Вы находитесь на странице: 1из 10

Multi-objective PFDE algorithm for solving the optimal siting and sizing

problem of multiple DG sources


Mohammad H. Moradi

, S.M. Reza Tousi, Mohammad Abedini


Department of Electrical Engineering, Faculty of Engineering, Bu-Ali Sina University, Hamedan, Iran
a r t i c l e i n f o
Article history:
Received 8 July 2013
Received in revised form 22 October 2013
Accepted 6 November 2013
Keywords:
Distributed generation (DG)
Multi-objective optimization
Pareto frontier differential evolution
Power losses
Voltage stability index (VSI)
a b s t r a c t
The upward trend in global warming has led to increase in utilization of renewable energy based distrib-
uted generation sources (DGs) in electricity production. This research presents a solution technique for
the distribution expansion planning problem using DGs. In this paper, three main factors associated with
the multiple DG sizing and placement procedure is scrutinized through a multi-objective optimization
approach. These factors include voltage stability, power losses and network voltage variations. In order
to solve this multi-objective optimization problem the Pareto Frontier Differential Evolution (PFDE) algo-
rithm is presented. The proposed method is implemented and tested on 69-buses and 33-buses IEEE test
systems. Results prove that the proposed method exhibits higher capability and efciency in nding opti-
mum solutions.
2013 Elsevier Ltd. All rights reserved.
1. Introduction
Distribution systems usually include feeders congured radially
and exclusively fed by a utility substation [1]. Incorporating DGs
within the distribution level have an overall positive impact on
reducing the power losses as well as improving the network volt-
age prole. Due to advances in small generation technologies, elec-
tric utilities have begun to change their electrical infrastructure
and have started utilization of DGs. In order to maximize the
resulting benets, careful attention has to be paid to location of
DGs as well as the appropriate amount of power injected by them
[1]. In other word to achieve the best results from DG deployment,
the DGs are to be both optimally located and sized in the corre-
sponding distribution systems. There are many advantages of
installing DGs in distribution systems. Some of these advantages
are listed below.
- Postponing the upgrade of an existing system.
- Peak shaving.
- Reduction of power losses.
- Low maintenance cost.
- High reliability.
- Power quality improvement in some cases.
- Possibility to exploit CHP generation.
- Meeting the increasing demand without requirement of extrav-
agant investment.
- Shorter construction schedules [24].
Solution techniques for DG deployment are attained via optimi-
zation methods. These techniques can be simply divided into size
and location sub problems. Many researches proposed different
methods such as analytic procedures as well as deterministic and
heuristic methods to solve the problem. Celli et al [5] introduce a
new software procedure on genetic algorithm (GA) which is capa-
ble of establishing the optimal DG location on an existing medium
voltage distribution network. The objective function is evaluated
based on the total cost of plant, residual, management and the cost
of losses. The paper aims to minimize the power loss costs in
distribution systems during a prex period. In Hegazy et al [6] an
integrated model is proposed for solving comprehensive DG opti-
mization problems with different solution alternatives, generating
power from DG, purchasing power from the main grid. GAMS
method is used with binary decision variables. Akbarimajd et al
[7] utilized continuous power ow methodology to locate the
buses that are more sensitive to voltage collapse. The sensitive
bus set is ranked based on their severity which is used accordingly
to indicate potential bus locations for placement of single and mul-
tiple DG sources. Lee et al [8] and Vakilian et al [9] applied the Tabu
search (TS) method to optimally size the DG, as well as the reactive
source within the distributed systems. They formulated their con-
strained nonlinear optimization problem by minimizing an objec-
tive function that sums the total cost of active power losses, line
loading and the cost of the added reactive sources. A combined
0142-0615/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijepes.2013.11.014

Corresponding author. Tel.: +98 811 2526078; fax: +98 811 8352828.
E-mail addresses: mh_moradi@yahoo. co. uk (M. H. Moradi),
smr_tousi@yahoo.com (S.M. Reza Tousi), m_abedini_dr@yahoo.com (M. Abedini).
Electrical Power and Energy Systems 56 (2014) 117126
Contents lists available at ScienceDirect
Electrical Power and Energy Systems
j our nal homepage: www. el sevi er . com/ l ocat e/ i j epes
method that included the GA and TS technique in order to solve the
DG sizing optimization problem was developed by Ehsan et al [10].
The authors restricted the number of DGs, as well as their gross
capacity, to be revealed prior to executing the optimization proce-
dure, but they augmented the objective function with penalty
terms in their formulation to handle the constraint violations. Hag-
hifam [11] used the ant colony optimization methodology for solv-
ing the distributed generation capacity and location problems. In
this research, they used a penalty factor to handle the violated con-
straints. Ajay et al [12] applied the particle swarm optimization
(PSO) to minimize the system real power losses. Constraint viola-
tions were handled via a penalty factor which was augmented with
the objective function. Borges et al [13] solved the distributed gen-
eration problem by using the GA method. They utilized the meta-
heuristic methods in solving multiple and single DG capacity and
location. In our latest published paper, GA and PSO were used to
nd the optimum location and sizing of DG respectively. A penalty
factor was used to handle the violated constraints [14]. Barin et al
[15] studied the electricity distribution system planning, aiming to
nd an adequate DG for steady-state operation of distribution sys-
tems. They utilized the Bellman-Zadeh method to apply weights to
each parameter. Dehghanian et al [16] solved the distributed gen-
eration problem by using the Non-dominated sorting genetic algo-
rithm II (NSGAII) method. They formulated their constrained
nonlinear optimization problem by minimizing an objective func-
tion that had three main factors. Kang et al [17] solved the problem
by PSO and group search optimizer for coping with the difcult
scenarios in a power system. Also, in [18], Kang et al proposed no-
vel efcient population-based heuristic approach for optimal loca-
tion and capacity of DGs in distribution networks. In their work the
objectives were fuel cost minimization, power loss reduction, and
voltage prole improvement.
In this paper, three main factors associated with the procedure
of multiple DGs sizing and placement is studied through a multi-
objective optimization approach. In the multiple-objective case,
there does not necessarily exist a solution that would best with re-
spect to all objectives because of incommensurability and conict
among the objectives. In the developed Pareto Frontier Differential
Evolution (PFDE) algorithm methodology, for maintaining diversity
in the population, the non-dominated ranking was combined with-
in the tness sharing method. The results showed that the proposed
PFDE algorithm is better than the single objective function in con-
ditions of solution worth and number of iterations. The rest of this
paper is structured as follows: discussion of the problem in Sec-
tion 2, differential evolution algorithmdifferential in Section 3, pro-
posed method PFDE Optimization Algorithm in part 4, application
study and numerical results in Section 5, discussions in part 6 and
conclusion in Section 7.
2. Discussion of the problem
There are two main aspects to the optimal DG integration prob-
lem. The rst is the optimal DG location, while the second is opti-
mal DG sizing. Optimal DG sizing is a highly nonlinear constrained
optimization problem represented by a nonlinear objective func-
tion that is subject to nonlinear inequality and equality constrains
as well as to boundary restrictions imposed by the system planner.
The full formulation of the DG optimization problem is shown in
the following section.
2.1. The objective functions
Three main factors associated with the multiple DGs sizing and
placement procedure are scrutinized through a multi-objective
optimization approach. One of the main factors considered in this
research is the voltage stability index. The network losses and im-
proves the network voltage prole are the other two criteria to be
investigated in this paper.
2.2. Active power losses
The objective function to be minimized in the distributed gen-
eration problem is the distribution systems active power losses
formulated as:
f
1
MinP
loss
1
p
loss

X
NB
i1
X
NB
j>1
fY
ij
gV
2
mi
V
2
i
2V
mi
V
mj
cosd
mi
d
mj
2
2.3. Voltage stability index (VSI)
When DGs are joined to a radial distribution system, the voltage
stability index is modied. The bus, which the value of voltage
Nomenclature
i,j bus indices
P
loss
the real active power losses
P
gni
generated active power delivered to bus ni
Q
gni
generated reactive power delivered to bus ni
P
dni
demand active power bus ni
Q
dni
demand reactive power bus ni
Y
ij
magnitude of the ijth element of admittance bus matrix
Y
Vm
i
magnitude of bus mi complex voltage (pu)
V
ni
magnitude of bus ni complex voltage (pu)
V
min
minimum bus voltage
V
max
maximum bus voltage
NB number of buses
d
ni
phase angle of voltage at bus n
i
(V
ni
= V
ni
< d
ni
)
d
mi
phase angle of voltage at bus m
i
d
min
minimum bus phase angle
d
max
maximum bus phase angle
h
ni
phase angle of the ijth element of admittance bus ma-
trix Y
Pf
DG
DG operating power factor
pf
DG
min
minimum permissible value DG operating power factor
pf
DG
max
maximum permissible value DG operating power factor
|S
ni
| apparent power at bus n
i
dij phenotype distance between two individuals
@
share
the maximum phenotype distance allowed
I
ni
current of branch n
i
R
ni
resistance of branch n
i
X
ni
reactance of branch n
i
VSI (n
i
) the voltage stability index
.
(n
i
= 2, 3, . . . , n)
Fig. 1. One line diagram of a radial distribution network.
118 M.H. Moradi et al. / Electrical Power and Energy Systems 56 (2014) 117126
stability index is poor, is further sensitive to voltage collapse. The
equations utilized to formulate this index are proposed in [19] to
solve the load ow for distribution networks. The VSI of the net-
work is given by Eq. (2). Fig. 1 also shows a branch of radial system.
By using this voltage stability index, one could measure the level of
stability of radial distribution networks and thus appropriate ac-
tion may be taken if the index indicates a poor level of stability.
After the voltages at each bus are considered for load ow study,
the VSI for all the receiving end buses of radial distribution systems
can be easily calculated using:
VSIni j V
mi
j
4
4P
ni
niR
ni
Q
ni
niX
ni
j V
mi
j
2
4P
ni
niR
ni
Q
ni
niX
ni

2
3
The bus, at which the range of the stability index is, minimum is
more sensitive to the voltage collapse. The objective function stabil-
ity index is given by Eq. (4).
f
2

1
VSIni
ni 2; 3; :::; n
n
4
VSI (n
i
) should be maximized for developing voltage stability as
its consequence the presented objective function will be
minimized.
2.4. Total voltage variation (TVV)
One of the chief rewards of siting distributed generation in the
distribution network is the maintaining the voltage at the load ter-
minals within a reasonable value and development in the voltage
prole systems. By introducing distributed generation into existing
network, voltage can be enhanced because the distributed genera-
tion can provide a part of the reactive and real power to the cus-
tomer and decrease the losses in the systems; thus the voltage
variation is enhanced. The total voltage variation in the network
is given by Eq. (5).
f
3

X
NB
i1
1 V
mi
j j 5
2.5. Equality constraints
The equality constraints are the nonlinear power ow equations
which state that all the real and reactive powers at any distribution
system bus must be conserved [20].
P
gni
P
dni
V
ni
X
N
j1
V
nj
Y
nj
cosd
ni
d
nj
h
nj
6
Q
gni
Q
dni
V
ni
X
N
j1
V
nj
Y
nj
sind
ni
d
nj
h
nj
7
2.6. Inequality constraints
2.6.1. Voltage
The bus voltage phase angle and magnitudes are bounded be-
tween two extreme levels forced by physical limitations.
V
min
6 V
ni
6 V
max
8
d
min
6 d
ni
6 d
max
9
2.6.2. Power factor
The DG power factor is permitted for values within lower and
upper limits determined by the nature and type of the DG to be in-
stalled in the distribution systems.
pf
DG
min
6 pf
DG
ni
6 pf
DG
max
10
2.6.3. Thermal limit
Thermal constraint of distribution lines of the systems must not
be exceeded from following value.
jS
ni
j 6 jS
max
ni
j ni 1; ; N 11
2.7. DG modeling
Most of the methods offered so far model distributed genera-
tion as a generator that is able of delivering just active real power.
Though, there are other models of distributed generation being
incorporated into distribution networks [21]. Different models
were proposed in the paper to represent the DG in the distribution
systems. In this paper, using three types of DG shave the capability
of satisfying such a requirement. One of them the DG model is rep-
resented by injecting a constant real power similar the photovol-
taic type and other implies that the DG model is represented by
injecting a constant real and reactive power at a designated power
factor into a distribution bus similar the CHP type and tertiary DG
model is represented by injecting only reactive power into bus
alike synchronous condenser (SC). The distributed generation gen-
erated constant real and reactive power should be within the DG
capacity.
P
min
gi
6 P
gi
6 Pg
max
gi
12
Q
min
gi
6 Q
gi
6 Qg
max
gi
13
3. Differential evolution algorithm (DE)
The single-objective evolutionary algorithm proposed by Price
and Storn in 1995 draws upon ideas from several genetic algo-
rithms and evolutionary methods [22]. One of them is a relatively
new element to the general class of evolutionary methods called
differential evolution. As other evolutionary methods and genetic
algorithms, DE is a population based technique for nding global
optima. The three main ingredients are mutation, crossover and
selection. Much of the power of this method is resulted from a very
useful mutation operator that is simple and elegant. Mutations are
obtained by computing the difference between two randomly cho-
sen parameter vectors in the population and adding a portion of
this difference in a third randomly chosen parameter vector to ob-
tain a candidate vector. The resulting magnitude of the mutation in
each of the parameters is different and close to optimal. Two
parameters are adopted in DE: crossover (CR) and mutation (F).
CR controls the inuence of the parent in the generation of the
offspring. Higher values mean less inuence of the parent in the
features of its offspring. F scales the inuence of the set of pairs
of solutions selected to calculate the mutation value.
This method is considered as one of the most powerful evolu-
tionary algorithms for real number function optimization nowa-
days. The DE algorithm can be explained in the following steps
which shown in Fig. 2.
Step 1 (initialization): puts the iteration t = 0 and generates ran-
domly m population. [Xi (0), i = 1,. . ., m], where x
j
(0) = [x
j,1
(0),
x
j,2
(0), . . ., x
j,n
(0)]x
j,r
(0), will generate in searching space
[x
min
r
; x
max
r
] randomly.
Step 2 (mutation): Generate a random number for F.
Step 3 (tness): evaluating each population in the initial popu-
lation using the objective function.
Step 4 (time update): updates the time counter t = t + 1.
M.H. Moradi et al. / Electrical Power and Energy Systems 56 (2014) 117126 119
Step 5 (new population): generates a new population by repeat-
ing the following steps until the new population is done:
Generate a random number for CR (crossover).
(Mutation) Randomly pick three populations from x
j
(t) such
that x
j,r1
(t) x
j,r2
(t) x
j,r3
(t). A trial vector y
i
(t) is dened as
y
i
(t) = x
j,r1
(t) + F (x
j,r2
(t) x
j,r3
(t)).
(Crossover) candidate vector [x
j
(t), j = 1, . . . ,n] is obtained via
crossover operator involving the vectors x
j,i
(t) and y
i
(t) and
is dened as:
x
j
t
y
i
t if rand0; 1 6 CR
x
j;i
t if rand0; 1 CR

14
- (Selection) the selection process involves a simple replacement
of the original parameter vector with the candidate vector if the
objective function decreases by such an action. If new popula-
tion less than old population, the operation will be continued.
Else go to step 6.
Step 6 the operation will stop if one of the stopping criteria was
detected, else going back to step 2.
4. Proposed method PFDE optimization algorithm
In theory multiple-objective optimization is very different with
the single objective optimization [23]. In the single-objective case,
one attempts to obtain the best solution, which is absolutely supe-
rior to all other alternatives. In the multiple-objective case, there
does not necessarily exist a solution that is best with respect to
all objectives because of incommensurability and conict among
the objectives. For a given non-dominated point in the criterion
space, its image point in the decision space is called efcient or
non-inferior. A point in the criterion space is efcient if and only
if its image in the decision space is non-dominated. There usually
exists a set of solutions; non-dominated solutions or Pareto opti-
mal solutions, for the multiple objective cases which cannot simply
be compared with each other. There are different ways utilized in
practice, but this paper, proposed a differential evolution algorithm
to enumerate points along the Pareto front.
Over several iterations, and then employ method to rank the
quality of the trade-offs based on for optimal planning of DGs. In
this paper, an extension to differential evolution has been
proposed to handle multiple objectives. This technique is also a
Start
Initial set random variable settings
(Parent)
Load flow and evaluation of initial Solution fitness
Determine F (mutation rate) CR (crossover rate)
Rand< CR
Select randomly three population and perform
mutation (offspring)
Replace offspring to population vector
yes
No
Is fitness of offspring better
than fitness of parent?
Replace to new population vector
Select randomly one parent of old population
Discard the offspring in new
population
yes
No
Size of new population< old
population
Check the stop critien
Stop
yes
yes
No
No
Fig. 2. The DE method for optimal sizing and siting of DGs.
120 M.H. Moradi et al. / Electrical Power and Energy Systems 56 (2014) 117126
Pareto-based approach that utilizes non-dominated ranking and
selection procedures to calculate numerous Pareto-optimal solu-
tions simultaneously. The Pareto ranking approach consists of
assigning rank 1 of non-dominated population and removing them
from contention, then nding the non-dominated population
among the remaining ones, rank 2; and so on. Therefore this meth-
od, assigns all non-dominated solutions in identical tness value in
order to give them an equal probability of reproduction. This larger
population is subjected to the non-dominated sorting and ranking
procedure. The ranking is then used to subsequently decrease the
population to its original size. Solution diversity is reached by
the tness sharing proposed by Srinivas et al [24]. The DE algo-
rithm can be explained in the following steps which shown in
Fig. 3.
Step 1 (initialization): puts the iteration t = 0 and generates ran-
domly m Gaussian population, will generate in searching space
x
min
r
; x
max
r
randomly.
Step 2 (mutation) Generate a random number for F
Start
Initial set randomvariable settings
(Parent)
Load flow and evolution of initial Solution
fitness
Determine F ( rate mutation) CR ( rate
crossover )
Rand < CR
Select randomly three population from non-
dominated and perform mutation (offspring)
Replace offspring to population vector
yes
No
Is fitness of offspring
better than fitness of
parent?
Replace to new population vector
Select randomly one parent of old
population
Discard the offspring in
newpopulation
yes
No
Size of new population< old
population
Check the stop
critien
Stop
yes
yes
No
No
Front1
population
classified
Indentify non-dominated
individuals
Assign dummy fitness
Sharing in current front
Front= Front +1
Reproduction according to
dummy fitness
No
yes
Fig. 3. The PFDE method for optimal sizing and sitting.
M.H. Moradi et al. / Electrical Power and Energy Systems 56 (2014) 117126 121
Step 3 (population classied) removing all dominated solutions
by repeating the following steps until the non-dominated is
identied:
The non-dominated individuals are rst identied and assigned
a large dummy tness value.
To maintain diversity in the population, these individuals are
shared with their dummy tness values, that is dened as:
Shd
ij

1
d
ij
@
share

2
if d
ij
< @
share
0; otherwise
(
15
After sharing, these non-dominated individuals are ignored
temporarily
The second non-dominated front in the rest of the population is
identied and assigned a dummy tness value that is kept smal-
ler than the minimum shared dummy tness of the previous
front, and tness sharing applied.
This process is continued until the entire population is classied
into several fronts, and output tness evolution.
Step 4 (time updating): updates the time counter t = t + 1.
Step 5 (new population): generates a new population by repeat-
ing the following steps until the new population is done:
Generate a random number for CR (crossover).
(Mutation) Randomly pick three populations from x
j
(t) from
the non-dominated solution such that x
j, r1
(t) x
j, r2
(t) .
xj, r
3
(t). A trial vector y
i
(t) is dened as y
i
(t) = x
j
,r
1
(t) + F
(x
j
,r
2
(t) x
j
,r
3
(t)).
(Crossover) candidate vector [x
j
(t), j = 1, . . . ,n] is obtained via
crossover operator involving the vectors x
j,i
(t) and y
i
(t) and is
dened as:
x
j
t
y
i
t if rand0; 1 6 CR
x
j;i
t if rand0; 1 CR

16
(Selection) the selection process involves a simple replacement
if the candidate vector dominates the original parameter vector.
If new population less than old population the operation will be
continued going to step 5, else going to step 6.
Step 6 the operation will stop if one of the stopping criteria was
detected, else going back to step 3.
5. Application study and numerical results
5.1. Distribution system model
In this paper, two test networks have been utilized to test and
conrm the suggested technique. The rst network is a 33 bus dis-
tribution test system with a total real and reactive load of 3.7 MW
and 2.3 Mvar, respectively and it is shown in Fig. 4 [25]. The second
system is a 69 bus distribution test system with a total real and
reactive load of 3.8 MW and 2.69 Mvar, respectively as it is shown
in Fig. 5 [25]. The suggested technique has been executed in the
MATLAB software and tested for the two test networks. Table 1,
demonstrates objective function values per installed DGs. The
Fig. 4. Single line graph of the 33 bus distribution test network.
Fig. 5. Single line graph of the 69 bus distribution test network.
Table 1
Objective function values of the systems per installation of DGs.
System Losses (P.U.) TVV (P.U.) f
3
33 BUS 0.211 0.314 1.49
69 BUS 0.225 0.219 1.45
Table 2
The PFDE parameters.
Pop size Crossover (CR) Function evaluations bound Mutation (F)
10 0.2 1700 0.5
Table 3
Comparisons of six non- dominated solutions of Multi-objective PFDE method with a single objective.
Method solutions No. Bus (33 bus) [bus(1)-bus(2)-bus(3)] Type of DGs Losses (P.U.) f
3
(P.U.) f
2
PQ (Mwat, Mvar) Bus (1) Q (Mvar) Bus (2) P (Mwat) Bus (3)
PFDE (1) [30-9-15] 1.0330 , 0.7284 0.5327 1.2107 0.0511 0.0080 1.0522
PFDE (2) [18-12-13] 0.3826 , 0.1387 1.1779 1.4051 0.1322 0.0004 1.2017
PFDE (3) [15-31-30] 1.0654 , 0.7335 1.1513 0.5625 0.0643 0.0070 1.0649
PFDE (4) [30-15-11] 1.0448 , 0.6793 0.5586 1.2349 0.0782 0.0134 1.0517
PFDE (5) [14-30-31] 0.9380 , 0.6845 0.6331 1.2317 0.0681 0.0047 1.0529
PFDE (6) [32-13-12] 0.9621 , 0.5842 0.6849 1.2801 0.1351 0.0035 1.0521
GA/PSO[14] [32-16-11] 1.2 0.863 0.925 0.1034 0.0124 1.0519
GA [14] [11-29-30] 1.5 0.4228 1.0714 0.1063 0.0407 1.0537
PSO[14] [13-32-8] 0.9816 0.8297 1.1768 0.1053 0.0335 1.0804
122 M.H. Moradi et al. / Electrical Power and Energy Systems 56 (2014) 117126
parameters of PFDE technique utilized for solving the problems
presented in this research are provided in Table 2.
5.2. 33 Bus system distribution
The optimal solutions in case of six non-dominated solutions are
shown in Table 2. The decision maker can select the best solution
based on the signicance of the operating constraints and different
objectives. In [14] the GA/PSO, GA and PSOthat with weighted sum-
mation technique is applied to convert the objective functions to a
single-objective optimization problem and solving the distributed
generation capacity and location problems. The solutions are
shown in Table 3 and compared with PFDE method. In Fig. 6 Pareto
front solutions are shown. Red asterisks are dominated and Black
diamonds are non-dominated solutions. Table 4 describes the opti-
mal solutions in case of six non-dominated Solutions that the DG
model is represented by injecting a constant real power.
Fig. 7 describes the voltage prole of whole buses in the 33 bus
network. It compares the voltage prole magnitudes before and
after installation of DGs. Prior to installation of DGs, the voltage
magnitude of bus 17 was poor, which is improved with installation
of DGs. In addition, the voltage ranks at whole buses of the net-
work have been improved. Fig. 8 shows the VSI. The VSI at the
buses for distribution network were improved after installation
of DGs.
5.3. 69 Bus system distribution
Considering 69 bus radial distribution system, the voltage
prole pre and after placing voltage regulators is shown in
Fig. 10. The voltage stability index for 69-node system is shown
in Fig. 9. The minimum voltage is also improved using PFDE. The
simulation result of the multi objective function for 69 bus radial
distribution system is shown in Table 5. The results compared with
GA/PSO, GA and PSO seems to offer improvement to optimal
solutions with Multi-objective function. In this network, siting
candidates have a discrete search space, while search space for
capacity of DGs is wide and continuous causing more difcult
search for the GA/PSO method compared with PFDE. Minimum
voltage and maximum voltage stability index of the system have
been improved and the network losses are reduced in each differ-
ent solution level.
6. Discussion
6.1. Multi-objective functions
The PFDE has a faster convergence speed compared with GA/
PSO, GA and PSO. In addition, the single objective methods have
performed weaker in both nding the capacity and siting of DGs
compared to multi-objective method. The obtained results show
that the PFDE method behaves robustly and is able to converge
to the correct Pareto front with an acceptable coverage and spread
of results. The major advantages of the proposed method are that
the technique is relatively easy to implement with a few tuning
parameters and also it can handle interdependence relationships
0.05
0.1
0.15
0.2
0.25
0
0.02
0.04
0.06
0.08
1.05
1.1
1.15
1.2
1.25
Loss (P.U.)
Voltage Profile (P.U.)
V
o
l
t
a
g
e

S
t
a
b
i
l
i
t
y

I
n
d
e
x
Fig. 6. The Pareto front solutions for multi-objective function for 33 bus systems.
Table 4
Comparisons of six non-dominated solutions of Multi-objective PFDE method with a single objective. (The DG injected only real power).
Method Solutions No. Bus (33 bus)[bus(1)-bus(2)-bus(3)] Type of DGs Losses (P.U.) f
3
(P.U.) f
2
P (Mwat)
PFDE (1) [13-26-32] 0.91
1.25
0.88
0.088 0.0566 1.0774
PFDE (2) [30-29-16] 1.01
1.05
1.2
0.1293 0.0103 1.0481
PFDE (3) [14-15-32] 0.5924
1.2185
1.1983
0.145 0.0015 1.0514
PFDE (4) [10-15-32] 0.7750
0.9916
1.3336
0.1250 0.0062 1.0506
PFDE (5) [11-16-32] 1.2129
0.6748
1.2135
0.1252 0.0054 1.0505
PFDE (6) [15-29-31] 1.2691
1.1179
0.9680
0.1322 0.003 1.0467
GA/PSO[14] [32-16-11] 1.20
863
0.925
0.1034 0.0124 1.0519
GA [14] [11-29-30] 1.5
0.4228
1.0714
0.1063 0.0407 1.0537
PSO[14] [13-32-8] 0.9816
0.8297
1.1768
0.1053 0.0335 1.0804
M.H. Moradi et al. / Electrical Power and Energy Systems 56 (2014) 117126 123
0 5 10 15 20 25 30
0.9
0.92
0.94
0.96
0.98
1
1.02
1.04
BUS
V
o
l
t
a
g
e

P
r
o
f
i
l
e
GA / PSO (with weighted sum method) per instalitin DGs PSO PFDE (1) PDE(2)
Fig. 7. Voltage prole for a 33 bus network.
5 10 15 20 25 30
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
BUS
V
o
l
t
a
g
e

S
t
a
b
i
l
i
t
y

I
n
d
e
x
per instalition DGs PFDE (2) PFDE (1) PSO GA / PSO
Fig. 8. Voltage stability index (VSI) for a 33 bus network.
10 20 30 40 50 60
0.9
0.92
0.94
0.96
0.98
1
1.02
Bus
V
o
l
t
a
g
e

p
r
o
f
i
l
e
Per installation DG GA / PSO PSO PFDE(2) PFDE(1)
Fig. 10. Voltage prole for a 33 bus network.
10 20 30 40 50 60
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
1.05
1.1
1.15
BUS
V
o
l
t
a
g
e

s
t
a
b
i
l
i
t
y

i
n
d
e
x
PFDE(1) per installation DG PFDE(2) GA/PSO
Fig. 9. Voltage stability index (VSI) for a 69 bus network.
124 M.H. Moradi et al. / Electrical Power and Energy Systems 56 (2014) 117126
among the decision variables effectively. Thus, it will result in
quick and acceptable Pareto optimal solutions.
6.2. Voltage stability index
Using the voltage stability index, one could measure the level of
voltage stability of radial distribution networks and thus appropri-
ate action can be taken if the index indicates a poor level of stabil-
ity. After DG installation, the proposed methods showed
considerable improvement in voltage stability index. (Figs. 8 and
9).
6.3. Voltage prole
The TVV index is a good indicator for determination of voltage
deviation from nominal values. Closer the index to zero means
the better situation of the network. So, this index must be mini-
mized to improve the voltage prole. Also higher VSI for each
bus means proper condition of that bus regarding voltage stability.
The proposed multi-objective optimization has been solved using
PFDE algorithm to calculate the optimum solutions for location
and sizing of multi DGs. Following DG installation, all methods
have shown major improvement in the system voltage prole
and VSI, especially in case of using proposed Pareto Frontier Differ-
ential Evolution method (Figs. 7 and 10 and Table 35).
6.4. DG modeling
Tables 3 and 4 demonstrate the optimal sizing and siting of dis-
tributed generation sources for various DG models by PFDE meth-
od for 33 bus test network. One of these models deals with only
delivering real power (Tables 3 and 4), while in the other model
both real and reactive power are delivered. Results show that the
installation of different types of DG directly inuences the power
loss reduction. Furthermore, the voltage prole and voltage stabil-
ity index improvements depend on the optimal power factor of DG.
7. Conclusion
Integrating DGs in electric power system networks is gaining
popularity globally due to its overall positive impacts. In this paper,
multiple DG sizing and siting problem is formulated as a con-
strained nonlinear programming optimization problem. The volt-
age stability index, active power losses and voltage prole are
the issues considered as objective functions to be optimized. The
optimal multiple DG sizing and siting problem was solved by the
developed PFDE method. In the developed PFDE methodology, for
maintaining diversity of solutions, the non-dominated ranking
was combined with tness sharing method. The efciency of the
proposed method has been demonstrated through 33-buses and
69-buses radial distribution networks and the obtained results
were compared with results of using single objective weighted
sum method in GA/PSO, GA and PSO. The results showed that the
PFDE technique had a higher potential in discovering optimal
solutions.
Appendix
See Table 6.
References
[1] Willis LH. Power distribution planning reference book ABB Power T&D
Company Inc.. Cary, North Carolina: Published by Marcel Dekker Inc.; 1997.
[2] Brown RE, Jiuping P, Xiaorninf F, Koutlev K. Siting distribution generation to
defer T&D expansion. IEEE transmission and distribution conference and
exposition 2001.
[3] Mndez VH, Rivier J, de la Fuente JI, Gmez T, Arceluz J, Marn J, et al. Impact of
distributed generation on distribution investment deferral. Int J Elect Power
Energy Syst May 2006;28(4):24452.
[4] Arya L, Atul Koshti SC. Distributed generation planning using differential
evolution accounting Voltage stability consideration. Int J Elect Power Energy
Syst 2012;42(1):196207.
[5] Cali G, Pillow F. Optimal distributed generation allocation in MV distribution
networks. IEEE power engineering society international conference on power
industry computer appcations, pp. 18, innovative commuting for power-
electric energy meets the market 2001.
[6] ElKhattam W, Hegazy YG, Salana MMA. An Integrated distributed generation
optimization model for distribution system planing. IEEE transaction on power
systems 2005;20(2):115865.
[7] Hedayati H, abaviniaki SAN, Akbarimajd A. A Method for placements for
service reliability important. IEEE Power Eng Soc Summer Meeting
2002;2:71924.
[8] Y, Lee, Rhee, S.-K. Lee, and S.-K.You,dispersed generator placement using
fuzzy-ga in distribution systems IEEE Power engineering society summer
meeting, Chicago, USA, Vol. 3, pp. 11481153, 2002.
Table 5
Comparisons of six non-dominated solutions of Multi-objective PFDE method with a single objective.
Method Solutions No. Bus (69 bus) [bus(1)-bus(2)-bus(3)] Type of DGs Losses (P.U.) f
3
(P.U.) f
2
PQ (Mwat, Mvar) Bus (1) Q(Mvar) Bus (2) P (Mwat)Bus (3)
PFDE (1) [13-62-63] 0.7566, 0.4339 0.6766 1.4824 0.0227 0.0425 1.0634
PFDE (2) [20-64-61] 0.9622, 0.4178 0.5146 1.4564 0.0360 0.0042 1.0718
PFDE (3) [18-64-63] 0.5985, 0.5902 0.8147 1.4562 0.0339 0.0303 1.0624
PFDE (4) [17-63-61] 0.6975, 0.4653 0.7300 1.4987 0.0376 0.0249 1.0229
PFDE (5) [64-61-20] 1.2368, 0.4608 0.4885 1.1179 0.0420 0.0005 1.0347
PFDE (6) [20-64-61] 0.9623, 0.4424 0.5146 1.4676 0.0680 0.0033 1.0624
GA/PSO[14] [63-61-21] 0.8849
1.1926
0.9105
0.0811 0.0031 1.0237
GA [14] [21-62-64] 0.9297
1.0752
0.9848
0.089 0.0012 1.0303
PSO[14] [61-63-17] 1.1998
0.7956
0.9925
0.0832 0.0049 1.0335
Table 6
The numerical values used for the constraints in equations
V
min
0.9 (p.u) S
max
ni
5 (MVA)
V
max
1.04 (p.u) Pg
max
gi
2 (Mwat)
dmin
0.01 (rad)
Pg
min
gi
0
dmax
0.01 (rad) Qg
max
gi
1 (Mvar)
pf
DG
max
1
Qg
min
gi
0
pf
DG
min
0.80
M.H. Moradi et al. / Electrical Power and Energy Systems 56 (2014) 117126 125
[9] Gandomkar M, Vakiliyan M, Ehsan M. A combination of genetic algorithm and
simulated annealing for optimal DG allocation in distribution networks.
Canada conference on electrical and computer emerging, Saskatoon, pp. 645
648, 2005.
[10] Gandomkar M, Vakiliyan M, Ehsan M. A genetic-based Tabu search algorithm
for optimal DG allocation in distribution networks. electrical power
components and systems, Vol. 33, pp. 13511362, Dec. 2005.
[11] H. Foliage and M. Haghifam,ACO Based algorithm for distributed generation
sources allocation and sizing in distribution systems, IEEE Lausanne power
Tech, pp. 555560, 2007.
[12] P.Ajay-D-Vimal Raj, S.Senthilkumar, J.raja, S.Ravichandran, and T.G.
Palanivelu,optimization of distributed generation capacity for line loss
reduction and voltage prole improvement using PSO Elecktrika: Journal of
electrical engineering, g, Vol 1, n.., p..41-48 2008.
[13] Falcao DM, Borges CT. Optimal distributed generation allocation for reliability,
looses, and voltage improvement. Int J Elect Power Energy Syst
2006;28(6):41320.
[14] Moradi MH, Abedini M. A combination of genetic algorithm and particle
swarm optimization for optimal DG location and sizing in distribution
systems. Int J Elect Power Energy Syst 2012;34(1):6674.
[15] A. Barin, L. Pozzatti, L. Canha, R.Machado, A.Abaide, F. Ferret, C. Carvalho.,
Analysis of Multi-Objective Methods Applied to Distributed Generation
Systems POWERENG 2007, April 1214, 2007, Setbal, Portugal.
[16] M. Aghtaie, P. Dehghanian, Optimal Distributed Generation Placement in a
Restructured Environment via a Multi-Objective Optimization Approach IEEE
Power engineering society summer meeting, pp. 16, 2008.
[17] Kang Q, Zhou MC, An J, Wu Q. Swarm intelligence approaches to optimal
power ow problem with distributed generator failures in power networks.
IEEE Trans Automation Sci Eng 2013;10(2):34353.
[18] Kang Q, Lan T, Yan Y, Wang L, Wu Q. Group search optimizer based optimal
location and capacity of distributed generations. Neurocomputing
2012;78(1):5563.
[19] Charkravor ty M, Das D. Voltage stability analysis of radial distribution
networks. Int J Elect Power Energy Syst 2001; 23(2):12935.
[20] El-Khattam, W., Hegazy, Y. G., and Salama, M. M. A., An Integrated Distributed
Generation Optimization Model For Distribution System Planning, IEEE
Transactions On Power Systems, Vol. 20, No. 2, May 2005.
[21] D. Quoc, Hung, and N. Mithulananthan. Analytical Expressions for DG
Allocation in Primary Distribution Networks IEEE transactions on energy
conversion. Vol. 25.no.3.September 2010.
[22] Storn, R., and Price, K. Differential evolution a simple and efcient heuristic
for global optimization over continuous spaces. J. Global Optim., 11, 341359.
[23] Deb K. Multi-objective optimization using evolutionary
algorithms. Chichester, U.K.: Wiley; 2001.
[24] N. Srinivas and K. Deb.,Multi-objective optimization using non-dominated
sorting in genetic algorithms, evolutionary computation vol2, no. 3, pp. 221
248, 1994.
[25] Baran ME, Wu FF. Optimal sizing of capacitors placed on a radial distribution
system. IEEE Trans Power Deliv 1989;4(1):73543.
126 M.H. Moradi et al. / Electrical Power and Energy Systems 56 (2014) 117126

Вам также может понравиться