Electric Power Systems Research 80 (2010) 893–897

Contents lists available at ScienceDirect

Electric Power Systems Research

journal homepage: www.elsevier.com/locate/epsr

A fuzzy expert system for loss reduction and voltage control in radial
distribution systems
A.Y. Abdelaziz ∗ , S.F. Mekhamer, M.H. Nada
Electrical Power and Machines Department, Faculty of Engineering, Ain Shams University, 1 Elsarayat Street, Abdo Basha Square, Abbassia, Cairo, Egypt

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, two methods for improving voltage profile and minimizing total system losses in radial
Received 8 June 2007 distribution feeders are presented. The first method concerns with the capacitor allocation problem.
Received in revised form Fuzzy expert system (FES) is used to select the best candidate nodes for capacitors to be installed in order
29 September 2009
to maximize total loss reduction and total net savings. The second method illustrated the voltage regulator
Accepted 30 December 2009
problem. In this method, the location and the tapping ratio of the voltage regulators are determined in
Available online 20 January 2010
order to minimize total system losses while keeping the voltage within specified limits. The two methods
have been applied to two test feeders. Comparison with other techniques is included to ensure methods
Keywords:
Capacitor placement
validity and superiority.
Fuzzy expert system © 2010 Elsevier B.V. All rights reserved.
Voltage regulator
Power loss reduction
Voltage profile

1. Introduction
Optimal reactive power and voltage control for distribution
system involves the installation of fixed shunt capacitors and tap-
changing voltage regulators along distribution feeders. This can
improve voltage profiles for all end-use customers and reduce real
power losses. Many studies and several algorithms have been pre-
sented for reactive power compensation problem solution. These
algorithms are classified into analytical, numerical programming;
heuristic techniques and artificial intelligence based (AI-based)
methods [1–3]. Also, many papers have dealt with the problem of
capacitor allocation in distribution systems using artificial neural
networks (ANN). Some of these papers are devoted to ANN-based
control of capacitors in distribution systems for the purpose of
power loss reduction with and without considering voltage reg-
ulators [4–6].
Although these previous methods to solve the capacitor alloca-
tion problem have various merits, their efficiency rely entirely on
the goodness of the data used. Fuzzy-set theory (FST) provides a
remedy for any lack of uncertainty in the data. Furthermore, fuzzy
logic has the advantage of including heuristics and respectively
engineering judgements into the capacitor allocation optimization
process.
∗ Corresponding author. Tel.: +20 10 1372930.
E-mail addresses: almoatazabdelaziz@hotmail.com, ayabdelaziz@gawab.com
(A.Y. Abdelaziz).
Mekhamer et al. [7] have presented two new heuristic tech-
niques by assuming every node in the system as a candidate node.
They selected the node that gives maximum cost reduction (first
technique) and the node that gives maximum loss reduction (sec-
ond technique).
In Ref. [8], a fuzzy-based approach for capacitor placement in
a distribution system has been presented. The problem as a fuzzy-
set optimization problem has been formulated to minimize the real
power loss and the system cost with voltage limiting constraints.
Two membership functions of voltage sensitivity and real power
loss have been stated and the intersection principle was used as a
decision making to find the optimum locations at which capacitors
should be installed.
In Ref. [9], the authors presented the same membership func-
tions used in Ref. [8] but they replaced the real losses by reactive
losses membership function and instead of using the intersection
principle as a decision making, they used the dot product to deter-
mine the suitable locations of capacitors.
In Ref. [10], the authors made some modifications in the weight-
ing factors of the membership functions presented in Refs. [7,8]
which give better results by giving an importance of 20–30% to the
voltage sensitivity index and an importance of 70–80% to the power
loss index.
A fuzzy logic based algorithm to select the best candidate nodes
by creating two input membership functions for voltage and power
loss index and an output one for capacitor placement suitability
has been presented in Ref. [11]. Capacitor sizing algorithm and
variational technique are used to determine the optimal size of

0378-7796/$ – see front matter © 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.epsr.2009.12.020
894 A.Y. Abdelaziz et al. / Electric Power Systems Research 80 (2010) 893–897
capacitor to be placed at the chosen node for most power loss
reduction.
In Ref. [12], a new method for optimum capacitor allocation
is also presented based on successive combinatorial optimization
technique. This technique finds the optimum capacitor sizes by
scanning the possible locations without ordering or replacement.
In Ref. [13] and its two companion Refs. [14,15] by the same
authors, the problem of Volt/VAr control on general radial distri-
bution systems is formulated, simplified and solved. The objective
was to minimize the peak power and energy losses while keeping
the voltage within specified limits under varying load conditions.
It is shown in Ref. [13] that the regulator (Volt) and the capacitor
(VAr) problem may be treated as two decoupled problems. The two
decoupled problems are expressed as two independent optimiza-
tion problems. The solution of the problem is given in Ref. [14]; the
numerical results on typical distribution system of 30 buses with 6
lateral branches are given in Ref. [15].
In Ref. [16], the authors presented a new method for determin-
ing the minimal changes in transformer taps to control voltage
levels. They solved a non-linear constraint optimization. The con-
straints are tackled in the minimization algorithm by forming the
Lagrangian. Necessary conditions for optimality are derived and
solved using the Newton–Raphson method.
In this paper, a fuzzy logic based technique is applied to two
standard test feeders in order to obtain the suitable nodes for capac-
itor placement such that the total system losses are minimized.
Also, a proposed technique for determination of voltage regulators
locations and their taping ratios is used and applied to the feeders
mentioned above. The results show the success of the two meth-
ods in improving voltage profile and maximizing total power loss
reduction.
2. Problem formulation
The overall approach to reduce the total power loss, minimize
the total cost function and improve the voltage profile by keeping
the bus voltages within specified limits can be divided into two
separate (individual) problems:
(A) The capacitor problem
(B) The voltage regulation problem
2.1. The capacitor problem
The capacitor problem is to determine the location and size of
fixed shunt capacitors to be installed in distribution system such
that the total system power loss is minimized and the voltage pro-
file is improved while the net annual dollar savings are reduced.
Considering investment cost, there is a finite number of stan-
dard capacitor sizes that are integer multiples of the smallest size
Qco . Generally, larger sizes are cheaper than smaller ones. The max-
imum permissible capacitor size is
Qc max = L × Qco
where L is an integer. The total annual cost function due to capacitor
placement and real power loss change is written as follows:

k Table 1
Cost = Kp Ploss + Kjc Qjc
j=1
where Kp is the annual cost of kW losses in $/kW, j = 1, 2, . . ., k are
the indices of buses selected for compensation, K c corresponds to
the capital investment per kVAr and Q c represents the standard
capacitor sizes considering investment cost.
The entire framework to solve the optimal capacitor allocation
problem includes the use of fuzzy expert system. First, a load flow
program of Ref. [17] calculates the power loss in each section of
the feeder. The power loss factor is then computed by dividing
the power loss in each section by the total system power loss.
These power loss factors with the per unit node voltages are the
inputs to the FES which determines the node most suitable for
capacitor installation by fuzzy inferencing. Finally, a numerical pro-
cedure (variational method) is used to determine the optimal size
of capacitor to be placed at the chosen node for maximum power
loss reduction. The above procedure is repeated until no additional
power loss reduction from the installation of capacitors is achieved.
The algorithm of the variational method is summarized in the
following steps:
(1) Assume that a capacitor (Qc ) is to be placed at bus 1. Try all the
available values of Qc at this bus. For each value of Qc , perform
a load flow computation and evaluate the cost function. Record
Qc that gives minimum cost and this cost as well.
(2) Repeat step (1) for all buses. Then place Qc1 at the bus (bus K1 )
that has the lowest minimum active power losses.
(3) With the optimal Qc placed at bus K1 , repeat steps 1 and 2 to
select the next candidate bus (K2 ) where the next optimal Qc2
shall be placed.
(4) With Qc1 and Qc2 placed at bus K1 and K2 , repeat steps 1 and 2,
and so on until no further candidate bus can be found.
2.1.1. FES implementation
The FES contains a set of rules which are developed from qual-
itative descriptions [18]. In a FES, rules may be fired with some
degree using fuzzy inferencing; whereas, in a conventional expert
system, a rule is either fired or not fired. For the capacitor allocation
problem, rules are defined to determine the suitability of a node
for capacitor installation. Such rules are expressed in the following
form:
IF premise (antecedent), THEN conclusion (consequent).
For determining the suitability of capacitor placement at a par-
ticular node, a set of multiple-antecedent fuzzy rules have been
established. The inputs to the rules are the voltage and power loss
factors, and the output consequent is the suitability of capacitor
placement. The rules, their consequents are summarized in the
fuzzy decision matrix in Table 1. The fuzzy variables, power loss
factor, voltage, and capacitor placement suitability are described
by the fuzzy terms high, medium/normal, low. These fuzzy vari-
ables described by linguistic terms are represented by membership
functions are graphically shown in Figs. 1–3.
2.1.2. Fuzzy inferencing and defuzzification techniques
After the FES receives inputs from the load flow program, sev-
eral rules may fire with some degree of membership. The fuzzy
inference method based on the Mamdani maximum–minimum
implication method of inference involves truncating the conse-
quent membership function of each fired rule at the minimum
membership value of all the antecedents. A final aggregated mem-
(1) bership function is achieved by taking the union of all the truncated
consequent membership functions of the fired rules. For the
capacitor allocation problem, the resulting capacitor placement
(2) Decision matrix for determining suitable capacitor locations.
AND Voltage
Low Normal High
Power loss factor Low Medium Low Medium
Medium Medium Medium Medium
High High Medium High
 A.Y. Abdelaziz et al. / Electric Power Systems Research 80 (2010) 893–897
Fig. 1. Power loss factor membership functions.
Fig. 2. Voltage membership functions.
suitability membership function, s of node i for k fired rules is
given by:
s (i) = max[min[p (i), v (i)]]
k
where p and v are the membership functions of the power loss
factor and voltage level, respectively.
Once the suitability membership function of a node is calculated,
it must be defuzzified in order to determine the node’s suitability
ranking. The centroid method of defuzzification has been chosen.
This method of defuzzification finds the center of area of the mem-
bership function. Thus, the capacitor placement suitability index is
determined by:

s (z).zdz
S= 
s (z)dz
2.2. The voltage regulation problem
The voltage regulator problem is to determine the minimum
required number of voltage regulators, their locations and their tap
positions in order to minimize the total power losses while keeping
the voltage within prespecified tight limits Vmin ≤ Vi ≤ Vmax where
Vmin , Vmax are the minimum and maximum voltage limits. In the
Fig. 3. Capacitor placement suitability membership functions.
895
Fig. 4. The ␲ equivalent representation for the voltage regulator.
solution of this problem, it is assumed that the distribution system
is considered to be optimally compensated by fixed capacitors as
already described and the cost of the voltage regulators is not taken
into consideration.
2.2.1. The regulator model
A voltage regulator is usually represented by a ␲ equivalent net-
work [12] as shown in Fig. 4. In general the relationship between
the unregulated voltage profile and the regulated voltage profile is
non-linear after an adjustment of the tap ratio and is determined
by a load-flow analysis.
(3) In what follows a simple scheme is described to place the min-
imum number of voltage regulators on the system considering the
optimal tap control.
STEP-1: Choose the location of the regulator as the closest bus
to the substation at which the voltage drops below Vmin .
STEP-2: Set the tap of the STEP-1 regulator to the optimal setting
as follows:
Vmax
a= (5)
Vbi
where a and bi are the tap ratio and the bus location of the ith
(4) voltage regulator, respectively.
Remark: Usually the maximum tap ratio amax which may be user
specified, should be considered during the placement scheme. If it
should happen that at a candidate location, say of the ith regulator,
the following occurs:
Vmax
ai = > amax (6)
Vbi
Then this implies that the optimal control will not be achieved if
the candidate location is chosen. Therefore, a nearby bus at which
the limit is not yet violated may be chosen and Eq. (6) is rechecked
until the proper location is found.
STEP-3: Perform load flow and find all bus voltages and total
system losses.
STEP-4: Check that all bus voltages are satisfying Vmin ≤ Vbi .
If the check shows that this is true, stop. If not, repeat STEPS-1–3
for the next regulator.
3. Implementation and results
In the following section, the implementation results of the two
problems (capacitor allocation problem and voltage regulator prob-
lem) applied to practical feeders are presented and comparisons to
other proposed approaches are included.
896 A.Y. Abdelaziz et al. / Electric Power Systems Research 80 (2010) 893–897
Fig. 5. The 9-bus test feeder.
Table 2
Results for different methods applied to the 9-bus feeder.
Reference no. Qc locations (node no.) Total kW losses
Ref. [7] 5, 4, 9, 8 688.677
Ref. [11] 4, 7, 6, 8 686.306
Ref. [12] 2, 4, 5, 9 682.074
Proposed FES method 4, 8, 5, 2 675.820
3.1. The capacitor allocation problem
3.1.1. First feeder
The 9-bus 23 kV radial distribution feeder presented in Ref.
[8] is the first feeder to be studied. The total active losses and
cost function before compensation are 783.8563 kW and $131,690
respectively with minimum voltage of 0.837 p.u. occurs at node
nine assuming the substation voltage to be 1 p.u. In this feeder Kp
is selected to be $168/kW. The system is shown in Fig. 5.
The FES applied to this feeder chooses nodes 4, 8, 5, 2 as candi-
date nodes for capacitors to be installed. The variational technique
chooses 3450, 450, 600, 4050 kVAr for nodes 4, 8, 5, 2 respectively.
This results in total loss reduction of 108.036 kW, thus a cost reduc-
tion of $16,530 with a minimum voltage of 0.8845 p.u. at bus (9).
The results of applying FES method and other proposed approaches
are shown in Table 2. Fig. 6 shows the difference in total cost
between other approaches and the proposed FES approach.
3.1.2. Second feeder
The proposed FES method discussed before is applied to another
22-bus, 11 kV radial distribution test feeder. The feeder data is pre-
sented in Ref. [12]. Before compensation, the total active losses are
157.147 kW, cost function is $26,401 and the minimum voltage is
0.8934 p.u. occurs at node 22. Applying the proposed FES method to
this feeder, nodes 7, 12, 22 are recommended for capacitor place-
ment with rated capacitor sizes obtained by variational method of
600, 750, 450 kVAr for each node respectively. The evaluated total
power loss and cost function are 95.1483 kW and $16,438 respec-
tively. The results implemented using other approaches and the
proposed FES approach is illustrated in Table 3. Fig. 7 shows the
total active losses using the proposed FES and other approaches.
Fig. 6. Total cost for different methods applied to the 9-bus feeder.
Table 3
Results for different methods applied to the 22-bus feeder.
Reference no. Qc locations (node no.) Total cost ($)
Ref. [7] 6, 14 16,424
Ref. [11] 17, 7 16,344
Proposed FES method 7, 12, 22 16,438
3.1.3. Comments on the results
For the first feeder, the total cost using the proposed FES method
is less than that using other techniques. However, for the second
feeder, it is higher, as the techniques in Refs. [7,12] recommended
only 2 nodes for capacitors to be placed, but the proposed FES
method suggested three nodes (7, 12, 22) for capacitors to be
installed with capacitor sizes of 600, 750, 450 kVAr for each node,
respectively.
There are two factors that affect the total system cost; the first
one which has the major effect is the total active power losses while
the second one which has the minor effect is the size of the installed
capacitors. Therefore, for the first feeder, the cost of the proposed
FES method has the lowest value as the total active power loss
resulted from using this method is the lowest one compared with
other methods. But for the second feeder, the size of the installed
capacitors has a great effect in the total system cost as the total
active power losses in all techniques are almost the same. So, the
proposed FES method leads to higher cost as the installed capacitors
in this method have the largest sizes among all other methods.
A comparison with the heuristic techniques presented in Ref.
[7]: derivation of the equations, problem formulation and problem
modifications cited in appendix of Ref. [7] show that the proposed
technique presented in this paper is simple and needs less mathe-
matical formulations. In addition, as known the heuristic searches
may reach a solution and sometimes may not.
In Ref. [11], the authors used Mamdani max.-product impli-
cation method of inference while in this paper the Mamdani
max.–min. implication method of inference is used which leads to
better results. Also in this paper, the range of power loss factor
membership function is more accurate and precise (0–0.3) while in
Ref. [11] the range is wide and less accurate (0–1). Therefore, better
results are obtained.
3.2. The voltage regulator problem
After compensation with fixed shunt capacitors, the scheme
described in Section 2.2.1 for voltage regulation problem is applied
to the two feeders. For the first feeder (9-bus feeder) the required
minimum and maximum voltages are 0.95 and 1.05 respectively
with maximum voltage regulator tap ratio of (1.1). The nearest
bus to the substation at which the voltage drops below 0.95 is
Fig. 7. Total active losses for different methods applied to 22-bus feeder.
 A.Y. Abdelaziz et al. / Electric Power Systems Research 80 (2010) 893–897
Table 4
Voltage control results for the two test feeders.
Voltage profile
Using capacitors only
Minimum voltage (p.u.) Maximum voltage (p.u.)
First feeder 0.8845 1.0041
Second feeder 0.9573 1.0000
Table 5
Power loss results for the two test feeders.
kW losses
Using capacitors only Using capacitors and
tap-changing voltage
regulators
First feeder 675.82 627
Second feeder 95.15 92
bus 6, however, if the first regulator is placed at the bus, the tap
ratio to boast the voltage to Vmax = 1.05, in order to provide opti-
mal control, will require a tap ratio of (1.05/0.947), i.e. greater
than the allowed ratio (1.1). Thus this location is not appropri-
ate. The next possible place is bus 5; the optimal tap ratio is
obviously a = 1.05/0.956 = 1.098. Performing load flow program
using Newton–Raphson method results in total active losses of
627 kW, minimum voltage of 0.9792 p.u. at bus 4 and maximum
voltage of 1.0495 p.u. at bus 5. For the second feeder (22-bus feeder)
the required minimum and maximum voltages are 0.98 and 1.02
respectively with the same voltage tap ratio of (1.1). The nearest
bus to the substation at which the voltage drops below 0.98 is bus
4. The optimal tap ratio is a = 1.02/0.9773 = 1.044. After performing
load flow program, the total active power loss is 92 kW, minimum
voltage is 0.9819 p.u. at bus 3 and maximum voltage is 1.0193 at
bus 4.
Tables 4 and 5 present a comparison between the results
obtained for voltage control and power loss reduction using:
(1) Capacitors only.
(2) Capacitors and tap-changing voltage regulators.
The results obtained from applying tap-changing voltage regula-
tor show the success of this method in minimizing the total power
loss and minimizing the voltage profile within prespecified tight
limits.
4. Conclusions
This paper presented two methods for minimizing total power
loss and improving voltage profile in radial distribution feeders.
The first method illustrated the capacitor allocation problem using
fuzzy expert system. In this method candidate nodes are selected
by FES for maximizing total loss and cost reduction. The second
method deals with the voltage regulation problem for minimizing
the total system losses and improving the system voltage profile.
The two methods are applied to two practical feeders. The results
compared to other previous work show the superiority of these pro-
posed methods in voltage improvement and power loss reduction.
897
Using capacitors and tap-changing voltage regulators
Minimum voltage (p.u.) Maximum voltage (p.u.)
0.9792 1.0495
0.9819 1.0193
References
[1] H.N. Ng, M.M.A. Salama, A.Y. Chikhani, Classification of capacitor allocation
techniques, IEEE Transactions on Power Delivery 15 (January (1)) (2000)
387–392.
[2] K.P. Wong, Artificial Intelligence and Neural Network Applications in Power
Systems, in: Proceedings of the IEE 2nd International Conference on Advances
in Power System Control, Operation and Management, Hong Kong, December,
1993, pp. 37–46.
[3] H.N. Ng, M.M.A. Salama, A.Y. Chikhani, A Survey of the Application of AI in
Capacitor Allocation and Control, in: Proceedings of IEEE Canadian Conference
on Electrical and Computer Engineering, vol. 1, May, 1997, pp. 161–164.
[4] N.I. Santoso, O.T. Tan, Neural Net-based real-time control of capacitors installed
on distribution systems, IEEE Transactions on Power Delivery 5 (January (1))
(1990) 266–272.
[5] A.Y. Abdelaziz, T.S. Abdel-Salam, B.I. Saad, M.A. El-Sharkawy, A neural network-
based approach for control of capacitors installed on distribution systems for
power loss reduction, in: Proceedings of the Eighth International Middle-East
Power Systems Conference MEPCON’2001, Helwan University, Egypt, Decem-
ber, 2001, pp. 255–259.
[6] B. Das, P.K. Verma, Artificial neural network-based optimal capacitor switching
in a distribution system, Electric Power Systems Research 60 (December (2))
(2001) 55–62.
[7] S.F. Mekhamer, S.A. Soliman, M.A. Mostafa, M.E. El-Hawary, New heuristic
strategies for reactive power compensation of radial distribution feeders, IEEE
Transactions on Power Delivery (October) (2002) 1128–1135.
[8] H.C. Chin, W.M. Lin, Capacitor placement for distribution systems with fuzzy
algorithm, in: Proceedings of the IEEE Ninth Annual International Conference,
TENCON ‘94, vol. 2, August, 1994, pp. 1025–1029.
[9] H.N. Ng, M.M.A. Salama, A.Y. Chikhani, Capacitor placement in distribution sys-
tems using fuzzy technique, in: Proceedings of the IEEE Canadian Conference
on Electrical and Computer Engineering, vol. 2, May, 1996, pp. 790–793.
[10] S.F. Mekhamer, S.A. Soliman, M.A. Mostafa, M.E. El-Hawary, Application of fuzzy
logic for reactive power compensation of radial distribution feeders, IEEE Trans-
actions on Power Systems (February) (2003) 206–213.
[11] N.N. Abdel Gawad, S.F. Mekhamer, Y. Galal, E. Zakzouk, Fuzzy Logic Application
for Capacitor Allocation in Radial Distribution Feeders”, MEPCON ‘05, Port Said,
December, 2005, pp. 613–619.
[12] N.N. Abdel Gawad, S.F. Mekhamer, Y. Galal, E.E. Zakzouk, Application of artificial
intelligence on radial distribution feeders: capacitor allocation problem, in:
Proceedings of the IEEE Portugal Power Conference on Artificial Intelligence in
Energy Systems and Power, AIESP, February, 2006, pp. 7–13.
[13] J.J. Grainger, S. Civanlar, Volt/Var Control on Distribution Systems with Lateral
Branches using Shunt Capacitors and Voltage Regulators. Part I. The Overall
Problem, in: Proceedings of the IEEE Transactions on Power Apparatus and
Systems, vol. PAS-104, No. 11, November, 1985, pp. 3278–3283.
[14] S. Civanlar, J.J. Grainger, Volt/Var Control on Distribution Systems with Lateral
Branches using Shunt Capacitors and Voltage Regulators; Part II: The Solu-
tion Method, in: Proceedings of the IEEE Transactions on Power Apparatus and
Systems, vol. PAS-104, No. 11, November, 1985, pp. 3284–3290.
[15] S. Civanlar, J.J. Grainger, Volt/Var Control on Distribution Systems with Lateral
Branches Using Shunt Capacitors and Voltage Regulators. Part III. The Numer-
ical Results, in: Proceedings of the IEEE Transactions on Power Apparatus and
Systems, vol. PAS-104, No. 11, November, 1985, pp. 3291–3297.
[16] M.R. Salem, L.A. Talat, H.M. Soliman, Voltage control by tap-changing
transformers for radial distribution network, in: Proceedings of the IEE,
Generation–Transmission–Distribution, vol. 144, No. 6, November, 1997, pp.
517–520.
[17] S.F. Mekhamer, S.A. Soliman, M.A. Mostafa, M.E. El-Hawary, Load flow solu-
tion of radial distribution feeders: a new contribution, International Journal of
Electrical Power and Energy Systems 24 (November (9)) (2002) 701–707.
[18] E. Cox, The Fuzzy System Handbook: A Practitioner’s Guide to Building, Using,
and Maintaining Fuzzy Systems, Academic Press, USA, 1994.