples 0 -1 -2 -3 -4 -5 0 1 2 3 Time(s) 12000 14000 (d) Evolution of the difference between the reference signal r2 and the system output y2 Evolution of the difference between the reference signal r1 and the system output y1 (c) 10000 4 5 5 0 -5 -10 -15 -20 -25 -30 -35 -40 0 5 x 10 (e) 1 2 3 Time(s) 4 5 5 x 10 (f) Figure 12 – Results of neuronal indirect adaptive control where r1 and r2 are respectively a triangular signal and a triangular signal: (a) control signal u1 applied to the system; (b) Control signal u2 applied to the system; (c) reference signal r1 and the system output y1 ; (d) reference signal r2 and the system output y2 ; (e) Evolution of the difference between the reference signal r1 and the system output y1 ; (f) Evolution of the difference between the reference signal r2 and the system output y2 . Page 717 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 Secondly, the proposed robust neural adaptive control by the technique H  is applied to the system using the following procedure: - We chose:   2  0.1   3  0.02 (115) - The offline calculation of the matrix A , B , C , K , P , Q : Based on the neural identification results of the system, we have: 0 0 A  0  0 1 0 0 0  1 0 1 0 , B  0 0 0 1   0 0 0 0 0 1  0 0 , C 0 0   1 0 0 0  1  0 (116) So that all the eigenvalues of matrix A0 are less than 1:  0.002 0.05 1 0.01  K 0.007 0 0.003  0 (117) The resolution of the Riccati equation (97) gives: 10 0 0 0   0 10 0 0   P  0 0 10 0     0 0 0 10  (118) If : Page 718 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 10.2 0 0   10  10.2 21 0 0.17   Q  0 0 19 10    0.17 10 20.6  0 (119) The figures (13), (14) and (15) show the results of the robust neural adaptive control by technical H  . We notice that the specified constraint of attenuation is verified: - where r1 and r2 are respectively a signal with random uniformly distributed amplitudes and a triangular signal:  e i  N i 0 2  450.37 2  5629.82 r (120)   i  k i 0 l (121) H er ,l  z  0.08  0.1 (122) - Where r1 and r2 are respectively a sinusoidal signal and a sinusoidal signal:  e i  N i 0 2  516.23 2  5735.82 r (123)   i  k i 0 l (124) H er ,l  z  0.09  0.1 (125) Page 719 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 - Where r1 and r2 are respectively a triangular signal and a triangular signal:  e i  N i 0  325.17 2 r (126)   i  k i 0 2 l  5419.82 (127) H er ,l  z 0.06  0.1  (128) 2 Control signal u1 applied to the system 1.5 1 0.5 0 0 1 2 3 Time(s) 4 3 Time(s) 4 5 5 x 10 80 Control signal u2 applied to the system 60 40 20 0 -20 -40 -60 0 1 2 5 5 x 10 Page 720 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 (a) 6 (b) Signal reference r1 System output y1 Signal reference r1 and System output y1 5 4 3 2 1 0 0 70 1 2 3 Time(s) 4 3 Time(s) 4 5 5 x 10 Signal reference r2 System output y2 60 Signal reference r2 and System output y2 50 40 30 20 10 0 -10 -20 -30 0 1 2 (c) 5 5 x 10 (d) Page 721 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 Evolution of the difference between the reference signal r1 and the system output y1 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 Evolution of the difference between the reference signal r2 and the system output y2 0 1 2 3 Time(s) 4 3 Time(s) 4 5 5 x 10 10 8 6 4 2 0 -2 -4 -6 -8 0 1 2 (e) 5 5 x 10 (f) Figure 13 – Results of the robust neural adaptive control by technique H  where r1 and r2 are respectively a signal with random uniformly distributed amplitudes and a triangular signal : (a) control signal u1 applied to the system; (b) Control signal u2 applied to the system; (c) reference signal r1 and the system output y1 ; (d) reference signal r2 and the system output y2 ; (e) Evolution of the difference between the reference signal r1 and the system output y1 ; (f) Evolution of the difference between the reference signal r2 and the system output y2 . Page 722 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 14 0.7 12 0.6 10 Control signal u2 applied to the system Control signal u1 applied to the system 0.5 0.4 0.3 0.2 8 6 4 0.1 0 2 -0.1 0 1 2 3 Time(s) 4 0 5 0 1 2 5 x 10 (a) 3 Time(s) 4 3 Time(s) 4 5 5 x 10 (b) Signal reference r1 System output y1 Signal reference r2 System output y2 30 2.5 2 Signal reference r2 and System output y2 Signal reference r1 and System output y1 25 1.5 1 0.5 0 20 15 10 5 0 1 2 3 Time(s) (c) 4 0 5 0 1 2 5 x 10 5 5 x 10 (d) Page 723 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 Evolution of the difference between the reference signal r1 and the system output y1 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 Evolution of the difference between the reference signal r2 and the system output y2 0 1 2 3 Time(s) 4 3 Time(s) 4 5 5 x 10 10 8 6 4 2 0 -2 -4 -6 -8 0 1 2 5 5 x 10 (e) (f) Figure 14 – Results of the robust neural adaptive control by technique H  where r1 and r2 are respectively a sinusoidal signal and a sinusoidal signal : (a) Control signal u1 applied to the system; (b) Control signal u2 applied to the system; (c) reference signal r1 and system output y1 ; (d) reference signal r2 and system output y2 ; (e) Evolution of the difference between the reference signal r1 and system output y1 ; (f) Evolution of the difference between the reference signal r2 and the system output y2 . Page 724 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 1.4 30 1.2 25 0.8 Control signal u2 applied to the system Control signal u1 applied to the system 1 0.6 0.4 0.2 20 15 10 0 5 -0.2 0 -0.4 0 1 2 3 Time(s) (a) 4 5 0 1 2 5 x 10 3 Time(s) 4 5 5 x 10 (b) Page 725 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 Signal reference r1 System output y1 5 4.5 Signal reference r1 and System output y1 4 3.5 3 2.5 2 1.5 1 0.5 0 0 1 2 3 Time(s) 4 3 Time(s) 4 5 5 x 10 60 Signal reference r2 System output y2 Signal reference r2 and System output y2 50 40 30 20 10 0 0 1 2 (c) 5 5 x 10 (d) Page 726 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems Evolution of the difference between the reference signal r1 and the system output y1 International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 0 -1 -2 -3 -4 -5 Evolution of the difference between the reference signal r2 and the system output y2 0 1 2 3 Time(s) 4 5 5 x 10 5 0 -5 -10 -15 -20 -25 -30 -35 -40 0 1 2 3 Time(s) 4 5 5 x 10 (e) (f) Figure 15 – Results of the robust neural adaptive control by technique H  where r1 and r2 are respectively a triangular signal and a triangular signal : (a) Control signal u1 applied to the system; (b) Control signal u2 applied to the system; (c) Reference signal r1 and system output y1 ; (d) Reference signal r2 and system output y2 ; (e) Evolution of the difference between the reference signal r1 and the system output y1 ; (f) Evolution of the difference between the reference signal r2 and the system output y2 . Page 727 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 Based on the results of the proposed neural adaptive controls, we can conclude: - Robust neural adaptive control by technique - The control signals are bounded. uncertainties compared with neural indirect adaptive control. H  reduces the effect of disturbances and / or - Abrupt changes of system parameters involve sudden changes of the amplitudes of commands laws and the outputs of the controlled system. - The proposed neural adaptive control guarantees the stability of control structures and show robustness in the presence of parameter changes of the controlled system. XI. CONCLUSION In this work, the purpose of the command is solving problems tracking given trajectories. The principal contribution of this work lies in developing new methodologies of adaptive control based on neural network. Two techniques of neural adaptive control have been proposed, developed and tested successfully. The first technique which is indirect neural adaptive control has the advantage of being simple to the use. It uses the neural model of the system to be controlled and Lyapunov theory for make online learning of neural controller and to maintain stability of the controlled system. On the other hand, this technique risks not to mitigate the effects of disturbance and therefore the controlled system cannot follow the trajectories of references of good performance. To solve this problem, we propose a robust neural adaptive control by the technique H  . In the second technique, the control law implemented is [1] [2] [3] [4] Ge, S.S., Jing Wang, “Robust adaptive neural control for a class of perturbed strict feedback nonlinear systems,” Neural Networks, IEEE Transactions on, Vol. 13, No. 6, 2002, pp. 1409 - 1419. Ruliang Wang, Chaoyang Chen, “Robust adaptive neural control for a class of stochastic nonlinear systems,” Computational Intelligence and Security (CIS), 2010 International Conference on, Nanning, 11-14 Dec. 2010, pp. 511 - 514. Hu, Tingliang, Zhu, Jihong, Sun, Zengqi, “Robust adaptive neural control of a class of nonlinear systems,” Tsinghua Science and Technology, Vol. 12, No. 1, 2007, pp. 14 - 21. Mou Chen, Shuzhi Sam Ge, Bernard Voon Ee How, “Robust adaptive neural network control for a class of uncertain mimo the sum of two terms. Indeed, the first term is an approximation of the inputs-outputs linearization control of the system to be controlled based on neural network. On the other hand, the second term is a term of robustification. The two proposed techniques of neural adaptive control ensure the stability of control structure and present robustness in the presence of parameter changes of the system to be controlled. Robust neural adaptive control by the technique H  reduces the effect of disturbances and / or uncertainties compared with the neuronal indirect adaptive control. The robust neuronal adaptive control by the technique H  allows to the controlled system outputs to follow correctly the reference signals relative to the adaptive control neuronal indirect. References [5] [6] [7] nonlinear systems with input nonlinearities,” IEEE Transactions on Neural Networks, Vol. 21, No. 5, 2010, pp. 796-812. Xiaoou Li and Wen Yu, “Robust adaptive control using neural networks and projection,” Advances in neural networks, Vol. 3174, 2004. Y. Zuo, Y. N. Wang, Y. Zhang, Z. L. Shen, Z. S. Chen, J. Chen, Q. Y. Xie, “RBF neural network robust adaptive control for wind generator system,” MECHANIKA, Vol. 17, No. 5, 2011, pp. 557-561. Lewis, F.L., Liu, K., Yesildirek, A, “Multilayer neural net robot controller with guaranteed tracking performance,” Intelligent Control, 1993., Proceedings of the 1993 IEEE International Symposium on, Chicago, IL , USA, 25-27 Aug Page 728 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 [8] [9] [10] [11] [12] [13] 1993 , pp. 225 - 231. Yangmin Li and Yugang Liu, “Sliding Mode Adaptive Neural-network Control for Nonholonomic Mobile Modular Manipulators,” 2nd International conference on Autonomous Robots and Agents, Palmerston North, New Zealand, 13-15 Dec. 2004, pp. 511 - 514. Tairen Sun, Hailong Pei, Yongping Pan, Hongbo Zhou, Caihong Zhang, “Neural network-based sliding mode adaptive control for robot manipulators,” Neurocomputing, Vol.74, No. 14–15, 2011, pp. 2377-2384. Tsung-Chih Lin, “Based on interval type-2 fuzzy-neural network direct adaptive sliding mode control for SISO nonlinear systems,” Communications in Nonlinear Science and Numerical Simulation, Vol.15, No. 12, 2010, pp. 40844099. A. Nied, S.I. Seleme Jr., G.G. Parma, B.R. Menezes, “Online neural training algorithm with sliding mode control and adaptive learning rate,” Neurocomputing, Vol.70, No. 16-18, 2007, pp. 2687-2691. Slotine JJ. and Lie W, “Applied Nonlinear Control,” Prentice-Hall.Inc, 1991. Yi Zuo, Yaonan Wang, Xinzhi Liu, Simon X. Yang, Lihong Huang, Xiru Wu, Zengyun Wang, “Neural network robust H  tracking control strategy for robot manipulators,” Applied Mathematical Modelling, Vol. 34, No. 7, 2010, pp. 1823-1838. [14] Xingjian Jing, “An H  control approach to robust learning of feedforward neural networks,” Neural Networks, Vol.24, No. 7, 2011, pp. 759-766. [15] Miyasato, Y., “Adaptive H  formation control for EulerLagrange by utilizing neural network approximators,” American Control Conference (ACC), June 29 2011-July 1 2011, San Francisco, CA, pp. 1753 - 1758. [16] Sitthidet Vachirasricirikul, Issarachai Ngamroo, “Robust controller design of heat pump and plug-in hybrid electric vehicle for frequency control in a smart microgrid based on specified-structure mixed [17] [18] [19] [20] [21] H2 / H control technique,” Applied Energy, Vol. 88, No. 11, 2011, pp. 3860-3868. Psaltis D., Sideris A.,Yamamura, A. A., “A multilayered neural network control,” IEEE Control Systems Magazine, 1988, Vol. 8, No. 2, pp. 17-21. Baltersee J, Chambers JA, “Nonlinear adaptive prediction of speech with a pipelined recurrent neural network,” IEEE Trans Signal Process, 1998, Vol. 46, No. 8, pp. 2207–2216. Stavrakoudis DG, Theocharis JB, “Pipelined recurrent fuzzy neural networks for nonlinear adaptive speech prediction,” IEEE Trans. Systems, Man and Cybernetics, (Part B), 2007, Vol. 37, No. 5, pp. 1305-1320. Haiquan Zhao, Jiashu Zhang, “A novel adaptive nonlinear filter based pipelined feedforward second-order Volterra architecture,” IEEE Trans Signal Process, 2009, Vol. 57, No. 1, pp. 237–246. Po-Rong Chang, Jen-Tsung Hu, “Optimal nonlinear adaptive prediction and modeling of MPEG video in ATM networks using pipelined recurrent neural networks,” IEEE J Select Areas Commun, 1997, Vol. 15, No. 6, pp. 1087–1100. [22] Yih-Shen Chen, Chung-Ju Chang, Yi-Lin Hsieh, Nat. Chiao Tung Univ, Hsinchu, Taiwan, “A channel effect prediction-based power control scheme using PRNN/ERLS for uplinks in DS-CDMA cellular mobile systems,” IEEE Trans Wireless Commun, 2006, Vol. 5, No. 1, pp. 23–27. [23] Mandic DP, Chambers JA, Toward, “an optimal PRNN based nonlinear prediction,” IEEE Transactions on Neural Networks, 1999, Vol. 10, No. 6, pp. 1435– 1442. [24] Mandic DP, Chambers JA, “On the choice of parameters of the cost function in nested modular RNNs,” IEEE Transactions on Neural Networks, 2000, Vol. 11, No. 2, pp. 315–322. [25] Haiquan Zhao, Jiashu Zhang, “Pipelined Chebyshev functional link artificial recurrent neural network for nonlinear adaptive filter,” IEEE transactions on systems man and cybernetics Part B Cybernetics a publication of the IEEE Systems Man and Cybernetics Society, 2010, Vol. 40, No. 1, pp. 162–172. [26] Williams RJ, Zipser D, “A learning algorithm for continually running fully recurrent neural networks,” Neural Computation, 1989, Vol. 1, No. 2, pp. 270–280. [27] Barak A. Pearlmutter, “Dynamic recurrent neural networks,” Technical report CMU-CS-90-196, 1990. [28] B. A. Pearlmutter, “Gradient calculations for dynamic recurrent neural networks: A survey,” IEEE Transactions on Neural networks, 1995, Vol. 6, No. 5, pp. 1212 – 1228. [29] Irina Klevecka, Janis Lelis, “Pre-Processing of Input Data of Neural Networks: The Case of Forecasting Telecommunication Network Traffic,” Telektronikk, 2008, Vol. 3, No. 4, pp. 168 – 178. [30] Haykin, S., “Neural Networks - A Comprehensive Foundation,” 2nd Ed.Prentice-Hall, Upper Saddle River, NJ. 1998. [31] Jalil Asadisaghandi,Pejman Tahmasebi, “Comparative evaluation of back-propagation neural network learning algorithms and empirical correlation for prediction of oil PVT propreties in Iran oilfields,” Journal of Petroleum Science and Engineering, 2011, Vol. 78, No. 2, pp. 464-475. [32] H. Al-Duwaish, M.N. Karim, and V. Chandrasekar, “Use of multilayer feedforward neural networks in identification and control of wiener model,” IEE Proceedings, Control theory and applications, 1996, Vol. 143, No. 3, pp. 255-/258. [33] K. S. Narendra and K. Parthasarathy, “Identification and Control of Dynamical Systems Using Neural Networks,” IEEE Transactions on Neural Networks, 1990, Vol. 1, No. 1, pp. 4-27. [34] Chao-Chee Ku and Kwang Y. Lee, “Diagonal recurrent neural networks for dynamic systems control,” IEEE Transactions on Neural Networks, 1995, Vol. 6, No. 1, pp. 144-156. [35] T. Denoeux, R. Lengellé, “Initializing back propagation networks with prototypes,” Neural Networks, 1993, Vol. 6, No. 3, pp. 351-363. [36] G.P. Drago, S. Ridella, “Statistically controlled activation weight initialization (SCAWI),” Neural Networks, IEEE Transactions on, 1992, Vol. 3, No. 4, pp. 627-631. Page 729 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 671-730 ISSN 2078-2365 [37] J.-P. Martens, “A stochastically motivated random initialization of pattern classifying MLPs,” Neural Processing Letters, 1996, Vol. 3, No. 1, pp. 23-29. [38] T. Masters, “Practical Neural Network Recipes in C++,” Academic Press, Boston, 1993. [39] D. Nguyen, B. Widrow, “Improving the learning speed of 2layer neural networks by choosing initial values of the adaptive weights,” International Joint Conference on Neural Networks, San Diego, CA , USA, 1990, Vol. 3, pp. 21-26. [40] S. Osowski, “New approach to selection of initial values of weights in neural function approximation,” Electronics Letters, 1993, Vol. 29, No. 3, pp. 313-315. [41] J.F. Shepanski, “Fast learning in artificial neural systems: multilayer perceptron training using optimal estimation,” Neural Networks, IEEE International Conference on, San Diego, CA , USA, 1988, Vol. 1, pp. 465-472. [42] H. Shimodaira, “A weight value initialization method for improving learning performance of the back propagation algorithm in neural Networks,” Tools with Artificial Intelligence, Proceedings, Sixth International Conference on, New Orleans, LA, USA, 1994, pp. 672-675. [43] Y.F. Yam, T.W.S. Chow, “Determining initial weights of feedforward neural networks based on least squares method,” Neural Processing Letters, 1995, Vol. 2, No. 2, pp. 13-17. [44] Y.F. Yam, T.W.S. Chow, C. T. Leung, “A new method in determining initial weights of feedforward neural networks for training enhancement,” Neurocomputing, 1997, Vol. 16, No. 1, pp. 23-32. [45] L.F.A. Wessels, E. Barnard, “Avoiding false local minima by proper initialization of connections,” Neural Networks, IEEE Transactions on, 1992, Vol. 3, No. 6, pp. 899-905. [46] N. Weymaere, J.P. Martens, “On the initialization and optimization of multilayer perceptrons,” Neural Networks, IEEE Transactions on, 1994, Vol. 5, No. 5, pp. 738-751. [47] Jim Y. F. Yam and Tommy W. S. Chow, “A weight initialization method for improving training speed in feedforward neural network,” Neurocomputing, 2000, Vol. 30, No. 1-4, pp. 219-232. [48] J.E. Nash, J.V. Sutcliffe, “River flow forecasting through conceptual models part I — A discussion of principles,” Journal of Hydrology, 1970, Vol. 10, No. 3, pp. 282-290. [49] S.A. Billings, Q.M. Zhu, “Nonlinear model validation using correlation tests,” International Journal of Control, 1994, Vol. 60, No. 6, pp. 1107-1120. [50] S.A. Billings, H.B. Jamaluddin, S. Chen, “Properties of neural networks with applications to modelling non-linear dynamical systems,” International Journal of Control, 1992, Vol. 55, No. 1, pp. 193-224. [51] Dokla, Mahmoud, Osman, Mohammed, UAE U, “Correlation of PVT Properties for UAE Crudes,” SPE Formation Evaluation, 1992, Vol. 7, No. 1, pp. 41-46. [52] Glaso, Oistein, SINTEF/NTH Trondheim, “Generalized Pressure-Volume-Temperature Correlations,” Journal of Petroleum Technology, 1980, Vol. 32, No. 5, pp. 785-795. Page 730 Farouk Zouar, Robust Neuronal Adaptive Control for a Class Of Uncertain Nonlinear Complex Dynamical Multivariable Systems 

International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 642-649 ISSN 2078-2365 Dynamic Analysis of A Fossil-Fueled Steam Electric Power Plant Using Fuzzy PID Controller BEELA.RAJESH Department of Electrical and Electronics Engineering GIT, GITAM University Visakhapatnam, 530045, Andhra Pradesh, India Email: rajesh38@gmail.com Abstract – Boiler-steam turbine generation units are used in the power system especially for some special features of fossil-fuelled power plants. When a load disturbance occurred in the system, a frequency variation will cause a primary regulation action on generation units. The units will automatically adjust their outputs to fit for the new load demand. Variation of the governing valve position may exceed to the outlet pressure of the related boiler but boiler often has a long control time cycle after the pressure error was observed. Fuzzy PID controller is usually used to speed up the regulation procedure of boiler and to improve the stability of the steam parameters upstream of steam turbine. Model parameter identification is one of the most reliable tools to estimate the model parameters. In the proposed work, a general model of power plant with PID & fuzzy PID control system is built for power system dynamic analysis. The model responses will be compared to the model without a PID & fuzzy PID model to evaluate the impact of PID & fuzzy PID model on system frequency stability. Keywords: PID controller, Fuzzy PID control system, dynamic model, Parameter identification, power plant I. INTRODUCTION Boiler-steam turbine generation units are used in the power system for some special features of fossil-fuelled power plants. When a load disturbance occurred in the system, frequency variation will cause a primary regulation action on generation units. The units will automatically adjust their outputs to fit for the new load demand. Variation of the governing valve position may exceed to the outlet pressure of the related boiler but boiler often has a long control time cycle after the pressure error was observed. PID controller and FPID controller is usually used to speed up the regulation procedure of boiler and to improve the stability of the steam parameters upstream of steam turbine. Many researchers have studied the SMT. T. PADMAVATHI Department of Electrical and Electronics Engineering GIT, GITAM University Visakhapatnam, 530045, Andhra Pradesh, India Email: tadipadma@gmail.com mathematic models of power plant for power system dynamic analysis. According to their research, low order models for turbine units are more popular for power system dynamic analysis. According to huge test experiences, single turbine model is not without a consideration of a main stream pressure variation. Boiler model is also needed for some circumstances. Control System of boiler and the PID controller and FPID controller acting on both the boiler and turbine systems will have great impact on the pressure stability even output power of turbine units. But these control systems are not well considered in relative research .In this paper, a fossil-fuel power plant model is presented with PID controller and FPID controller power system analysis. The model parameters are identified for a turbine coal fired generation unit. The model responses are compared to the model without PID controller and FPID controller model to evaluate the impact of PID controller and FPID controller model on system frequency stability. Frequency response models have received limited treatment in the literature. The basic concept of the model derived here is based on the idea of uniform or average frequency, where synchronizing oscillations between generators are filtered out, but the average frequency behaviour is retained. The synchronizing oscillations are, taken from the simulations of reference [l].We seek to average these individual machine responses with a smooth curve that can be used to represent the average frequency for the system. Such a filtered or average frequency.. Similar and related approaches have been pursued more recently [3, 4]through work on energy functions. The basic ideas are also important in the work on system Area control simulators [5, 6],as well as the work on long term dynamics [7, 8].In addition to these resources, certain ideas have also been adopted from the work on coherency based dynamic equivalents [9, 10], as well as the work on transient energy stability analysis. A Genetic Algorithm (GA) represents a heuristic search technique based on the evolutionary ideas of natural selection and genetics. Page 642 Beela and Padmavathi, Dynamic Analysis of A Fossil-Fueled Steam Electric Power Plant Using Fuzzy PID Controller International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 642-649 ISSN 2078-2365 Although randomized, using the historical information they direct the search into the region of better performance within the search space. . In this paper, the PID and FPID controller is developed and compared with respect to their overshoot or undershoot and settling time under various operating conditions for a two area steam turbine and boiler model. II. MODELING OF STEAM TURBINE In a steam turbine the stored energy of high temperature and high pressure steam is converted into mechanical (rotating) energy, which then is converted into electrical energy in the generator. The original source of heat can be a furnace fired by fossil fuel (coal, gas, or oil) or biomass. The turbine can be either tandem compound or cross compound. In a tandem compound unit all sections are on the same shaft with a single generator, while a cross compound unit consists of two shafts each connected to a generator. The cross compound unit is operated as one unit with one set of controls. The power output from the turbine is controlled through the position of the control valves, which control the flow of steam to the turbines. The valve position is influenced by the output signal of the turbine controller. High Pressure (HP), Intermediate Pressure (IP) and Low Pressure (LP) are the different turbine sections. The turbine considered for study in this paper is reheating type .Reheating improves efficiency [8].The effects of steam chest; reheated and nonlinear characteristics of control valve are considered. The fraction of turbine power generated by intermediate section is assumed as negligible on base value Fig.2.1 steam turbine model Steam flow entered into steam turbine gs is proportional to sum of the product of governing valve position variation PGV and steam pressure variation of superheater Ps and two variations themselves. TH, TR and TL are time constants of three equivalent steam volume as high pressure volume, reheated volume and crossover volume, and Pg, Pr and Pc are average steam pressures of three volumes. Output power is a sum of output by three kinds of turbine cylinder. Power of each cylinder is considered to be proportion to its inlet steam pressure due to high pressure ratio. Relative with the rated output power the output portions of three cylinders are KH, KR and KL respectively. III. MODELING OF BOILER The modelling of a general fossil fuelled boiler. Two equivalent storage volumes and an equivalent resistant component are assumed for the steam transmission process. The output power of a steam turbine generation unit is critical to the power system analyses. It depends greatly on the steam flow gs entered into steam turbine, which may be changed rapidly due to a variation of governing valve position P GV or the super heater steam pressure PS.PGV is controlled by governing system of steam turbine while P S will be affect by PGV and/or the firing command PF controlled by a boiler governing system. The supplied steam pressure P S respond fast to the variation of governing valve position P GV but very slowly to the change in firing command P F because of huge time lag of the heat transfer and steam transmission processes. Then the relationship between pressure drop and flow is assumed to be linear for a small disturbance. Page 643 Beela and Padmavathi, Dynamic Analysis of A Fossil-Fueled Steam Electric Power Plant Using Fuzzy PID Controller International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 642-649 ISSN 2078-2365 controller (PID controller) is a generic control loop feedback mechanism controller) widely used in industrial control systems – a PID is the most commonly used feedback controller. A PID controller calculates an "error" value as the difference between a measured process variables and a desired set point. The controller attempts to minimize the error by adjusting the process control inputs. The PID controller calculation algorithm involves three separate constant parameters, and is accordingly sometimes called three-term control: the proportional, the integral and derivative values, denoted P,I, and D. Heuristically, these values can be interpreted in terms of time: P depends on the present error, I on the accumulation of past errors, and D is a prediction of future errors, based on current rate of change. The weighted sum of these three actions is used to adjust the process via a control element. Fig.3.1 Simplified boiler dynamic model Where K is the proportion gain, PD is the drum steam pressure Variation and gD is the flow rate variation discharge from the drum. Pressure in the equivalent storage volume is generally proportional to the integral of the mass flow difference between its input and output interfaces. Where gsand gW are the flow rate variation discharge from the super heater and the water wall respectively, while Ts and T D are their relative time constants. The dynamic process of the fuel feed and burring system and water walls are both considered as first-order inertia and pure delay. The effect on reheated pressure of fuel feed variation is ignored. According to the mathematic models shown before, the whole boiler system can be merged as a model. Where gQ is the release heat variation from the fired fuel or the Mass flow rate variation of the fuel. Tw and TF are the time Constants of water walls and fuel feed system respectively. The effect on reheated pressure of fuel feed variation is ignored. According to the mathematic models shown before, the whole boiler system can be merged as a model. Where gQ is the release heat variation from the fired fuel or the Mass flow rate variation of the fuel. T w and TF are the time Constants of water walls and fuel feed system respectively. IV. PID CONTROLLER One of the usual compensators that are widely used is PID controller. The combination of lead and lag compensators is used to achieve desired transient behaviour and low steady state error. The structure of this compensator that has been used in this study A proportional–integral–derivative In this tutorial we assume the controller is used in closed loop unity feedback system the variable de note the tracking error which is send to the PID controller. The controller single u form the controller to plant equal to the Proportional gain ( ) time magnitude error gain Integral gain ( ) time the integral of the time pulse the Derivative gain ( ) times the derivative of the error. Fig.4.1 block diagram of a PID controller V. FUZZY PID CONTROLLER Although it is possible to design a fuzzy logic type of PID controller by a simple modification of the conventional ones, via inserting some meaningful fuzzy logic IF- THEN rules into the control system, these approaches in general complicate the overall design and do not come up with new fuzzy PID controllers that capture the essential characteristics and nature Page 644 Beela and Padmavathi, Dynamic Analysis of A Fossil-Fueled Steam Electric Power Plant Using Fuzzy PID Controller International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 642-649 ISSN 2078-2365 of the conventional PID controllers. Besides, they generally do not have analytic formulas to use for control specification and stability analysis. The fuzzy PID controllers to be introduced below are natural extensions of their conventional versions, which preserve the linear structures of the PID controllers, with simple and conventional analytical formulas as the final results of the design. Thus, they can directly replace the conventional PID controllers in any operating control systems (plants, processes). The conventional design of PID controller was somewhat modified and a new hybrid fuzzyPID controller was designed. Instead of summation effect a mamdani based fuzzy inference system is implemented. The inputs to the mamdani based fuzzy inference system are error and change in error. The main difference is that these fuzzy PID controllers are designed by employing fuzzy logic control principles and techniques, to obtain new controllers that possess analytical formulas very similar to the conventional digital PID controllers. 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 e PB PB PB PB PB PS PS PS PS PS ZO ZO ZO ZO ZO NS NS NS NS NS NB NB NB NB NB ec PB PS ZO NS NB PB PS ZO NS NB PB PS ZO NS NB PB PS ZO NS NB PB PS ZO NS NB Kp PB PB PB PB PB PM PM PB PB PB PM PM PM PM PB PB PM PM PM PB PB PB PB PB PB Ki ZO ZO ZO ZO ZO PS PS ZO ZO ZO PS PS PS PM PM PM PM PM PM PM ZO ZO ZO ZO ZO Kd ZO PS ZO PS ZO PS PS PM PS PS ZO PS PM PS ZO ZO PS PB PS ZO ZO PS PB PS PS VI. 1. RESULTS AND DISCUSSION TWO AREA STEAM SIMULATIONS RESULTS: TURBINE MODEL Fig.5.1Structure of FUZZY Table 1 Rule base for FPID Controller Fig 6.1Frequency deviation inarea2with1%distrubancearea1 Page 645 Beela and Padmavathi, Dynamic Analysis of A Fossil-Fueled Steam Electric Power Plant Using Fuzzy PID Controller International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 642-649 ISSN 2078-2365 Fig Fig 6.2Frequency deviation inarea1with1%distrubancearea2 6.4Frequency deviation inarea1with2%distrubancearea2 In the above graphsfirstly, a step load disturbance occurs in two areas (area1&area2) with step load increasing of 0.01p.u. With the use of a PID controller with a step load disturbance of 0.01p.u rise time has been reduced to 0.33s, settling time to 9.97 and overshoot is reduced to 146.With the use of a fuzzy PID controller with a step load disturbance of 0.01p.u rise time has been reduced to 0.31s, settling time to 9.97 and overshoot is reduced to 132.6. In the above graphs firstly, a step load disturbance occurs in two areas (area1&area2) with step load increasing of 0.02p.u. With the use of a PID controller with a step load disturbance of 0.02p.u rise time has been reduced to 0.33s, settling time to 9.97 and overshoot is reduced to 142.With the use of a fuzzy PID controller with a step load disturbance of 0.02p.u rise time has been reduced to 0.3s, settling time to 9.97 and overshoot is reduced to 128.3. Fig 6.5Frequency deviation inarea2with3%distrubancearea1 Fig 6.3Frequency deviation inarea2with2%distrubancearea1 Page 646 Beela and Padmavathi, Dynamic Analysis of A Fossil-Fueled Steam Electric Power Plant Using Fuzzy PID Controller International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 642-649 ISSN 2078-2365 Fig 6.6Frequency deviation inarea2with3%distrubancearea1 Fig 6.8Frequency deviation inarea1with4%distrubancearea2 In the above graphs firstly, a step load disturbance occurs in two areas (area1&area2) with step load increasing of 0.03p.u. With the use of a PID controller with a step load disturbance of 0.03p.u rise time has been reduced to 0.32s, settling time to 9.97 and overshoot is reduced to 137.With the use of a fuzzy PID controller with a step load disturbance of 0.03p.u rise time has been reduced to 0.3s, settling time to 9.97 and overshoot is reduced to 125.2. In the above graphs firstly, a step load disturbance occurs in two areas (area1&area2) with step load increasing of 0.04p.u. With the use of a PID controller with a step load disturbance of 0.04p.u rise time has been reduced to 0.31s, settling time to 9.97 and overshoot is reduced to 132.With the use of a fuzzy PID controller with a step load disturbance of 0.03p.u rise time has been reduced to 0.3s, settling time to 9.97 and overshoot is reduced to 123. Fig 6.9Frequency deviation inarea2with5%distrubancearea1 Fig 6.7Frequency deviation inarea2with4%distrubancearea1 Page 647 Beela and Padmavathi, Dynamic Analysis of A Fossil-Fueled Steam Electric Power Plant Using Fuzzy PID Controller International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 642-649 ISSN 2078-2365 super heater steam pressure is controlled by the governing system. TW Flow rate variation TD Drum steam temperature TS Steam temperature K Proportion gain TF Firing command Fig 6.10Frequency deviation inarea2with5%distrubancearea1 In the above graphs firstly, a step load disturbance occurs in two areas (area1&area2) with step load increasing of 0.05p.u. With the use of a PID controller with a step load disturbance of 0.05p.u rise time has been reduced to 0.33s, settling time to 9.97 and overshoot is reduced to 128.With the use of a fuzzy PID controller with a step load disturbance of 0.05p.u rise time has been reduced to 0.31s, settling time to 9.97 and overshoot is reduced to 120.5. 2. SYSTEM DATA 2.1 Steam turbine model: The steam turbine is mostly large power reheat units. There are multi low pressure cylinders and even multi intermediate pressure cylinders. The intermediate pressure cylinders can be considered as for the dynamic analysis. TH High pressure temperature TR Reheat temperature TL Lower pressure temperature Table7. 1 Time constant value for turbine model TH 0.26s TR 18.5s TL 0.69s Table.2.1 Time constant values of boiler model TW 22.3s TD 46.1s TS 0.9s TF 1.2s K 1.8 2.2s Table.2.2 Parameters fortwo area steam turbine model with PID&FUZZY PID controller KP -1.5 KI -1 KD -1.5 Table.2.3 Parameters for two area steam turbine&boiler modelwith PID&FUZZYPIDcontroller KP -7.67 KI -0.99 KD -3.5 Normal operating conditions T Turbine G Governor KP Constant Power R Feedback gain TT1=0.03, TG1=0.08 TP1=20, R1=2.4 KP1=120, T12=0.545 B1=0.425, K1=1, a12=-1 2.2 Boiler model: The out power of the steam generation units is critical to the power system analyses. Its depends greatly on the steam flow entered in to the steam turbine, which may be changed rapidly due to a variation of governing valve position. And Page 648 Beela and Padmavathi, Dynamic Analysis of A Fossil-Fueled Steam Electric Power Plant Using Fuzzy PID Controller International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 642-649 ISSN 2078-2365 4. Table 7.2. Parameter Comparison of a Two area system for PID and FPID Controllers. Operating Conditions 0.01 0.02 0.03 0.04 0.05 Type of Controller PID FPID PID FPID PID FPID PID FPID PID FPID V. Rise Time 0.33 0.31 0.33 0.3 0.32 0.3 0.31 0.3 0.33 0.31 Settling Time 9.97 9.97 9.97 9.97 9.97 9.97 9.97 9.98 9.97 9.97 5. Overshoot 146 132.6 142 128.3 137 125.2 132 123 128 120.5 6. 7. 8. 9. CONCLUSIONS: 10. For supplying stable and reliable electric power, load frequency control is an important issue in power system operation and control. Automatic load frequency control is used to maintain the generator power output and frequency within the prescribed values. In this work the two area load frequency controller is considered. The simulated study shows the frequency response and steady state response of two area systems by using fuzzyPID& PID controller. The conventional fuzzyPID is compared withPIDcontroller; two similar areas are given with a disturbance of 0.05p.u. The simulation study shows that the stability of the system improved the frequency response and less settling time and steady state responses. Hence from the results we conclude that the fuzzy PID controller is said to be better compensating then conventional PID controller 11. 12. 13. 14. VI. REFERENCES 1. Power Prabhakundur. 2. 3. Power System Engineering byNagrath Kothari IEEE Committee Report, "Dynamic Models for Steam and Hydro Turbines in Power System Studies", IEEE system Stability and Control Trans Power Apparatus & Systems, Vol. 92, No. 6, 1973, pp. 1904- 1915. L. N. Bize, J. D. Hurley, "Frequency Control Considerations for Modem Steam and Combustion Turbines", IEEE Engineering Society Winter Meeting, 1999, pp. 548-553. Dai Yipping, Zhao Ting, TianYunfeng, Gao Lin, "Research on the influence of primary frequency control distribution on power system security and stability", Second IEEE Conference on Industrial Electronics and Applications, 2007, pp.222-226. P. M. Anderson, M. Mirheydar, "A low-order system frequency response model", IEEE Transaction on Power Systems, Vol. 5, No. 3, 1990, pp. 720-729 K. J. Astrom, K.Eklund, "A simplified non-linear model of a drum boiler-turbine unit", International Journal of Control, Vol. 16, No. I, 1972,pp. 145-169. de Mello, F.P., "Dynamic models for fossil fuelled steam units in power system studies", IEEE Transactions on Power Systems, Vol. 6, No. 2,1991, pp. 753-761. Dai Yipping, Zhao Ting, TianYunfeng, Gao Lin, "Research on the primary frequency control characteristics of generators in power system“,Second IEEE Conference on Industrial Electronics and Applications,2007, pp. 569-574. QH WU, "Learning coordinated control of power systems using interconnected learning automata", Electrical power & Energy systems, Vol. 17, No. 3, 1995, pp.91-99. Gao Lin, Dai Vi-ping, Xia lun-rong, "A New Framework for Power System Identification Based on an Improved Genetic Algorithm", 2009 4th IEEE Conference on Industrial Electronics and Applications, May 25-27,2009, Xi'an, China. Gao Lin, Dai Vi-ping, Xia Jun-rong, "Parameter Identification of Hydro Generation System with Fluid Transients Based on Improved Genetic Algorithm", 2009 Fifth International Conference on Natural Computation, 14-16 August 2009, Tianjin, China H.R.Berenji, Fuzzy logic controllers, in: R.R. Yager,L.A.Zadeh(Eds.), An Introduction to Fuzzy Logic Application in Intelligent System, Kluwer Academic Publisher, Boston, MA, 1992. W.R.HwangandS.ZeinSabatto.“FuzzyControllerDesignUs ing Genetic Algorithm ” in Engineering newcentury, Proceeding, IEEE, April 1997 by Page 649 Beela and Padmavathi, Dynamic Analysis of A Fossil-Fueled Steam Electric Power Plant Using Fuzzy PID Controller 

International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 764-768 ISSN 2078-2365 Solid State Lighting Reliability from Failure Mechanisms Perspective: A Review of Related Literature Shailesh K R, Ciji Pearl Kurian, Savitha G Kini Department of Electrical & Electronics Engineering, Manipal Institute of Technology, Manipal University – 576104, India shailesh9348@rediffmail.com ABSTRACT - Remarkable long-life makes LED lighting systems a long-term investment, and great energy and maintenance savings easily give good reason for the higher initial cost. LEDs are similar in construction to microelectronics devices, but there are functional requirements, materials, and interfaces in LEDs that make their failure modes and mechanisms distinctive. Over the last few years, considerable effort has gone into the study of the failure mechanisms and reliability of Solid State lighting systems (SSL). Although still very incomplete, our knowledge of the reliability issues relevant to SSL is growing. This paper provides an overview of SSL failure modes and mechanisms that are commonly encountered. It focuses on the reliability issues of LED devices. KEYWORDS: Solid State Lighting, LED Reliability, SSL reliability, SSL Failure modes, SSL failure mechanisms 1. INTRODUCTION Remarkable long-life makes LED lighting systems a long-term investment, and great energy and maintenance savings easily give good reason for the higher initial cost. All LED systems do not perform equally over their years of operation. Substandard quality products can prematurely fail or degrade in light output far below initial claims - so much so that they fail to provide the value initially promised. The construction of LEDs is somewhat similar to other semiconductor devices, but their applications and construction make their failure modes and mechanisms distinctive. Hindrance to the large scale adoption of LEDs in traditional applications is the lack of information available on their reliability. Another obstacle is the lack of worldwide accepted standards, because all commercial properties of an LED lighting system, such as luminous flux output, chromaticity, and lifetime, are functions of the junction temperature. All LED systems do not perform equally over their years of operation. Substandard quality products can prematurely fail or degrade in light output far below initial claims - so much so that they fail to provide the value initially promised. Accelerated Life testing in a short time can predict the life characteristics of LED products under the conditions of normal stress; it is the effective way of the reliability evaluation of LED lighting products for long-term use. The literature review presented in this paper helps LED lighting designers and LED product manufacturers to understand LED failure mechanisms and reliability thus helping them to design efficient LED lighting products. 2. LED PRINCIPLE OF WORKING The LED-chip is the main component of the LED device. This chip is a semiconductor that generates light in a PN-junction by electron p-hole recombination. The active region in the LED-chip is a complex structure of epitaxial layers. For different colours differential material-combinations are used: InAlGaP - red, InGaN - blue, GaAlAs - IR, AlGaN UV. The material and the quality of the epitaxial layers effectively determine the efficiency factor of the generation of light. Page 764 Shailesh et. Al., Solid State Lighting Reliability from Failure Mechanisms Perspective: A Review of Related Literature International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 764-768 ISSN 2078-2365 The principal parameters for the function are forward driving current IF and junction temperature Tj within the active layer, influencing both the power consumption and the colour, which significantly determine the lifetime as well. LEDs are commercially available in different technologies. Critical and significant criteria are a stable current path through bonding- solder- and glued connection, an appropriate heat sink for sufficient cooling of the chip by a good thermal and a high extraction of light from the LED by optical elements and areas of reflection. LEDs are encapsulated in general with transparent material like silicone or epoxy. temperature within LED and reducing the life of LEDs [9]. Under Electromigration, high drive currents or excessive current density can cause contact migration between the electrical contact and the surface of the LED die, which causes short circuit [10].Researchers have reported electromigration of contact metals occurrence along crystalline defects [11]. Proper thermal management and improvement of thermal conductivities of interface materials must be improved. As it is seen that contact resistances of interface materials is responsible for overall thermal resistance [8]. 3. CAUSES FOR LED FAILURE 4.3. Metal and dopant diffusion LEDs fail as there is a gradual lowering of light output and loss of optical efficiency due to aging. Catastrophic failures, however rare, can occur as well. LED failure modes can be broadly classified as Semiconductor-related and Packaging-related. 4. SEMICONDUCTOR RELATED FAILURES 4.1. Nucleation and growth of dislocations In this type of failure degradation happens in the active region where the radiative recombination occurs [1]. This type of failure happens, if there is an existing defect in the crystal and this defect is accelerated by heat or high current density or emitted light [2-5]. GaAs and AlGaAs are more susceptible to this mechanism than GaAsP and InP. Due to different properties of the active regions, GaN and InGaN are almost insensitive to this kind of defect [6, 7]. Ionizing radiations are also responsible for defect creation. Future research should focus on improved internal thermal management handling of thermal resistance from junction to the package to decrease the formation of crystal defects and dislocation movements caused by high-current-induced thermal effects and high ambient temperature [8]. 4.2. Electromigration It caused by high current density can move atoms out of the active regions, leading to emergence of dislocations and point defects, acting as nonradiative recombination centers and producing heat instead of light. Improperly designed LEDs may develop areas of uneven thermal resistances leading to current crowding, causing thermal runaway resulting in increasing Movement of metal atoms from the electrodes into the active region is caused by high electrical currents or voltages at high temperatures can move metal atoms this is metal diffusion [12, 13]. In some cases, especially with GaN/InGaN diodes, a barrier metal layer is used to hinder the electromigration effects [14]. Some materials, notably indium tin oxide and silver, are subject to electromigration which causes leakage current and non radiative recombination along the chip edges. It is frequently observed in LEDs diffusion of dopants into the active region during operation can cause reduction in light output [15]. The main reason for light degradation is current density, temperature, and current distribution, which causes an increase in series resistance [16-19]. LEDs with low dopant concentration in active region degrade most rapidly. Contaminants like oxygen can be intentionally introduced to form complexes to prevent doping migration [20]. 4.4. Cracking of Die Severe thermal shocks can cause breaking of dies of LEDs. Due to differences in material properties, LED packages can be subjected to mechanical stress when a high drive current is applied or when high ambient temperature conditions are suddenly applied. The high electrical stress and extreme thermal shock are the causes of die cracking [21]. It is necessary to control die cracking by fine-tuning thermal expansion coefficients between the substrate and epitaxial layers. The growth of the optimal medium layer between the substrate and the epitaxial layer is a important breakthrough to prevent cracking of die [22]. It is reported that cases light output and electrical degradation were due to die cracking. The way in Page 765 Shailesh et. Al., Solid State Lighting Reliability from Failure Mechanisms Perspective: A Review of Related Literature International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 764-768 ISSN 2078-2365 which die are made has a very critical impact on their cracking. Early defects caused by the sawing or grinding process may act as a starting point for die cracking [23, 24]. 5. PACKAGE RELATED FAILURES 5.1. Epoxy / Encapsulant degradation Prolonged exposure to light from LEDs can cause epoxy materials to be degraded [25, 26]. Yellowing of epoxies is due to prolonged exposure to UV light. This type of discoloration results in a reduction in the transparency of the encapsulant and causes a decrease in LED light output [27]. Further, it has been demonstrated that degradation and the associated yellowing increases exponentially with exposure energy. The thermal effects associated with excessive junction temperature also play a role in encapsulant yellowing [27, 28]. Yellowing is also due to a combination of ambient temperature and LED self-heating. It is found that that a temperature of around 150oC was sufficient to change the transparency of the epoxy, causing the attenuation of the light output of LEDs [29]. While phosphors are a necessary component for producing white light, their presence causes a decrease in reliability [28]. 5.2. Thermal stress Sudden failures are most often caused by thermal stresses and shocks [21]. Researchers have found number of cracks introduced from thermal expansion in the centre of the lens surface and on the inside of the polymer encapsulation when high power LED samples aged at different temperatures [30].Prolonged exposure to high condensing moisture causes cloudiness of the epoxy lenses in LEDs due to hygro-mechanical stresses [31]. 5.3. Phosphor degradation The different phosphors used in white LEDs tend to degrade with heat and age, but at different rates causing changes in the produced light colour. The driving forces are high drive current and excessive junction temperature, which are attributed to increases in temperature of the inside of the package [10] there by efficiency of the phosphor is degraded when the temperature rises. 5.4. Encapsulant carbonization Studies indicate that carbonization of the plastic encapsulation material on the diode surface leads to the formation of a conductive path across the LED and subsequently to the destruction of the diode itself [10, 35]. Carbonisation is responsible for light output degradation. Carbonization of the encapsulant decreases the encapsulant’s insulation resistance, significantly inhibiting its ability to provide electrical insulation between adjacent bond wires and leads [35]. The loss in insulation resistance at temperatures above threshold temperature can initiate a thermal runaway process leading to carbonization of the encapsulant. In this process, the fusing of the bond wires at high current causes the current to be shunted through the plastic, leading to joule heating of the plastic [36]. Under carbonization of the encapsulant there will be light output degradation. 5.5. Other failures Further reliability of solder interconnects in a LED package is influenced by environmental loads, solder material properties, and the intermetallics formed within the solder and the metal surfaces where the solder is bonded [32, 33]. The reliability of the interconnects between packages and circuit boards connections depends on the magnitude of the temperature swing, electrical power of LED packages and board design [34]. Higher electrical power in LEDs accelerated the rate of interconnect failures at solder joints. Using an active cooling device improved the cycles to failure and made them longer than did passive cooling methods [34]. Delamination happens when repeated cyclic stresses can cause the material layers of LED packages to separate, causing significant loss of mechanical toughness. Delamination can either occur between the die and silicone encapsulant, between the encapsulant and packaging lead frame, or between the LED die and die attach [8]. Delamination causes decreased light output. Delamination increases the thermal resistance of the delamination layer leading to increased junction temperature, which also affects many other types of failures and eventually decreasing the life of LEDs. Interface contamination during the LED manufacturing can result in poor bonding of interfaces, which can initiate delamination. Page 766 Shailesh et. Al., Solid State Lighting Reliability from Failure Mechanisms Perspective: A Review of Related Literature International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 764-768 ISSN 2078-2365 6. CONCLUSIONS As LED prices fall, designers are increasingly using them in their product designs, especially for lighting. LEDs have a reputation for being tireless workers that that never need replacing and require little payment in terms of power consumption. The main LED failure mechanisms are mechanical and thermal in nature. They involve thermal cycles, thermal shock, and LEDs operating at high temperatures so the wire bond ages. As the metal oxidizes and becomes brittle over time, the likelihood of an LED failure increases. Another cause of LED open circuits is electrostatic discharges. Better understanding of the causes responsible for failures in LEDs with respect to improving material properties and fabrication technology is the need of the hour. A deeper understanding of various process variables and associated environments critical for LED quality must form part of LED reliability studies. Failure analysis of LEDs has been performed through conventional microelectronics failure analysis approaches and off-line analysis techniques. There is a need to develop advanced failure analysis techniques for LEDs. [7] [8] [9] [10] [11] [12] Collaboration between standards bodies and professional societies is required to arrive at internationally accepted standards to ensure a fair comparison of published performance and reliability data. Reliability study of complete LED luminaires is the need of the hour. [13] [14] REFERENCES [1] [2] [3] [4] [5] [6] Uddin A, Wei AC, Andersson TG. “Study of degradation mechanism of blue light emitting diodes”. Thin Solid Films 2005;483:378–81. Tharian J. “Degradation- and failure mode analysis of III-V nitride devices”, IPFA 2007. 14th international symposium; 2007. p. 284–7. Hoang T, LeMinh P, Holleman J, Schmitz J. “The effect of dislocation loops on the light emission of silicon LEDs”. 35th European solidstate device research conference; 2005. p. 359– 62. Lu G, Yang S, Huang Y. “Analysis on failure modes and mechanisms of LED”. 8th ICRMS; 2009. p. 1237–41. Yanagisawa T. “The degradation of GaAlAs red light-emitting diodes under continuous and lowspeed pulse operations”. Microelectronics Reliability 1998; 38: 1627–30. Sugiura L. “Comparison of degradation caused [15] [16] [17] [18] [19] [20] by dislocation motion in compound semiconductor light-emitting devices”. Applied Physics Letters 1997; 70: 1317–9. Sugiura L. “Dislocation motion in GaN lightemitting devices and its effect on device lifetime”. Journal of Applied Physics 1997;81:1633–8 M.-H. Chang et al. “Light emitting diodes reliability review”, Microelectronics Reliability, 2012, vol: 52,p.762–782 Arnold J. “When the light goes out: LED failure mode and mechanisms”. DfR solutions; 2004. p. 1–4. Barton DL, Osinski M, Perlin P, Eliseev PG, Lee J. “Single-quantum well InGaN green light emitting diode degradation under high electrical stress”. Microelectronics Reliability, 1999;39:1219–27. Barton DL, Zeller J, Phillips BS, Chiu P-C, Askar S, Lee D-S, et al. “Degradation of blue AlGaN/InGaN/GaN LEDs subjected to high current pulses”. 33rd Annual Proceedings of IEEE international Reliability Physics Symposium; 1995. p.191–9. Ott M.“Capabilities and reliability of LEDs and laser diodes”. International NASA Parts and Packaging Publication; 1996. p. 1–7. Hwang N, Naidu PSR, Trigg A. “Failure analysis of plastic packaged optocoupler light emitting diodes”. 5th conference EPTC; 2003. p. 346–9. Mobile Reference, “Electronics Quick Study Guide for Smartphones and Mobile Devices”, 2007 Pavesi M, Manfredi M et al. “Optical evidence of an electro thermal degradation of InGaN-based light-emitting diodes during electrical stress”. Applied Physics Letters 2004;84:3403–5. Meneghesso G, Levada S, Zanoni E. “Failure mechanisms of GaN-based LEDs related with instabilities in doping profile and deep levels”. IEEE 42nd Annual International Reliability Physics Symposium; 2004. p. 474–8. Kozodoy P, DenBaars SP, Mishra UK. “Depletion region effects in Mg-doped GaN”. Journal of Applied Physics, 2000;87:770–5. Meneghini M, Trevisanello L-R, Levada S, Meneghesso G, Tamiazzo G, Zanoni E, et al. “Failure mechanisms of gallium nitride LEDs related with passivation”. IEEE IEDM Technical Digest 2005:1009–12. ASM International, “Microelectronics Failure Analysis Desk Reference”, 2011 Altieri-Weimar P, Jaeger A, Lutz T, Stauss P, Page 767 Shailesh et. Al., Solid State Lighting Reliability from Failure Mechanisms Perspective: A Review of Related Literature International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 764-768 ISSN 2078-2365 [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] Streubel K, Thonke K, et al. “Influence of doping on the reliability of AlGaInP LEDs”. Journal of Materials Science: Materials in Electronics 2008;19:S338–41. Lumileds. Luxeon reliability, Reliability Datasheet RD25, Philips Lumileds; 2006. Barton DL, Osinski M, Perlin P, Helms CJ, Berg NH. “Life tests and failure mechanisms of GaN/AlGaN/InGaN light emitting diodes”. IEEE 35th Annual Proceedings of Reliability Physics Symposium; 1997. p. 276–81. Wu JD, Huang CY, Liao CC. “Fracture strength characterization and failure analysis of silicon dies”. Microelectronics Reliability 2003;43:269– 77. Iksan H, Lin K-L, Hsieh J. “Fracture analysis on die crack failure”. IMAPS Taiwan; 2001. p. 35– 43. Narendran N, Gu Y, Freyssinier JP, Yu H, Deng L. “Solid-state lighting: failure analysis of white LEDs”. Journal of Crystal Growth 2004;268:449–56. Torikai A, Hasegawa H. “Accelerated photodegradation of poly(vinyl chloride)”. Polymer Degradation and Stability,1999;63:441– 5. Zhang Q, Mu X, Wang K, Gan Z, Luo X, Liu S. “Dynamic mechanical properties of the transient silicone resin for high power LED packaging”. ICEPT-HDP; 2008. p. 1–4. Meneghini M, Trevisanello L-R, Meneghesso G, Zanoni E. “A review on the reliability of GaNbased LEDs”. IEEE Transactions on Device and Materials Reliability, 2008;8: 323–31. Barton DL, Osinski M. “Degradation mechanisms in GaN/AlGaN/InGaN LEDs and LDs”. Proceedings of the 10th SIMC-X; 1998. p. 259–62. Hsu Y-C, Lin Y-K, Chen M-H, Tsai C-C, Kuang J-H, Huang S-B, et al. “Failure mechanisms associated with lens shape of high-power LED modules in aging test”. IEEE Transaction on Electronics Devices 2008;55:570–1. Hewlett Packard. “Reliability of precision optical performance AlInGaP LED lamps in traffic signals and variable message signs”. Application Brief I-004; 1997. Choubey A, Yu H, Osterman M, Pecht M, Yun F, Yonghong L, et al. “Intermetallics characterization of lead-free solder joints under isothermal aging”. Journal of Electronic Materials, 2008;37: 1130–8. Li GY, Chen BL. “Formation and growth [34] [35] kinetics of interfacial intermetallics in Pb-free solder joint”. IEEE Transactions on Component Packing Technology 2003;26:651–8 Chang M-H, Das D, Lee SW, Pecht M. “Concerns with interconnect reliability assessment of high power light emitting diodes (LEDs)”. SMTA China south technical conference; 2010. p. 63–9. McCluskey P, Mensah K, O’Connor C, Lilie F, Gallo A, Pink J. “Reliability of commercial plastic encapsulated microelectronics at temperatures from 125oC to 300oC”. Proceedings of HITEN 1999; 1999. p. 155–62. Page 768 Shailesh et. Al., Solid State Lighting Reliability from Failure Mechanisms Perspective: A Review of Related Literature 

International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 776-783 ISSN 2078-2365 Grid Connected DFIG With Efficient Rotor Power Flow Control Under Sub & Super Synchronous Modes of Operation D.Srinivasa Rao T.Ajaykumar EEE Department Gudlavalleru Engineering College, Gudlavalleru Andhra Pradesh, INDIA E-Mail:dsrinivasarao1993@yahoo.com EEE Department Gudlavalleru Engineering College, Gudlavalleru Andhra Pradesh, INDIA E-Mail:tajay.gec@gmail.com Abstract—To harness the wind power efficiently the most reliable system in the present era is grid connected Doubly fed induction generator (DFIG). The DFIG brings out the advantage of utilizing the turns ratio of the machine and hence the converter does not need to be rated for the machine’s full rated power. Depending on wind speed, a DFIG based variable speed wind turbine is capable of operating in sub-synchronous or super-synchronous mode of operation using power electronic converters. The power flow in the rotor circuit is controlled for maintaining the stator power constant by effecting rotor voltage through IGBT in sub-synchronous mode and in the case of supersynchronous mode it is controlled by current sequence through LCI. The operation of the proposed scheme is illustrated in different operating conditions i.e. above and below synchronous speeds using computer simulations. Keywords— DFIG, Sub&Super synchronous, Commutated Inverter (LCI), Sinusoidal PWM Inverter. I. W Line INTRODUCTION ind energy has become one of the most important and promising sources of renewable energy. With increased penetration of wind power into electrical grids, Doubly-Fed Induction Generator (DFIG) based wind turbines are largely deployed due to their variable speed feature and hence influencing system dynamics. This has created an interest in developing suitable models for DFIG to be integrated into power system studies. In standalone induction generator, both the terminal voltage and frequency will vary with variation in wind speed and load and an excitation capacitor will be required. In grid connected induction generator, control of the terminal voltage and frequency under change in load and wind speed, is possible and reactive power can be supplied by the grid. With this DFIG based Variable-speed wind turbines, an increased energy capture, improved power quality and reduced mechanical stress on the wind turbine. It consists of a wound rotor induction machine with slip rings, and power electronic converters between the rotor slip-rings and the grid. In this paper how we can obtain constant power for variable wind speeds under sub & super synchronous speed operation of a DFIG is investigated. The stator of DFIG is directly connected to the grid while the rotor fed at variable frequency through converter cascade (AC/DC/AC) via slip rings and brushes to allow the DFIG to operate at variable wind speeds in response to changing wind speeds. Both the stator and rotor windings are able to supply power to the grid. The direction of the power flow in the rotor circuit depends on the variation of the wind speed. The power electronic converters control both the direction and magnitude of the power flow of the machine. In sub-synchronous mode, the converter feeds the rotor windings from the grid, whereas the rotor supplies power to the grid in super-synchronous mode of operation. To ensure variable speed operation, and maintain the stator power constant both converters need to be controlled under sub- synchronous and super-synchronous modes of operation [3]. Most, if not all, of the published papers on the application of DFIG for wind energy conversion systems using force commutated inverters in the rotor circuit and d-q axis control for maintain stator power is constant. However, in this paper another approach is used which is the power flow approach and a very simple control technique by employing line commutated SCR inverter in the rotor circuit of the DFIG. In this approach the inter relations among the Page 776 Srinivasa and Ajaykumar, Grid Connected DFIG With Efficient Rotor Power Flow Control Under Sub & Super Synchronous Modes of Operation International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 776-783 ISSN 2078-2365 rotor power (slip power sPs), the air gap power Ps and the mechanical power Pm are used for analysis of DFIG based wind energy conversion system. This paper is organized as follows. In section II Power flow in DFIG wind energy conversion system and steady state model of DFIG are described. The operation of the open and closed loop systems of the proposed scheme employing sub-synchronous and super-synchronous modes by using power electronic converters for the grid interface has been analyzed in section III. And the development of simulation models of the proposed scheme along with simulation results is presented in section IV. Finally main observations are concluded in section V. II. parameters. Fig.2 illustrates the standard per-phase equivalent circuit of DFIG in which rotor circuit parameters are referred to the stator frequency, so that all machine reactances are determined at supply frequency. Fig.2. Per-phase equivalent circuit of a DFIG POWER FLOW & STEADY STATE MODEL OF DFIG A. Power flow in DFIG DFIG can be operated in two modes of operation namely; sub-synchronous and super-synchronous mode depending on the rotor speed below and above the synchronous speed. The power flowing in the rotor of a doubly fed induction machine (i.e. of the wound rotor type) has three components. These are a) the electromagnetic power transferred between the stator and the rotor through the air gap which is known as the air gap power Ps; b) the mechanical power Pm transferred between the rotor and shaft; c) the slip power Pr which is transferred between the rotor and any external source or load (e.g. a converter) through the rotor slip-rings. These three components of rotor power are interrelated, under sub- and super-synchronous modes of operation, as shown in figure.1 When machine is doubly-fed, the per unit power into the rotor circuit comes from two sources Pr, in1 = Re ([V2'(I2')*]) (1) Pr, in2 = T (ωr/ωb) = T (1-S) And (2) Where (*) denotes the complex conjugate operator. Since the machine is a generator, positive ‘T ‘denotes generator operation. The power lost in the rotor circuit is 2 Pr, loss= │I2'│ Rr' (3) The power output of the circuit is * Pr, out= Re [E (I2') ] (4) Conservation of power requires that Pr, in1+Pr, in2 = Pr, loss+ Pr, out (5) So that * * 2 Re [V2'(I2') ] + T (1-S) = Re [E (I2') ] +│Is│ Rr' Or * * 2 T (1-S) = Re [E (I2') ] - Re [V2'(I2') ] +│Is│ Rr' But (8) ̅ = –I2' + j (6) (7) Xlr' Substituting Eq. (8) into Eq.(7) , Fig.1. Power flow in DFIG wind energy conversion system B. Steady State Model The standard steady-state per-phase equivalent circuit can be utilized for assessing the performance of doubly fed induction machine subject to the usual assumptions of a three-phase balanced supply, fixed rotor speed, and constant machine T (1-S) = Re -1 V2'(I2')* +│I2'│2 Rr' 1- (9) Or T (1-S) = Re V2'(I2') * - │I2'│2 Rr' (10) Page 777 Srinivasa and Ajaykumar, Grid Connected DFIG With Efficient Rotor Power Flow Control Under Sub & Super Synchronous Modes of Operation International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 776-783 ISSN 2078-2365 Having obtained the rotor current from Eq. (17) it is now possible to obtain the air gap voltage E from Eq. (8). The stator current can then be found from, Cancelling out the (1-s) term (I2') T=Re * -│I2'│ 2 This resulting equation represents the basic torque equation for a doubly fed induction generator. Solution of Eq.11 in terms of the rotor current has been developed by Smith et. al [4]. Expanding Eq. (11), T= ' ' I2, re + ' 2 I2, im - (I2, re ) -(I2, im')2 (12) In general, the phase position of the rotor voltage is typically defined as its relative phase position with respect to the stator terminal voltage V1. Hence, and can be assumed to be known or specified quantities. Assuming that T and S are also specified, Eq. (12) can be solved for the currents by also assuming that their ratio (power factor) is specified. An alternative approach to solving Eq. 12 is to assume that the phase position of the rotor current is known rather than the rotor voltage. In this case, assuming the real part of the stator current as reference, ' I2, im = 0 (13) And ' I2, re = I2 ' (14) Eq. (12) becomes ' '2 I2 – (I2 ) (15) ' Which is simply a quadratic in terms of I2 .Upon solving Eq. (15) T= I 2' = √ (16) Or I2' = √( ) I 1 = I 2' – E (11) (17) The voltage can also be written as V2'cosΦ2 where Φ2 represents the phase angle of the rotor terminal voltage V2' with respect to the rotor input current I2'. Hence the rotor current I2' can be determined as a function of slip for any desired torque and specified value of rotor voltage and phase. (18) The stator voltage can then be obtained by the stator loop equation V1 = E-I1 (Rs+jXls) (19) In general the voltage obtained will not be identical to the available terminal voltage except for specific combinations of rotor voltage and slip. Hence, iteration is necessary to converge on the correct values which correspond to the specified stator terminal voltage. III. OPERATION UNDER SUB AND SUPER SYNCHRONOUS MODES Depending on wind speed, a doubly fed induction generator (DFIG) based variable speed wind turbine is capable of operating in sub-synchronous or super-synchronous mode of operation using power electronic converters. Traditional Wound Rotor Induction Generator (WRIG) will never produce power at sub-synchronous mode of operation. In this mode, it produces motoring torque which can be utilized by controlling rotor voltage or current. The component of rotor side converter must need to be controlled properly for reliable operation of the machine under sub-synchronous and supersynchronous modes. Rotor side converter controls the imposed voltage and current for the rotor circuit of the machine. The control of imposed current is necessary for creating generating torque in sub-synchronous mode of operation. The control of voltage or current is necessary to utilize extra generating torque in super-synchronous mode. During sub-synchronous mode, the speed of the rotor is less than the machine synchronous speed. As a result the slip is positive (s > 0), and a motoring torque is produced. To utilize this torque, negative power (according to the positive slip) is required in the rotor circuit of the machine. These can be achieved by the changing the magnitude of the injected voltage to the rotor circuit and the rotor receives power form the grid through grid side converter and DC-link. In supersynchronous mode, the rotor speed is greater than the machine synchronous speed and slip is negative (s< 0). The rotor voltage/current sequence has to be reversed to supply extra generating power to the grid through DC-link and grid side converter. The magnitude of the rotor current and voltage is changing according to the wind variations. The mechanical power and the stator electric power output are computed as follows: Pr  Tm *r Page 778 Srinivasa and Ajaykumar, Grid Connected DFIG With Efficient Rotor Power Flow Control Under Sub & Super Synchronous Modes of Operation International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 776-783 ISSN 2078-2365 Ps  Tem * s dr  Tm  Tem dt For a loss-less generator, the mechanical equation is: J In steady-state at fixed speed for a loss-less generator Tm  Tem and pm  Ps  Pr And it follows that pr  Pm  Ps  Tmr  Tems  sPs where s  (s  r ) / s is defined as the slip of the generator. Generally the absolute value of slip (s) is much lower than 1 and, consequently, P r is only a fraction of Ps. Since Tm is positive for power generation and since ωs is positive and constant for a constant frequency grid voltage, the sign of Pr is a function of the slip sign. Pr is positive for negative slip (speed greater than synchronous speed) and it is negative for positive slip (speed lower than synchronous speed). For super-synchronous speed operation, Pr is transmitted to DC bus capacitor and tends to raise the DC voltage. For sub-synchronous speed operation, Pr is taken out of DC bus capacitor and tends to decrease the DC voltage. PCgrid is used to generate or absorb the power P g in order to keep the DC voltage constant as shown in Fig.3. In steady-state for a lossless AC/DC/AC converter Pg is equal to Pr and the speed of the wind turbine is determined by the power Pr absorbed or generated by PCrotor. The phasesequence of the AC voltage generated by PCrotor is positive for sub-synchronous speed and negative for super synchronous speed. The frequency of this voltage is equal to the product of the grid frequency and the absolute value of the slip. PCrotor and PCgrid have the capability for generating or absorbing reactive power and could be used to control the reactive power or the voltage at the grid terminals. Between the two converters, a dc-link capacitor is placed, as energy storage, in order to keep the voltage or the speed of the DFIG and also the power factor at the stator terminals, while the main objective for the grid-side converter is to keep the dc-link voltage constant. IV. SIMULATION STUDIES OF PROPOSED SCHEME This section discusses the modeling of DFIG, power electronic converters and the simulation results of the overall scheme in both sub-synchronous and super-synchronous modes of operation. A. Open Loop Super-Synchronous Mode The block schematic for open loop super-synchronous mode is shown in Fig.3. Ratings of DFIG used in the proposed scheme are: Nominal power (P) = 2.65kW, V L-L = 400V, f = 50Hz, synchronous speed (Ns) = 1000 rpm, number of poles (P) = 6 [7]. In open loop super-synchronous mode firing angle ( ) of the line commutated inverter is varied manually to maintain the stator power constant at 2.65kW for speeds varying from 1050 rpm to 1200 rpm. As the speed varies, the rotor power delivered to the grid is varied but stator power is maintained constant. The parameters chosen for the simulation study are : stator resistance : 0.8285Ω stator leakage inductance : 3.579 mH rotor resistance : 0.7027Ω rotor leakage inductance : 3.579 mH magnetizing inductance : 62.64 mH The simulation model for this mode of operation is developed and the simulation results obtained are given in Table 1. Table 4.1 Simulation results for open loop super-synchronous mode Speed (Nr) in rpm 1200 1175 1150 1125 1100 1075 1050 Firing angle (α) in deg. 99.87 98.52 97.18 95.83 94.49 93.14 91.83 Stator power (Ps) in watts 2642 2666 2678 2694 2647 2631 2600 Rotor power (Pr) in watts 548.1 474.0 400.3 323.3 247.5 171.3 93.79 Rect.volt age in volts 93.17 80.56 68.23 55.40 42.72 31.23 17.96 LCI current (Iact) in amp 5.977 5.990 6.005 5.988 5.984 5.899 5.800 Fig.3. DFIG system with power electronic converters variations(or ripple) in the dc-link voltage small. With the machine-side converter it is possible to control the torque Page 779 Srinivasa and Ajaykumar, Grid Connected DFIG With Efficient Rotor Power Flow Control Under Sub & Super Synchronous Modes of Operation International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 776-783 ISSN 2078-2365 Fig. 4.2 Variation of active power delivered at the rotor side (a)Nr=1200 rpm (b) Nr=1100 rpm Fig. 4.1 Variation of active power delivered at the stator side Fig. 4.1 shows the variation of active power of the stator for varying rotor speeds of 1200 rpm and 1100 rpm. It can be seen that the stator power is delivered to the grid and is maintained at around 2.65kW for both speeds by controlling the firing angle of line commutated inverter. Similarly from Fig. 4.2 we observe that the rotor power delivered to the grid is maintained at slip times the stator power in both speeds i.e., 1200 rpm and 1100 rpm by controlling the firing angle of line commutated inverter. Speed(Nr) in rpm 1200 1175 1150 1125 1100 1075 1050 Firing angle (α) in deg. 99.87 98.52 97.18 95.83 94.49 93.14 91.83 Stator power (Ps) in watts 2653 2667 2679 2688 2687 2660 2650 Rotor power (Pr) in watts 550.1 474.6 399.1 324.0 249.8 176.1 99.5 Rect.voltage in volts 93.17 80.55 68.00 55.18 42.63 30.58 17.57 B. Closed Loop Super-Synchronous Mode Fig. 4.3 shows the closed loop super synchronous mode, in which the firing angle ( ) of the line commutated inverter is varied automatically i.e., the actual DC link current, Iact is compared with the reference current, Iref and any mismatch is used to change the firing angle α, of the inverter as follows α = (Iref - Iact)*[Kp+KI/s] where Kp and KI are the proportional and integral stage gains respectively. The optimum values for Kp and KI have been arrived at by trial and error method [6]. The values have been chosen taking into account the range of mechanical torque of the wind turbine. This range will represent the variation in wind speed with which the system has to operate. In the proposed scheme, the P and I controller gains (KP = 0.5 and KI = 100) have been chosen for operating the system with rotor speed varying from 1050 rpm to 1200 rpm, to maintain the stator power constant at 2.65kW. LCI current (Iact) in amp 6.000 5.998 5.991 6.007 6.050 6.010 6.020 Fig. 4.3 Simulation model of the closed loop super-synchronous mode The simulation model for this mode of operation is developed and is shown in Fig. 4.3. The simulation results obtained are given in Table 4.2. Table 4.2 Simulation results for closed loop super-synchronous mode (a)Nr=1200 rpm (b) Nr=1100 rpm Page 780 Srinivasa and Ajaykumar, Grid Connected DFIG With Efficient Rotor Power Flow Control Under Sub & Super Synchronous Modes of Operation International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 776-783 ISSN 2078-2365 Fig. 4.4 shows the variation of active power of the stator for varying rotor speeds of 1200 rpm and 1100 rpm. It can be seen that the stator power is delivered to the grid and is maintained at around 2.65kW for both speeds by controlling the firing angle of line commutated inverter. (a)Nr=1200 rpm (b) Nr=1100 rpm Fig. 4.4 Variation of active power delivered at the stator side Similarly from Fig. 4.5 we observe that the rotor power delivered to the grid is maintained at slip times the stator power for both speeds i.e., 1200 rpm and 1100 rpm by controlling the firing angle of line commutated inverter. Speed (Nr) in rpm Modula-tion index (m) 800 825 850 875 900 925 950 0.2500 0.2195 0.1893 0.1594 0.1299 0.1008 0.0720 (a)Nr=800 rpm Stator power (Ps) in watts 2648 2650 2650 2650 2652 2655 2647 Rotor power (Pr) in watts 651.1 574.6 498.0 421.5 345.8 270.1 193.6 Rotor freq (fr) in Hz 10.0 8.75 7.50 6.25 5.00 3.75 2.50 Rotor voltage (RMS) in volts 85.62 74.96 64.99 54.86 44.73 34.79 24.88 (b) Nr=900 rpm Fig. 4.6 Variation of active power delivered at the stator side (a)Nr=1200 rpm (b) Nr=1100 rpm Fig. 4.5 Variation of active power delivered at the rotor side C. Open Loop Sub-Synchronous Mode In open loop sub-synchronous mode, modulation index of the sinusoidal pulse width modulation inverter is varied manually to maintain the stator power constant at 2.65kW for speeds varying from 800 rpm to 950 rpm. As the speed varies, the rotor power absorbed from the grid is varied but stator power is maintained constant. The simulation model for this mode of operation is developed and the simulation results obtained are given in Table 4.3. Fig. 4.6 shows the variation of active power of the stator for varying rotor speeds of 800 rpm and 900 rpm. It can be seen that the stator power delivered to the grid is maintained at 2.65kW for both speeds by controlling the modulation index of the sinusoidal PWM inverter. Table 4.3 Simulation results for open loop sub-synchronous mode (a)Nr=800 rpm (b) Nr=900 rpm Fig. 4.7 Variation of active power absorbed from the grid at the rotor side Similarly from Fig. 4.7 we observe that the rotor power absorbed from the grid is maintained at slip times the stator power for both speeds i.e., 800 rpm and 900 rpm by controlling the modulation index of the sinusoidal PWM inverter. Page 781 Srinivasa and Ajaykumar, Grid Connected DFIG With Efficient Rotor Power Flow Control Under Sub & Super Synchronous Modes of Operation International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 776-783 ISSN 2078-2365 D. Closed Loop Sub-Synchronous Mode In closed loop sub-synchronous mode, the modulation index of the sinusoidal pulse width modulation inverter is varied automatically i.e., the actual rotor voltage, V2 is compared with the reference voltage, V2 ref = s*V1 and any mismatch is used to change the modulation index m, of the inverter as follows. m = (V2 – V2 ref)*[Kp+KI/S]. The optimum values for Kp and KI have been arrived at by trial and error method. The values have been chosen taking into account the range of mechanical torque of the wind turbine. This range will represent the variation in wind speed with which the system has to operate. In the proposed scheme, the P and I controller gains (KP = 0.05 and KI = 2.38) have been chosen for operating the system with rotor speed varying from 800 rpm to 900 rpm, to maintain the stator power constant at 2.65kW, though the rotor power absorbed from the grid is varied. The simulation model for this mode of operation is developed and is shown in Fig. 4.8. The simulation results obtained are given in Table 4.4 (a) Delivered to the grid (b) Absorbed from the grid Fig. 4.9 Variation of active power for Nr = 800 rpm Fig. 4.9 (a) shows the variation of active power of the stator for speed of 800 rpm. It can be seen that the stator power is delivered to the grid and is maintained at 2.65kW by controlling the modulation index of the sinusoidal PWM inverter. Similarly from Fig. 4.9 (b), we observe that the rotor power absorbed from the grid is maintained at slip times the stator power. V. CONCLUSION In this paper the operation of a double-fed woundrotor induction machine, coupled to a wind turbine, as a generator at different speeds is investigated. A very simple and easy to implement configurations of DFIG for wind driven applications have been demonstrated. The power flow in the rotor circuit has been controlled for maintaining the stator power constant in both sub & super-synchronous modes of operation. The simulation results depict the smooth control of active power fed to the grid with variation in rotor speed of the DFIG. Such a system allows the utilization of wind power in different operating conditions i.e. above and below synchronous speeds, thus leading to higher power harvesting and consequently higher efficiency of wind energy conversion system. Fig. 4.8 Block diagram for closed loop sub-synchronous mode Table 4.4 Simulation results for closed loop sub-synchronous mode Speed (Nr) in rpm Modulation index (m) 0.2497 Stator power (Ps) in watts 2610 Rotor power (Pr) in watts 641.7 Rotor freq (fr) in Hz 10.0 Rotor voltage in volts 825 0.2187 2560 554.6 8.75 75.0 850 0.1894 2652 498.5 7.50 64.9 875 0.1603 2748 437.7 6.25 55.0 900 0.1320 2690 318.0 5.00 44.7 800 85.5 REFERENCES [1] T.A.Lipo University of Wisconsin Madison WI USA, “A Super synchronous Doubly Fed Induction Generator Option for Wind Turbine Applications” IEEE Conference proceedings, 24-26 June 2009, pages 1-5. [2] The Electric Generators Handbook “Variable Speed Generators” by Ion Boldea Polytechnic Institute Timisoara, Romania, 2006. [3] M.Aktarujjaman, M.E.Haque, K.M.Muttaqi, M.Negnevitsky, and G.Ledwich, sch.of.Eng., Univ.of Tasmania, Hobart, TAS, “Control Dynamics of a Doubly Fed Induction Generator Under Sub and Super- Page 782 Srinivasa and Ajaykumar, Grid Connected DFIG With Efficient Rotor Power Flow Control Under Sub & Super Synchronous Modes of Operation International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 776-783 ISSN 2078-2365 Synchronous Modes of Operation” IEEE Conference Proceedings, 20-24 July 2008, pages 1-9. [4] “Design and Test of DC Voltage Link Conversion System and Brushless Doubly-Fed Induction Generator for Variable-Speed Wind Energy Applications” by T.A.Lipo, D.Panda, and D.Zarko University of Wisconsin Madison, Wisconsin, August 1999 - May 2003. [5] W. Leonhard, Control of Electrical Drives, 2nd ed. Berlin, Germany: Springer- Verlag, 1996. [6] “A Simple Controller using Line Commutated Inverter with Maximum Power Tracking for Wind-Driven Grid-Connected Permanent Magnet Synchronous Generators” by V.Lavanya, N.Ammasai Gounden, and Polimera Malleswara Rao Department of Electrical and Electronics Engineering, NIT, Tiruchirappalli, IEEE Conference Proceedings, 2006. [7] “An Improved Control Strategy of Limiting the DC-Link Voltage Fluctuation for a Doubly Fed Induction Wind Generator” by Jun Yao, Hui Li, Yong Liao, and Zhe Chen, Chongqing University, Chongqing, China. IEEE transactions on power electronics, vol. 23, no. 3, may 2008. [8] “Design and Test of DC Voltage Link Conversion System and Brushless Doubly-Fed Induction Generator for Variable-Speed Wind Energy Applications” by T.A.Lipo, D.Panda, and D.Zarko University of Wisconsin Madison, Wisconsin, August 1999 - May 2003. Page 783 Srinivasa and Ajaykumar, Grid Connected DFIG With Efficient Rotor Power Flow Control Under Sub & Super Synchronous Modes of Operation 

International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems Abd El–Naser A. Mohamed1, Ahmed Nabih Zaki Rashed2*, and Amina E. M. El-Nabawy3 1,2,3 Electronics and Electrical Communications Engineering Department Faculty of Electronic Engineering, Menouf 32951, Menoufia University, EGYPT 1 E-mail: Abd_elnaser6@yahoo.com, 2*E-mail: ahmed_733@yahoo.com Abstract: Optical wireless links offer gigabit per second data rates and low system complexity. For ground space and or terrestrial communication systems, these links suffer from atmospheric loss mainly due to fog, scintillation and precipitation. Optical Wireless link provides high bandwidth solution to the last mile access bottleneck. However, an appreciable availability of the link is always a concern. Wireless Optics (WOs) links are highly weather dependent and fog is the major attenuating factor reducing the link availability. Optical wireless links offer gigabit per second data rates and low system complexity. For ground space and or terrestrial communication scenarios, these links suffer from atmospheric loss mainly due to fog, scintillation and precipitation signals and then to upgrade the transmission bit rate distance product for ultra long transmission links. This paper has presented the bad weather effects such as rain, fog, snow, and scattering losses on the transmission performance of wireless optical communication systems. It is taken into account the study of bit error rate, maximum signal to noise ratio, maximum transmission optical path lengths and maximum transmission bit rates under these bad operating conditions. Keywords: Wireless Optics (WOs), Specific attenuation, Visibility, Rain Scattering, Rain Attenuation, Empirical model, and Bad weather effects. I. INTRODUCTION The optical wireless communication (OWC) system has attracted significant interest because it can solve the last mile problem in urban environments. The last mile problem is when Internet providers cannot connect the fiber optic cables to every household user because of the high installation costs. The only disadvantage of the OWC system is that its performance depends strongly on weather conditions. Fog and clouds scatter and absorb the optical signal, which causes transmission errors. Most previous studies consider only single-scattering effects and assume that the received signal has no intersymbol interference (ISI), which is true only for light-fog conditions [1]. Maintaining a clear line of sight (LOS) between transmit and receive terminals is the biggest challenge to establish optical wireless links in the free space especially in the troposphere [2]. The LOS is diminished due to many atmospheric influences like fog, rain, snow, dust, sleet, clouds and temporary physical obstructions like e.g., birds and aeroplanes [3]. Moreover, the electromagnetic interaction of the transmitted optical signal with different atmospheric effects results in complex processes like scattering, absorption and extinction that are a function of particle physical parameters. Hence the local atmospheric weather conditions mainly determine the availability and reliability of such optical wireless links since there is always a threat of downtime of optical wireless link caused by adverse weather conditions [4]. Optical wireless links are also influenced by atmospheric temperature that varies both in spatial and temporal domains. The variation of temperature in the optical wireless channel is a function of atmospheric pressure and the atmospheric wind speed. This effect is commonly known as optical turbulence or scintillation effect and causes received signal irradiance or power fades in conjunction with the variation of temperature along the propagation path. As a result of this scintillation phenomenon, the optical wireless channel distance and the capacity are reduced. Thereby restricting the regions and times where optical wireless links can be used potentially. In order to take full advantage of the tremendous usefulness of optical wireless technology require a proper characterization of different atmospheric effects influences and a meaningful interpretation of the filed measurements in such adverse conditions [5]. Optical Wireless communication, also known as free space optical (FSO), has emerged as a commercially viable alternative to radio frequency (RF) and millimeter wave wireless for reliable and rapid deployment of data and voice networks. RF and millimeter wave technologies Page 650 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 allow rapid deployment of wireless networks with data rates from tens of Mbit/sec (point-to-multipoint) up to several hundred Mbit/sec (point-to-point). Though emerging license free bands appear promising, they still have certain bandwidth and range limitations [6]. Optical wireless can augment RF and millimeter wave links with very high (>1 Gbit/sec) bandwidth. In fact, it is widely believed that optical wireless is best suited for multi Gbit/sec communication. The general acceptance of free space laser communication (lasercom) or optical wireless as the preferred wireless carrier of high bandwidth data has been hampered by the potential downtime of these lasercom systems in heavy, visibility limiting, weather. There seems to be much confusion and many preconceived notions about the true ability of lasercom systems in such weather. There still is some confusion over how different laser wavelengths are attenuated by different types of weather [7]. In the present study, optical wireless communication is now a well established access technology, better known for its robustness in transmitting large data volumes in an energy efficient manner. However the bit error rate (BER) performance of a wireless optical communication ground link is adversely affected by cloud coverage, harsh weather conditions, and atmospheric turbulence. Fog, clouds and dry snow play a detrimental role by attenuating optical energy transmitted in terrestrial free space and thus decrease the link availability and reliability. II. BLOCK DIAGRAM OF OPTICAL WIRELESS COMMUNICATION SYSTEM There are three key function elements of optical wireless communication system as shown in Fig. 1. The transmitter, the atmospheric channel and the receiver. The transmitter converts the electrical signal into light signal. The light propagates through the atmosphere to the receiver, which converts the light back into an electrical signal. The transmitter includes a modulator, a laser driver, a light emitting diode (LED) or a laser, and a telescope [8]. Fig. 1. General block diagram of optical wireless communication system. The modulator converts bits of information into signals in accordance with the chosen modulation method. The driver provides the power for the laser and stabilizes its performance, it also neutralizes such effects as temperature and aging of the laser or LED. The light sources convert the electrical signal into optic radiation. The telescope aligns the laser LED radiation to a collimated beam and directs it to the receiver. In the atmospheric channel, the signal is attenuated and blurred as a result of absorption, scattering and turbulence. This channel maybe the traversed distance between a ground station and a satellite or a path of a few kilometers through the atmosphere between two terrestrial transceivers [9]. The receiver includes a telescope, filter, photo detector, an amplifier, a decision device, and a clock recovery unit. The telescope collects the incoming radiation and focus it onto filter. The filter removes background radiation and allows only the wavelength of the signal to pass through the electronic signal. The decision unit determines the nature of the bits of information based on the time of arrival and the amplitude of the pulse. The clock recovery unit and synchronizes the data sampling to the decision making process. Page 651 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 III. SYSTEM MODEL ANALYSIS In a terrestrial FSO, the communication transceivers are typically located in the troposphere. Troposphere is home to all kinds of weather phenomena and plays a very detrimental role for FSO communications in lower visibility range conditions mainly due to rain, snow, fog and clouds. The estimated of fog, snow and rain attenuation effects using empirical models as mentioned in Ref. [10]:  fog    3.912      V  55x104  q , (1) Where V is visibility range in km, is transmission wavelength in m. αfog( ) is the total extinction coefficient and q is the size distribution coefficient of scattering related to size distribution of the droplets. In case of clear or foggy weather with no rain or snow, Ref. [11, 12] approximations of the q parameter to compute the fog attenuation, that are very accurate for the narrow wavelength range between 1.3–1.65 m. 1.6  V  50 km    q  1.3  6 km V  50 km , 0 V  0.5 km    (2) Transmitted optical pulses in free space are mainly influenced by two main mechanisms of signal power loss, absorption and scattering. Absorption is mainly due to water vapours and carbon dioxide, and depends on the water vapour content that is dependent on the altitude and humidity. By appropriate selection of optical wavelengths for transmission the losses due to absorption can be minimized. It was found that scattering (especially Mie scattering) is the main mechanism of optical power loss as the optical beam looses intensity and distance due to scattering. The beam loss due to scattering can be calculated from the following empirical, visibility range dependent formula [13]:  sca t    17  550    V   0.195V , dB/km (3) Where V is visibility range in km, is transmission wavelength in m. Then the total attenuation of wireless medium communication system can be estimated as:    fog     snow  rain   scat   , dB/km (4) When the optical signal passes through the atmosphere, it is randomly attenuated by fog and rain. Although fog is the main attenuation factor for optical wireless links, the rain attenuation effect cannot be ignored, in particular in environments where rain is more frequent than fog. As the size of water droplets of rain increases, they become large enough to cause reflection and refraction processes. These droplets cause wavelength independent scattering [13]. It was found that the resulting attenuation increases linearly with rainfall rate; furthermore the mean of the raindrops size is in the order of a few millimeters and it increases with the rainfall rate [14]. Let R be the rain rate in mm/h, the specific attenuation of wireless optical link is given by [15]:  ra in  1.076 R0.67 dB/km (5) If S is the snow rate in mm/h then specific attenuation in dB/km is given by [16, 17] as:  snow  a Sb dB/km (6) If is the wavelength, the parameters a and b for dry snow are given as the following: a  5.42x104   5.495876, b  1.38 (7) The parameters a and b for wet snow are as follows [18, 19]: a  1.023x104   3.7855466, b  0.72 (8) In order to estimate the coverage at millimeter wavelengths under direct Line of Sight (LOS) conditions, the free space propagation model is used. The SNR requirements for modulation scheme at a fixed data rate of one Gbit/sec is obtained from the following formula [20]:  4 SNR  PT  30  GT  GR  20log  c    10 logkB B.W T     NF  Fm ,   dB (9) Where PT is the transmitter power, GT is the transmitter antenna gain, GR is the receiver antenna gain, c is the carrier wavelength, kB is the Boltzmann’s constant (1.38X10-23 J/K), Receiver Bandwidth (B.W= 1 MHz), T is the ambient temperature in K, , Receiver Noise Figure, Fm is the Fade margin, and α is the total attenuation in dB/km. The maximum propagation distance (L) for meeting the SNR requirements [21]: L  10 / 20 , (10) The transmitter and receiver antenna gains can be expressed as the following: Page 652 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365   Dr  GR    ,    2 GT  32 2  div , (11) Where θdiv is the transmitter divergence of the beam in radians can be expressed as follows: 4  div  ,  Dt (13) The basic formula for a typical optical link is an exponential decaying function as function of the path length L as the following expression [22, 23]: PR  PT Dt  div  L2 Dr e L , (14) Where PR is the received power after traveling the path length L through the lossy medium, PT is the initial transmitted power, and α is the total attenuation coefficient of the medium. The bit error rate (BER) essentially specifies the average probability of incorrect bit identification. In general. The higher the received SNR, the lower the BER probability will be. For most PIN receivers, the noise is generally thermally limited, (12) which independent of signal current. The bit error rate (BER) is related to the signal to noise ratio (SNR) as follows [24, 25]:  2    SNR   . exp  BER    ,  SNR .   8    (15) The maximum transmission bit rates BRmax. which is a losses limited one, and is given by [26]: BRmax .  Bu exp   L    m) (16) Where Bu is the maximum available transmission bit rate without any limitations, and αm is the system marginal loss. IV. Simulations Results and Performance Evaluation The model have been deeply investigated to present the bad weather effects on the transmission performance and system operation characteristics of wireless optical communication systems for different visibility ranges over wider range of the affecting parameters. Table 1: Proposed operating parameters for wireless optical communication systems [2, 5, 13, 20]. Operating parameter Room temperature, T=T0 Signal transmitted power, PT Operating signal wavelength, System marginal loss, αm Transmitter lens diameter, Dt Bit rate (max.) without any limitations, Bu Detector electronic bandwidth (B.W) Value and unit 300 K 100 m Watt 1.3 ≤ , µm ≤ 1.65 3 dB 100 cm 1 Gbit/sec 1 MHz High visibility, Vhigh 50 ≤ Vhigh, km ≤ 80 Medium visibility, Vmedium 6 ≤ Vmedium, km ≤ 50 0 ≤ Vlow, km ≤ 0.5 Low visibility, Vlow Receiver aperture diameter (antenna size) Dr 50 cm System marginal loss, αm 3 dB Receiver noise figure, NF 5 dB Fade margin, Fm 20 dB Snow rate, S 0.2 mm/h Rain rate, R Carrier wavelength, 1 mm/h c 1.55 m Page 653 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 windows under both bad weather effects of dry and wet snow. It is observed that second optical transmission window has presented higher maximum propagation distance compared to third optical transmission window. As well as low visibility range has presented the highest propagation distance compared to both medium and high visibility. Based on the modeling equations analysis and the assumed set of the operating parameters as shown in Table 1. The following facts are assured as shown in the series of Figs. (2-31): i) Figs. (2-7) have assured that maximum propagation distance decreases with increasing visibility ranges for both optical transmission Maximum propagation distance, L, km 45 42.5 2' nd o ptical transmissio n windo w =1.3 m 40 3' rd o ptical transmissio n windo w =1.55 m 37.5 35 32.5 30 27.5 25 22.5 Dry snow 20 17.5 15 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Low visibility, VLow, km Fig. 2. Maximum propagation distance in relation to low visibility at different optical transmission windows at the assumed set of the operating parameters. Maximum propagation distance, L, km 46 42 Wet snow 38 34 30 26 2' nd o ptical transmissio n windo w =1.3 m 3' rd o ptical transmissio n windo w =1.55 m 22 18 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Low visibility, VLow, km Fig. 3. Maximum propagation distance in relation to low visibility at different optical transmission windows at the assumed set of the operating parameters. Page 654 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 Maximum propagation distance, L, km 12 11 2' nd o ptical transmissio n windo w =1.3 m 10 3' rd o ptical transmissio n windo w =1.55 m 9 8 7 6 5 Dry snow 4 3 2 6 10 14 18 22 26 30 34 38 42 46 50 Medium visibility, VMedium, km Fig. 4. Maximum propagation distance in relation to medium visibility at different optical transmission windows at the assumed set of the operating parameters. Maximum propagation distance, L, km 14 13 Wet snow 12 11 10 9 8 7 6 2' nd o ptical transmissio n windo w =1.3 m 5 4 3' rd o ptical transmissio n windo w =1.55 m 3 2 6 10 14 18 22 26 30 34 38 42 46 50 Medium visibility, VMedium, km Fig. 5. Maximum propagation distance in relation to medium visibility at different optical transmission windows at the assumed set of the operating parameters. Page 655 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 Maximum propagation distance, L, km 4 2' nd o ptical transmissio n windo w =1.3 m 3.5 3' rd o ptical transmissio n windo w =1.55 m 3 2.5 2 1.5 1 Dry snow 0.5 0 50 52.5 55 57.5 60 62.5 65 67.5 70 72.5 75 77.5 80 High visibility, VHigh, km Fig. 6. Maximum propagation distance in relation to high visibility at different optical transmission windows at the assumed set of the operating parameters. Maximum propagation distance, L, km 4.5 2' nd o ptical transmissio n windo w =1.3 m 4 3' rd o ptical transmissio n windo w =1.55 m 3.5 3 2.5 2 1.5 Wet snow 1 0.5 0 50 52.5 55 57.5 60 62.5 65 67.5 70 72.5 75 77.5 80 High visibility, VHigh, km Fig. 7. Maximum propagation distance in relation to high visibility at different optical transmission windows at the assumed set of the operating parameters. Page 656 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 100 Received signal power, PR, Watt 2' nd o ptical transmissio n windo w =1.3 m 90 3' rd o ptical transmissio n windo w =1.55 m 80 70 60 50 Dry snow 40 30 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Low visibility, VLow, km Fig. 8. Received signal power in relation to low visibility at different optical transmission windows at the assumed set of the operating parameters. Received signal power, PR, Watt 85 80 2' nd o ptical transmissio n windo w =1.3 m 75 3' rd o ptical transmissio n windo w =1.55 m 70 65 60 55 50 45 40 Wet snow 35 30 25 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Low visibility, VLow, km Fig. 9. Received signal power in relation to low visibility at different optical transmission windows at the assumed set of the operating parameters. Page 657 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 Received signal power, PR, Watt 200 185 2' nd o ptical transmissio n windo w =1.3 m 170 3' rd o ptical transmissio n windo w =1.55 m 155 140 125 110 95 Dry snow 80 65 50 6 10 14 18 22 26 30 34 38 42 46 50 Medium visibility, VMedium, km Fig. 10. Received signal power in relation to medium visibility at different optical transmission windows at the assumed set of the operating parameters. 180 Received signal power, PR, Watt 2' nd o ptical transmissio n windo w =1.3 m 160 3' rd o ptical transmissio n windo w =1.55 m 140 120 100 80 Wet snow 60 40 6 10 14 18 22 26 30 34 38 42 46 50 Medium visibility, VMedium, km Fig. 11. Received signal power in relation to medium visibility at different optical transmission windows at the assumed set of the operating parameters. Page 658 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 475 Received signal power, PR, Watt 450 2' nd o ptical transmissio n windo w =1.3 m 425 3' rd o ptical transmissio n windo w =1.55 m 400 375 350 325 300 275 250 225 Dry snow 200 175 150 50 52.5 55 57.5 60 62.5 65 67.5 70 72.5 75 77.5 80 High visibility, VHigh, km Fig. 12. Received signal power in relation to high visibility at different optical transmission windows at the assumed set of the operating parameters. Received signal power, PR, Watt 400 375 2' nd o ptical transmissio n windo w =1.3 m 350 3' rd o ptical transmissio n windo w =1.55 m 325 300 275 250 225 200 Wet snow 175 150 50 52.5 55 57.5 60 62.5 65 67.5 70 72.5 75 77.5 80 High visibility, VHigh, km Fig. 13. Received signal power in relation to high visibility at different optical transmission windows at the assumed set of the operating parameters. Page 659 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 16 2' nd o ptical transmissio n windo w =1.3 m Signal to noise ratio, SNR, dB 15 3' rd o ptical transmissio n windo w =1.55 m 14 13 12 11 10 9 8 Dry snow 7 6 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Low visibility, VLow, km Fig. 14. Signal to noise ratio in relation to low visibility at different optical transmission windows at the assumed set of the operating parameters. 14 2' nd o ptical transmissio n windo w =1.3 m Signal to noise ratio, SNR, dB 13 3' rd o ptical transmissio n windo w =1.55 m 12 11 10 9 8 7 Wet snow 6 5 4 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Low visibility, VLow, km Fig. 15. Signal to noise ratio in relation to low visibility at different optical transmission windows at the assumed set of the operating parameters. Page 660 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 Signal to noise ratio, SNR, dB 25 22.5 20 Dry snow 17.5 15 2' nd o ptical transmissio n windo w =1.3 m 12.5 3' rd o ptical transmissio n windo w =1.55 m 10 6 10 14 18 22 26 30 34 38 42 46 50 Medium visibility, VMedium, km Fig. 16. Signal to noise ratio in relation to medium visibility at different optical transmission windows at the assumed set of the operating parameters. 23.5 2' nd o ptical transmissio n windo w =1.3 m Signal to noise ratio, SNR, dB 22 3' rd o ptical transmissio n windo w =1.55 m 20.5 19 17.5 16 Wet snow 14.5 13 11.5 6 10 14 18 22 26 30 34 38 42 46 50 Medium visibility, VMedium, km Fig. 17. Signal to noise ratio in relation to medium visibility at different optical transmission windows at the assumed set of the operating parameters. Page 661 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 Signal to noise ratio, SNR, dB 35 34 2' nd o ptical transmissio n windo w =1.3 m 33 3' rd o ptical transmissio n windo w =1.55 m 32 31 30 29 28 27 Dry snow 26 25 24 50 52.5 55 57.5 60 62.5 65 67.5 70 72.5 75 77.5 80 High visibility, VHigh, km Fig. 18. Signal to noise ratio in relation to high visibility at different optical transmission windows at the assumed set of the operating parameters. 33 Signal to noise ratio, SNR, dB 32 2' nd o ptical transmissio n windo w =1.3 m 31 3' rd o ptical transmissio n windo w =1.55 m 30 29 28 27 26 25 24 Wet snow 23 22 21 20 50 52.5 55 57.5 60 62.5 65 67.5 70 72.5 75 77.5 80 High visibility, VHigh, km Fig. 19. Signal to noise ratio in relation to high visibility at different optical transmission windows at the assumed set of the operating parameters. Page 662 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 0.007 2' nd o ptical transmissio n windo w =1.3 m Bit error rate, BERx10-9 0.006 3' rd o ptical transmissio n windo w =1.55 m 0.005 0.004 0.003 0.002 Dry snow 0.001 0 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Low visibility, VLow, km Fig. 20. Bit error rate in relation to low visibility at different optical transmission windows at the assumed set of the operating parameters. 0.007 2' nd o ptical transmissio n windo w =1.3 m 0.006 Bit error rate, BERx10-9 3' rd o ptical transmissio n windo w =1.55 m 0.005 0.004 0.003 0.002 Wet snow 0.001 0 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Low visibility, VLow, km Fig. 21. Bit error rate in relation to low visibility at different optical transmission windows at the assumed set of the operating parameters. Page 663 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 0.12 3' rd o ptical transmissio n windo w =1.55 m Bit error rate, BERx10-10 0.108 2' nd o ptical transmissio n windo w =1.3 m 0.096 0.084 0.072 0.06 0.048 0.036 Dry snow 0.024 0.012 6 10 14 18 22 26 30 34 38 42 46 50 Medium visibility, VMedium, km Fig. 22. Bit error rate in relation to medium visibility at different optical transmission windows at the assumed set of the operating parameters. 0.16 3' rd o ptical transmissio n windo w =1.55 m 0.14 Bit error rate, BERx10-10 2' nd o ptical transmissio n windo w =1.3 m 0.12 0.1 0.08 0.06 Wet snow 0.04 0.02 0 6 10 14 18 22 26 30 34 38 42 46 50 Medium visibility, VMedium, km Fig. 23. Bit error rate in relation to medium visibility at different optical transmission windows at the assumed set of the operating parameters. Page 664 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 0.018 0.016 Bit error rate, BERx10-12 Dry snow 0.014 0.012 0.01 0.008 2' nd o ptical transmissio n windo w =1.3 m 0.006 3' rd o ptical transmissio n windo w =1.55 m 0.004 0.002 50 52.5 55 57.5 60 62.5 65 67.5 70 72.5 75 77.5 80 High visibility, VHigh, km Fig. 24. Bit error rate in relation to high visibility at different optical transmission windows at the assumed set of the operating parameters. 0.035 3' rd o ptical transmissio n windo w =1.55 m Bit error rate, BERx10-12 0.03 2' nd o ptical transmissio n windo w =1.3 m 0.025 0.02 0.015 Wet snow 0.01 0.005 50 52.5 55 57.5 60 62.5 65 67.5 70 72.5 75 77.5 80 High visibility, VHigh, km Fig. 25. Bit error rate in relation to high visibility at different optical transmission windows at the assumed set of the operating parameters. Page 665 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 Maximum bit rate, BRmax., Gbit/sec 0.11 0.1 Dry snow 0.09 0.08 0.07 0.06 2' nd o ptical transmissio n windo w =1.3 m 0.05 3' rd o ptical transmissio n windo w =1.55 m 0.04 0.03 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Low visibility, VLow, km Fig. 26. Maximum transmission bit rate in relation to low visibility at different optical transmission windows at the assumed set of the operating parameters. Maximum bit rate, BRmax., Gbit/sec 0.1 0.09 2' nd o ptical transmissio n windo w =1.3 m 0.08 3' rd o ptical transmissio n windo w =1.55 m 0.07 0.06 0.05 0.04 0.03 Wet snow 0.02 0.01 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Low visibility, VLow, km Fig. 27. Maximum transmission bit rate in relation to low visibility at different optical transmission windows at the assumed set of the operating parameters. Page 666 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 Maximum bit rate, BRmax., Gbit/sec 1.4 2' nd o ptical transmissio n windo w =1.3 m 1.2 3' rd o ptical transmissio n windo w =1.55 m 1 0.8 0.6 Dry snow 0.4 0.2 6 10 14 18 22 26 30 34 38 42 46 50 Medium visibility, VMedium, km Fig. 28. Maximum transmission bit rate in relation to medium visibility at different optical transmission windows at the assumed set of the operating parameters. Maximum bit rate, BRmax., Gbit/sec 1 0.9 2' nd o ptical transmissio n windo w =1.3 m 0.8 3' rd o ptical transmissio n windo w =1.55 m 0.7 0.6 0.5 0.4 Wet snow 0.3 0.2 0.1 6 10 14 18 22 26 30 34 38 42 46 50 Medium visibility, VMedium, km Fig. 29. Maximum transmission bit rate in relation to medium visibility at different optical transmission windows at the assumed set of the operating parameters. Page 667 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 Maximum bit rate, BRmax., Gbit/sec 3.25 2' nd o ptical transmissio n windo w =1.3 m 3 3' rd o ptical transmissio n windo w =1.55 m 2.75 2.5 2.25 2 1.75 1.5 Dry snow 1.25 1 50 52.5 55 57.5 60 62.5 65 67.5 70 72.5 75 77.5 80 High visibility, VHigh, km Fig. 30. Maximum transmission bit rate in relation to high visibility at different optical transmission windows at the assumed set of the operating parameters. 2.75 Maximum bit rate, BRmax., Gbit/sec 2' nd o ptical transmissio n windo w =1.3 m 2.5 3' rd o ptical transmissio n windo w =1.55 m 2.25 2 1.75 1.5 1.25 1 Wet snow 0.75 0.5 50 52.5 55 57.5 60 62.5 65 67.5 70 72.5 75 77.5 80 High visibility, VHigh, km Fig. 31. Maximum transmission bit rate in relation to high visibility at different optical transmission windows at the assumed set of the operating parameters. observed that second optical transmission window ii) As shown in Figs. (8-13) have proved that received has presented lower signal to noise ratio compared signal power increases with increasing visibility to third optical transmission window. As well as ranges for both optical transmission windows high visibility range has presented the highest under both bad weather effects of dry and wet signal to noise ratio compared to both medium and snow. It is also observed that second optical low visibility. transmission window has presented lower received iv) As shown in Figs. (20-25) have assured that bit signal power compared to third optical error rate increases with increasing visibility ranges transmission window. As well as high visibility for both optical transmission windows under both range has presented the highest received signal bad weather effects of dry and wet snow. It is power compared to both medium and low observed that second optical transmission window visibility. has presented higher bit error rate compared to iii) Figs. (14-19) have indicated that signal to noise third optical transmission window. As well as low ratio increases with increasing visibility ranges for visibility range has presented the highest bit error both optical transmission windows under both bad rate compared to both medium and high visibility. weather effects of dry and wet snow. It is also Page 668 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 v) Figs. (26-31) have indicated that maximum transmission bit rate increases with increasing visibility ranges for both optical transmission windows under both bad weather effects of dry and wet snow. It is also observed that third optical transmission window has presented higher transmission bit rate compared to second optical transmission window. As well as high visibility range has presented the highest transmission bit rate compared to both medium and low visibility. V. Conclusions In a summary, the wireless optical communication systems have deeply investigated under the bad weather of rain, fog, scattering dry and wet snow over wide range of the affecting parameters. Maximum propagation distance, received signal power, signal to noise ratio, bit error rate, and transmission rates for different visibility ranges are the major interesting design parameters as a measurement of the system performance under different optical transmission windows. It is theoretically found that wet snow has presented bad effects on the wireless optical communication systems compared to dry snow. As well as optical wireless communication systems have presented the highest received signal power, signal to noise ratio, transmission bit rates, and the lowest propagation distance and bit error rate for different visibility ranges at third optical transmission window compared to second optical transmission window. REFERENCES [1] A. Nkansah and N. J. Gomes, “A WDM/SCM Star/Tree Fiber Feed Architecture for Pico-cellular Broadband Systems”, International Topical Meeting on Microwave Photonics, pp. 271-274, Sept. 2003. [2] Abd El-Naser A. Mohammed, Mohamed M. E. ElHalawany, Ahmed Nabih Zaki Rashed, and Mohammed S. F. Tabour “High Transmission Performance of Radio over Fiber Systems over Traditional Optical Fiber Communication Systems Using Different Coding Formats for Long Haul,” International Journal of Computer Science and Telecommunications (IJCST), Vol. 2, No. 3, pp. 2942, June 2011. [3] C. Lim, D. Novak, “Capacity Analysis for WDM Fiber-Radio Backbones with Star Tree and Ring Architecture Incorporating Wavelength Interleaving”, Journal of Lightwave Tech., Vol. 21, No. 12, pp. 3308-3315, Dec. 2003. [4] S. Sarkar, S. Dixit, and B. Mukherjee, “Hybrid Wireless Optical Broadband Access Network (WOBAN)μ A Review of Relevant Challenges,” IEEE/OSA Journal on Lightwave Technology, , Vol. 25, No. 11, pp. 1233-1239, Nov. 2007. [5] H. Hennier, O. Wilfert, “An Introduction to Free space Optical Communications,” Radio Engineering Journal, Vol. 19, No. 2, pp. 203-212, June 2010. [6] Abd El-Naser A. Mohammed, Ahmed Nabih Zaki Rashed, and Mohammed S. F. Tabour “Transmission Characteristics of Radio over Fiber (ROF) MillimeterWave Systems in Local Area Optical Communication Networks,” International Journal of Advanced Networks and Applications, Vol. 2, No. 6, pp. 876-886, May/June 2011. [7] H. Manor and S. Arnon, “Performance of an Optical Wireless Communication System As A function of Wavelength,” Applied Optics , Vol. 42, No. 21, pp. 4285–4294, July 2003. [8] C. C. Davis, I. I. Smolyaninov and S. D. Milner, “Flexible Optical Wireless Links and Networks”, IEEE Communication Magazine, Vol. 3, No. 2, pp. 51-57, Mar. 2003. [9] A. K. Majumdar and J. C. Ricklin, “Free Space Laser Communications, Principles and advantages”, Springer Science 2008. [10] J. C. Juarez, A. Dwivedi, A. R. Hammons, S. D. Jones, V. Weerackody and R. A. Nichols, “Free Space Optical Communications for Next Generation Military Networks”, IEEE Communication Magazine, Vol. 2, No. 1, pp. 46-51, November 2006. [11] D. Giggenbach, J. Horwath and B. Epple, “Optical Satellite Downlinks to Optical Ground Stations and High-Altitude Platforms,” IST Mobile & Wireless Communication Summit, Budapest, Hungary, July 2007. [12] M. Aliba, K. Oawa, and S. Ito, “Measurement and Simulation of the Effect of Snow Fall on Free Space Optical Propagation,” Applied Optics, Vol. 47, No. 31, pp. 5736-5743, 2008. [13] H. Hemmati, “Deep Space Optical Communications”, John Wiley & Sons, 2006. [14] Ibrahim M. El-dokany, Abd El–Naser A. Mohamed, Ahmed Nabih Zaki Rashed, and Amina M. ElNabawy, “Upgrading Efficiency and Improvement of the Performance of Broadband Wireless Optical Access Communication Networks” International Journal of Communication Networks and Information Security (IJCNIS), Vol. 3, No. 2, pp. 149-162, Aug. 2011. Page 669 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 650-670 ISSN 2078-2365 [15] S. Muhammad, B. Flecker, E. Leitgeb and M. Gebhart, “Characterization of Fog Attenuation in Terrestrial Free Space Optical Links”, Journal of optical Engineering, Vo. 46, No. 4, pp. 1-9, June 2007. [16] M. Akiba, K. Ogawa, K. Wakamori, K. Kodate and S. Ito, “Measurement and Simulation of the Effect of Snowfall on Free Space Optical Propagation”, Journal of Applied Optics, Vol. 47, No. 31, pp. 57365743, Nov. 2008. [17] S. Betti, V. Carrozzo, E. Duca, “Over Stratospheric Altitude Optical Free Space Links: System Performance Evaluation Transparent Optical Networks”, in ICTON, International Conference on Transparent Optical Networks, July 2007. [18] P. F. J. Lermusiaux, C.-S. Chiu, G. G. Gawarkiewicz, P. Abbot, A. R. Robinson, R. N. Miller, P. J. Haley, W. G. Leslie, S. J. Majumdar, A. Pang and F. Lekien, “Quantifying Uncertainties in Ocean Predictions” in Oceanography, special issue on “Advances in Computational Oceanography”, T. Paluszkiewicz and S. Harper, Eds., Vol. 19, No. 1, pp. 92-105, 2006. [19] Ahmed Nabih Zaki Rashed, “High Transmission Bit Rate of Multi Giga Bit per second for Short Range Optical Wireless Access Communication Networks” International Journal of Advanced Science and Technology, Vol. 32, pp. 23-32, July 2011. [20] D. Giggenbach, J. Horwarth, M. Knapek, “Optical Data Downlinks From Earth Observation Platforms”, Free Space Laser Communication Technologies XXI, Proc. of SPIE, Vol. 19, No. 3, pp. 12-25, 2009. [21] P. Corrigan, R. Martini, E. A. Whittaker, C. Bethea, “Quantum Cascade Lasers and the Kruse model in Free Space Optical Communication”, Optics Express, Vol. 17, No. 6, pp. 432-444, 2009. [22] Z. Bielecki, W. Kolosowski, J. Mikolajczyk, “Free Space Optical Data Link Using Quantum Cascade Laser”, PIERS Proceedings, Cambridge, USA, July 2-6, 2008. [23] Abd El-Naser A. Mohammed, Mohamed M. E. ElHalawany, Ahmed Nabih Zaki Rashed, and Mohamoud M. Eid “Optical Add Drop Multiplexers with UW-DWDM Technique in Metro Optical Access Communication Networks,” International Journal of Computer Science and Telecommunications (IJCST), Vol. 2, No. 2, pp. 513, April 2011. [24] P. S. Andre, A. N. Pinto, J. L. Pinto, T. Almeida, and M. Pousa, “Selective Wavelength Transparent Optical Add-Drop Multiplexer Based on Fiber Bragg Gratings ,” J. Opt. Commun., Vol. 24, No. 3, pp. 222229, 2006. [25] M. S. Ab-Rahaman, S. Suliana, K. Mat, and B. Ng, “The Hybrid Protection Scheme in Hybrid OADM/OXC/MUX ,” Australian J. of Basic and Applied Sciences, Vol. 2, No. 4, pp. 968-976, 2008. [26] A. A. Aboul Enein, F. Z. El-Halafawy, M. H. A. Hassan, A. A. Mohammed, “Thermal Environmental Effects,” Alex. Eng. J., Alex. Univ., Vol. 28, No. 2, pp.169-183, 1989. Page 670 Abd El–Naser et. Al., The Effects of the Bad Weather on the Transmission and Performance Efficiency of Optical Wireless Communication Systems 

International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 769-775 ISSN 2078-2365 Combined Economic and Emission Dispatch for a Wind Integrated System Using Particle Swarm Optimization K.S. Linga Murthy1, G.V.S.Subramanyam2, K.SriChandan3 E.E.E. Department, Gitam University, India. (Email: 1 hod_eee@gitam.edu, 2subbu.venkatasatya@gmail.com, 3srichandank@gmail.com) Abstract—This paper deals with the problems of Economic Dispatch, Emission Dispatch and Combined Economic and Emission Dispatch problems for an integrated system having thermal and wind units. Particle Swarm Optimization and Genetic Algorithm methods are used to solve the problems of Economic dispatch, Emission dispatch and Combined Economic and Emission Dispatch problems. The effectiveness of PSO and GA methods are demonstrated by comparing the results obtained with both the methods. Keywords —Emission level, Fuel cost, Particle Swarm optimization, Wind integration. I. INTRODUCTION A Power system is a mix of different types of generations, out of which thermal, hydro and nuclear power generations have the maximum contribution. However, economic operation has conveniently been considered by proper scheduling of thermal and hydro-generation only. The nuclear stations are run at their base loads keeping safety in mind [13]. The purpose of economic dispatch is to find out the most economical schedule of the generating units while satisfying load demand and operational constraints. Economic dispatch is a familiar problem pertaining to the allocation of the amount of power to be generated by different units in the system on an optimum economy base [1]. This problem has been tackled by many researchers in the past. Recently the problem which has attracted much attention is pollution minimization due to pressing public demand for clean air. Environmental pollution is a direct consequence of industrial advancement. Technology, which has made economic development possible, produces enormous quantities of harmful by-products and wastes. Thermal power stations are major causes of atmospheric pollution, because of high concentration of pollutants they cause. It is utmost important to protect our environment from harmful emissions out from thermal power plants. Power utilities using fossil fuels as a primary energy source, give rise to particulates and gaseous pollutants apart from heat. The particulates as also the gaseous pollutants such as carbon dioxide (CO2), oxides of sulphur (SOX) and oxides of nitrogen (NOx) cause detrimental effects of CO2 on the environment is not yet precisely known. Pollution control agencies (Municipal/Governmental regulatory bodies) restrict the amount of emission of pollutants depending upon their relative harmfulness to human beings [1]. So, the emission dispatch has been formulated. Of the pollutants emitted, NOx is of major concern and hence it has been considered. The objective of emission dispatch is to minimize the total environmental degradation or the total pollution emission due to burning of fuels for production to meet the load demand [14]. Hence, there is a need to formulate the combined emission and economic dispatch (CEED) problem.The idea behind combined emission and economic dispatch is to compute the optimal generation for individual units of the power system by minimizing the fuel cost and emission levels simultaneously, Page 769 Linga et. al., Combined Economic and Emission Dispatch for a Wind Integrated System Using Particle Swarm Optimization International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 769-775 ISSN 2078-2365 subjected to the system constraints. E. Constraints 1. Power Balance Constraint P ng II. Problem formulation A. Economic dispatch The generation cost function is usually expressed as a quadratic polynomial and can be represented as below for an ith generator. + bi Pi +ci Rs/hr ( 1) In the expression above, Pi is the output power in MW and ai , bi, ci are the fuel cost-coefficients of the ith generating unit. B. Emission Dispatch Pollutant Emissions from the generating units such as oxides of nitrogen can be expressed as a quadratic polynomial and can be represented as below for an ith generator. Ei = αi Pi2 + βi Pi + i Kg/hr (2) In the expression above, Pi is the output power in MW and αi, βi, i are the emission coefficients of the ith generating unit. C. Constraints 1. Power Balance Constraint P ng gi = PD + PL (3) i1 Where, ng = number of generating units, P gi is the power generated by ith unit in MW, Pw is the wind power that is available in MW, PD is the load demand in MW, PL is the transmission loss in MW. 2. Generating Limits Pgimin ≤ Pgi ≤ Pgimax (i=1,2,3,....ng) ng= number of generators. D. CEED Cost Function In this formulation both fuel cost objective and emission level objective are combined to form a single objective with the introduction of factor called ‘The Price Penalty Factor’, hm (Rs/Kg). Minimize FT =( a i P i2 + bi P i + ci ) + hm( αi P i2 + i P i + i) Where, FT is the total cost of generation(RS/hr). (4) (5) Where, ng = number of generating units, Pgi is the power generated by ith unit in MW, PD is the load demand in MW, PL is the transmission loss in MW. Losses can be calculated by B coefficients, which can be expressed as PL = Fi =ai Pi2 = PD + PL gi i1  ng ng i 1 j 1 Pi Bij Pj (6) Where, Bij is generation loss coefficient. Pi and Pj are The real power injections at ith and jth buses respectively. 2. Generating Limits Pgimin ≤ Pgi ≤ Pgimax (i=1,2,3,....ng) ng= number of generators. (7) III. WIND INTEGRATION This paper deals with a multi-objective generation dispatch problem that considers environment and fuel cost under substantial penetration of wind energy has been proposed [11]. Wind plants are different from conventional generation plants in that their fuel supply is neither steady nor controllable, and as a result, they exhibit greater uncertainty and variability in their output [10]. Wind plants naturally operate when the wind blows, and their power levels vary with the strength of the wind. The turbine power output is controlled by pitching the blades. With each new generation of wind turbines, the size has increased and reductions in the life-cycle cost of energy have been achieved through economies of turbine scale and a larger rotor to increase energy capture. However, there are constraints to this continued growth in size. At some point, it will cost more to build a larger turbine than the benefit of increased energy benefit is worth. In addition, land transport restrictions, cost as well as crane requirements, can impose size limits for wind turbines installed on land. A misconception about wind power [9] is that wind plants will cause the entire power system to collapse. But, because abrupt wind-related changes in plant output do not occur, this fear is unfounded. In fact, a modern wind plant will actually help a power system handle a major outage or contingency elsewhere on the system. Reactivepower control and low-voltage ride-through capabilities of modern wind plants actually improve system stability. Wind-energy generation only occurs when the wind is Page 770 Linga et. al., Combined Economic and Emission Dispatch for a Wind Integrated System Using Particle Swarm Optimization International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 769-775 ISSN 2078-2365 blowing [12]. Wind power is therefore not dispatchable like conventional energy sources and delivers a variable level of power depending on the wind speed. Wind is primarily an energy resource and not a capacity resource. Its primary value is to offset fuel consumption and the resulting emissions. The output of output of wind power plant, or multiple wind power plants, is variable over time. Each megawatt generated by wind reduces the required generation of other units. Therefore, the remaining nonwind generation units only need to supply the load that is not supplied by the wind. This remaining load is often called the net load. Therefore, the non-wind portion of the power system is operated to the net load, which is the difference between load and wind. Although wind is a variable resource, operating experience and detailed wind integration studies have yet to find a credible and firm technical limit to the amount of wind energy that can be accommodated by electrical grids. Some countries already receive a significant amount of electricity from wind power. There is not a technical limit to increased penetration of wind energy but there might be an economic limit, a point at which it is deemed too expensive to accommodate more energy from wind in comparision with the value that it adds to the system. IV. Problem formulation with Wind Integration A.CEED problem formulation with WIND Integration Minimize FT = (ai Pi2 + bi Pi +ci) + hm ( αi Pi2 + βi Pi + i) Where, FT is the total cost of generation (RS/hr). B. Constraints 1 Power Balance Constraint P ng gi +Pw = PD + PL (8) i1 where, ng = number of generating units, Pgi is the power generated by ith uint in MW, Pw is the wind power that is available in MW, PD is the load demand in MW, PL is the transmission loss in MW. 2. Generating Limits Pgimin ≤ Pgi ≤ Pgimax (i=1,2,3,....ng) ng= number of generators V. GENETIC ALGORITHM The aim of optimization is to determine the best-suited solution to a problem under a given set of constraints. Since the beginning of the nineteenth century, a significant evolution in optimization theory has been noticed [15]. Classical linear programming and traditional non-linear optimization techniques such as Lagrange’s Multiplier was prevalent until this century. Unfortunately, these derivative based optimization techniques can no longer be used to determine the optima on rough non-linear surfaces. One solution to this problem has been put forward by the evolutionary algorithms research community. Genetic algorithm (GA), enunciated by Holland, is one such popular technique which comes under evolutionary algorithms. Genetic Algorithm consists of a string representation of points in the search space, a set of genetic operators for generating new search points, a fitness function to evaluate the search points and a stochastic assignment to control the genetic operations [12]. It typically consists of three phases. 1. Initialization 2. Evaluation 3. Genetic Operation Initialization is the generation of initial population of chromosomes i.e. initial search points. Fitness function is so selected that the most fit solution is the nearest to the global optimum point. For minimization type problems, fitness function can be function of variables that bear inverse proportionality relationship with the objective function. The genetic operators are reproduction, crossover, and mutation. Reproduction is simply an operator where by an old chromosome is copied into a mating pool according to its fitness value. The commonly used method for selecting chromosomes for parents to cross over is Roulette Wheel selection, in roulette wheel selection technique, selection is usually implemented as a linear search through roulette wheel with slots weighed in proportion to string fitness values. The crossover is mainly responsible for the global search property of the GA. It is recombination operation. Here the gene information (information in a bit) contained in the two selected parents is utilized in certain fashion to generate two children who bear some of the useful characteristics of parents and expected to be more fit than parents. Usually, the probability of Crossover (PC) is high and chosen to be in between (0.6 to 0.8). Mutation operator is capable of creation new genetic material in the population to maintain the population diversity. It is nothing but random alteration of a bit value at a particular bit position in the chromosome. Usually, the probability of Mutation (PM) is very less and is chosen to be in between (0.001 to 0.01). We have another operator in GA, called Elitism. The copying of best population to next population is called Elitism. If the probability is high, then the convergence rate increases. Usually, the probability of Elitism (P E) is chosen to be 0.15 Page 771 Linga et. al., Combined Economic and Emission Dispatch for a Wind Integrated System Using Particle Swarm Optimization International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 769-775 ISSN 2078-2365 A. Implementation of CEED with Genetic Algorithm: Step7: Repeat the procedure from Step 2 until chromosome count > population size. Proposed Algorithm for solving CEED problem: Step1: Generate initial population of chromosome of binary bits using random generation technique. Step2: Implementation of a problem in a GA starts from the parameter encoding. Proposed approach uses the equal system λ (equal system incremental cost) criterion as its basis[11]. The only encoded parameter is the normalized system incremental cost, λnm. Decode the chromosomes of the population and determine normalized system incremental cost, λnm yj =  l i 1 2i-1 *bij (j=1,β,…L) (9) (10) (11) (12) (13) obtained (14) Step4:Calculate the generation output of all the units for each chromosome from its λf,e value and enforce P i limits. Pi = (bi + hm *βi) ng - 2 Bij Pj λ f,e i=1 2α +βhmαi ( i ) λ f,e (15) Step5: Calculate transmission losses using B-coefficient equation (5) and compute the error ε P ng ε = |PD-PL- gi | (16) i1 Step6: Calculate the fitness value of the chromosome, using the equation fitness= 1/ (1  50* ferror ) ferror= abs(Pgi-PD-Ploss) Step10: Copy the PE % chromosomes of old population to new population starting from the best ones from the top. Step11: Perform crossover on selected parents and generate new child chromosomes, repeat it to get required number of chromosomes. Step13: Perform mutation on all chromosomes. Step14: Replace old population with new population. l=length if string. bij =ith binary digit of jthstring. L=population size. 1- Step9: Check if the error is less than ε. if yes, go to Step 15. Step12: Add all the generated child chromosomes to new population. Step3:Calculate the actual system incremental cost λf,e Initial point in search space, λf,e is calculated as, λf,e = λmin + λnm(λmax – λmin) Calculating λmin and λmax values: α(i)=b(i)+hm*β(i) β(i)=β*(a(i)+hm*α(i)) λmin=α(i)+(β(i)*Pmin(i)) λmax=α(i)+(β(i)*Pmax(i)) The equivalent decimal integer of binary string λ is from: Step8: Sort the chromosomes and all their related data in the descending order of fitness. (17) (18) Step15: Calculate the total cost, fuel cost, emission release, emission cost, power generated by units. VI. Particle Swarm Optimization The aim of optimization is to determine the best-suited solution to a problem under a given set of constraints. Since the beginning of the nineteenth century, a significant evolution in optimization theory has been noticed [15]. Classical linear programming and traditional non-linear optimization techniques such as Lagrange’s Multiplier was prevalent until this century. Unfortunately, these derivative based optimization techniques can no longer be used to determine the optima on rough non-linear surfaces. One solution to this problem has been put forward by the evolutionary algorithms. Genetic algorithm (GA), enunciated by Holland, is one such popular algorithm which is a guided search technique. When it comes to evolutionary programming, techniques like Particle Swarm optimization (PSO) and Differential Evolution (DE) have been proposed. These algorithms are inspired by biological and sociological motivations and can take care of optimality on rough, discontinuous and multimodal surfaces. Particle Swarm Optimization (PSO) has been developed through simulation of simplified social models [2]. This algorithm is motivated by the behavior of organisms such as bird flocking and fish schooling and it utilizes a population based search procedure. The algorithm searches a space by adjusting the trajectories of individual vectors, called “particles” as they are conceptualized as moving points in multidimensional space. The individual particles are drawn stochastically towards the positions of their own previous best Page 772 Linga et. al., Combined Economic and Emission Dispatch for a Wind Integrated System Using Particle Swarm Optimization International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 769-775 ISSN 2078-2365 performance and the best previous performance of their neighbors. The particles are thought of as collision-proof birds and the original intent is to graphically simulate the graceful but unpredictable choreography of a bird flock. PSO is initialized with a group of random particles and then searches for optima by updating generations. Each particle in PSO represents a feasible solution. In other words, each particle represents a point in multi-dimensional search space, in which optimal point is to be determined. Each particle changes its state by ‘flying’ around the multi-dimensional search space until a relatively unchanging state (optimal state) has been obtained. In every iteration, each particle is updated by following two “best” values. The first one is the best solution it has achieved so far. This value is called “localbest”. Another “best” value that is tracked by the particle swarm optimizer is a global best and called “globalbest”. In PSO, the coordinates of each particle represent a possible solution that has two vectors associated with it, position (xi) and velocity (vi) vectors [16]. The size of the vectors xi and vi is equal to the problem space dimension. Each particle updates its position based on its own best exploration, best swarm overall experience, and its previous velocity vector according to the following equations: vik+1= wvik+c1r1(localbesti-xik)+c2r2(globalbesti-xik) (19) xik+1 = xik+vik+1 (20) where, c1and c2 are two positive constants, r1and r2 are two randomly generated numbers with a range of [0,1]. The first term of right-hand side of (19) corresponds to global search. The second and third terms of equation (19) corresponds to local search. So, this method has a well-balanced mechanism to utilize global and local search efficiently [2]. VII. Algorithm for solving Combined Economic and Emission Dispatch problem using PSO Power outputs from each generator are taken as the particles of the PSO [2]. The PSO algorithm for dispatch problems is stated as follows: Step1:The particles are randomly generated between the maximum and minimum operating limits of the generators. Step2:The particle velocities are generated randomly. Step3:Objective function values of the particles are evaluated. Penalties are given for violations of demand constraint (2).These values are set the localbest value of the particle. Step4:The best value among all the localbest values (globalbest) is identified. Step5:New velocities for the particles are calculated using (19). Step6:The positions for each particle are updated using (20). Step7:New objective function values are calculated for new positions of the particles. If the new value is better than the previous localbest, the new value is set to localbest. If the stopping criterion is met, the positions of the particles represent the optimum solution. If the stopping criteria is not met, the procedure is repeated from Step4. VIII. Results In GA, the population size is taken as 60,String length =16, P c =0.70, Pm =0.01, Pe =0.15 In the PSO technique, the population is taken as 40 and the values of c1 and c2 are c1=2 and c2=2. The techniques are tested on IEEE 30-bus system [17], having 6 generators and a total demand of 900MW. The cost coefficients for the generators and their capacities, the corresponding emission coefficients for the generators and the B-coefficients considered in[17], are mentioned in the appendix. The Economic Dispatch problem is solved using GA and PSO and the results are tabulated in Table I. The Emission Dispatch problem is solved using GA and PSO and the results are tabulated in Table II. Later, CEED problem is solved using GA and PSO and the results are tabulated in Table III. The CEED problem is solved using PSO taking into account wind integration [11]. For a total demand of PD=900MW, the Transmission losses, Fuel Costs, Emission release and Total cost are calculated without and with wind integration. Results are compared in Table IV. The parameters mentioned above are calculated with increasing wind penetration and the results are tabulated in Table V. Computations have been carried out in MATLAB 7.5 environment. Table I. Results comparing GA and PSO for Economic Dispatch problem Method Transmission losses (MW) Fuel cost (Rs/hr) Emission release (Kg/hr) GA 27.0153 50691.60 793.319 PSO 20.5518 45464.00 708.54 Page 773 Linga et. al., Combined Economic and Emission Dispatch for a Wind Integrated System Using Particle Swarm Optimization International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 769-775 ISSN 2078-2365 Table II. Results comparing GA and PSO for Emission Dispatch problem Emission release(Kg/hr) Method Total cost(Rs/hr) Transmission losses (MW) Emission release (Kg/hr) Fuel cost (Rs/hr) 452.228 406.069 61518.00 58022.00 IX. Conclusions GA 26.12 647.11 49634.30 PSO 16.96 646.13 48069.00 Table III. Results comparing GA and PSO for Combined Economic and Emission Dispatch problem Method Transmission losses (MW) Fuel cost (Rs/hr) Emission release (Kg/hr) Total cost (Rs/hr) GA 23.41 48029.00 687.52 78983.7 PSO 20.55 45464.00 603.61 72651.0 Table IV: Results comparing without and with wind integration. Parameters Transmission losses (MW) Fuel Cost(Rs/hr) Emission release(Kg/hr) Total cost (Rs/hr) Without wind integration With PW = 90 MW 20.55 16.728 45464.00 41150.00 603.61 452.228 72651.00 61518.00 Table V: Results with increasing wind penetration. WindPower integrated(MW) Transmission losses (MW) Fuel Cost(Rs/hr) PW = 90 PW = 120 16.728 15.533 41150.00 39732.00 In this paper, GA and PSO techniques are used to solve the Economic Dispatch problem, the Emission Dispatch problem and the Combined Economic and Emission Dispatch problems. The results showed that in Economic Dispatch problem, the objective being to reduce the fuel cost, the emission level was higher, also, PSO yielded in better results when compared with GA. Similarly, in Emission Dispatch problem, the objective being to reduce the pollution level, the fuel cost was higher, also, PSO yielded in better results when compared to GA. Hence, the CEED problem is formulated whose objective is to reduce both fuel cost and emission release. Results have shown that PSO yielded in better results when compared to GA. Later, when wind integration was taken into account, the results have shown that the transmission losses, fuel cost, emission levels and total cost can be reduced. Also, it was observed that the transmission losses, fuel cost, emission level and total cost can be reduced further with increase in wind penetration level. However, there may be a limit for the amount of wind power that can be integrated to a system. X. References [1]J.Nanda, D.P.Kothari, K.S.Linga Murthy ‘Economic-Emission load dispatch through goal programming techniques’, IEEE Trans. on Energy conversion, Vol.3, No 1, March1988. [2]T.Thakur, Kanik Sem, Simedha Saini and Sudhanshu Sharma ‘A Particle Swarm Optimization solution to NO2 and SO2 Emissions for environmentally constrained Economic dispatch problem’, β006 IEEE. [γ]James Kennedy and Russell Eberhart ‘Particle Swarm optimization’, 1995 IEEE. [4]J.W.Lamont and E.V.Obessis ‘Emission dispatch models and algorithms for the 1990’s’, IEEE Trans. on Power systems, Vol.10, No.2, May 1995. [5]A.A.El-Keib, H.Ma, J.L.Hart ‘Environmentally constrained Economic dispatch using Lagrangian relaxation method’, IEEE Trans. on Power systems, Vol.9, No.4, Nov. 1994. [6]Yong-Lin Hu and William G.Wee ‘A Hierarchical system for Economic dispatch with environmental constraints’, IEEE Trans. on Power systems,Vol. 9,No.2, May 1994. Page 774 Linga et. al., Combined Economic and Emission Dispatch for a Wind Integrated System Using Particle Swarm Optimization International Electrical Engineering Journal (IEEJ) Vol. 3 (2012) No. 2, pp. 769-775 ISSN 2078-2365 [7]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 Trans. on Power systems, Vol. 8, No.β, Mayβ00γ. [8]Chao-Lung Chiang ‘Improved genetic algorithm for power Economic dispatch of units with valve-point effects and multiple fuel’, IEEE Trans. on Power systems, Vol.β0, No.4, Nov.β005. [9]‘Wind plant integration’, IEEE power and energy magazine, Nov./Dec.2005. [10]‘A mighty wind’, IEEE power and energy magazine, March/April 2009. [11]H.T.Jadhav, Ananya Deb, Ranjit Roy ‘A craziness based Differential Evolution algorithm for thermal-wind generation dispatch considering Emission and Economy with valve-point effect’, 2011 IEEE. [12]‘Wind power myths debunked’, IEEE power and energy magazine Nov./Dec. 2009. [1γ] ‘Power system optimization’ D.P.Kothari, J.S.Dhillon [14]Dr.M.Sudhakaran,Dr.S.M.R.Slochanal,R.Sreeram, N.Chandrasekhar ‘Application of refined Genetic algorithm to Combined Economic and Emission dispatch’, β004 IEEE. [15]Swagatam Das, Ajith Abraham, Amit Konar ‘Particle Swarm optimization and Differential evolution algorithms: technical analysis, applications and hybridization perspectives’ Springer-Verlag Berlin Heidelberg 2008. [16] M.A.AlRashidi and M.E. El-Hawary ‘Economic dispatch with environmental considerations using Particle Swarm optimization’, 2006 IEEE. [17]M.Rajkumar,S.Kannan,K.Mahadevan,S.Ramesh 3 4 5 6 0.00683 0.00683 0.00461 0.00461 -0.54551 -0.54551 -0.51116 -0.51116 40.26690 40.26690 42.89553 42.89553 The Transmission loss coefficients matrix 0.00014 0.000017 0.000015 0.000019 0.000026 0.000022 0.000017 0.000060 0.000013 0.000016 0.000015 0.000020 0.000015 0.000019 0.000026 0.000022 0.000013 0.000016 0.000015 0.000020 0.000065 0.000017 0.000024 0.000019 0.000017 0.000071 0.000030 0.000025 0.000024 0.000030 0.000069 0.000032 0.000019 0.000025 0.000032 0.000085 ‘A Nondominated sorting genetic algorithm-II technique for environmental/economic power dispatch’, β010 IEEE. APPENDIX Fuel Cost Coefficients and Generator Capacity Limits Generator 1 2 3 4 5 6 a b 0.1524 0.1058 0.0280 0.0354 0.0211 0.0179 c 38.5397 46.1591 40.3965 38.3055 36.3278 38.2704 756.798 451.325 1049.99 1243.53 1658.55 1356.65 Pmin (MW) 10 10 35 35 130 125 Pmax (MW) 125 150 225 210 325 315 Emission Coefficients Generator 1 2 α 0.00419 0.00419 0.3276 0.3276 13.85932 13.85932 Page 775 Linga et. al., Combined Economic and Emission Dispatch for a Wind Integrated System Using Particle Swarm Optimization