You are on page 1of 12

Application of particle swarm optimization for distribution

feeder reconguration considering distributed generators


J. Olamaei
a
, T. Niknam
b,
*
, G. Gharehpetian
c
a
Technical Engineering Department, Ph.D Student, Science & Research Branch, Islamic Azad University-(IAU), Teharn, Iran
b
Electrical Engineering Department, Shiraz University of Technology, Shiraz, Iran
c
Electrical Engineering Department, Amirkabir University of Technology, Teharn, Iran
Abstract
In many countries the power systems are going to move toward creating a competitive structure for selling and buying
electrical energy. These changes and the numerous advantages of the distributed generation units (DGs) in term of their
technology enhancement and economical considerations have created more incentives to use these kinds of generators than
before. Therefore, it is necessary to study the impact of DGs on the power systems, especially on the distribution networks.
The distribution feeder reconguration (DFR) is one of the most important control schemes in the distribution networks,
which can be aected by DGs. This paper presents a new approach to DFR at the distribution networks considering DGs.
The main objective of the DFR is to minimize the deviation of the bus voltage, the number of switching operations and the
total cost of the active power generated by DGs and distribution companies. Since the DFR is a nonlinear optimization
problem, we apply the particle swarm optimization (PSO) approach to solve it. The feasibility of the proposed approach is
demonstrated and compared with other evolutionary methods such as genetic algorithm (GA), Tabu search (TS) and dif-
ferential evolution (DE) over a realistic distribution test system.
2008 Elsevier Inc. All rights reserved.
Keywords: Distributed generator; Distribution feeder reconguration; Particle swarm optimization; Distribution networks
1. Introduction
Distributed generation units (DGs) are grid-connected or stand-alone electric generation units located
within the electric distribution system at or near the end user. It is generally accepted that centralized electric
power plants will remain the major source of the electric power supply in future. DGs, however, can comple-
ment central power systems by providing the incremental capacity to the grid or to an end user. Installing
DGs at or near the end user can also, in some cases, benet the electric utility by avoiding or reducing the
cost of transmission and distribution system upgrades. Taking into account the consumers considerations,
a potentially lower cost, higher service reliability, high power quality, increased energy eciency, and energy
0096-3003/$ - see front matter 2008 Elsevier Inc. All rights reserved.
doi:10.1016/j.amc.2007.12.053
*
Corresponding author.
E-mail addresses: taher_nik@yahoo.com, niknam@sutech.ac.ir (T. Niknam).
Available online at www.sciencedirect.com
Applied Mathematics and Computation 201 (2008) 575586
www.elsevier.com/locate/amc
independence can be the key points for the interest in DGs. The use of the renewable distributed energy gen-
erations such as wind, photovoltaic, geothermal or hydroelectric power can also provide signicant environ-
mental benets. It is expected that the penetration of the distributed generation will surpass more than 25% of
the total generation, in the foreseeable future [1]. Therefore, it is of crucial importance to study their impacts
on distribution systems. The distribution feeder reconguration (DFR) is one of the most signicant control
schemes in the distribution networks which can be aected by DGs.
Generally, the DFR is dened as altering the topological structure of distribution feeders by changing the
open/closed states of sectionalization and tie switches so that the objective function is minimized, and the con-
straints are met.
In recent years, many researchers have investigated loss minimization in the area of network recongura-
tion of distribution systems. The problem of minimizing losses through distribution system reconguration
was rst reported in 1975 by Merlin and Back [2], who modeled the distribution system as a spanning tree
structure, with line sections represented by the arcs of a graph, and the buses by the nodes. The nal cong-
uration that minimized losses was determined from the values found for binary variables associated with
switch status in which system constraints were neglected. In [36], the authors have suggested employing a
method based on heuristic algorithm to determine the conguration of radial distribution networks, which
nally led to loss minimization. In [7], the authors have proposed a solution procedure, employing simulated
annealing (SA), to search an acceptable non-inferior solution. In [8,9], the authors have presented articial-
intelligence-based applications. In [10], the authors have discussed time-varying load analysis to reduce losses.
In [11,12], the authors have combined the optimization techniques with heuristic rules and fuzzy logic for
higher eciency and robust performance. Recently, genetic algorithm (GA) and evolutionary programming
have been used [1320]. In [21,22], the authors have also proposed a solution procedure regarding DGs,
but the impacts of DGs on distribution system performance have not been studied in detail.
In this paper, a novel DFR approach is presented for a distribution network containing DG units. A cost-
based control methodology is proposed as a proper criterion for the real/reactive-power control of the DG
units of a distribution system. In the proposed DFR approach four objective functions have been considered
as follows:
v The cost of the active power generated by DGs.
v The cost of the active power generated by distribution companies.
v The number of switching operations.
v The deviation of the voltage of buses.
Since the tie and sectionalizing switches and DG units are non-dierentiable and nonlinear, respectively, the
distribution feeder reconguration problem is conventionally considered a mixed-integer nonlinear program-
ming problem. To solve such a problem, classical methods, e.g. linear programming, mixed-integer program-
ming, quadratic programming, etc., can be applied. However, in some cases, the mentioned methods fail to
provide the global minima and only reach local minima. Moreover, some classical methods cannot handle
the integer problems [23]. The two foregoing shortcomings can be overcome if an evolutionary method is uti-
lized. It is independent of the objective function type and constraints renders. In this paper, an evolutionary
optimization method, based on particle swarm optimization (PSO), has been utilized to solve the DFR prob-
lem.The paper has been organized as follows: In Section 2, the proposed DFR has been formulated, and a cost
evaluation method applicable to distributed generation has been presented. In Sections 3 and 4, the basic prin-
ciples of the PSO and its application in the DFR control problem have been discussed, respectively. In Section
5, the eectiveness of the proposed approach has been demonstrated and compared with other evolutionary
methods. Lastly, the conclusion of the paper is provided in Section 6.
2. Modeling of distribution feeder reconguration
In this section, the distribution feeder reconguration considering DG units has been modeled as a multi-
objective, non-dierentiable optimization problem.
576 J. Olamaei et al. / Applied Mathematics and Computation 201 (2008) 575586
2.1. Objective function
In the proposed DFR approach, the objective function consists of four terms: (i) the cost of the active
power generated by the distribution companies, (ii) the cost of the active power generated by DG units,
(iii) the number of switching operations, and (iv) the deviation of the bus voltage. The objective function
of the DFR with its four terms has been demonstrated by the following equation:
f (X) =
X
N
sub
i=1
P
sub;i
Price
i

X
N
g
i=1
C
Pgi
(P
gi
) w
1

X
N
sw
i=1
[S
i
S
o;i
[ w
2

X
N
bus
i=1
[V
i
V
rat
[; (1)
X = [Sw; P
sub
; P
G
; Q
G
[;
Sw = [Sw
1
; Sw
2
; . . . ; Sw
N
sw
[;
P
sub
= [P
sub;1
; P
sub;2
; . . . ; P
sub;N
sub
[;
P
G
= [P
g1
; P
g2
; . . . ; P
gN
g
[;
Q
G
= [Q
g1
; Q
g2
; . . . ; Q
gN
g
[;
where N
sub
, N
g
, N
sw
and N
bus
are the number of substations, DGs, switches and buses, respectively. X is the
state variable vector. P
sub,i
is the active power of the ith substation. P
sub
is the substation active power vector
including active power of all substations. Sw
i
is the state of the ith switch specied in terms of on/o status,
taking 0 or 1 as its value. Sw is the switching state vector including state of all switches. Price
i
is the electrical
energy price at the ith substation. C
Pgi
(P
gi
) is the cost of active power generated by the ith DG. Q
G
is the DGs
reactive power vector including reactive power of all DGs. Q
gi
is the reactive power of the ith DG.P
G
, the DGs
active power vector including active power of all DGs. P
gi
is the active power of the ith DG. S
i
and S
o,i
are the
new and original states of switch i, respectively. And V
i
and V
rat
, the real and rated voltage on bus i, whilst w
1
and w
2
represent weighting factors.
2.2. Cost model for DG units
In general, the cost of each kWh of electric energy generated by a DG unit is assumed to be composed of (i)
the initial investment, including the cost of equipment, infrastructure, commissioning, etc., (ii) operation and
maintenance (O&M) costs, and (iii) fuel cost [24]. This can be expressed by the following equation:
C
Pg
(P) = a b P: (2)
Coecients a and b are calculated as follows:
a =
capital cost ($=kW) capacity (kW) Gr
life time (Year) 365 24 LF
; (3)
b = fuel cost ($=kWh) O&M cost ($=kWh); (4)
where Gr and LF are the annual rates of benet and DG loading factor, respectively.
2.3. Constraints
The constraints can be listed as follows:
v Active power constraints of DGs:
(P
gi
)
2
(Q
gi
)
2
6 S
2
gi;max
; (5)
P
gi
and S
gi,max
are the active and apparent power of the ith DGs.
v Distribution line limits:
[P
Line
ij
[ < P
Line
ij;max
; (6)
J. Olamaei et al. / Applied Mathematics and Computation 201 (2008) 575586 577
[P
Line
ij
[ and P
Line
ij;max
are absolute power owing over distribution lines and maximum transmission power be-
tween nodes i and j, respectively.
v Distribution power ow equations:
P
i
=
X
N
bus
i=1
V
i
V
j
Y
ij
cos(h
ij
d
i
d
j
);
Q
i
=
X
N
bus
i=1
V
i
V
j
Y
ij
sin(h
ij
d
i
d
j
);
(7)
P
i
and Q
i
are the net injected active and reactive power at the ith bus. V
i
and d
i
are the amplitude and the
angle of voltage at the ith bus, respectively. And Y
ij
and h
ij
are the amplitude and the angle of the branch
admittance between the ith and jth buses.
v Capacity margin of transformers:
[S
i
[ < S
i;max
; (8)
|S
i
| and S
i,max
are absolute and maximum powers of the ith transformer, respectively.
v Radial structure of the network.
3. The basic principle of PSO algorithm
PSO is a population-based optimization tool, where the system is initialized with a population of random
particles and the algorithm searches for optima by updating generations and is originally developed by Ken-
nedy and Eberhart [24,25] and inspired by the paradigm of birds ocking. PSO consists of a swarm of particles
and each particle ies through the multi-dimensional search space with a velocity, which is constantly updated
by the particles previous best performance and by the previous best performance of the particles neighbors.
In an n-dimensional search space, let the position and velocity of the ith individual be represented as vectors
x
i
= (x
i1
, . . ., x
id
, . . ., x
in
) and v
i
= (v
i1
, . . ., v
id
, . . ., v
in
), respectively. The best previous experience of the ith par-
ticle is recorded and represented by Pbest
i
= (Pbest
i1
, . . ., Pbest
id
, . . ., Pbest
in
). The global best position of the
swarm found so far is denoted by Gbest
i
= (Gbest
i1
, . . ., Gbest
id
, . . ., Gbest
in
). The modied velocity of each
particle can be rst calculated regarding the personal initial velocity, the distance from personal (local) best
position and the distance from global best position as expressed by the following equation:
v
(t1)
id
= x v
(t)
id
c
1
rand
1
() (Pbest
id
x
(t)
id
) c
2
rand
2
() (Gbest
d
x
(t)
id
); (9)
x
(t1)
id
= x
(t)
id
v
(t1)
id
: (10)
Eq. (9) determines the direction in which the ith particle should be taken along. Therefore, the new position of
that particle can be determined by applying Eq. (10).
In these equations i = 1, 2, . . ., m is the index of each particle, t is the iteration number,rand
1
() and rand
2
()
are the random numbers between 0 and 1.The constants c
1
and c
2
are the weighting factors of the stochastic
acceleration terms, which pull each particle toward Pbest and Gbest positions. Low values allow particles to
roam far from the target region before being togged back. On the other hand, high values result in abrupt
movements toward, or backward the target region. Hence, the learning factors c
1
and c
2
are often set to
2.0 according to early experiences [26]. We bear in mind that the PSO has been found to be robust and fast
in solving nonlinear, non-dierentiable, multi-objective problems. Shi and Eberhart [27] has introduced the
parameter x into the PSOs equation to improve its performance. The appropriate selection of inertia weight
x in (7) provides a balance between global and local explorations, requiring less iteration on average to nd a
suciently optimal solution. As originally developed, x often decreases linearly from about 0.9 to 0.4 during a
run. In general, the inertia weight x is set according to the following equation [27]:
x
(t1)
= x
max

x
max
x
min
t
max
t: (11)
In Eq. (11), t
max
is the maximum number of iterations and t is the current iteration number. Fig. 1 represents a
graphical depiction of the basic idea of the particle swarm optimizer.
578 J. Olamaei et al. / Applied Mathematics and Computation 201 (2008) 575586
4. Application of PSO in the proposed DFR
In this section, the PSO algorithm has been applied to the DFR problem. It should be noted that the state
variables are real and reactive power values of the DG units, status of switches and active power value of sub-
stations. To apply the PSO algorithm, the following steps have to be taken:
Step 1: The input data including network conguration, line impedance and status of DGs and switches are
to be read.
Step 2: The proposed DFR problem needs to be transformed into an unconstrained one by constructing an
augmented objective function incorporating penalty factors for any value violating the constraints:
F (X) = f (X) k
1
X
N
eq
j=1
(h
j
(X
i
)
2
)
!
k
2
X
N
ueq
j=1
(Max[0; g
j
(X
i
)[)
2
!
; (12)
f (X) is the objective function values of DFR problem. N
eq
and N
ueq
are the number of equality and
inequality constraints, respectively. h
i
(X
i
) and g
i
(X
i
) are the equality and inequality constraints. k
1
and k
2
are the penalty factors, respectively.
Step 3: The initial population and initial velocity for each particle should be generated randomly.
Step 4: The augmented objective function (Eq. (12)) is to be evaluated for each individual by using the result
of distribution load ow.
Step 5: The individual that has the minimum objective function should be selected as the global position.
Step 6: The ith individual is selected.
Step 7: The best local position (Pbest) is selected for the ith individual.
Step 8: The modied velocity for the ith individual needs to be calculated based on the local and global posi-
tions and Eq. (9).
Step 9: The modied position for ith individual should be calculated based on Eq. (10) and then checked
with its limit.
Step 10: If all individuals are selected, go to the next step, otherwise i = i + 1 and go to Step 6.
Step 11: If the current iteration number reaches the predetermined maximum iteration number, the search
procedure is stopped, otherwise go to Step 4.
The last Gbest is the solution of the problem. The owchart of PSO algorithm is shown in Fig. 2.
5. Simulation results
In this section, the PSO algorithm is utilized to solve the DFR of a distribution test feeder, whose one-line
diagram is given in Fig. 3. It is assumed that every branch has a sectionalizing switch. This system has three
X
i
X
i + 1
V
G b e s t
V
L b e s t
V
i
V
i + 1
X
i
: Current searching Point X
i+1
: Modified searching Point
V
i
: Current Velocity V
i+1
: Modified Velocity
V
Gbest
: Velocity based on Gbest V
Lbest
: Velocity based on Lbest
Fig. 1. Concept of a searching by PSO. X
i
: current searching point; X
i+1
: modied searching point; V
i
: current velocity; V
i+1
: modied
velocity; V
Gbest
: velocity based on Gbest; V
Lbest
: Velocity based on Lbest.
J. Olamaei et al. / Applied Mathematics and Computation 201 (2008) 575586 579
feeders and four tie switches. The tie switches and sectionalizing switches are normally open and closed,
respectively.
It is assumed that there are three generators, whose specications are given in Table 1.
The active power prices at each substation are 0.05, 0.045 and 0.051 $/kW, respectively.
Read data including network configuration, line impedance
and status of DGs and switches
Create initial population and initial velocity based on
DGs, load values
Convergence condition is satisfied
Stop and print the results.
Yes
No
Calculate the next position for each individual based on eq. (10)
Calculate the modified velocity for each individual based on eq. (9)
Calculate the objective function for each individual,
based on distribution load flow results
Select the global position based on the objective
function values
Calculate the next position for each individual based on eq. (10)
Check the new position with its limits
i=i+1
Select the i
th
individual
Select the local position for the i
th
individual
All individuals are selected
No
Yes
Fig. 2. PSO owchart.
580 J. Olamaei et al. / Applied Mathematics and Computation 201 (2008) 575586
The parameters required for the implementation of the PSO algorithm are c
1
, c
2
, x
min
and x
max
. In this
paper, the best values for the mentioned parameters are c
1
= c
2
= 2, x
min
= 0.4 and x
max
= 0.9 determined
by 100 times running of the algorithm. The number of particles is 20.
Table 2 provides the simulation results carried out on the test system of Fig. 3, using MATLAB software,
and includes the best and the worst solutions. Furthermore, Table 2 provides the simulation results based on
the particle swarm optimization (PSO) algorithm, genetic algorithm (GA), Tabu search (TS) and dierential
evolution (DE) whilst their simulation parameters are presented in the Appendix, for 300 times running of the
algorithms. In Table 2, the smallest and the largest values of the minimized objective function are referred to
as the Best Solution and the Worst Solution, respectively. Comparison of the best and the worst solutions
of the proposed optimization algorithm with the other evolutionary methods conrms the eectiveness of the
proposed method. In addition to the best and the worst solutions, Table 2 provides the standard deviation and
the average value of the objective function (minimized) value for the PSO, GA, TS and DE.
Figs. 47 illustrate the convergence performance of the PSO algorithm, GA, DE and TS for the best solu-
tions. It can be evidently seen from these gures that the value of the objective function settles at the minimum
after about 130 iterations.
The average computing times for the PSO, DE, TS and GA algorithms are ~30, 210, 250 and 200 s in run-
ning on a P4 1.8 GHz/512 MB RAM, respectively.
Substation 1
Substation 2
Substation 3
1 2 3 8
10
11
12
20
21
22
23 24 25 26 27 28 29 30 31
19
18
17 16 15 14 13
9
7 6 5 4
Tie Sw1
Tie Sw2
Tie Sw3
Tie Sw4
Fig. 3. Distribution test feeder.
Table 1
Characteristic of generators
Capacity
(kW)
Max reactive
power (kVar)
Min reactive
power (kVar)
Capital cost
($/kW)
Life time
(year)
Fuel cost
($/kWh)
O&M cost
($/kWh)
Location
G1 300 240 180 3674 12.5 0.029 0.01 17
G2 1000 800 600 1500 20 0 0.005 31
G3 1000 800 600 715 20 0.067 0.006 8
Table 2
Simulation result based on dierent optimization methods
Method Best solution Worst solution Average Standard deviation
PSO 122.45 126.78 124.31 2.01
TS 123.78 128.36 126.01 3.43
DE 124.95 128.73 126.58 2.78
GA 123.49 127.67 125.73 2.47
J. Olamaei et al. / Applied Mathematics and Computation 201 (2008) 575586 581
0
100
200
300
400
500
600
700
800
1 21 41 61 81 101 121 141
Iteration
O
b
j
e
c
t
i
v
e

F
u
n
c
t
i
o
n

V
a
l
u
e
Fig. 4. Convergence performance of PSO for the best solution.
123
124
125
126
127
128
129
130
1 51 101 151 201 251 301
Iteration
O
b
j
e
c
t
i
v
e

F
u
n
c
t
i
o
n

V
a
l
u
e
Fig. 5. Convergence performance of GA for the best solution.
100
120
140
160
180
200
220
1 51 101 151 201
O
b
j
e
c
t
i
v
e

F
u
n
c
t
i
o
n

V
a
l
u
e
Iteration
Fig. 6. Convergence performance of DE for the best solution.
100
200
300
400
500
600
700
800
1 51 101 151 201
Iteration
O
b
j
e
c
t
i
v
e

F
u
n
c
t
i
o
n

V
a
l
u
e
Fig. 7. Convergence performance of TS for the best solution.
582 J. Olamaei et al. / Applied Mathematics and Computation 201 (2008) 575586
For the best solution, nal simulation results for two cases (dispatchable DGs (case 1) and non-dispatch-
able DGs (case 2)) are shown in Table 3. For the non-dispatchable DG units case, it is assumed that each DG
unit delivers half of its rated real power.
Before reconguration, the total power loss of the system is 276.2 kW and the minimum voltage is
0.8821 p.u.
As shown in Table 3, the total real losses and minimum voltage in case 1 are less than case 2 and those of
both cases are less than the case without reconguration. Considering Table 3, it can also be seen that in case 1
only one tie switch and one sectionalizing switch have been changed, whereas the remaining tie switches and
sectionalizing switches have not been changed.
6. Discussion
In this section, with reference to the tables and gures presented in the previous sections, the following dis-
cussion can be put forward:
v The comparison amongst convergence times of solutions based on PSO, GA, TS and DE shows that the
PSO has the capability of being used in practical systems.
v The PSO algorithm is precise. In other words, not only does this method reach a much better optimal solution
in comparison with GA, DE and TS, but also the standard deviation for dierent executions is very small.
v Since most of DG units have private ownership, the cost of active power generation can be used as a signal
for generating active power by DG units.
v When DG units are dispatchable, active power loss is 56.44 kW. This amount is 97.46 kW less than that of
DG units when they are not dispatchable. On the other hand, it can be concluded that the system perfor-
mance can be improved under proper control.
v The active power losses in cases 1 and 2 are less than those before the reconguration.
v The minimum voltage in case 2 (non-dispatchable DGs) is less than case 1 (dispatchable DGs).
v In case 1, only one tie switch and one sectionalizing switch are changed whilst in case 2, four sectionalizing
switches and four tie switches are changed. In other words, number of switching operations in case 1 is less
than case 2.
v The active power of DG1 is zero. The corresponding reason can be explained from the data of Table 1;
DG1 has the shortest life-time, the largest capital cost, and the largest O&M cost.
v The method can be applied to a wide variety of similar optimization problems with non-dierential and
non-continuous objective functions and constraints.
Table 3
Optimal dispatch schedule for the two cases
Dispatchable DGs (case 1) Non-dispatchable DGs (case 2)
Electrical energy losses (kW) 56.44 153.9
Tie switch 1 Open Close
Tie switch 2 Open Close
Tie switch 3 Close Close
Tie switch 4 Open Close
Sectionalizing switch of branch 2229 Open Open
Sectionalizing switch of branch 96 Close Open
Sectionalizing switch of branch 67 Close Open
Sectionalizing switch of branch 1617 Close Open
Minimum voltage 0.9351 0.9111
Number of switching operations 2 8
P
g1
(kW) 0
P
g2
(kW) 899.44
P
g3
(kW) 856.90
Q
g1
(kVar) 77.0
Q
g2
(kVar) 31.50
Q
g3
(kVar) 267.57
J. Olamaei et al. / Applied Mathematics and Computation 201 (2008) 575586 583
7. Conclusion
As pointed in the previous sections, issues such as environmental pollution, restructuring in electrical indus-
try and technology advancement have resulted in an increase in the usage of distributed generators which most
of the time connect to the distribution networks. Therefore, with an increase in connecting these generators to
the distribution networks, it is a necessity to study the eects of these generators on distribution systems. This
paper has presented an ecient approach to distribution feeder reconguration in presence of DG units. Due
to private ownership of DG units, a cost-based compensation method has been used to dispatch DGs. The
simulation results have prominently shown that the performance of the system is much better when DG units
are dispatchable. Also the PSO algorithm has been applied to nd the best solution for the DFR problem. The
results have demonstrated that PSO has better performance than GA.
Appendix. Genetic algorithm
In this paper, integer strings instead of binary coding are used to represent value of variables, and include
these processes:
1. Representation and initialization.
2. Fitness function.
3. Reproduction operation.
4. Crossover operation.
5. Mutation operation.
Simulation conditions are:
Initial population = 1000.
Selected population = 300.
Mutation = 4%.
Crossover probability = 0.20.3.
Dierential evolution
Dierential evolution (DE) is similar to genetic algorithm, but diers from GA with respect to the mech-
anism of mutation, crossover, and selections. In an overall view, the DE, based on mutation rules, is classied
into ve dierent strategies as follows [28]:
v Best/Rand:
Xp
i;j1
= X
best
a
+
(X
r1;j
X
r2;j
): (A:1)
v Rand/Rand:
Xp
i;j1
= X
r1;j
a
+
(X
r2;j
X
r3;j
): (A:2)
v Old/Best/Rand:
Xp
i;j1
= X
i;j
a
+
(X
best
X
i;j
X
r1;j
X
r2;j
): (A:3)
v Best/RandRand:
Xp
i;j1
= X
best
a
+
(X
r1;j
X
r2;j
X
r3;j
X
r4;j
): (A:4)
v Rand/RandRand:
Xp
i;j1
= X
r1;j
a
+
(X
r2;j
X
r3;j
X
r4;j
X
r5;j
): (A:5)
584 J. Olamaei et al. / Applied Mathematics and Computation 201 (2008) 575586
where X is the set of population; Xp
i,j+1
is a partner to mate with the next generation; X
i,j
is the ith member
of the current population; X
best
is the best member in population; a is constant factor in range of [0, 2]; j is
the current generation; j + 1 is the next generation; X
r1,j
, X
r2,j
, X
r3,j
, X
r4,j
and X
r5,j
are the randomly selected
populations in the current generation.
After the creation of the partner for mating, the crossover is applied on the member as follows:
X
i;j1
= X
+
i;j
(1 P
c
) Xp
+
i;j1
P
c
(A:6)
where P
c
is the crossover probability.
New generation is compared with its limits:
Initial population = 1000.
Selected population = 300.
Mutation = 4%.
Crossover probability = 0.20.3.
Tabu search
Tabu search is a heuristic algorithm for guiding the search to nd a good solution to a combinatorial prob-
lem. It is derived from the works of Fred Glover with seminal ideas and contributions from various other
sources. Tabu search has been successfully applied to obtain optimal or sub-optimal solutions to problems
such as timetables, the traveling salesman and so on [29].
To apply the Tabu search algorithm to solve the DFR problem the following steps should be repeated [29]:
Step 1: Generation of initial population.
Step 2: Selection of good population and generation of the Tabu list.
Step 3: Creation of new population.
Step 4: Evaluation and selection.
Step 5: Updating the Tabu list.
Step 6: Checking the convergence.
Simulation conditions are:
Initial population = 1000.
Selected population = 300.
Tabu list = 30.
References
[1] T. Ackerman, G. Anderson, L. Soder, Distributed generation: a denition, Elsevier Sci. (2003) 195204.
[2] A. Merlin, H. Back, Search for a minimal-loss operating spanning tree conguration in an urban power distribution system, in:
Proceedings of the Fifth Power System Computation Conference, Cambridge, UK, 1975, pp. 118.
[3] S. Civanlar, J.J. Grainger, H. Yin, S.S.H. Lee, Distribution feeder reconguration for loss reduction, IEEE Trans. Power Del. 3 (3)
(1988) 12171223.
[4] V. Gomes, s. Carneiro, A new reconguration algorithm for large distribution systems, IEEE Trans. Power Del. Syst. 20 (3) (2005)
13731378.
[5] D. Shirmohammadi, H.W. Hong, Reconguration of electric distribution networks for resistive line loss reduction, IEEE Trans.
Power Syst. 4 (1) (1989) 14921498.
[6] E. Lopez, h. Opaso, Online reconguration considering variability demand: applications to real networks, IEEE Trans. Power Syst.
19 (1) (2004) 549553.
[7] V. Parada, J.A. Ferland, Optimization of electrical distribution feeders using simulated annealing, IEEE Trans. Power Del. 19 (3)
(2004) 11351141.
[8] A. Augugliaro, L. Dusonchet, M. Ippolito, E.R. Sanseverino, Minimum losses reconguration of MV distribution networks through
local control of tie-switches, IEEE Trans. Power Del. 18 (3) (2003) 762771.
J. Olamaei et al. / Applied Mathematics and Computation 201 (2008) 575586 585
[9] H. Kim, Y. Ko, Articial neural network based feeder reconguration for loss reduction in distribution systems, IEEE Trans. Power
Del. 8 (3) (1993) 13561367.
[10] R. Taleski, D. Rajicic, Distribution network reconguration for energy loss reduction, IEEE Trans. Power Syst. 12 (1) (1997) 398
406.
[11] Q. Zhou, D. Shirmohammadi, W.H.E. Liu, Distribution feeder reconguration for service restoration and load balancing, IEEE
Trans. Power Syst. 12 (2) (1997) 724729.
[12] Debaprya Das, A fuzzy multi-objective approach for network reconguration of distribution systems, IEEE Trans. Power Del. 21 (1)
(2006) 202209.
[13] I.Z. Zhu, Optimal reconguration of electrical distribution network using the rened genetic algorithm, Elect. Power Syst. Res. 62
(2002) 3742.
[14] Y.C. Huang, Enhanced genetic algorithm-based fuzzy multi-objective approach to distribution network reconguration, Proc. Inst.
Elect. Eng. 149 (5) (2002) 615620.
[15] J.P. Chiou, F.S. Wang, Hybrid method of evolutionary algorithms for static and dynamic optimization problems with application to
fed-batch fermentation process, Comput. Chem. Eng. 23 (1999) 12771291.
[16] S. Ching-Tzong, C. Lee, Network reconguration of distribution systems using improved mixed-integer hybrid dierential evolution,
IEEE Trans. Power Del. 18 (3) (2003) 10221027.
[17] Y. Hong, S. Hu, Determination of network conguration considering multi-objective in distribution systems using genetic algorithm,
IEEE Trans. Power Syst. 20 (2) (2005) 10621069.
[18] J. Chiou, C. Chang, Variable scaling hybrid dierential evaluation for solving network reconguration of distribution system, IEEE
Trans. Power Syst. 20 (2) (2005) 668674.
[19] A. Delbem, A. Carvalho, N. Bretas, Main chain representation for evolutionary algorithms applied to distribution system
reconguration, IEEE Trans. Power Syst. 20 (1) (2005) 425436.
[20] K. Prasad, R. Ranjan, Optimal reconguration of radial distribution system using a fuzzy mutated genetic algorithm, IEEE Trans.
Power Del. 20 (2) (2005) 12111213.
[21] J. Chio, J. Kim, Integration operation of dispersed generations to automated distribution networks for network reconguration,
IEEE Bologna Power Tech. Conf. (2003) 23632367.
[22] M. Oliveria, L. Ochoa, Network reconguration and loss allocation for distribution systems with distributed generation, IEEE/PES
Trans. Distrib. Conf. Expos. (2004) 206211.
[23] T. Niknam, A.M. Ranjbar, A.R. Shirani, A new approach based on ant algorithm for Volt/Var control in distribution network
considering distributed generation, Iranian J. Sci. Technol. Trans. B 29 (B4) (2005) 115.
[24] J. Kennedy, R. Eberhart, Particle swarm optimization, in: IEEE International Conference on Neural Networks, vol. 4, Piscataway,
NJ, 1995, pp. 19421948.
[25] R. Eberhart, Yuhui Shi, Particle swarm optimization: development, application and resources, IEEE Congr. Evol. Comput. 1 (2007)
8186.
[26] X. Hu, Y. Shi, R. Eberhart, Recent advances in particle swarm, IEEE Congr. Evol. Comput. 1 (2004) 9097.
[27] Y. Shi, R. Eberhart, A modied particle swarm optimizer, in: Proceeding of the IEEE World Congress on Computational Intelligence,
May 1998, pp. 6973.
[28] B.V. Babu, M.M.L. Jehan, Dierential evolution for multiobjective optimization, in: The 2003 Congress on Evolutionary
Computation, vol. 4, 2003, pp. 26962703.
[29] W.M. Lin, F.-S. Cheng, M.T. Tsay, An improved Tabu search for economic dispatch with multiple minima, IEEE Trans. Power Syst.
17 (1) (2002).
586 J. Olamaei et al. / Applied Mathematics and Computation 201 (2008) 575586