World Academy of Science, Engineering and Technology 37 2008

Using Genetic Algorithm for Distributed

Generation Allocation to Reduce Losses and
Improve Voltage Profile
M. Sedighizadeh, and A. Rezazadeh

size of the DG units in large distribution systems is a complex

Abstract—This paper presents a method for the optimal combinatorial optimization problem.
allocation of Distributed generation in distribution systems. In this The optimal placement and sizing of generation units on the
paper, our aim would be optimal distributed generation allocation for distribution network has been continuously studied in order to
voltage profile improvement and loss reduction in distribution achieve different aims. The objective can be the minimization
network. Genetic Algorithm (GA) was used as the solving tool, of the active losses of the feeder [1], [2]; or the minimization
which referring two determined aim; the problem is defined and of the total network supply costs, which includes generators
objective function is introduced. Considering to fitness values operation and losses compensation [3], [4], [5], [6]; or even
sensitivity in genetic algorithm process, there is needed to apply load the best utilization of the available generation capacity [7]. As
flow for decision-making. Load flow algorithm is combined
a contribution to the methodology for DG economical
appropriately with GA, till access to acceptable results of this
operation. We used MATPOWER package for load flow algorithm analysis, in this paper it is presented an algorithm for the
and composed it with our Genetic Algorithm. The suggested method allocation of generators in distribution networks, in order to
is programmed under MATLAB software and applied ETAP voltage profile improvement and loss reduction in distribution
software for evaluating of results correctness. It was implemented on network. The Genetic Algorithm is used as the optimization
part of Tehran electricity distributing grid. The resulting operation of technique. In Section 2 it is presented a brief discussion about
this method on some testing system is illuminated improvement of distributed generation issues and Section 3 is an introduction
voltage profile and loss reduction indexes. to the Genetic Algorithm. The problem formulation is
presented in Section 4 and the proposed solution method is
Keywords—Distributed Generation, Allocation, Voltage Profile, discussed in Section 5. It is presented an application example
losses, Genetic Algorithm. in Section 6 and finally, the conclusions.

I. INTRODUCTION II. DISTRIBUTED GENERATIONS

B EFORE installing distributed generation, its effects on

voltage profile, line losses, short circuit current, amounts
of injected harmonic and reliability must be evaluated
A general definition was then suggested in [8] which are
now widely accepted as follows: ‘‘Distributed Generation is
an electric power source connected directly to the distribution
separately. The planning of the electric system with the network or on the customer site of the meter’’. The definitions
presence of DG requires the definition of several factors, such of DG do not define the technologies, as the technologies that
as: the best technology to be used, the number and the can be used vary widely. However, a categorization of
capacity of the units, the best location, the type of network different technology groups of DG seems possible, such as,
non-renewable DG and renewable DG. From distribution
connection, etc. The impact of DG in system operating
system planning point of view, DG is a feasible alternative for
characteristics, such as electric losses, voltage profile, stability
new capacity especially in the competitive electricity market
and reliability needs to be appropriately evaluated. The
environment and has immense benefit such as [6]: Short lead-
problem of DG allocation and sizing is of great importance. time and low investment risk since it is built in modules,
The installation of DG units at non-optimal places can result Small-capacity modules that can track load variation more
in an increase in system losses, implying in an increase in closely, Small physical size that can be installed at load
costs and, therefore, having an effect opposite to the desired. centers and does not need government approval or search for
For that reason, the use of an optimization method capable of utility territory and land availability, Existence of a vast range
indicating the best solution for a given distribution network of DG technologies.
can be very useful for the system planning engineer. The For these reasons, the first signs of a possible technological
selection of the best places for installation and the preferable change are beginning to arise on the international scene,
which could involve in the future the presence of a
Manuscript received January 5, 2008 consistently generation produced with small and medium size
The Authors are with Faculty of Electrical and Computer Engineering, plants directly connected to the distribution network (LV and
Shahid Beheshti University, Tehran, 1983963113, Iran (phone: +98-21- MV) and characterized by good efficiencies and low
29902290, fax: +98-21-22431804, e-mail: [email protected]).

251
 World Academy of Science, Engineering and Technology 37 2008

emissions. This will create new problems and probably the 2. Voltage constraint
need of new tools and managing these systems. V bus min ≤ V bus ≤ V bus max
The objective function, with composing constraints and
III. GENETIC ALGORITHM goals, is determined as following:
Genetic Algorithm is a general-purpose search techniques
based on principles inspired from the genetic and evolution ⎧ 1 n ⎫
mechanisms observed in natural systems and populations of MaxF = k1 ⎨Max[0, ∑ (Voltage% iwithDG −Voltage% iwithoutDG )]⎬
⎩ n i =1 ⎭
living beings. Their basic principle is the maintenance of a
population of solutions to a problem (genotypes) as encoded ⎧ m ⎫
+ k 2 ⎨Max[0, (∑ PjwithoutDG − PjwithDG )]⎬
information individuals that evolve in time [6]. ⎩ j =1 ⎭
Generally, GA comprises three different phases of search:
⎧ m ⎫
Phase 1: creating an initial population; phase 2: evaluating a k 3 ⎨Max[0, (∑ Q jwithoutDG − Q jwithDG )]⎬
fitness function; phase 3: producing a new population. A ⎩ j =1 ⎭
genetic search starts with a randomly generated initial
population within which each individual is evaluated by First term in this expression, related to difference between
means of a fitness function. Individual in this and subsequent average of voltage profile percentage in base case and other
generations are duplicated or eliminated according to their cases according to DG’s locations. By this way summation of
fitness values. Further generations are created by applying GA active and reactive power losses difference are computed and
operators. This eventually leads to a generation of high objective function is established. Max operator is used for
performing individuals. enforcing the constraints. The negative values influence is
There are usually three operators in a typical genetic forbidden by this operator. Mentioned parameters are listed
algorithm [6]: the first is the production operator (elitism) below:
which makes one or more copies of any individual that posses
Voltage% iwithDG : Voltage Percent in ith bus with DG resource.
a high fitness value; otherwise, the individual is eliminated
from the solution pool; the second operator is the Voltage%iwithoutDG : Voltage Percent in ith bus without DG
recombination (also known as the 'crossover' ) operator. This resource.
operator selects two individuals within the generation and a PiwithDG : Active Power Losses in jth branch with DG resource.
crossover site and carries out a swapping operation of the
string bits to the right hand side of the crossover site of both PiwithoutDG : Active Power Losses in jth branch without DG
individuals. Crossover operations synthesize bits of resource.
knowledge gained from both parents exhibiting better than QiwithDG : Reactive Power Losses in jth branch with DG
average performance. Thus, the probability of a better
offspring is greatly enhanced; the third operator is the resource.
'mutation' operator. This operator acts as a background QiwithoutDG : Reactive Power Losses in jth branch without DG
operator and is used to explore some of the invested points in resource.
the search space by randomly flipping a 'bit' in a population of k1, k 2 , k3 : Emphasis or penalty factors
strings. Since frequent application of this operator would lead n : Number of Buses
to a completely random search, a very low probability is
m : Number of Branches.
usually assigned to its activation.

V. PROPOSED ALGORITHM
IV. PROBLEM FORMULATION
Although using algorithm genetic is applied to this
The main goal of the proposed algorithm is to determine the
optimization problem, but we think objective functions are
best locations for new distributed generation resources by
different so this algorithm is chosen for Distributed
minimizing different function, related to project aims. In this
Generation allocation in this research. GAs are able to reach a
work, we are following two goals for determining the formula
good solution - with high probability to be the best one – by a
that are used in point of start:
finite steps of evolution steps performed on a finite set of
1. loss Reduction possible solutions. Some objective function for optimization
2. voltage profile Improvement problem is the minimization of:

These items should compose with constraints to obtain the 1. The cost due to system losses
proper objective functions. The main constraints in the 2. The network upgrading investments
optimization process in the proposed methodology are: 3. The generation cost (investments fuel, operation and
maintenance)
1. Losses before installing DG in power grid should
be less than losses after installing of it. The flow chart of our manner is shown in Fig. 1. Genetic
Loss with DG ≤ Loss without DG Algorithm sets in the core of it and load flow algorithm is

252
 World Academy of Science, Engineering and Technology 37 2008

used in everywhere that evaluation process is essential. First

of all, base case network is evaluated and its indexes are
registered. In the main path of it, Genetic Algorithm
meaningful search to find the optimum place for installing
DG. This routine is programmed under MATLAB software.

Start

Read Data From Network

Coding of Data Network

Run Base Case Power Flow

Create initial population and Fitness Function

Registry the base case characteristic

Crossover

Mutation

Fig. 2 Single Line Diagram of the Khoda Bande Loo feeder in ETAP
space
Matting Pool
No
TABLE I
LINE INFORMATION
Satisfy the
constraint Line Characteristics
Yes
From To Rohm X ohm
No
Create new generation
1 2 0.176 0.138

Run load flow and set fitness value for each

chromosome
2 3 0.176 0.138

3 4 0.045 0.035
Compare fitness value with base case characteristic

4 5 0.089 0.069
Is it
Satisfy the 5 6 0.045 0.035
objective
function?
5 7 0.116 0.091
YES
7 8 0.073 0.073
solution

8 9 0.074 0.058
Fig. 1 Flow chart of implemented methodology 8 10 0.093 0.093

7 11 0.063 0.05
VI. CASE STUDY
Other system was selected from one part of Tehran 11 12 0.068 0.053
distribution network. Single line diagram of the network is
7 13 0.062 0.053
shown in Fig. 2. ETAP software (ETAP PowerStation is a
strong analyzer program that is used for checking the obtained
result from MATLAB). This is MV feeder with 13 buses from Two DG resources are used for improving of network
63/20 kV Khoda-Bande-Loo substation. Table I illustrates line indexes. Both of them have 1600kw and 0.01kVar. For
and bus information. The advantage of using this network normalizing fitness values that are attained during performing
should be its practicality. the routine, following points are intended:

• Three constant values are:

12
SAP = ∑ Pi withoutDG (1)
i =1

253
 World Academy of Science, Engineering and Technology 37 2008

12 1000 ⎧⎪ 1 12 ⎫⎪
SRP = ∑ Qi MaxF = ⎨Max[0, ( ∑ Voltage%iwithDG − AVP )]⎬
withoutDG AVP ⎪⎩ 13 i =1 ⎪⎭
i =1 (2)
700 ⎧⎪ 12 ⎫⎪
+ ⎨Max[0, ( SAP − ∑ PjwithDG )]⎬
SAP ⎪ ⎪⎭
1 13 ⎩ j =1
AVP = ∑ Voltage% j withoutDG ⎧ 12 ⎫
13 j =1 200 ⎪ ⎪
⎨Max[0, ( SRP − ∑ Q jwithDG )]⎬
SRP ⎪ j =1 ⎪⎭
⎩
TABLE II
BUS INFORMATION Emphasis factor for each terms are: k1 =1000; k2 = 700 and
k3 = 200. With this expression, the best location is bus 9 and
Bus Characteristics 13.
P kw Q kvar
Genetic Algorithm has a matting pool that new generations
Bus Number
are exist with GA operators. However the number of
1 0 0 chromosomes which are injected into matting pool be more,
the number of offspring will be highness. If this number is not
2 890 468 proper, a purpose location, close to absolute optimal point is
selected or time is grown up.
3 628 470 Voltage percent and power losses in base case and all
possible cases with installing DG are illustrated in Fig. 3 and
4 1112 764 4.
5 636 378

6 474 344

7 1342 1078

8 920 292

9 766 498

10 662 480

11 690 186

12 1292 554
Fig. 3 Comparing Total Power Losses for all possible cases
13 1124 480

In this expressions SAP is summation of branches Active

Power Losses in base case, SRP is summation of branches
Reactive Power Losses in base case and AVP is Average of
Voltage Percent in base case.

• Maximum amounts of active and reactive power

losses summation for buses are occurred when
DG resources not installed in the network (base
case).
• Minimum amounts of voltage percent average for
Fig. 4 Comparing Average of Percent for all possible cases
buses are occurred in base case.
• With this constant values, objective function Results for installing of two DG on bus 3 and 9 are shown
normalized as following expression: in following figures:

254
 World Academy of Science, Engineering and Technology 37 2008

Fig. 5 Average voltage percent for base case and DG installed on bus Fig. 8 Surface diagram of fitness values
9 and 13
With analyzing of simulations we can conclude:

1. The appropriate selection of the first population is effected

on algorithm convergence. So be sure about grid coding
correctness and them fitness values.

2. For Distributed Generation resource placing in network

with lower number of buses, disabling the GA operators,
cause to convert GA to direct search. Absolute optimal
point is given by this way and in case of increasing the
number of buses, even with one distributed generation
resource, direct research finished by increasing of
Fig. 6 Total active power losses for base case and DG installed on calculating and passing time.
bus 9 and 13
3. Considering the load flow algorithm, one disadvantage of
For assurance of result validity, all possible cases are GA in this routine is increasing of computing level and
calculated. You should see them in Fig. 7. If fitness values are passing time. The maximum style of installing DG for 13
indicated in a matrix, because of the same selected capacities buses case study with two distributed generation source is
for both DG resources, it is symmetrical. For example the
78 time. Finding the best solution without performing
element 2, 4 is same amount as 4, 2. Element 2, 4 point to the
fitness value which comes from installing DG on bus 2 and 4. routine for all possible cases is art of GA by using its
Fig. 8 shows this value in surface diagram. It is clear that, operator and valuing method. Part of computing and
these values are increased, with closing to the last buses. passing time is related to use multi objective genetic
algorithm.

4. Lack of updatable and available information bank which is

contained specification of distribution network in
different part of the country would be a failure in
continuous connection between university researches and
electricity industry. Existing such information based on
reality need for applying university theory researches
practically.
5. In running our algorithm, whatever amounts of
Fig. 7 Comparing fitness values for all possible cases chromosomes which enter in matting pool are higher, it
closes to absolute solution, but if this amounts to be low,
it has a point near the absolute solution.

6. To comprehensive evaluate of this system, we compose tow

DG sources and cerate a matrix for all available cases.
For example element 22 of this matrix illuminate fitness
of installing DG with double capacity on bus number 2.

255
 World Academy of Science, Engineering and Technology 37 2008

After analyzing this matrix best values are yield as distribution networks", CCECE/CCGEI, Saskatoon, May 2005 IEEE,
PP.645-648.
following:
[7] A. Keane, and M. O'Malley, "Optimal allocation of embedded
generation on distribution networks," IEEE Trans. Power Systems, vol.
TABLE III 20, pp. 1640-1646, Aug. 2005.
COMPARE RESULTS [8] Distributed generation: a definition; Thomas Ackermann , Göran
Total Total Andersson, Lennart Söder; Electric Power Systems Research 57 (2001)
Bus Voltage Active Reactive 195–204; December 2000.
Fitness
No. % mean Power Power
Losses Losses (kw)
13 98.823 100.5 74 361.47
M. Sedighizadeh received the B.S. degree in
Electrical Engineering from the Shahid Chamran
13 & 9 98.81 92.9 72.3 396.35
University of Ahvaz, Iran and M.S. and Ph.D. degrees
in Electrical Engineering from the Iran University of
12 & 9 98.803 93.5 71.7 394.60 Science and Technology, Tehran, Iran, in 1996, 1998
and 2004, respectively. From 2000 to 2007 he was
with power system studies group of Moshanir
As you see in this Table III best amounts are shown. The Company, Tehran, Iran. Currently, he is an Assistant
algorithm has been used this values for its decision Professor in the Faculty of Electrical and Computer
criteria. Maximum of mean voltage percent, occur when Engineering, Shahid Beheshti University, Tehran,
Iran. His research interests are Power system control and modeling, FACTS
DG installed at bus 13 (3200kw = 1600kw+1600kw).
devices and Distributed Generation.
Minimum total of active power losses occur when DG
install at buses 9 and 13, also minimum amount of total
reactive power occur when DG install at buses 12 and 9. A. Rezazade was born in Tehran, Iran in 1969. He
The last column shows fitness value of this best point, but received his B.Sc and M.Sc. degrees and Ph.D. from
Tehran University in 1991, 1993, and 2000,
notes that, weight factors have clear effects on fitness. respectively, all in electrical engineering. He has two
years of research in Electrical Machines and Drives
7. This surface diagram in Fig. 8 is shown that: laboratory of Wuppertal University, Germany, with
a. Installing DG in every bus improve network the DAAD scholarship during his Ph.D. and Since
2000 he was the head of CNC EDM Wirecut
characteristic, in other hand totally installing DG has machine research and manufacturing center in
proper effects in this system. Pishraneh company. His research interests include
b. Whatever you close to last bus, fitness value is more application of computer controlled AC motors and EDM CNC machines and
important. computer controlled switching power supplies. Dr. Rezazade currently is an
assistant professor in the Power Engineering Faculty of Shahid Beheshti
University. His research interests are Power system control and modeling,
VII. CONCLUSION Industrial Control and Drives.
In this paper the results of application of GA algorithm to
the optimal allocation of DGs in distribution network is
presented. The effectiveness of the proposed algorithm to
solve the DG allocation problem is demonstrated through a
numerical example. The Khoda Bande Loo distribution test
feeder in Tehran has been solved with the proposed algorithm
and, the simple genetic algorithm.

REFERENCES
[1] K. Nara, Y. Hayashi, K. Ikeda,and T. Ashizawa, "Application of tabu
search to optimal placement of distributed generators," in Proc.
2001IEEE Power Engineering Society Winter Meeting, pp. 918-923.
[2] T. K. A. Rahman, S. R. A. Rahim, and I. Musirin, "Optimal allocation
and sizing of embedded generators," in Proc. 2004 National Power and
Energy Conference, pp.288-294.
[3] G. Celli, and F. Pilo, "Optimal distributed generation allocation in MV
distribution networks," in Proc.2001 IEEE PICA Conference, pp. 81-86.
[4] W. El-Khattam, K. Bhattacharya, Y. Hegazy, and M. M. A. Salama,
"Optimal investment planning for distributed generation in a competitive
electricity market," IEEE Trans. Power Systems, vol. 19, pp. 1674-1684,
Aug.2004.
[5] W. El-Khattam, Y. G. Hegazy, and M. M. A. Salama, "An integrated
distributed generation optimization model for distribution system
planning," IEEE Trans. Power Systems, vol. 20, pp. 1158-1165, May
2005.
[6] M. Gandomkar,M. Vakilian,M. Ehsan, " A combination of genetic
algorithm and simulated annealing for optimal DG allocation in

256