U.P.B. Sci. Bull., Series C, Vol. 72, Iss.

3, 2010 ISSN 1454-234x

OPTIMAL LOSS REDUCTION OF DISTRIBUTION

NETWORKS USING A REFINED GENETIC ALGORITHM

Vlad TUDOR1

Reconfigurarea reţelelor în sistemele de distribuţie a energiei este realizată

prin modificarea stării dispozitivelor de comutaţie şi are ca scop reducerea
pierderilor de energie sau repartizarea cât mai echilibrată a sarcinii în sistem. O
soluţie optimă implică menţinerea în limite acceptabile a tensiunii în nodurile de
reţea, precum şi a valorii curentului care trece prin liniile şi echipamentele
electrice, care să asigure necesarul de sarcină al consumatorilor. Complexitatea
topologică a reţelelor de distribuţie reale implică un număr foarte mare de
configuraţii posibile. Această problemă de optimizare combinatorică poate fi
rezolvată eficient utilizând algoritmul genetic rafinat propus în această lucrare.
Algoritmul lui Prim este utilizat pentru a asigura radialitatea potenţialelor soluţii
propuse spre evaluare, folosind muchii cu cost aleator.

The network reconfiguration in power distribution systems is realized by

changing the status of sectionalizing switches, and is usually done for loss reduction
or for load balancing in the system. An optimal loss reduction is obtained to
maintain acceptable voltage at customer loads as well as to assure sufficient
conductor and substation current capacity to handle load requirements. The
topological complexity of real distribution networks implies searching through many
possible configurations. This combinatorial optimization problem can be resolved
using the efficient refined genetic algorithm proposed in this paper. Prim’s
algorithm is used with random costs for generating different spanning trees for
evaluation.

Keywords: power distribution network, graph theory, loss reduction,

optimization, refined genetic algorithm

1. Introduction

Power distribution systems have a complex topology and usually contain a

great number of electrical devices, which leads to a wide solution space in the
problem of loss reduction using reconfiguration. While the transmission and
subtransmission lines are configured in a meshed network, the distribution feeders
are configured radially in almost all cases. Network reconfiguration is an
important operation problem in on-line configuration management, due to remote
control capability to the switches.

1
Phd student, Faculty of Automatic Control and Computers, University POLITEHNICA of
Bucharest, Romania. E-mail: [email protected]
30 Vlad Tudor

The distribution system should be operated at minimum cost, subject to a

number of constraints:
- radial configuration;
- all loads are served;
- lines, transformers and other equipment within current capacity limits;
- voltage drop within limits.

A global optimal solution to the problem was proposed by Merlin and

Bach [1] using an exhaustive search with branch-and-bound technique. This
method works fine if the initial is close to the global optimum. Being a
combinatorial method, it’s too slow for real distribution systems. Shirmohammadi
and Hong [2] tried to overcome some disadvantages of the algorithm, developing
a new power flow calculation formula.
Civanlar [3] estimated with an innovative algebraic expression the load
loss effected by operating a pair of switches, decreasing the time spent calculating
the power flow in the network at each iteration. Moreover, many heuristic
methods have been developed using branch exchange technique, including two
approximate power flow methods proposed by Baran and Wu [4]. This method is
very fast and good for use in real-time operations. Even though it is easily trapped
in a local minimum, it can be used to provide a loss reduction after an emergency
reconfiguration.
Other methods include neural networks, which were used in several papers
and they serve as a state estimator but don’t analyze the topologies. Simulated
annealing is not a reconfiguration algorithm by itself, but it can be used as a
modification to some other basic algorithms.
Many papers proposed genetic algorithms (GA) in order to obtain a fast
and efficient solution to this combinatorial optimization problem. The
convergence is very fast if the genetic parameters are adapted for the network in
case. Therefore, a new type of algorithms were developed, named refined genetic
algorithms (RGA), which have the ability to adapt the parameters at each
generation, without needing human interaction.
In this paper we propose a refined genetic algorithm which uses Prim’s
algorithm in order to obtain spanning trees (radial configurations), with random
costs for every branch. At every generation, the power flow calculations are made
only if the candidate solution is uniqe, i.e. the time spent evaluating the network is
notably decreased.

2. Mathematical model

A power distribution network can be modeled topologically as an oriented

graph and operationally as an unoriented graph, governed by Kirchhoff's laws.
 Optimal loss reduction of distribution networks using a refined genetic algorithm 31

Let G(N,M) be an unoriented graph with n vertices and m edges, where

N = {n1 , n2 , K, n n } is a finite non-empty set and M , a finite set of unordered sets
(mi , m j ), i ≠ j , i, j ∈ {1,2, K , m}, mi , m j ∈ M .
Also let k (k ≥ 1) special vertices named substation nodes.

Every vertex has associated three data (costs):

- vi ∈ V , V = {v1 , v2 , K, v n } - customer service voltage at node i;
- pi ∈ P, P = {p1 , p 2 , K , p n } - customer load’s real power at node i;
- qi ∈ Q, Q = {q1 , q 2 , K , q n } - customer load’s reactive power at node i.
Every edge has associated three data (costs):
- ii ∈ I , I = {i1 , i2 , K , im } - complex current in edge i;
- ri ∈ R, R = {r1 , r2 , K , rm } - resistance of edge i;
- xi ∈ X , X = {x1 , x 2 , K , xm } - reactance of edge i.

For generalisation purposes, every edge contains a tie switch which can be
opened or closed. Let C = {c1 ,..., c m }, ci ∈ {0,1} be a set that defines the current
configuration of the network (1-closed, 0-opened). The process of reconfiguration
implies taking the distribution network from one state to another.

The optimization problem can be stated as follows:

m
2
minimize f c = ∑ ii ⋅ ci ⋅ ri (1)
i =1

subject to the following restrictions

Vmin ≤ v j ≤ Vmax , ∀j ∈ {1,..., n} (2)

ii ≤ I max , ∀i ∈ {1,..., m} (3)

where Vmin and Vmax are the acceptable limits for customer service voltage at node i
and I max , the maximum current that can flow through edge j.

Discussion. If k = 1, only one vertice will be a substation node and then

the solution will be represented by a minimal spanning tree (radial restriction). If
k > 1, then the final configuration would be a spanning forest consisting of k
spanning trees (one for every substation node).
32 Vlad Tudor

3. The proposed algorithm

The refined genetic algorithm uses Prim’s algoritm with random costs for
every branch in order to obtain minimal spanning trees (radial configurations) as
shown in the flowchart of Fig. 1:

begin

initialize population
with Prim’s algorithm

gen = 1

evaluation

termination no
condition met
reproduction
yes

stop adaptive crossover

adaptive mutation

gen = gen + 1 run Prim’s algorithm

with candidate costs

yes unique no
candidate

Fig. 1. Algorithm’s flowchart

Each generation must ensure that the candidate solution will be unique in
the current population. Only a viable configuration would be eligible for
evaluation, decreasing the number of power flow calculations.
 Optimal loss reduction of distribution networks using a refined genetic algorithm 33

Chromosome Representation. Similar to the tree configuration in graph

theory, each node in distribution networks has only one father node, because of
the open loop operation of distribution networks. The connections among load
nodes and source nodes in radial distribution networks are represented by tree or
forest configurations. Each chromosome will have the form of the incidence
matrix with random values for non-zero elements. This leads to a random starting
point in running Prim’s algorithm.
Initial population. Empirically, I have found that the optimal number of
individuals (nc) that forms the population is
n
nc = , (4)
3
where n is the number of nodes in the network. So, for each chromosome, the
proposed algorithm generates an incidence matrix, with random costs, and then
runs Prim’s algorithm to create the asociated spanning tree.
Reproduction. At each generation, two individuals are randomly selected
for reproduction. If the new chromosome, obtained with crossover and mutation,
is not unique in the current population, the process is repreated [5].
Crossover. A new form of operator was developed due to the specific
form of the chromosme. Let be CR ∈ [1,100] and CR _ pos ∈ [1,100], two random
values at each generation. The new chromosome will be obtained as follows:
- if CR is less than 50, then the first CR _ pos lines will be taken from
the first chromosome, and the remaining ones, from the second;
- if CR is greated than 50, the first CR _ pos lines will be taken from
the second chromosome, and the remaining ones, from the first.
This form of crossover ensures that some of the tree structure remains unchanged
after this step in the algorithm.
Mutation. It is a very high probability that two different individuals would
lead to the same configuration, so mutation is needed to ensure diversity. For each
non-zero value in the chromosome structure, a randomly value named
CM ∈ [1,100] is generated. If this value is less than 10, the cost (cij) will be
replaced with 100 − cij .
Adaptive crossover and mutation. Ass the generations pass and the
individuals converge to the global solution, the intervals for the randomly
generated variables decrease proportionally with the number of generations left to
evaluate.
Evaluation. For each valid configuration, the structure of the network
must be tested to see if the imposed restrictions have been violated. A full load
flow calculation is made and if the solution is within the limits, it tests if the
34 Vlad Tudor

power loss is less than the most unoptimal solution in the current population. If so,
the new candidate solution takes it’s place.
Stopping criteria. The number of generations (gen) is initially set to three
times the number of nodes. Becouse of the condition that every new valid
chromosome must be unique, the number of possible configurations could be less
than the initial evaluation. In order to stop the process, a new variable have been
set for counting the ununique generated solutions. If this number exceeds gen,
then the program halts.

5. Test cases

Both systems were tested using a Pentium 4 2.4GHz processor and the
algorithm was implemented in standard C++ using MinGW Developer Studio.
The first system was proposed by Glamocanin [6] and it is used to verify
the algorithm’s computer implementation. Table 1 contains the branch impedance
and bus load data for the circuit shown in Fig. 2.

Table 1

Branch impedance and bus load data for test case 1

Bus Impedance [Ω] Load at To Bus [kVA]
From To R X P Q
1 2 0.7820 0.2120 600.0 400.0
1 3 0.7820 0.2120 500.0 300.0
1 4 1.5640 0.4240 100.0 90.0
2 5 0.7820 0.2120 600.0 400.0
2 6 1.1730 0.3180 1300.0 1100.0
3 7 1.3685 0.3710 1300.0 1000.0
4 8 1.1730 0.3180 100.0 90.0
3 9 0.7820 0.2120 800.0 600.0
2 10 1.1730 0.3180 300.0 100.0
3 5 1.1730 0.3180
6 9 0.7820 0.2120
8 9 1.1730 0.3180
4 10 1.1730 0.3180
 Optimal loss reduction of distribution networks using a refined genetic algorithm 35

Fig. 2. First test case’s circuit

Each line segment is assigned a conductor with a rating of 194A. The

nominal substation voltage is 10kV line-to-line and the distribution transformer
ratio is 48.11:1 in each line segment. The voltage drop constraint is 5%,
corresponding to a customer service voltage of 114V.
The global optimum (247.436kW – active loss) was obtained after 3
generations in 2ms. The tie switches left open after configuration are: 2-10, 3-5,
6-9 and 8-9.
The second system was proposed by Baran and Wu [7] and is more severe
than the first, because of the low voltage conditions, the large reactive load at bus
30, and the number of tie switches within a single feeder. Table 2 contains the
branch impedance and bus load data for the circuit.
The substation voltage is 12.66kV line-to-line, which requires a
transformer ratio of 60.91:1 in each segment.
The large reactive load at bus 30 can make a traditional algorithm to find a
sub-optimal solution. The real and reactive loads at bus 30 must be served
together. Considering only the real power, the losses increase by a
disproportionate amount due to the current drawn by the reactive load.
The global optimum (127.847kW – active loss) was obtained after 34
generations in 171ms (see Fig. 3).
36 Vlad Tudor

Table 2
Branch impedance and bus load data for test case 2
Bus Impedance [Ω] Load at To Bus [kVA]
From To R X P Q
1 2 0.0922 0.0470 100.0 60.0
2 3 0.4930 0.2511 90.0 40.0
3 4 0.3660 0.1864 120.0 80.0
4 5 0.3811 0.1941 60.0 30.0
5 6 0.8190 0.7070 60.0 20.0
6 7 0.1872 0.6188 200.0 100.0
7 8 0.7114 0.2351 200.0 100.0
8 9 1.0300 0.7400 60.0 20.0
9 10 1.0440 0.7400 60.0 20.0
10 11 0.1966 0.0650 45.0 30.0
11 12 0.3744 0.1238 60.0 35.0
12 13 1.4680 1.1550 60.0 35.0
13 14 0.5416 0.7129 120.0 80.0
14 15 0.5910 0.5260 60.0 10.0
15 16 0.7463 0.5450 60.0 20.0
16 17 1.2890 1.7210 60.0 20.0
17 18 0.7320 0.5740 90.0 40.0
2 19 0.1640 0.1565 90.0 40.0
19 20 1.5042 1.3554 90.0 40.0
20 21 0.4095 0.4784 90.0 40.0
21 22 0.7089 0.9373 90.0 40.0
3 23 0.4512 0.3083 90.0 50.0
23 24 0.8980 0.7091 420.0 200.0
24 25 0.8960 0.7011 420.0 200.0
6 26 0.2030 0.1034 60.0 25.0
26 27 0.2842 0.1447 60.0 25.0
27 28 1.0590 0.9337 60.0 20.0
28 29 0.8042 0.7006 120.0 70.0
29 30 0.5075 0.2585 200.0 600.0
30 31 0.9744 0.9630 150.0 70.0
31 32 0.3105 0.3619 210.0 100.0
32 33 0.3410 0.5302 60.0 40.0
8 21 2.0000 2.0000
9 15 2.0000 2.0000
12 22 2.0000 2.0000
18 33 0.5000 0.5000
25 29 0.5000 0.5000
 Optimal loss reduction of distribution networks using a refined genetic algorithm 37

Fig. 3. Second test case circuit – final configuration

The global optimal configuration has lower system losses, but still does
not meet the normal operating voltage criteria of at least 114V (see Table 3). This
can be done using voltage regulators and, in practice, some shunt capacitor banks
installed. The algorithm was using an emergency limit of 112.5V in order to
obtain viable solutions in rapid time for evaluation.
Table 3
Node-voltage profile for optimal configuration
1 2 3 4 5 6 7 8 9 10 11
120.0 119.7 118.5 118.3 118.1 117.8 117.7 115.7 115.3 115.7 115.7
12 13 14 15 16 17 18 19 20 21 22
115.8 115.5 115.4 114.6 114.4 114.1 113.9 119.4 117.5 117.0 116.6
23 24 25 26 27 28 29 30 31 32 33
117.7 116.0 114.7 117.7 117.7 117.6 114.1 113.7 113.3 113.3 113.9
38 Vlad Tudor

The final population contains a set of sub-optimal solutions which can be

used if the global optimal configuration cannot be utilized:
Table 4
Final set of configurations in losses [kW]
1 2 3 4 5 6 7 8 9 10
130.5 127.8 128.5 130.1 131.5 130.7 131.6 130.8 129.8 130.4

Using this technique leads to a list of optimal configurations, becouse all

the individuals that form the final generation are close to the global optimal.

7. Conclusions
In this paper a fast and reliable refined genetic algorithm has been
proposed by the author for configurating a power distribution network. For
reconfiguration, it can be very easily adapted by putting the initial configuration
in one of the individuals that forms the initial population of chromosomes.
For future work, the algorithm could be adapted for multiple substations
and tested with more complex topologies, using multiprocessor machines.

REFERENCES
[1] A. Merlin, H. Back, "Search for a mininimal loss operating spanning tree configuration for an
urban power distribution system". Proc. of the Power Systems Comp. Conf., Cambridge 1975
[2] D. Shirmohammadi, H.W. Hong, Reconfiguration of electric distribution networks for resistive
line losses reduction, IEEE Transactions on Power Delivery, vol. 4, no. 4, April 1989, pp.
1492-1498
[3] S. Civanlar, J.J. Grainger, Y. Yin, S.S. Lee, “Distribution Feeder Reconfiguration for Loss
Reduction”, IEEE Transactions on Power Delivery, vol. 3, no. 3, July 1988, pp. 1217-1223
[4] M.E. Baran, F.F. Wu, Network reconfiguration in distribution systems for loss reduction and
load balancing, IEEE Transactions on Power Delivery, vol. 4, no. 2, April 1989, pp. 1401-
1407
[5] C. Soare, S.St. Iliescu, V. Tudor, I. Fǎgǎrǎşan, O. Dragomir, F. Dragomir, “Computer Aided
Modelling in Matlab and Simulink - Advanced Control System for Processes” (Proiectarea
asistatǎ de calculator în Matlab şi Simulink – conducerea avansatǎ a proceselor), 2006,
AGIR Publishing House; pp. 199-226
[6] V. Glamocanin, Optimal loss reduction of distribution networks, IEEE Transactions on Power
Systems, vol. 5, no. 3, August 1990, pp. 774-782
[7] M.E. Baran, F.F. Wu, "Network Reconfiguration in Distribution Systems for Loss Reduction
and Load Balancing", IEEE Transactions on Power Delivery, vol. 4, no. 2, April 1989, pp.
1401-1407.