1426
3, August 1995
A NEW FUZZY REASONING APPROACH
FOR LOAD BALANCING IN DISTRIBUTION SYSTEM
B. Naga Raj K.S.Prakasa Rao, Senior Member, IEEE.
Department of Electrical Engineering
Indian Institute of Technology, Delhi
Hauz Khas, NEW DELHI - 110 016. INDIA
Abstract - Transformer and feedef load balancing in a
distribution system reduces the risk of overloads due to load
changes. The possible out of service area fobwing the
occurrenceof a fault is also reduced. However, keeping in view
the l i e expectancy of the switches, it is desirable that the
number of switching operations is kept to a minimum. A new
algorithm for load balancing based on fuzzy set decision theory
is presented in this paper. The decision regarding switching
operations is amved at by considering transformer and feeder
load balancing together. The proposed method is illustrated
through an example.
1. INTRODUCTION
Primary distribution systems are usually operated in a radial
configuration, with each load-point being supplied by one end only
one transformer. It is always desirable to operate the system with
the loads on the transformers and feeders kept balanced in order to
reduce overloading and the possible out-of-service area. Load
balancing also helps in optimal utilization of transformers and
feeders so that investments for capacity enhancement could be
deferred.
In normal operation, load balancing of a distribution system is
achieved by reconfiguring the feeders thereby redistributing the
loads among feeders and transformers. Some loads can be
transferred from heavily loaded feedersltransformers to relatively
lightly loaded neighbouring feedersltransformers. In this way, by
changing the status of openlclosed switches on distribution
feeders,the loads on the system get evenly distributed among the
various feedersltransformers.
Essentially, load balancing is a combinatorial optimization
problem involving a decision making regarding the position of all
the sectionalizing switches in a distribution system. Usually there
are a large number of sectionalizing switches in a typical distribution
95 WM 126-3 PWRS A paper recommended and approved
by t h e IEEE Power Syatem Engineering Committee o f t h e
IEEE Power Engineering S o c i e t y f o r p r e s e n t a t i o n a t
t h e 1995 IEEE/PES Winter Meeting, January 29, t o
February 2, 1995, New York, NY. Manuscript submitted
J u l y 28, 1994; made a v a i l a b l e f o r p r i n t i n g
January 18, 1995.
0885-8950195/$04.00 0 1995 IEEE
IEEE TI-tions on Power Systems, Vol. 10. No.
system and hence the possible switching options are also extremely
large. Hence it is very difficult to obtain an optimal solution to this
problem in a short time.
A number of heuristic approaches have been proposed in the
past [1-71 to obtain a near optimal solution to the above problem in
a short time. A simple search technique for service restoration and
load balancing was proposed by Castro et al. (11 considering the
data base and implementation requirements given by the operators
for on-line distribution automation application. In the method
proposed by Aoki et al. [2], load transfer is camed out initially for
a pair of transformers which have the highest and the least load
indices (load to capacity ratio). By appropriate switching
operations, the load indices of other transformers are equalised as
closely as possible. Load balancing for feeders is also performed in
a similar way, through open loop switches. Baran and Wu [3]
proposed a method, by which a gradual reduction of system load
index is achieved through a search process. A heuristic method for
load balancing was proposed by Hsu et a1.[4]. The method is
applicable to bwh constant as well aswhanging load conditions.
Chen and Cho [S] evaluated the optimal switching operations based
on the hourly load patterns. The critical switches are identified by
investigating the optimal switching patterns. Expert systems have
also been used for load balancing [6-71. Most of the above
approaches consider transformer and feeder load balancing
independent of each other.
Loads on any transformer are non-homogenous in nature. That
is, a lightly loaded transformer might have some lightly loaded
feedersllaterals and some heavily loaded feedersllaterals connected
to it. Similarly, a heavily loaded transformer might have some
heavily loaded feedersllaterals and some lightly loaded
feedersllaterals connected to it.
In any load transfer from a heavily loaded transformer (source
transformer) to a lightly ,loaded transformer (sink transformer),
preference should be given to load transfer from a heavily loaded
feederllateral on the source transformer. to lightly loaded
feederllateral on the sink transformer.
System operators do not usually consider the loading level of
transformers, main feeders and lateral feeders as equally significant.
Overloading of a transformer necessitates a switching operation
more urgently than the overloading of a main feeder. Likewise,
alleviation of overload on a main feeder is more critical than that on
a lateral feeder. Load balancing is thus a multiple objective decision
making problem and a compromise is required between the number
of switching operations and the degree of balancing achieved in the
process.

An operator's preference in finding a compromise solution is
required in such an environment where different options satisfy the
various requirements to varying degrees. Classical approaches do
not have a mechanism to incorporate the vague or "fuzzy"
preference of the operator in obtaining an optimal solution in the
presence of such multiple objectives.
In the proposed method, the degree of satisfaction of various
components (such as transformers, feeders and laterals) is
considered along with the desirability of minimum switching options
in a fuzzy set theoretic framework during the load balancing. The
effectiveness of the proposed method for load balancing is
demonstrated on a typical distribution system which consists of four
transformers, six main feeders, seventy eight sections and a hundred
and six sectionahzing switches.
11. FORMULATION OF THE PROBLEM
The following assumptions [4] have been made in this work:
1. All feeder section loads are known.
2. The loads are 3-phase and are balanced.
3. Loads are modelled as constant current sinks.
NOMENCLATURE:
1. - Difference operator.
2'. p Membership function.
3. CAPTG), CAPF(i)
and CAPLF (k) : Rated capacities (in A) of transformer
j , feeder i and lateral feeder k
respectively.
4. LOADTG), LOADF(i)
and LOADLF(k): Loads incident (in A) on transformer
j, feeder i and lateral feeder k.
5. SLI System Load Index which is obtained
as the ratio of the total system load
and the sum of capacities of the
system transformers.
6. TLlG), FLI(i) and
LFLI(k) Load indices of transformer j , feeder
i and lateral feeder k, given as load on
the component divided by the rating of
the component.
7. ILTTQ) Ideal load transfer (in A) that would
make the transformer load index equal
to the system load index, and i s given
by
LOADTG) - CAPTG) * SLI
8. ILTF(i) Ideal load transfer (in A) that would
make the feeder load index equal to
the system load index, and is given by
LOADF(i) - CAPF(i) * SLI
9. 1 Amount of load transferred by a
switching operation.
10. To Source transformer.
I
1427
11. T, Sink transformer.
12. ns Number of switching operations.
Ideal Balanced Distribution Svstem: An ideal balanced distribution
system is that system in which every transformer and feeder is
loaded to the same extent so that the load indices of all components
are identically equal to the system load index.
It widely acknowledged that due to the discrete nature of
system loads, an ideally balanced system state can seldom be
attained in practice.
Load Balancing: The objective of reconfiguration for load balancing
is to identify a proper set of switches that should b e closed or
opened such that appropriate load transfer among transformers and
feeders results in a practical balanced system which is as-close to the
ideal balanced system as possible.
111. SYSTEM MODELLING IN FUZZY FRAMEWORK
Applicability of fuzzy set theory to different powel: system
problems is being investigated widely over the past several years.
A representative list of published work can be found in [8-131.
Optimization in Fuzzy Environment 1141: In fuzzy domain, each
objective is associated with a membership function. The
membership function indicates the degree of satisfaction of the
objective. In the crisp domain, either the objective is satisfied or it
is violated, implying membership values of unity and zero
respectively. On the contrary, Fuzzy sets entertain varying degrees
of membership function values from zero to unity. Thus fuzzy set
theory is an extension of standard set theory[8].
When there are multiple objectives to be satisfied
simultaneously, a compromise has to be made to get the best
solution. One solution methodology for multi-objective optimization
in fuzzy framework is based on Max-min principle which is
described as follow:
- For each option considered, the degrees of satisfaction of all
the different objectives are evaluated.
- The degree of overall satisfaction for this option is the
minimum of all the above degrees of satisfaction.
- The optimal solution for the system is the maximum of all such
degrees of satisfaction.
This has been proved by Zimmermann [14].
In the proposed method for load balancing, the terms DOSF),
D O S O , DOS(LF) and DOS@) indicate the degree of satisfaction
for load balancing of Transformer, Feeder, Lateral Feeder and the
number of switching operations respectively.
A brief explanation of the above terms will be in order:
i) Degree of Satisfaction for Transformers, DOSQ:
Consider Fig.1 which is a part of a large distribution system with
SLI = 0.5.
1428

n
l-
TO sw

5 Rg.1 Example to Illustrate

TS
; If--, f3 5
f2
DOS (1) Fig.3 Example to Illustrate DOS(F)
1
1

L o a d Transfer91 ( A )
Fig.2 Membership Function
L
IJ DOS(T) for Transformers Load Transfer, I ( A )
Fig.4 Membership Function
Assume
~ D O S ( FFor
) Feeders
CAPT(To) = 1 W A LOADT(T,J = 700A
CAPT(TJ = lo00 A LOADT(TJ = 400 A Assume

Then SLI = 0.5

CAPT(T0) = 1000 A LOADT(To) = 675 A
ILT(T,) = 700 - (lo00 * 0.5) = 200 A CAPT(TJ = lo00 A LOADT(T,) = 400 A
ILT(TJ = 400 - (lo00 * 0.5) = 100 A
CAPF(fl) = 500 A LOADF(fl) = 400 A
Define a ‘Target Load Transfer for Transformers’, TLTT as CAPF(f2) = 500 A LOADF(f2) = 275 A
CAPF(f3) = lo00 A LOADF(f3) = 400 A
TLTT = Min ( ILT(TiJ, ILT(T,) )
From the above system, a balanced feeder loading would be
For this case, obtained if the ideal loading of transformer To is equitably
distributed among feeders fl and f2, even while performing a load
TLTT = Min (200,100) = 100 A transfer from Toto T..

This implies that one should attempt to transfer the minimum Define the Target Load Transfer for a Feeder ‘i’ as
load that would take one of the two transformers to a TLI = SLI.
Here, the target for load transfer is 100 A. However, a load Rating of the feeder (Ideal
transfer of exactly 100 A may not be available for switching. In TLTF(i)=LOADF(i) - *
Transformer
order to quantify the degree of satisfaction for various switching (Sum of the ratings of Load.)
options, a fuzzy membership function for DOS(T) is formulated as the feeders connected
shown in Fig.2. to the same transformer)

A load transfer of 1 = TLTT is given a membership value of
unity. The larger the deviation from the target, the lesser is the
degree of satisfaction. For all load transfers greater than twice the
TLTT, the membership function is zero, which is completely
undesirable. This can be represented as
PDOql.) = 1/TLTT for I<TLTT
= 2-UTLTT for TLTT<1<2 * TLTT
For example, a load transfer of 80 A would have a
membership value of 0.8 while a transfer of 110 A would have a
value of 0.9. Thus, a transfer of 110 A is more desirable than a
transfer of 80 A when the TLTT is 100 A.
ii) Degree of Satisfaction for Feeder, DOS@) : Consider Fig.3
which is a part of a large distribution.
To quantify the degree of satisfaction for switching operations
on feeders, a membership function pDWF)is defined as shown in
Fig.4
Here also, as in the case of transformers, a load transfer
of 1 = TLTF(i) is given a degree of satisfaction of unity. The
greater the divergence from this target, the lesser is the degree of
satisfaction. For 1 > 2 * TLTF(i), the value is zero. This can be
expressed as
PrnB = UTLTF(i) for 1 < TLTF(i)
= 2-UTLTF(i) for TLTF(i) < 1 < TLTF(i)
Considertwo points in the search process for the system shown
in Fig.3 the details of which are given in Table 1.
 1429

Table I shows that a load transfer of 125 A on f l is more number of switching operations is given a lower memebership value
preferable than a load transfer of 100 A on feeder a, though 100 A vol.
is the target load transfer between the two transformers. This is due
to the fact that the first case offers more compensation with regard TABLE I. Details of the switching options ( for system in Fig.3 )
to feeder load balancing. The final loads on the transformers and to illustrate DOS(F).
feeders, for the above options, are indicated in Table U, for the sake
of comparison.

From Table 11, it can be seen that the first option allows
greater feeder load balancing compared to the second option and
hence is preferred.

iii) Degree of Satisfaction for Later Feeders, DOS(LF):

The balancing of lateral feeders is also considered for the TABLE 11. Final load level of transformers and feeders for the
following reason: In typical distribution system, each transformer is svstem in Fig.3
supported by another source, not only via an open sectionalizing
switch on the main feeders, but also by a similar arrangement on the Feeder h d ( A )
lateral feeders. Usually, the lateral feeders have lesser capacity I I I I
n R r3
compared to the main feeders.
1. 550 525 275 275 525
2. 575 MO 400 I75 MO
For load balancing, the case of lateral feeders is somewhat
different from the transformers and main feeders. Since lateral
feeders are spread out along each main feeder and their capacities n
are diverse, it is not appropriate to prescribe a definite 'target for LL
load transfer' for lateral feeders.

Hence, a membership function for degree of satisfaction for 0

lateral feeders is defined as in Fig.5. 0
2,
The degree of satisfaction for lateral feeders is equal to unity .-a '
as long as the lateral load remains less than 50% of its rated r
capacity ( LFLI(LF) < 0.5 ). From 50% rated capacity to 150 5%
rated capacity, the membership function value keeps decreasing.
!!!
a,
More than 150% loading ( LFLI(LF) > 1.5 ) is absolutely n
undesirable. This is due to the fact that in most utilities, a higher E
loading is allowed on lateral feeders as compared to the a,
transformers and main feeders. This can be described as z Lateral Feeder Load Index
ILDOs(LF3 = 1 for UCAP(LF) < 0.5 LFLI (If)
= 1.EUCAP(LF) for 0.5< UCAP(LF) < 1.5
Fig. 5 Membership Function
~ D O S ( L F .For
) Lateral
iv) Degree of Satisfaction of Number of Switching Operation,
PDOS,,:
n
v1
Feeders
As stated above, one of the objectives of load balancing is to
have a minimal number of switching operations. Hence a
membership function for DOS(ns) is also defined as in Fig.6 which
decreases with an increase in the number of switching operations.
Each switching option involves the operation of two switches: O1
0
closing an open sectionalizing switch and opening another switch in
the loop that is formed. This can be written as .-
Pm-) = 1 for ns 5 2
= 1.25 - (ns/8) for 2<ns<10 a,
n
Switching options in which only a single pair of switches are c
operated are known as single-switching options. Cases where two a, 2 10
or more pairs of switches are operated are known as double and
multiple switching options respectively.
No of Switchings ns
Fig.6 Me mbership Function
It is noted that, in fuzzy set rotation, a high membership valve
indicates a desirable situation. For example, the degree of DOS (ns) For No. of
satisfaction for a single switching operation which is a highly
desirable situation, is assigned a value equal to unity. A larger Swit chings
1430
1V. PROPOSED FUZZY-REASONING APPROACH FOR
LOAD BALANCING
The proposed fuzzy reasoning algorithm for load balancing is
described as follows:
STEP 1 : Compute the system load index (SLI) and the
transformer load indices (TLI) of all the transformers in
the system.
STEP 2 : Select the two transformers which have the highest and
the least load indices. Evaluate the Target Load Transfer
TLTT and TLTF (wherever applicable).
STEP 3 : Execute load balancing as explaining below:
For all single switching options, calculate D O S O ,
DOS(F) and DOS(LF) wherever applicable.
Extract the best possible switching option through Max-
Min principle of fuzzy numbers.
If the best possible option obtained as above has a
degree of satisfaction less then p,, for ns = 2 then
search for double - switching options and so on. Else,
go to step e).
Under multiple switching options for load balancing
between a pair of transformers, evaluate the best
compromise amongst the best options obtained for the
various switching combinations.
Go to STEP 1, till all transformers are considered and
no further balancing is possible.
STEP 4 : After balancing of all transformers is camed out,
balancing of main feeders via. loop switches is
performed, in a similar fashion to that of the
transformer balancing, as follow:
FIG07 EXAMPLE SYSTEM
a) Select a transformer.
b) Evaluate the loads and the target load transfer for all
feeders provided with open loop switches.
c) Execute load balancing as in STEP 3.
d) Repeat for all transformers.
STEP 5 : If the position of any loop switch is altered in STEP 4,
then proceed to STEP 1. Else STOP.
It may be noted that the transformer load balancing is not
exhaustive since the loop switches remain fmed during the
calculation. Hence, if any loop switch is altered during
STEP 4, one has to search for the possibility of a greater load
balancing.
In the search process of transferring the load between a pair of
transformers, the load is gradually shifted from the source
transformer to the sink transformer by sequentially opening and
closing the appropriate switches until1 the desired load balancing has
been achieved. If at any juncture, the load transferred '1' exceeds
the target load transfer TLT, then further options along this path
need not be considered. This is justified due to the fact that when
one moves further along the path the amount of load transferred
would be increasing, there by decreasing the Degree of Satisfaction
of load balancing (refer Fig.:! and Fig.4).
V. APPLICATION OF FUZZY REASONING APPROACH
The effectiveness of the proposed fussy reasoning approach is
illustrated through an example system shown in Fig.7 [15]. For
this system, each transformer has a rating of 1000 A. Feeders f3
and f4 also have a rating of 1000 A while f l , f2, M and f6 have a
rating of 500 A each. The system load index SLI is 0.408. Load
balancing has been performed for this system using the proposed
method. The following switching operations are suggested to
achieve load balancing:
I
 143 1

Switches to be closed : 19,29,37 and 69 Table 111. Comparision of loads and load indices of
Switches to be opened : 14,28,35 and 67 transformers and feeders
I I I
S.NO. TrPNrama F d a Nta
A comparison of the loads on the transformers and feeders BaQ
before and after load balancing is provided in Table 111.
Losd
VI DISCUSSION - (4
1. Trl 413
The solution for the example system was obtained in 50 secs. fl 0.646 220
of CPU time on a PC 286. Aoki et a1 [2] have used a similar a 0.516 I93

technique on the Hiroshima City System for load balancing. In 2. Tn 0.323

crisp domain i.e., without incorporating the fuzzy operator's B 388
preference. They reported a reasonable CPU time (3.1 - 3.6 secs. 3. Tr3 0.298 421
on a HITAC M - 200H Computer) for obtaining solution in real i4 0.298 421 0.421
time. Since the fuzzy-logic based technique does not impose any 4. TA 410 0.410
extra significant computational burden, this technique can also be f5 0.536 0.504
used in real time Distribution Management Systems OMS). f6 0.316

VII. CONCLUSIONS

An approach based on fuzzy reasoning has been developed for

determining a proper set of switching operations to balance the loads
on a distribution system. A characteristic of the fuzzy reasoning
approach is that it considers the desirability of a switching option
vis-a-vis the components for which load balancing is contemplated.
In most of the existing works, the loads of various components are
viewed as being equally critical. Whereas, an experienced operator
would give a higher preference to balancing of transformer loads,
then to feeder loads and lastly to lateral loads. Such an intelligence
is incorporated in the proposed fuzzy reasoning approach and is
expected to give more realistic options in practice.
REFERENCES
C.H .Castro, J .B.Bunch and T.M .Topka, "Generalized
Algorithms for Distribution Feeder Deployment and
Sectionalization", IEEE Trans. on Power Apparatus and
Systems. Vol. PAS-99, MarlApril 1980, pp 549-557.
K.Aoki, H.Kuwabara, T.Satoh and M.Kanezashi, "An
Efficient Algorithm for Load Balancing of Transformers and
Feeders by Switch Operation in Large Scale Distribution
Systems", IEEE Trans. on Power Delivery, Vol. PWRD-3,
Oct. 1988, pp 1865-1872.
M.E.Baran and F.F.Wu, "Network Reconfiguration in
Distribution Systems for Loss Reduction and Load Balancing",
IEEE Trans. on Power Delivery, Vol. PWRD-3, April 1989,
pp 1401-1407.
[7] G.Chang, J.Zrida and J.D.Birdwell, "KnowledgeBased
Distribution System Analysis and Reconfiguration", IEEE
Tans. on Power Delivery, Vol. PWRS-5, Aug. 1990, pp 744-
749.
[SI K.Tomsovic, "A Fuzzy Linear Programming Approach to the
Reactive PowerNoltage Control Problem", IEEE Trans. on
Power System, Vol. PWRS-5, Feb. 1992, pp 287-293.
[9] H.C.Kuo and Y.Y.Hsu, "Distribution System Load Estimation
and Service Restoration Using a Fuzzy Set Approach", IEEE
Trans. on Power Delivery, Vol. PWRD-8, Oct. 1993, pp
1950-1957.
[lo] Y.Y.Hsu and H.C.Kuo, "A Heuristic Based Fuzzy Reasoning
Approach for Distribution System Service Restoration", IEEE
Trans. on Power Delivery, Vol. PWRD-9, April 1994, pp
948-953.
[ l l ] C.C.Su and Y.Y.Hsu, "Fuzzy Dynamic Programming: An
Application to Unit Commitment", IEEE Trans. on Power
Systems, Vol. PWRS-6, Aug. 1991, pp 1231-1237.
[12] Y.Y.Hsu and H.C.Kuo, "Fuzzy-set Based Contingency
Ranking", IEEE Trans. on Power System, Vol. PWRS-7,
Aug. 1992, pp 1189-1196.
[131 D. Srinivasan, C.S. Chang and A. C .Liew, "M ultiobjective

Generation Scheduling Using Fuzzy Optimal Search

Y.Y.Hsu, Y.Jwo-Hwu, S.S.Liu, Y.W.Chen, H.C.Feng and
Y.M.Lee, "Transformer and Feeder Load Balancing Using a
Heuristic Search Approach". IEEE Trans. on Power Systems,
Vol. PWRS-8, Feb. 1993, pp 184-190.
C.S.Chen and M.Y.Cho, "Determination of Critical Switches
in Distribution System", IEEE Trans. on Power Systems, Vol.
PWRD-7, July 1992, pp 1443-1447.
C.S.Chang and T.S.Chung, "An Expert System for On-line
Security-Economic Load Allocation On Distribution Systems",
IEEE Tans. on Power Delivery, Vol. PWRD-5, Jan. 1990, pp
467-473.
Technique", IEE Proceedings, Vol. 141, Pt. C. No.3, May
1994, pp 233-242.
[14] H.J.Zimmermann, "Fuzzy Programming and Linear
Programming with Several Objective Functions",
TIMSlStudies in the Management Sciences, Vol. 20, 1984,
North-Holland, pp 109-121.
[15] K.Aoki, K.Nara and T.Satoh, "New Reconfiguration
Algorithm for Distribution System-Priority Contrained
Emergency Service Restoration", Proc. IFAC Conference on
Power Systems and Power Plant Control, Seoul, 1989, pp 443-
448.
1432

B.Naea Rai was born in Tenali, A.P, India, on

.,a Feb 9, 1971. He received his B.E.(Electrical &
I
Electronics) from Andhra University,
I “I.\ Visakhapatnam, A.P, in 1992. From 1992 to
I
1993 he was with the Coromandel Fertilizers
Ltd., Visakhapatnam, A.P. He is currently
working for his M.Tech (Power Systems) degree
in the Dept. of Electrical Engg., IIT Delhi,
India. His research areas include applications of
AI techniques to Power System.

K.S.Prakasa Rao (S’70 - M’74 - SM’81) was born in Prakkilanka,

A.P, India, on July 15, 1942. He received his B.E.(Electrical) and
M.E.(Power Systems) from Osmania University, Hyderabad, in
1964 and 1966 respectively. He obtained his Ph.D degree from the
Indian Institute of Technology, Kanpur in 1974. He is presently a
professor in the Electrical Engineering Department at IIT Delhi. His
fields of interest are Power Systems Planning, Operation and
Reliability.