Virtual Power Plants, Distributed Generation, Microgrids, Renewables and Storage (IEEE SmartGridComm)

Assessment and mitigation of voltage violations by

solar panels in a residential distribution grid
Tom Verschueren∗ , Kevin Mets∗ , Bart Meersman† , Matthias Strobbe∗ , Chris Develder∗ , and Lieven Vandevelde†
∗ Dept. Information Technology – IBCN, Ghent University – IBBT, Ghent, Belgium
† Dept. Electrical Energy, Systems and Automation (EESA) – EELAB, Ghent University, Ghent, Belgium
Email: chris.develder@intec.ugent.be, bart.meersman@ugent.be

Abstract—Distributed renewable electricity generators, such these solar panels, since maximum production is often reached
as solar cells and wind turbines introduce bidirectional energy at moments when the inhabitants don’t need a lot of power.
flows in the low-voltage power grid, possibly causing voltage For example because they are working during daytime or they
violations and grid instabilities. The current solution to this
problem comprises automatically switching off some of the local are on holiday during the summer. When they arrive at home
generators, resulting in a loss of green energy. In this paper we and start cooking, washing, etc. solar production might be low,
study the impact of different solar panel penetration levels in an especially during the winter. Unfortunately, electricity has to
residential area and the corresponding effects on the distribution be used the moment it is generated as the power grid represents
feeder line. To mitigate these problems, we assess how effective the ultimate just-in-time product delivery [3].
it is to locally store excess energy in batteries. A case study on a
residential feeder serving 63 houses shows that if 80% of them When PV power production is high and consumption is low,
have photo-voltaic (PV) panels, 45% of them would be switched extra power is injected into the grid, possibly causing voltage
off, resulting in 482 kWh of PV-generated energy being lost. We violations and making the grid unstable. On the other hand, to
show that providing a 9 kWh battery at each house can mitigate ensure that all households can be served in case consumption is
some voltage violations, and therefor allowing for more renewable high and production is low, the presence of PV panels does not
energy to be used.
necessarily decrease the required amount of (spare) capacity
I. I NTRODUCTION that must be available all the time, requiring high investments
and resulting in expensive energy bills.
The power grid is moving away from the current centralized In this paper we study the impact of distributed solar power
power generation paradigm. With governments promoting lo- production in a distribution grid by determining power flow
cal renewable power generation at residential sites, distributed and voltage fluctuations at the grid connection of each house-
power generation is gaining in popularity. Environmental con- hold, taking available generation capabilities (in casu, PV
cerns and efforts to become less dependent on fossil fuels panels) and consumption patterns into account. Thus, we will
are the driving force for the replacement of traditional energy assess which PV panels will be turned off because of voltage
sources by green alternatives. The EU 20-20-20 targets aim violations, and as such assure grid stability. This has the
for a reduction in EU greenhouse gas emissions of at least undesirable side effect that some of the PV energy is actually
20%, 20% of EU energy consumption to come from renewable wasted, and not injected into the grid. We subsequently assess
energy sources, and a reduction in energy consumption of how we can alleviate these problems by providing local storage
20% [1]. Currently, a lot of wind farms are constructed, capacity with batteries.
offshore as well as on land, and in residential areas more and The remainder of this paper is structured as follows. In
more houses are covered with solar panels. Section II related work is presented. The simulation tool we
By combining power grid technologies with information and developed and used to obtain our case study results is briefly
communication technologies a future smart grid will be able to described in Section III. The set-up of our case study, including
deal with the unpredictable and distributed nature of these new the models for generation, consumption and storage, is detailed
forms of power generation. Typical issues of the future power in Section IV. The results and discussion of the observerd
grid stemming from the presence of distributed generation effects of distributed PV generation in a residential area is
(DG) include [2]: voltage and frequency instabilities as a result presented in Section V. The final Section VI summarizes our
of local power generation, power security issues resulting conclusions and provides some thoughts on future work.
from bidirectional energy flows. In addition, given the less
predictable nature of renewable energy sources, the demand- II. R ELATED W ORK
supply matching problem becomes more challenging, and With the ever increasing instalment of renewable energy
associated control algorithms that cater for flexibility of loads sources at residential areas, the impact of distributed genera-
(i.e., that can be shifted in time) become more complex. tion on electrical power systems has been of major interest [4]–
In residential neighbourhoods with houses equipped with [6].
photo-voltaic (PV) panels, there is often a mismatch between The effect of the capacity and location of distributed gen-
the production and consumption of the power generated by erators in the distribution grid has been discussed in [6].

978-1-4577-1702-4/11/$26.00 ©2011 IEEE 540

That analysis is based on the voltage stability at each of
the 69 nodes in the considered distribution system, which is Application
calculated using the steady state voltage index method [7].
Different approaches at determining system stability with Middleware
different penetration levels of DG are used in [5], which,
apart from the the voltage stability metrics, includes stability
Support
analysis and the assessment of the oscillatory behaviour of the
power system. The power system consists of a high voltage Network Electrical
and two low voltage networks, with a total number of 56 end-
user nodes, each connected to a fuel cell and micro turbine.
Battery energy storage systems are interesting to support the Fig. 1. Three-layer architecture of the simulation environment.
stability and reliability of the power grid and for efficiently
using the energy generated by unpredictable and intermittent
sources such as solar panels and wind turbines. Especially III. S IMULATION FRAMEWORK
electrical vehicles receive a lot of attention as possible power To study smart grid scenarios such as the one we are
storage and source as they have a higher economic potential addressing in the current paper, we resort to simulations.
than dedicated storage systems (V2G principle [8], [9]). To mitigate the supply and demand challenges stemming
The authors of [10] present recent advances in energy from e.g. the intermittent nature of renewable sources, it is
storage systems with respect to ease of maintenance, costs and clear that information and communication technologies (ICT)
size. Different types of energy storage systems are discussed can play a major role and will be crucial to enable smart
including different types of batteries, flywheels, superconduct- grid applications [15], [16]. Whereas both the communication
ing magnetic energy storage, ultracapacitors, microturbines network and the power system research communities have
and fuel cells. Also [11] gives an overview of different developed ample simulation tools, we believe there is a need
kinds of energy storage systems together with a number for an innovative framework that couples both worlds in an
of benefits, both technical (e.g. grid voltage support, peak integrated view. Therefore, we have developed a modular
shaving, reliability, etc.) as financial (e.g. reduction of need for simulation framework based on OMNeT++ [17], reusing some
extra generation capacity, increased revenue from renewable of the communication network modules already available, but
sources, etc.). Furthermore, some modelling and simulation adding extra models for the electrical network. Below, we give
approaches are discussed as well as a number of challenges a high level overview of the simulator design, and outline the
that need further research. power model used. For details, we refer to [18].
In [12], the siting and sizing of electricity storage is A. High-level design
optimized stochastically using an multi-objective genetic al- As sketched in Fig. 1, we designed our smart grid simulation
gorithm. The optimal integration of battery energy storage environment as a layered architecture, comprising three major
units is analysed with respect to several objectives including layers: application, middleware and support layers. The ap-
minimization of total installed power and energy capacity, plication layer consists of high-level applications or services,
minimization of large voltage deviation probabilities and max- for example advanced meter reading services, demand side
imization of total revenue on a power exchange market. management services, or billing services. The services in the
The work of [13] presents a game-theoretic framework for application layer make use of the middleware layer, which
modelling storage devices in large-scale systems where each provides generic functionality that can be used by any service.
storage device is owned by a self-interested party that aims to This includes e.g. service/device discovery functions as well
maximise its monetary profits. The designed storage strategy as a message-based communication interface to send messages
results in reduced peak demands and lower electricity bills. between service components, independent of underlying net-
working technology (e.g. ZigBee or PLC; TCP or UDP). The
Besides already mentioned benefits such as reducing peak
goal of this middleware layer is to support a broad range
demands, avoiding voltage violations, lowering grid invest-
of applications while reducing the effort required to develop
ment costs and reducing energy bills for the end users,
these services to a minimum. The support layer, composed
battery energy storage systems have the additional benefit of
of the network and electrical components, provides support
being able to protect the privacy of the end user by hiding
functions for the layers above. Communication between ser-
information contained in consumption data [14].
vices is simulated by the network component that provides
Our work discussed below also addresses issues of storage simulation models for multiple types of physical media and
in residential scenarios, but differs from the above works communication protocols. The simulation environment must
in that it assesses how storage can be exploited to counter be able to model and interact with (virtual) electrical devices.
voltage violation issues stemming from fairly high penetration This is supported by the electrical component of the simulator,
of residential photo-voltaic installations. as outlined below. Basic electrical models are provided (e.g.

541
PV panel, battery, electric vehicle), but the user can add his
own models. The characteristics of the specific models used
in this paper are discussed in Section IV.
B. Power network model
Our current research interest mainly pertains to the dis-
tribution grid, since the major research questions to be ad-
dressed are in this part of the power network. Therefore,
we implemented a distribution grid model to allow assessing (a) Overall set-up
potential problems, including in particular voltage violations,
stemming from distributed generation. Our current model
applies to radial feeder networks, comprising one or more
feeders that in turn may include other subfeeders. Loads
or generators can be connected anywhere along the feeders.
The model is based on the fast harmonic simulation method
presented in [19]. It uses an iterative forward/backward sweep
method to analyse the radial distribution system. Each iteration
consists of two parts: the backward and forward sweep. The
backward sweep determines the currents in every node, based
on known voltages in each node. Subsequently, the currents in
(b) House module
all network branches are determined. Next, the forward sweep
determines the voltage at every node. After each iteration, Fig. 2. The case study set-up in the simulator.
the voltage at every node is compared with the voltage in
the previous iteration. If the difference is below a certain
threshold, the iteration process is stopped. We realized the The overall layout of our case studies, as represented in our
iterative method as a MATLAB module, which is linked to the simulation framework is illustrated in Fig. 2. The main setup of
OMNeT++ modules through appropriate interfaces, to allow the OMNeT++ simulation environment is shown in Fig. 2(a),
e.g. specification of the number of houses, the loads etc. where each house icon on the right represents multiple (21)
House modules. All of those house modules are connected to
IV. C ASE STUDY SET- UP a specific feeder, coming from the powerGrid module. The
To assess the impact of PV panels on a distribution feeder, powerGrid module provides a high level representation of a
we have considered a radial feeder topology, as presented distribution grid, and allows for the configuration of several
in [19], comprising three feeders, with 21 houses on each of feeders that define the distribution grid.
them. All of the houses are connected on the same phase, A House module, detailed in Fig. 2(b), acts as a container
and are placed equidistantly on feeders with a length of 1000 for a number of devices that influence the electrical behavior of
meters. The low voltage lines used in this simulation are the house. The baseload models the consumption of electricity
underground cable lines. by the whole of all household appliances. All these devices, as
To assess the impact of the PV panels, we will compare well as the PV panel and the battery, are connected to the smart
three cases: meter of the house which keeps track of the net power supplied
• NoPV: will illustrate the baseline scenario without any to (or possibly by) the home. Thus, in our case studies, each
PV panels or batteries, as a reference case. household will be characterised by
• PV-ideal: will show the idealized case where we do not • A base load profile: the local electricity consumption
take into account any voltage violations when assessing profile stemming from typical household appliances;
energy flows. Thus, in this case all produced PV energy • A PV panel: the solar panel, if present;
can always be used (either locally in the house, or pushed • A battery, if present.
into the grid). The properties of these models used are described in more
• PV-real: will show the realistic case, where PV panels detail in the subsections below.
are shut off the grid when they cause voltage problems.
Therefore, in this case we will have no voltage violations, A. Base load profile
but part of the generated PV energy will be lost. The base load profiles are based on the synthetic load
• PV+batt: will indicate how the use of batteries may profiles (SLP), made available by the Flemish Regulation
mitigate the voltage violations. In this case, we will Entity for the Electricity and Gas market (VREG) [20]. These
assume each household with a PV installation is also SLP profiles model typical user consumption using statistical
equipped with a battery. This battery will then be used averages on real life data, as measured by the VREG: an SLP
during daytime to store excess energy produced by PV curve contains for each 15 minute interval, over a complete
panels, and to release it in the evening/night time. year, the amount of energy consumed (in kWh). We used such

542
SLP data representing a residential consumers. In particular,
we used data for the first of July of the 2011. profile published
by VREG. The total yearly power consumption was set at
3500kWh, corresponding with normal residential usage [20].

B. PV panel model
To model the production of the PV panels, the same
approach as for the base load profiles has been used. The
European PV GIS [21] database contains averaged solar power
irradiation profiles, similar to the (consumption) SLP files. The
the same generation profile (average solar power irradiance
in July at Ghent, Belgium) is used for all PV panels in the
distribution network, albeit altered with a random uncertainty
factor of 5 percent a each location. Next to this irradiance
profile, the PV model also takes a surface (squared meters)
as parameter. By multiplying the irradiance value (W/m2 ) (a) Percentage of PV installations shut off from the grid
with the surface (m2 ), we get a simplified model omitting
any efficiency factors. The results presented in the next section
used an average PV panel surface of 6m2 . To account for vari-
ances between households, we randomly assigned PV surfaces
and efficiency factors drawn from a uniform distribution with
aforementioned averages and a maximal deviation of ±10%.

C. Battery model
We used a simplified, lossless model of battery cells, which
we assume to be characterized by the following parameters:
(i) maximum storage capacity EB (in kWh), (ii) maximum
decharge power Pout (in kW), and (iii) maximum charge
power Pin (in kW). Thus, when charging a battery at time t,
it will draw p(t) ≤ Pin from the grid, while it will discharge
at p(t) ≤ Pout . For simplicity, in this work we disregard any
battery inefficiencies such as self-discharge, loss of capacity
(e.g. due to memory effects), etc. For the case study below, (b) Energy lost due to PV shut-off
we assumed Pin = 1kW and Pout = 0.7kW.
Fig. 4. Percentage of PV panels being shut down and the resulting renewable
V. R ESULTS AND DISCUSSION power lost, when no battery is being used.

Figure 3 below show the power levels (W) as observed at

the main feeder of the distribution grid, and hence showing
the total net power consumption or production of the entire that more PV panels are being shutdown in the 80% case and
distribution grid network. Three cases are presented, for dif- thereby missing out on potentialy green energy. This can be
ferent PV penetration levels: we consider 50%, 70% or 80% explained by the fact that more panels results in more power
of the houses to have a PV installation.. being injected in the grid, which results in a overvoltage risk.
Figure 3 shows both the PV-ideal and PV-real scenario. In To mitigate the problems of unwanted shutdowns of PV
the 50% penetration case, no voltage violations occurred and panels caused by the voltage instabilities, we introduce a
therefor no PV panels were disconnected from the grid, result- battery in the houses that have a PV panel. The battery we
ing in a match between the PV-ideal and PV-real measurements assumed is a rather large model of 9 kWh, corresponding to
(note that therefor the curves in Fig. 3(a) overlap). For 70% or two 4 kWh batteries such as used in [14]. A deterministic al-
80% penetration however, the injected energy by the PV panels gorithm for controlling the charging/discharging of the battery
reaches a sufficiently high threshold which causes overvoltage is implemented, charging the battery during the daytime (e.g.
problems in the grid. When a PV panel senses overvoltage at when solar power is available) and discharging the battery
its grid connection point, it will disconnect itself from the grid. completely during the night, when no energy is produced by
This will remain disconnected as long as voltage violations the PV panels: the charge interval is set to 08:00–16:00, while
persist. the remaining time the battery is decharged. When discharging
The percentage of PV panels that are shutdown over time the battery, we assume that the released energy is used to
are displayed in Fig. 4(a) for both cases. The actual power power the household appliances. Thus, only if the discharge
lost is shown in Fig. 4(b). These two figures clearly show rate of the battery is higher than the consumption rate of the

543
 (a) PV penetration 50% (b) PV penetration 70% (c) PV penetration 80%

Fig. 3. The net power consumed by the whole of the three feeders comprising the distribution grid spanning 63 houses.

TABLE I
P ERCENTAGE OF PV PANELS BEING SHUTDOWN & ENERGY LOST
Percentage shut down (%) Energy lost (kWh)
PV-Real 70% 34% 291 kWh
PV-Real: 80% 45% 482 kWh
PV-Battery 70% 16% 121 kWh
PV-Battery 80% 28% 289 kWh

house, excess power is injected into the grid.

Because part of the PV-generated power is now stored in the
battery, the impact of PVs on the voltage profile is alleviated.
This results in less PV panels having to shutdown due to
overvoltage problems, as seen in Fig. 5(a) and Fig. 5(b).
The resulting net flow at the main distribution grid’s low-
voltage transformer supplying the three feeders connecting
the houses, is now shown in Fig. 3. There, both the lowered
production during the day and reduced consumption at night
is clearly visible.
Table I gives an overview of the number of PV panels
(a) Percentage of PV shut down
having to shut down for the problematic penetration levels, for
both cases (with or without battery). Even the very straightfor-
ward (de)charging policy we adopted clearly shows the benefit
of providing energy storage. However, we note that there still
is an opportunity to optimize the storage capacity and charging
policy to further alleviate observed voltage problems (and
associated waste of captured solar energy by PV panels).
VI. C ONCLUSION AND F UTURE W ORK
The primary concern addressed in this paper is the voltage
violations caused by the presence of PV panels in a typical
residential distribution grid. For a radial grid layout comprising
63 houses, we assessed (i) the net power consumed/produced
on the feeders, (ii) the amount of PV panels that would be shut
off to avoid voltage violations and (iii) how these problems can
be mitigated by providing batteries for storing excess energy.
We analyzed this set-up using our simulation tool based on
(b) Energy lost due to PV shut-off
OMNeT++.
Considering an increasing amount of PV installations, we Fig. 5. Percentage of PV panels being shut down and the resulting amount
observed that for the case at hand, the grid could sustain 50% of power lost, when using a battery to mitigate voltage problems.
of houses having PV without any problems. However, when
PV penetration increases to 70% (respectively 80%), we note
that up 34% (resp. 45%) of the PV panels will be shut down

544
during daytime. This corresponds 291 kWh (resp. 482 kWh) [8] W. Kempton and J. Tomic, “Vehicle-to-grid power implementation: From
of energy that thus is lost. To investigate possible mitigation stabilizing the grid to supporting large-scale renewable energy,” Journal
of Power Sources, vol. 144, no. 1, pp. 280–294, 2005.
of this problem, we investigated the effect of adding batteries [9] K. Liyanage, A. Yokoyama, Y. Ota, T. Nakajima, and H. Taniguchi, “Im-
(9 kAV each) to each of the PV installations. Using a simple, pacts of communication delay on the performance of a control scheme
deterministic charge/decharge algorithm set to charge during to minimize power fluctuations introduced by renewable generation
under varying v2g vehicle pool size,” in Smart Grid Communications
daytime, and release the energy at nighttime, we saw that we (SmartGridComm), 2010 First IEEE International Conference on, oct.
can alleviate many of the unwanted PV shut-offs, and thus 2010, pp. 85 –90.
recover about half of the otherwise lost energy. [10] E. Furling, M. Piemontesi, P. Prasad, and D. Sukumar, “Advances in
energy storage techniques for critical power systems,” in The Battcon
Future work will address (i) investigation of selective/opti- 2002 proceedings, 2002.
mized battery installations at possibly a subset of the houses, [11] A. Mohd, E. Ortjohann, A. Schmelter, N. Hamsic, and D. Morton,
(ii) the study of ideal battery (de)charging schedules, using e.g. “Challenges in integrating distributed energy storage systems into future
smart grid,” in Proc. 2008 IEEE Int. Symp. Industrial Electronics, 2008.
optimisation techniques as in [22], and (iii) online distributed [12] F. Geth, J. Tant, E. Haesen, J. Driesen, and R. Belmans, “Integration of
coordination strategies to attain optimal battery (de)charging energy storage in distribution grids,” in Proc. IEEE Power and Energy
under uncertain load/production profiles. Society General Meeting, 2010.
[13] P. Vytelingum, T. D. Voice, S. D. Ramchurn, A. Rogers, and N. R.
ACKNOWLEDGMENT Jennings, “Agent-based micro-storage management for the smart grid,”
in Proc. 9th Int. Conf. Autonomous Agents and Multiagent Sys., ser.
Work presented in this paper has been supported by the AAMAS ’10. Richland, SC: International Foundation for Autonomous
Flemish Government through the LINEAR project (through Agents and Multiagent Systems, 2010, pp. 39–46.
[14] G. Kalogridis, C. Efthymiou, S. Z. Denic, T. A. Lewis, and R. Cepeda,
IWT) and the project SmartE (an IBBT ICON project). K. “Privacy for smart meters: Towards undetectable appliance load signa-
Mets would like to thank the Agency for Innovation by Sci- tures,” in Proc. 1st IEEE Int. Conf. Smart Grid Commun. (SmartGrid-
ence and Technology in Flanders (IWT) for financial support Comm 2010), Gaithersburg, MD, USA, 4–6 Oct. 2010, pp. 85–90.
[15] , “ICT for energy efficiency,” EU Commission, Tech. Rep., 2008.
through his Ph.D. grant. C. Develder is partially supported by [16] G. Andersson, M. D. Ilić and, V. Madani, and D. Novosel, “Network
the Research Foundation – Flanders (FWO – Vlaanderen) as systems engineering for meeting the energy and environmental dream,”
a post-doctoral fellow. Proc. IEEE, vol. 99, no. 1, pp. 7–14, Jan. 2011.
[17] A. Varga and R. Hornig, “An overview of the OMNeT++ simulation
R EFERENCES environment,” in Proc. 1st Int. Conf. Simulation Tools and Techniques for
Commun., Networks and Systems (Simutools 2008), Marseilles, France,
[1] M. Webb, SMART 2020: enabling the low carbon economy in the 3–7 Mar. 2008, pp. 60:1–60:10.
information age, a report by The Climate Group on behalf of the Global [18] K. Mets, T. Verschueren, C. Develder, T. L. Vandoorn, and L. Vande-
eSustainability Initiative (GeSI), 2008. velde, “Integrated simulation of power and communication networks for
[2] E. Coster, J. Myrzik, B. Kruimer, and W. Kling, “Integration issues of smart grid applications,” in Proc. 16th IEEE Int. Workshop Computer
distributed generation in distribution grids,” Proc. IEEE, vol. 99, no. 1, Aided Modeling, Analysis and Design of Commun. Links and Netw.
pp. 28–39, Jan. 2011. (CAMAD 2011), Kyoto, Japan, 10–11 Jun. 2011.
[3] US Department of Energy, “The smart grid: An introduction,” 2008. [19] L. Degroote, L. Vandevelde, and R. B., “Fast harmonic simulation model
[4] M. Reza, P. Schavemaker, J. Slootweg, W. Kling, and L. van der Sluis, for the analysis of network losses with converter-connecter distributed
“Impacts of distributed generation penetration levels on power systems generation,” Electric Power System Research, vol. 80, pp. 1332–1340,
transient stability,” in Power Engineering Society General Meeting, Nov. 2010.
2004. IEEE, june 2004, pp. 2150 –2155 Vol.2. [20] VREG, “Flemish regulation entity for the electricity and gas market,”
[5] A. Azmy and I. Erlich, “Impact of distributed generation on the stability Jan. 2010. [Online]. Available: http://www.vreg.be/
of electrical power system,” in Power Engineering Society General [21] PVGIS, “Photovoltaic geographical information system (pvgis),” Jan.
Meeting, 2005. IEEE, june 2005, pp. 1056 – 1063 Vol. 2. 2011. [Online]. Available: http://re.jrc.ec.europa.eu/pvgis/
[6] N. G. A. Hemdan and M. Kurra, “Distributed generation location and [22] K. Mets, T. Verschueren, W. Haerick, C. Develder, and F. De Turck,
capacity effect on voltage stability of distribution networks,” in Proc. “Optimizing smart energy control strategies for plug-in hybrid electric
Annual IEEE Conf. Student Paper, Aalborg, Denmark, 15 Feb. 2008. vehicle charging,” in Proc. 1st IFIP/IEEE Int. Workshop on Management
[7] M. Chakravorty and D. Das, “Voltage stability analysis of radial distri- of Smart Grids, at 2010 IEEE/IFIP Netw. Operations and Management
bution networks,” International Journal of Electrical Power & Energy Symp. (NOMS 2010), Osaka, Japan, 19–23 Apr. 2010, pp. 293–299.
Systems, vol. 23, no. 2, pp. 129 – 135, 2001.

545