Voltage Control Scheme Using Fuzzy Logic

for Residential Area Networks with PV
Generators in Saudi Arabia
R. A. Shalwala, Student Member, IEEE, and J. A. M. Bleijs, Member, IEEE

Abstract--One of the most important operational requirements
for any electrical power network for both distribution and
transmission level is voltage control. Many studies have been
carried out to improve or develop new voltage control techniques
to facilitate safe connection of distributed generation. In Saudi
Arabia, due to environmental, economical and development
perspectives a wide integration of photovoltaic (PV) generation in
distribution network is expected in the near future. This
development in the network may cause voltage regulation
problems due to the interface with the existing conventional
control system. Therefore, a new control system with PVs should
be developed. This paper introduces a new voltage control
scheme for residential area networks in Saudi Arabia based on
Fuzzy Logic concept (FL). The structure of two implementations
of FL controller to regulate the voltage by setting the on-load tap
changing transformer is proposed. In order to confirm the
validity of the proposed methods, simulations are carried out for
a realistic distribution network with real data for load and solar
radiation. Results showing the performance of each
implementation are presented and discussed.
Index Terms-- Distribution System, ETAP, Fuzzy logic
controller, Grid Connected, MATLAB, Photovoltaic Systems,
Saudi Arabia, Solar radiation, Voltage control

raditionally, the distribution network of the power system
is a passive network with a radial configuration.
Electricity flows one way from a substation to a large
distribution network. During normal operation or planning
period, a steady-state analysis of voltage regulation, system
losses, protection coordination, power quality, and system
reliability must be performed to ensure proper operation
within appropriate operating voltage range. Each utility has its
own operation and planning criteria depending on distribution
system characteristics and design criteria.
Currently, in Saudi Arabia, the exploitation of solar energy
as an alternative source of electric power is being considered
because of the abundant amount of irradiation and long hours
of sunshine. One way to achieve this is by using GridConnected
Photovoltaic systems (GCPV) on domestic dwellings directly
connected to the distribution network. This means that in the


This work was supported by the Ministry of High Education In Saudi
Arabia under Grant U208. The support of the Saudi Cultural Bureau in
London for sponsoring the fact-finding trip to Saudi Arabia to obtain the
required data is gratefully acknowledged.
R. A. Shalwala is with the Department of Engineering, University of
Leicester, Leicester, UK (e-mail: rs234@leicester.ac.uk).
J. A. M. Bleijs is with the Department of Engineering, University of
Leicester, Leicester, UK (e-mail: jamb1@leicester.ac.uk)

©2010 IEEE

future the system performance will be affected by PV
generators. According to [1-3] distributed generation has both
advantages and disadvantages for the system.
In this paper two implementations of fuzzy logic technique
have been used to maintain the voltage in a residential area
network with high penetration of PV generators in Saudi
Arabia. The ETAP simulation package has been used for
power flow calculation and the MATLAB software package
has been used to design the fuzzy logic controller.
Generally, distribution utilities deliver electric energy to
their customers within an appropriate voltage range to meet
customer requirements. For a radial configuration the bus
voltage, voltage drop, power flow, and power loss can be
calculated by using a simplified model such as the two-bus
system as shown in Fig. 1 [4].
| |





| |


Fig. 1. Model of a two-bus distribution system.
The model consists of a short distance line represented by a
series connection of resistance (R) and inductive reactance
(X). In this case, real and reactive power (
between bus #1 and bus #2 is described by (1) and (2).








is the voltage angle at bus #1
is the voltage angle at bus #2
is the admittance angle

Power loss between bus #1 and bus #2 is given by eq.(3):

| |

In addition, the voltage at bus #2 and the voltage drop
between theses buses can be calculated in terms of the voltage
at bus #1 by using (4) and (5), respectively.


In a typical distribution system there are many scenarios to
be considered, and to handle calculation in a large system,
power system simulation software is required. In this paper,
the power systems simulation package ETAP is used for
evaluating of steady-state performance under different load
and PV generation conditions.

1344 I1 I2 15 0.0 10.1344 Fig. I2. Load Conditions Real load data for the selected residential area has been collected from the Saudi Electricity Company in the Western Region (SEC-WR) who has also provided details of the transformers and lines of the network. a SG with unity power factor is used in ETAP to represent the PV generator.1344 I3 I4 153 0..128+j0.0 Average 0 2 4 6 8 10 12 14 16 18 20 22 24 Time (hour) Fig. Therefore.TABLE I III.0 Jun 50. 3 shows the average daily load profile for this area during each month.128+j0.0 Sep 30. this reduces the power factor (PF) of the total load to approximately 0.0 May 60. A considerable part of the load is due to air conditioning (A/C) systems and in general the load reaches its maximum between noon and 16:00 h in summer. the longest branch which is ISK11 will be the most sensitive because it will have the largest variations for the different load and irradiation scenarios. Based on this information a detailed model of the distribution network has been created in ETAP. The current generation of PV inverters operate at unity power factor.Peak load (Maximum Summer load) 3.0 Average Load Aug 40..1344 I7 I8 648 0. Test residential network with 3 branches There are 6 load nodes tapped off from each feeder.128+j0. their behavior during steady state is similar to that of a Synchronous Generator (SG) with unity power factor. 2.Normal load (Annual Average load) 4. A.1344 I6 I7 132 0. Since this study is concentrating on the effect of GCPV on the voltage regulation in the network under normal operation (no faults).2).38 kV transformer (see detail in Fig. The line parameters of this branch are shown in Table I. The following conditions of each consumer load in the network will be considered in this research: 1. Each branch is equipped with on-load tap changing transformer (LTCT) that has the ability of changing the voltage level of the branch at the main substation bus in small steps (0. This area includes 5000 residential properties.128+j0.. RSF04. Since in this research Building Integrating photovoltaic (BIPV) system will be used to address the effect of such .128+j0.1344 I5 I6 396 0. 2). Average Daily load for a residential Area Jan 100. connecting to a customer feeder through another stepdown11/0.. SYSTEM MODELING LINE PARAMETERS OF ISK11 The following model of a real distribution network in a residential area will be used as a base case in this paper (Fig.0 90. through a step-down 110/11 kV power transformer at each primary substation (ISK11.0 Apr 70.128+j0.5% of nominal voltages).128+j0.85. This type of load can reach 65 per cent of the total load during summer and since the AC systems are motor-driven. R1. The main branches can be interconnected through normally-open circuit breakers in the case of outage of one of the 110/11 kV transformers.1344 I8 I9 255 0.. 3. Average daily load of a residential area.. and ISK10) connecting to 3 branches.1344 I4 I5 513 0. PV Generators PV generators are connected to the grid through powerelectronic inverters.R7.0 Peak Load Feb Mar Laod (MW) 80. adjusted by the automatic voltage controller (AVC).0 Oct 20. Fig. So.128+j0.S1.. Each branch includes a number of secondary substations (labeled as I1.Extreme load (based on the maximum capacity of customer circuit breaker) 2.128+j0. The minimum load is equal to about 30% of summer peak load.1344 I2 I3 327 0.I9.Light load (30% of peak load) B.and S9).0 Jul Nov Dec Light Load 0. The distribution network starts from the Station bus at 110 kV. From Bus To Bus Lenghth (m) Impedance (Ω/km) ISK11 I1 1875 0.

0%. hence it will adjust its voltage level to 1.5% steps). The AVR will therefore assume that the branch load is at thee minimum level. 5. Fig. .-2. However. Fig. Also. So. the range of power that can be generated using concentrating PV system between 9:00 to 15:00 for each single house is between 14. together with the preferred position of tap chaanger. This figure comes from about 29 different designs of residential houses in Saudi Arabia. The output power of the PV generator which will be delivered to the customer or the network can be estimated as: (6) Based on the previous information and bby using (6).Max PV = 36 kW 2. a 90% efficiency of the inverter is considered in designing the PV ggenerator in order to increase the penetration level.5%. coordination between PVs and the LTCT larger penetration and better voltage control. The montthly average solar radiation that was recorded at Jeddah meteorology station for year 2002 is shown in Fig. The standard deviation for all vo oltage nodes in the system from the nominal is used to determine the best possible position of LTC (-5. This creates an unpredictable and un ncontrolled situation where the voltage level of all nodes migh ht or might not be within acceptable limits. The simulations show that both best b and preferred position for all scenarios improve the voltag ge level and keep it within the allowable limits for all customerss in the branch. Table T II shows the results for all scenarios.0%. The averaage available roof area for PV installation on houses in a residenntial area is about 100m . IMPACT OF PV ON PRESENT VOLTAG GE CONTROL At each branch of Fig. The solar radiation data for this study hhas been obtained from King Abdul-Aziz City for Science and Technology (KACST) which has 40 stations around the ccountry recording the solar radiation every 5 minutes. which is no longer a good indication of the feeder status. which estimates the voltage drop oover the branch by measuring the branch current at the mainn substation end.Voltage profile of all feeders co onnected to ISK11 branch. The worse cases are shown in Fiig. since it depends mainly on measuring the branch current att the main substation end.u. then the voltage profile along the feeders deepends mainly on Voltage % of Nominal Solar Radiation(W/m^2) 1100 1000 900 800 700 600 500 400 300 200 100 0 the PVs’ location. 4 shows clearly the large amount oof solar radiation between 9:00 to 15:00 throughout the yeaar with minimum value of 400 W/m^2 and maximum valuee of about 1000 W/m^2. and Fig. and the PV power factor. SCENARIOS AND ASSUMPTION In order to proof that the adjusttment of the LTC only is sufficient to keep the voltage level along a the branch within the permissible level. are carrying the most of the branch lo oad. It has been assumed that t the PVs are connected to each node in the system with equal power. concentrating PV modules (CPV) with 40% efficiency has been assumed.4 kW 3. Average daily Solar Radiation Jan Max Radiation Voltage profile ( Light load.No PV IV. The preseence of GCPV on feeders makes the power flow bi-directionaal. Also.0 00 p. this method assumes that the poweer is unidirectional from the main substation bus (where the AVR is located) flowing to the end of the branch. connected to feeders along a ISK11 in Fig. Monthly average solar radiation.0% to +5. in order to reduce the number of tap ch hanging operations the best results are rounded to the nearestt preferred position of (5. 5 shows the voltage profile of each feeeder in this case. the load is assumed to be same for all houses.0% in n 0. Jedddah 2002. 4. Therefore. to make the scenarios more realistic a further 3 random conditions of the load feeders are also considered. 106 105 104 103 102 101 100 99 98 97 96 95 94 I1 I2 I3 I4 I5 I6 I7 I8 I9 Upper limit 1 2 Lower Limit 3 4 5 6 Customer Nodes Fig. the available roof area of the selected buildings must be known. power generated. 6. Max x PV&0% tap) Feb Mar Apr Minimum Midday Radiation May Jun Jul Aug No Radiation Sep Oct Nov 0 2 4 6 8 10 12 14 16 18 20 22 24 Dec Time (hour) Fig 4. nventional voltage control It can be concluded that the con technique cannot properly adjust th he voltage level of feeders with various PVs connected to them m. The current flowing from the LTCT will be very low. A V. consider one of the worse scenarios at light load and a maximum PV. However. 2. and if the PVs connected to feeders are carrying most or alll the branch load. To illustrate such a problem. In this study. 7.+2.Minimum midday PV = 14. the followiing conditions for PV will be considered in this research: 1.4 kW to 36 kW.5%. However. PV status and combinations. (0% Tap).+5%). the system hass been simulated for all possible scenarios of load con nditions. This will w increase the life time for the tap changer and reduce the disturbances d in the system due to changing the LTC setting. a new technique must be developed to facilitate the L for safe integration.system on voltage regulation. In this case the PVs. 2 the voltage leveel is regulated by the AVR.

8.33 +22. Just as fuzzy logic caan be described simply as "computing with words rather than n numbers" fuzzy control can be described simply as "control with sentences rather than equations". v rather than forcing the use of precise statements and exact e values.12 00. High (H).0% +2.u.5% ( Extreme load.5% 0.31 +00.) Centralized system +5. By suitable selection of input-output linguistic variables and a rule base. High (H). Voltage profile of XLNPV scenario with prefferred position of tap ( Light load.5% XLMPV Extreme Max 99.32 +33.48 +22. This control system is simulated in MATLAB software. .44 +00.5% LLMPV Light Max 103.0% -2.91 +11.5% Average Voltage +5.0% I2 I3 I4 I5 I6 I7 I8 I9 1 2 3 4 5 6 Fig.0% NLMPV No Max 104. an nd simpler [6].5% 0.0% XLAPV Extreme Min 96. 9. Max PV & -5%tap) 106 105 104 103 102 101 100 99 98 97 96 95 94 I1 I2 I3 I4 I5 I6 I7 I8 I9 Upper limit Lower Limit 1 2 3 4 5 6 Customer Nodes Fig.04 -11.10. 7. Normal (N).0% ALMPV Average Max 102.39 +55.0% +2.19 +22.0% R3LMPV Random3 101.0% +2.5% 0.58 -11. 1st Implementation of FLC Fig.56 +00.5% -5.0% 0. Normal (N).5% +2.5% +5.5% +2. 6.35 -44. fuzzy logiic controllers allow the representation of imprecise human n knowledge in a logical way.0% 0.92 +33.0% -5.5% -2.5% -2.5% Avr V Fuzzy Logic Controller Tap position 0.0% -2. the preferred tap changer setting.07 -22.5% -2.5% +2. they conv vert the linguistic control strategy based on expert knowledgee into an automatic control strategy [5]. Very V High (VH).34 +22.43 -11.0% LLAPV Light Min 101.0% LLNPV Light No 99.0% -2.5% ALAPV Average Min 100. in contrast to o mathematical models or other expert systems.5% R3LAPV Random3 Min 99.5% R2LAPV Random2 Min 98. a broad range of desirable control outcomes caan be achieved. FUZZY LOGIC CONTROL ALNPV Average No 98. Low (L) and Very low (VL) as shown in Fig.0% R2LMPV Random2 Max 101. 1st Implementatiion of FLC 1) The membership of input and output o signals: Input: Average voltage (AvgV) of alll customers in the branch.5% XLNPV Extreme No 94. thus making them more robust.5% R1LAPV Random1 Min 99. The T controller consists of one input. Voltage profile of LLMPV scenario with preferrred position of tap Input variable “Avg gV” Fig. with approximate terms and values. the average customer vo oltage. No PV & +5% tap) 106 105 104 103 102 101 100 99 98 97 96 95 94 Voltage % of Nominal Max I1 Fuzzy Logic Control (FLC) offers an alternative to conventional controllers when theree is no available accurate model of the system to be controlled d.0% +2. Therefore.0% R1LNPN Random1 No 97.5% R2LNPV Random2 No 96. and one output. closer adherence to set point conditions if desired.0% 0. All customers voltages BEST & PREFERRED POSITION OF TAP CHANGER FOR R ALL SCENARIOS Average Scenario Load PV B Best Preferred V(p. Low (L) and Very Low (VL) as shown in Fig.71 -22.33 +00. Upper limit Lower Limit Voltage % of Nominal Customer Nodes Fig. and the ability y to explicitly set the tradeoff between energy costs and inteerior environment. Input membership function for 1st implementation of FLC Output: Tap changer setting (TC).9. By providing an algorithm.5% PLMPV Peak Max 101.0% PLAPV Peak Min 98.5% 0.51 +33. Very High (VH).TABLE II VI.0% +5. Possible features might include user-specified d overall control 'tightness' analogous to a control range. Fuzzy logic controllers consist of a set of o linguistic control rules based on fuzzy implications and th he rules of inference.5% R3LNPV Random3 No 97.95 -22. more compact.0% R1LMPV Random1 Max 102.94 -44. 8 shows the straight forward application of fuzzy logic controller based on the numerical solution for the preferred tap changer position at each scenario.0% PLNPV Peak No 96.

The small differences in Fig. it needs the installation of a digital voltmeter at each customer to send the voltage values to a centrralized system via a communication network.5% +5.0% NLMPV -4134 208 -5.0% LLAPV -1502 130 -2.0% PLNPV 1619 1004 +5.13. Table IV shows the measurements of active power (P) and reacttive power (Q) at ISK11.5% ALAPV -915 486 0.0% ALMPV -3382 638 -2. It is quite clear that there is a poositive correlation between the load of the branch and the reactiive power flow at ISK11.0% -2. 11 shows the difference between the numerically calculated and proposed FLC setting for the ttap changer.5% R3LNPV 1168 720 +2.TABLE IV POWER FLOW MEASUREMENTS @ ISK11 FOR F SCENARIOS USING ETAP mentation of FLC Fig. 5% 4% 3% 2% 1% 0% -1% -2% -3% -4% -5% Numerical Fuzzy 94 95 96 97 98 99 100 101 102 103 104 1055 Average Costomer voltages Fig.5% P @ ISK11 Q @ ISK11 Fuzzy Logic Controller Tap position 0. .5% R2LNPV 1327 824 +2. 12.0% R3LMPV -3046 841 -2. average load→ medium Q (ML) and light load→ light Q (LL) as shown in n Fig. no PV→ high P (NoPV).5% XLMPV -1859 1540 0. Scenario P@ISK11(kW) Q@ISK1 11(kVar) Preferred tap ALNPV 805 479 +2. Also.14. Fig. According to the new relationnships. Numerical and Fuzzy Logic settingg for LTC The FLC gives almost the same results as the numerical solutions. This makees this solution prohibitively expensive. 13. Input membership function 1 forr 2nd implementation of FLC Input2: Active power (P) @ ISK11. B. 11.5% +5. 12 can be set up. extreme and peak load→ high Q (HL).5% XLNPV 2444 1549 +5.0% XLAPV -670 1480 +2.5% LLMPV -3946 315 -5.0% LLNPV 200 99 0. Minimum PV→ medium P (MinPV)) and maximum Power→ light P (MaxPV) as shown in Fig. the system in Fig. Output membership function for 1st implem 2) The control rules: TABLE III RULE TABLE FOR 1ST IMPLEMENTATION O OF FLC Input: AvgV VH H N L VL Output: TC VL L N H VH Tap changer Setting Fig. 10. as the load increases the reactivee power increases and vice versa. Fuzzy Logic controllerr implementation 2 where: 1) The membership of input and output o signals: Input1: Reactive power (Q) @ ISK11. The m main advantage of this way of control is that it is independent oof branch and line parameters and can be applied to various tyypes of networks.0% R2LMPV -2899 927 -2. there are many techniqques other than FL can be used based on look-up tables for thee average voltage and then set the preferred tap changer possition which may does not give any advantages of using FLC.5% R1LAPV -609 683 0.5% R3LAPV -565 709 0.0% PLAPV -127 974 +2. However.0% R1LMPV -3088 818 -2.0% +2. 9 aree because the full range of variation in average voltage is not used in the numerical calculation (scenarios only). So. the correlation between the local measurements in the system and the preferred setting have to be established based on engineering sense. 2nd Implementation of FLC In order to find a better solution than prevviously introduced for controlling the voltage in the branch without using a communication network and take advantaage of the FLC features.0% Fig.0% R1LNPN 1122 691 +2. if all requiirements for this method are available.5% R2LAPV -411 805 0.5% PLMPV -2624 1082 0. On the other hand therre is a negative correlation between the PV generation and the active power flow at ISK11.

"Fuzzy Logic in Control Sy ystems: Fuzzy Logic Controller. in 2006. Low (L).S. from electrical generators and pow wer electronics to power systems and advanced controllers. M. High (H). F. 1. 22 – 29. Kirrschen. pp. Jul 21-25 2002.15.0% [3] Davis. Davis. UK.C. and G. Leicester. 1995. Shalwala a received the B. No. U. The first implementation is asssociated with significant costs due to hardware cost and the need to widespread communication infrastructure but itt can be applied to many networks since it is independent on the t network parameters.I. Very Highh (VH).. London . Issue 1. pp.5% 0.” Proceedings of Society Transmission and Distribution n Conference.K. His research interessts are in power system planning and operation. The Netherlands. 54-61. 404-43 35. 15. Vol. the membership for the input has to be tuned byy determining the preferred position numerically for the poweer flow values in the transition region. 1.131-140 IX. Santamou uris.B. Jan-Feb 2003. The second implementation sho ows a novel technique to control the LTCT based on the pow wer flow information at the transformer itself. 2. the main drawbacck of this method is that it depends on the network paraameters and the load characteristics. Jenkins. Jeddah. 16 shows the tap changer settings for poossible range of P and Q using the proposed FLC. VII. P. P. “Distributed Generation: Semantic Hype or the Dawn of a New N Era?. Jul 21-25 2002.0% 2010 470 -1070 -2. He is currently pursuing the Ph." IEEE Transactions on Sysstems. Guaraccino. Johannes (Hans) Bleijs B received his MSc degree in Electrical and Electronic E Engineering from Eindhoven Univeersity of Technology. pp. in 2002 and the M. Tap changer setting 5.mentation of FLC Fig. M degree from University of Nottingham. VIII. In order to optimizee this solution. Parts I & II. A. The main advantaage of this implementation is that all measurements are taken lo ocally and there is no need for remote communication with other o information in the system. ACKNOWLED DGMENT Fig. This makes this solution simple and cheapp compared with other control technique. Normal (N). So. "Comparison of conventional and fuzzy y control of indoor of indoor air quality in buildings. Michel. W. 20. pp 62-69. Output membership function for 2nd implem mentation of FLC [2] 2) The control rules: TABLE V OF FLC RULE TABLE FOR 2ND IMPLEMENTATION O P→PV Max PV No PV Min PV Q→Load HL VH H N ML H N L LL N L VL Fig. and SEC-WR main office in Jeddah for providing the utility grid data. March/April 1990. G. REFEREN NCES [1] Fig. degree at Un niversity of Leicester. C. Vol. where he teaches Electrical Machines and Power Systems. 1996. Saudi Arabia. It haas been found that both proposed FLCs have the ability to impprove the voltage profile of distribution network and keepp it within the permissible limits. the results are encouraging and warrant further investigation using the fuzzy logic concept in such problems. Bruant. Since 1991 he has been a Lecturer in Electricaal Engineering in the Department of Engineering at th he University of Leicester. P.S. CONCLUSIONS Two methods of implementation the oof a fuzzy logic controller for setting the tap changer positioon in distribution network were investigated in this paper. U. . Nottin ngham. HIES X. Laambert. His field of research covers a wide range of subjects in Renewablee Energy Conversion and Energy Storage. D. degree from King AbdulAziz Un niversity.R. “Distributed Resourrce Electric Power Systems Offer Significant Advantages Over Central Station Generation and T&D Power Systems Part II. Embedded generation: The Institution of Electricall Engineers. R. In general. M. 1. Murray. IEEE Vol. 16. In 1983 he joined Imperial College in London as a a Research Associate working on integration off wind turbines with diesel generators.D. Puttgen..0% 2.C. He wass awarded a PhD degree from Imperial College in 1990. Vol. Journal of Intelligent and Fuzzy Systems. Lee. both in electrical engineerin ng. MacGregor.” Proceedings of o the IEEE Power Engineering Society Transmission and Distribution n Conference.5% 1400 -4150 1550 950 1250 Reactive Power Q(kVar) 1100 650 800 350 [4] [5] [6] The authors would like to thanks KACST for providing the solar irradiance data. 14. pp. ren newable energy resources and distribution network k. Strbac. in 198 82. Murray W. Crossley. However. Allan.K. Input membership function 2 for 2nd implem Output: Tap changer setting (TC). The key benefit of this implementation is that all m measurements are taken locally and there is no need for remotee communication. Dounis.” Power and Energy Magazine. BIOGRAPH -2610 500 50 200 -5. H. Man and Cybernetics. for each network k the FLC need to be set up based on analysis of the network load l data. Raed A. Fuzzy Logic set for LTC based on powerr flow @ ISK11 The controller gives the preferred tap chaanger position for all scenarios when the values of P and Q value of each scenario have been used as an input to the this system.". Very low (VL) as show wn in Fig. N. “Distributed Resourrce Electric Power Systems Offer Significant Advantages Over Central Station Generation and T&D o the IEEE Power Engineering Power Systems Part I.