An Application of Genetic Algorithms to the Network
Reconfiguration in Distribution for Loss Minimization
and Load Balancing Problem (part 2)
b
Dai-Seub ChoAa Chang-Si& Kim and Jua Hasegawaa
a :Dept. of Electrical Eng. Hokkaido University,
North 13 West 8, Uta-ku, Sapporo 060, Japan
b: Dept. of Electrical Eng. Chosun University, 375
Seosuk Dong, Kwangju City, 501-759 South Korea
Abstract- Network reconfiguration in distribution
system is realized by changing the s h t u s of
sectionalizing switches, and is usually done for loss
reducing or for load balancing in the system. This
paper presents a new method which applies a genetic
algorithm for determining which sectionalizing
switch to operate in order to solve the distribution
system loss minimization reconfiguration problem. In
addition, the proposed method introduces a new
limited life feature for performing natural selection of
individuals. Simulations were carried out in order to
veri@ the effectiveness of the proposed method. These
results showed that the proposed method is effective
in dealing with the problems of homogeneity and
genetic drift associated with the population in the
initial state.
Keywords:
Distribu tionau tomation,Distributionsystem
operation, distribution system planning, power flow
analysis, genetic algorithm, limited life, combination
optimization.
1. INTRODUCTION
The distribution system loss minimization
reconfiguration problem is in essence a 0-1 planning
problem which means that for typical system scales
the number of combinations requiring searches
becomes extremely large. In order to deal with this
problem, a new approach which applies a genetic
algorithm (GA). [11 was presented. Briefly, genetic
algorithms are a type of random number search
method, however, they are not merely a one-point
search as they incorporate a multi-point search
feature.
h r t h e r , every point is not separately and
respectively renewed, therefore, if parallel
processingis applied, we can expect a fast solution
IEEE Catalogue No. 95TH8130
0-7803-2981-3/95/$4.00@1995IEEE
algorithm to result. In order to successfully apply a
genetic algorithm to this distribution system loss
minimization reconfiguration problem 121, a new
method that automatically amends the crossover and
mutation rate is proposed in this paper. .This is
referred to as 'dynamic parameter modification'
(DPM) .
However, as one GA behaviaral problem, in the case
where the difference of fitness within a population is
small, population homogenization is incurred, and due
to genetic drift [SI evolution becomes slow. in order to
improve the GA convergence characteristics for the
loss minimization reconfiguration problem, a ' limited
life feature [3] is introduced. More specifically, for
the population homogenization and genetic drift
problems, in addition to fitness dependent elimination,
we introduce a mechanism that carries out
elimination when the limited age is reached. Then, by
combining this technique with ' elitist policy' pi], we
can expect solution convergence to be significantly
faster.
We can carry out the search process efficiently
because faster convergence to an optimal solution is
achieved by the 'limited life' In addition, convergence
characteristics are improved because the DPM method
gradually limits the field of search from the initial
state.
2. PROPOSED METHOD
2.1 Formulation of a loss minimum problem
In an open loop radial distribution system, loss
minimum reconfiguration problem is to decide the
positions of open sectionalizing switches, which can
minimize distribution losses under the electric
current capacity limit and voltage drop limit
376
constraints. This problem can be formulated as 0-1
integer programming problem as follows.
To simplify the presentation, we will represent the
system on a per phase basis and the loads along a
feeder section as constant P,Q loads placed at the end
of the lines. Distribution system loss minimization .
The network reconfiguration problems for loss
reduction and load balancing involve the same type of
operation, namely the load transfer between the
feeders or substations by changing the positions of
switches. They only differ in their objective. Other
factors, such as the voltage profile of the system,
capacities of the lines/transformers, reliability
constraints can be considered as constraints.
To state these problems as optimizationproblems,
note that the radial configuration to a spanning
tree" of a graph representing the network topology.
Thus, we have a so-called minimal spanning tree
problem which can be stated as follows. Given a graph,
find a spanning tree such that the objective function
is minimized while the following constraints are
satisfied.
Constraints
- Voltage drop limit
.
- Line capacity limit
- Transformers
- Power supply constraint
Objective Terms
Having a network model, now we can express the
power loss and measure the balance in the system in
terms of system variables.
For loss reduction, the objective is to minimize the
totali2r losses in the system, which can be calculated
as follows pi.
LP
n-1
xri
pi' + Qi'
= (2) Selection: In accordance with the rules which
i -0 vi'
This will be the objective function, Lp of network
reconfiguration for loss reduction.
For load balancing, we will use the ratio of complex
power at the sending end of a branch, Siover its KVA
capacity, Si as a measure of how much that branch
is loaded. The branch can be a transformer, a tie-line
with a sectionalizing switch or simply a line section.
Lb = E(+)' 2 p i 2 + Qi'
=
.m*
SI
This will be the objective function, Lb of load
balancing. As noted before, the two problems are
377
similar. They both require the same data (system
parameters and load ) and load flow calculation to
evaluate the objectives for a given network topology.
2.2 Genetic Algorithm
The number of the open state of a sectionalizing
switch is denoted as a gene. The candidate solutions
for the optimization problem are expressed by the
strings (chromosome) pertaining to each gene. GA use
the following.
(1) Fitness function links genetic algorithms with the
problems we are asking them to solve. The fitness
of individual i that does not form a closed loop is not
limited to the usual satisfying of constraints, therefore,
it is necessary to consider this as a penalty from the
fitness function. That is to say, li;: is calculated by the
following equation.
1
Fi =
{Li - Lb + a* (gZl + gZ2) + b* gZ3)
where,
Li : sum of the distribution system loss
Lb : loss base
Gli: penalty pertaining to line capacity constraint
violation
ai: penalty pertaining to transformer capacity
constraint violation
G3i: penalty pertaining to voltage drop constraint
violation
a,b : weight constants
depend on the fitness A of each individual, the
multiplication and natural selection is performed.
(3) crossover: One-point crossover occurs when parts
of two parent chromosomes are swapped after a
randomly selected point, creating two children.
(4) Mutation: By mutation is one procedure carried
out by one-point mutate. When bit mutation is applied
to a bit string it sweeps down the list of bits,
replacing each by a randomly selected bit if a
probability test is passed. The selection of individuals
and the exchange point are randomly decided. The
child obtained replaces the parent.
(5) Elitist Policy: If no increasingly fit individual has
been discovered between generations, the elitist
policy simply carries forward the most fit individual
from the previous generation into the next.
When the individuals which proceed to next
generation are selected, the individuals which have
small differences in fitness are difficult to select. So,
a group of individuals which shows the same fitness
in each generation occurs. This phenomenon is called
genetic drift, and this controls the evolution speed,
that is, the convergence speed becomes slow.
2.3 Dynamic Parameter Modification
A special method is necessary to control genetic
operator for broad search during the early stages of
evolution and narrow search in the latter stages. We
propose the Dynamic Parameter Modification (DPM)
which offers the decision of crossover rate and
mutation rate by fitness. That is, crossover rate
Pc(t+l) and mutation rate Pm(t+l) in the (t+l)th
generation are determined by the average of fitness
f(t) in the t-th generation as follows.
a : crossover weight coefficient
p : mutation weight coefficient
2.4 Selection by Limited Life
We introduceThe introduce the concept of age in each
individual and a mechanism which applies natural
selection if an individual has aged. The offspring
which is born by the manipulation of early
circumstances and genes (crossover, mutation,
selection ) is set age 0, and simultaneously the
parent's age is increased by 1.
378
If an individual has attained the predetermined age, it
is selected. By repeating this manipulation, an
individual cannot live too long and can be picked out
at an early time. This results in increased evolution
speed.
3. NUMERICAL EXAMPLES
3.1 Simulation Conditions
In primary distribution systems, sectionalizing
switches are used for both protection, to isolate a fault,
and for configuration management, to reconfiguration
the network. Fig. 1 shows a schematic diagram of a
simplified primary circuit of a distribution system
together with sectionalizing switches.
Distribution systems are normally operated as
radial networks; however , configuration is changed
during operation by changing the state of some
sectionalizing switches.
Branching Point 8 Sectionalizing Switch(c1osed)
Sectionalizing Switch(open) ar> Transformer
aBreaker
Fig.1. Schematic diagram of a primary circuit of
,a distribution system
The system load is assumed to be constant.
Convergence condition is satisfied when the number
of identical individuals having the highest fitness
exceeds 90%. In the case where even after 5000
generations convergence has not been achieved, the
process is terminated.
Simulations were carried out for 10 case randomly
made initial states.

3.2 TEST RESULTS

The proposed solution method has been implemented

in C + + . The test results for loss reduction will be
presented to illustrate the performance of the
proposed method.

The test system is hypothetical system with a 4

feeder substation, 12 branches, 37 sectionalizing
switched and 49 sections .The system data is given in
Table 1.

(b) With Limited Life

Table 1. Model System Specification
Transformer Capacity
Section Load
Feeder Voltage Drop Constraint
Feeder Line Capacity Constraint
Section Resistance
Fig.2. Convergent Characteristics.
Values(pu)
With regard to this, in the case where natural
selection is considered at the point where the limited
0.1 life age of an individual
1.5
is reached,the improvement of convergence against
I 0.01
homogenization and genetic drift, can be obtained.

(1) Efficacy relating to limited life of an Variation in aging layer of a population (case for
individual
Limited Life of 4 years) is shown in Fig. 3.
The combination setting was crossover rate
0.4,mutation rate 0.06 [SIand given the same initial
state. The case where a limited life is not considered
is shown in Fig. 2(a) and the case where a limited life SELECTION BY LIMITED LIFE

is considered is shown in Fig. 2 0 ) . These results

compare the variation in the number of individuals for
the various types of individuals. In this figure, the
most numerous individuals having the highest fitness
become the optimum solution a t the final stage. In the
case where a limited life is not considered, due to
GENERATlON NUMBER
genetic drift the suppression in the increase of
individuals with high fitness can be understood.
Because of this, convergence is slower as 1838 Fig. 3. Variation in aging layer of a population
( case for Limited Life of 4 years).
generation were required.

379
(2) Comparison with DPM the convergence generation number ranged from 45 to

Fig. 4(a) compares different cases of parameter 86, which is to say that a number of infeasible paths

variation. Within DPM, the results for the case where existed in the solution search process is successfully

a limited life is not considered are shown. The vertical restricted meaning that a stable convergence feature

axis Is the convergence generation number, the is obtained.

horizontal axis is for the 10 cases resulting form Further, it can be seen from the optimum solutions
different initial state conditions which are plotted in and from fitness variation, crossover rate variation
order from the smallest case to the largest. Similarly, and situation of variation of mutation rate in Figures
Fig. 4@)gives the results obtained when a limited life 5(a), @) and (e) respectively, that the crossover rate
was considered. These 2 sets of results verify that on and mutation rate follow the generation progress and
average the proposed method with a limited life gradually decrease which contributes to convergence
converges approximately 12% faster. characteristic improvement.

In the case where a limited life was not considered, U.

:* .: .: ,pu.blm i i
.i
.....'.~.""".~.......,.......,...............,.......,.......,.......
depending on the initial state the convergence
generation number ranges from 48 to 101, thus,
comparing this with the fixed parameter easel71 a
large margin of improvement has been realized.
m

With the proposed method consideringa limited life

(a) Fitness

I
.............................................................
, , , I * I I I I I I

"I
, I I I I I I I . I I

.____.
, , , . , , , , , I I
, I I I I I t I I I I

....._I ..... _ _.....

___: .....:_ _ _
____.; _ __:_ _ _ . :
__.._:

cur

(a) Without Limited Life. (b) Crossover Rate

I!
.
I
L.....
.
,

: ....:
L.
.
,
L.....L.
,

!
.
..................................
. ,

!
. .
,

: !
I

:
I

;
.
I
.
m-:
I
.
I

cu* (c) Mutation Rate

(b) With Limited Life. Fig. 5. Changes of fitness crossover rate and
Fig. 4. DPM Dependent Characteristics.
mutation rate.

380
 4. CONCLUSIONS REFERENCES

In this paper, the authors presented a new method
which applies a genetic algorithm for determining
which sectionalizing switch to operate in order to
solve the distribution system loss minimization
reconfiguration problem.
In addition the proposed method introduces a new
limited life feature for performing natural selection of
individuals. Simulations were carried out in order to
verify the effectiveness of the proposed method. From
these results the following points were observed.
(1) For the purpose of solving the problems of
homogeneity and genetic drift associated with the
population in the initial state it can be concluded that
the proposed method is effective. The conventional
method which does not consider the limited life
feature and uses fixed parameters requires long
calculation time depending on initial state conditions .
In contrast, the proposed method is able to perform
solution searches effectively and results in improved
[l] J. H. Holland. "Adaptation in Natural and
Artificial Systems", Univ. Michigan Press, 1975.
[2] K. Nara, et al. "Implementation of Genetic ,
Algorithm for Distribution Systems Loss Minimum
Reeonfiguration", IEEE Trans. Power Systems,
I
Vol. 7', No. 3, pp. 1044,1992.
J. F.. Crow and M. Kimura. An Introduction to
Population Genetics Theory. Harper & Row, 1970.
K. DeJong. "An Analysis of the Behavior of a
Class of Genetic Adaptive Systems", Ph.D. Thesis,
University of Michigan, 1975.
M. N. Swamy & K. Thulasirmaman. Graphs,
Networks, and Algorithms, John Wiley & Sons,
1981.
D.S. Choi, J. S. Lee & J. Hasegawa. "An
Application of Genetic Algorithms to the Network
Reconfiguration in Distribution Systems for Loss
Minimization and Load Balancing Problem",
B E E SICICI '95 2-8 July pp.81-86,1995.

convergence.

(2) The GA based on DPM considering the limited life

feature when compared with a DPM only dependent
GA, demonstrates less dependence on initial state
condition and a more stable convergence can be
obtained.

381