IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 30, NO.

2, APRIL 2015 545

Multiobjective Optimization Algorithm for Switch

Placement in Radial Power Distribution Networks
José Roberto Bezerra, Student Member, IEEE,
Giovanni Cordeiro Barroso, Ruth Pastôra Saraiva Leão, and Raimundo Furtado Sampaio

Abstract—Customer satisfaction counts very much to the electric improvements in reliability indices of the utilities [3]. Different
power distribution companies. The improvement of reliability in- approaches have been pursued to achieve this target.
dices is an effective way to achieve this goal. The optimal placement Heuristic techniques have been successfully applied to the
of power switches is a desirable procedure because while it reduces
investment in network assets, it also reduces the number of cus- switch placement problem. The Immune Algorithm is applied in
tomers not supplied by outages and, accordingly, improves system [4] to derive optimal switch placement in a distribution network
reliability. This paper proposes a multiobjective optimization ap- using an objective function that minimizes the cost of customer
proach for switch placement based on the particle swarm optimiza- service outage and investment cost of line switches, considering
tion method applied to radial distribution networks of power util- the failure rates of load points involved.
ities. The proposed algorithm gives a set of solutions instead of a
single one. It intends to be a valuable and flexible tool for plan- The Ant Colony Optimization technique is employed in [5]
ning modern and reliable power distribution networks. Simula- to obtain the placement of sectionalizing switches in distribu-
tion results when using the IEEE 123-node test system and the Roy tion networks with distributed generation. Objectives are de-
Billinton Test System are presented as testbeds. fined as the improvement of reliability and minimization of the
Index Terms—Distribution planning, multiobjective optimiza- economic costs. A fuzzy approach is used to deal with the con-
tion, particle swarm optimization (PSO), power system reliability, flicting objectives.
switch placement. Fuzzy logic and genetic algorithms (GAs) are employed
in a hybrid algorithm in [6] for improvement of the SAIDI
index. The algorithm requires many network parameters for its
I. INTRODUCTION
application.
A three-state version of the particle swarm optimization

C USTOMER satisfaction is essential for a successful

business. The electric power companies are closely
supervised and subjected to heavy penalties by the regulatory
(PSO) method, called Trinary particle swarm optimization
(TPSO), is presented in [7]. The authors propose the TPSO
algorithm in order to obtain the optimal number and placement
agencies when the customer reliability indices, such as system
of sectionalizer switches in distribution networks using the
average interruption duration index (SAIDI), system average
ECOST as the objective function.
interruption frequency index (SAIFI), and cost of outage
The aforementioned techniques have performed relevant re-
(ECOST) are undermined. The improvement of such indices
sults for the switch placement problem. However, the disadvan-
is pursued especially by the distribution companies which are
tage of local optimum existing in heuristic algorithms [8] causes
the most affected amongst the generation, transmission, and
uncertainties about the accuracy of the results, what can lead to
distribution power sectors [1].
the application of nonheuristic strategies, such as the mixed-in-
The digital relays, digital automation, and the most recent
teger linear programming used in [9]. The authors ensure that
smart-grid concept have enhanced the operation of the distribu-
the proposed solution achieves global optimal results solving
tion power systems, prompting them to a self-healing condition
the local minimum problem.
[2]. These changes lead to the improvement of the power supply
Despite the results presented by the heuristic and nonheuristic
and satisfaction of customers. Much effort has been devoted to
algorithms for solving the switch placement problem, so far
circumvent power outages, and the optimal placement of power
they have not found practical applications due to the very high
switches is a valuable tool that is able to drastically reduce the
number of appointed switches, making it unfeasible to be ap-
impacts caused by outages in distribution networks, ensuring
plied in distribution networks or at least limited for small net-
works with a few feeders [10]. Moreover, these algorithms re-
quire reliability data, such as failure rates of the distribution net-
Manuscript received July 15, 2013; revised November 17, 2013 and January
27, 2014; accepted April 04, 2014. Date of publication May 06, 2014; date of work components, which are usually unavailable or unknown
current version March 20, 2015. This work was supported by the Brazilian Na- by the utilities.
tional Energy Agency (ANEEL). Paper no. TPWRD-00803-2013. This paper proposes a multiobjective switch placement
J. R. Bezerra is with the Federal Institute of Ceará, Fortaleza, Ceará 60115-
222, Brazil. (MOSP) algorithm to solve switch placement problems in
G. C. Barroso, R. P. Saraiva Leão, and R. F. Sampaio are with the Federal radial distribution networks. The MOSP algorithm is based
University of Ceará, Fortaleza 60455-760, Brazil. on the PSO method and adopts a multiobjective approach to
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. deal with the cost-reliability dilemma. The MOSP algorithm
Digital Object Identifier 10.1109/TPWRD.2014.2317173 achieves a set of solutions to the switch placement problem

0885-8977 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Sharda University. Downloaded on December 20,2024 at 10:00:05 UTC from IEEE Xplore. Restrictions apply.
546 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 30, NO. 2, APRIL 2015

instead of a single solution. This makes the proposed solution

more adaptable to different utilities' circumstances and eases
the decision making.
Case studies that validate the effectiveness of the MOSP algo-
rithm are presented using two test systems: the IEEE 123-Node
Test System and the Roy Billinton test system. The proposed
MOSP algorithm has achieved good performance when com-
pared to the proposals presented in [7] and [9].
The organization of this paper is as follows. A brief introduc-
tion of multiobjective optimization applied to the switch place-
ment problem is given in Section II. The problem formulation
is presented in Section III. The MOSP algorithm is presented in
Section IV. The results from the application of the MOSP algo- Fig. 1. Pareto-optimal front.
rithm are presented in Section V and the final conclusions are
discussed in Section VI.
small subset of the solutions can be assumed to be effective as
II. MULTIOBJECTIVE OPTIMIZATION will be explained. The switch placement problem will be taken
Optimization procedures refer to finding extreme values of as an example to introduce the dominance concept.
one or more parameters in order to achieve better performance The switch placement problem in power distribution net-
within a problem. These extreme points are called the optimal works presents a classical trade-off in the engineering domain:
solution of such a problem [11]. When only one parameter needs cost and reliability. Improvements in reliability can be achieved
to be optimized, this procedure is called single-objective opti- by suitably positioning switches in the power distribution
mization (SOO). Methods for solving SOO problems have been networks. However, the high cost of the switches and their
developed and are currently well understood to be applied in maintenance makes the installation of such equipment carefully
a wide range of problems. However, most of the problems in assessed by the utilities.
science and engineering have more than one objective to be op- As the number of installed switches increases, it is an ex-
timized. This has fostered the knowledge area called multiob- pected improvement in the network reliability due to greater
jective optimization (MOO). switch flexibility to isolate parts of the network and transfer
The most applied approach used to solve an MOO problem is loads to alternative feeders. However, the cost that is incurred
to convert it into an SOO problem. The weighted sum method increases with the number of power switches. The issue is to
(WSM) is a simple and efficient way to achieve this goal. In achieve maximum reliability with minimum costs. The relia-
the WSM, a composite objective function is derived such that bility can be evaluated using the number of customers not sup-
each objective is multiplied by a user-defined weight to form a plied (CNS)—one of the most important indices monitored by
weighted sum. The choice of the weights in most cases is not the regulatory agencies to evaluate the quality of service of the
a trivial matter; it might be subjective, nontechnical, or expe- companies. The lower the CNS is, the higher the reliability of
rience based. From the composite objective function, the algo- the system.
rithm then gives away a particular solution [11]. Fig. 1 shows a set of solutions, for instance, of a switch place-
Generally, more than one solution is obtained in MOO ment problem, represented by the letters from to . Com-
problems, and as the objectives are normally conflicting, for paring solutions and , it can be noted that is better than
instance, when comparing two distinct solutions, one can be from the standpoint of objective 1 (number of switches),
better than the other from a given viewpoint. For such cases, and, conversely, solution is better than from the stand-
a unique solution cannot be declared optimal. The ideal pro- point of objective 2 (CNS). Therefore, neither solution nor
cedure is to find a set of solutions that suits several situations can be assumed to be the optimal solution for both objectives
that matches different expectations, thereby easing the decision simultaneously.
making. The evaluation of solutions and is distinct. Both are of
Evolutionary algorithms (EAs) are largely applied to MOO- the same value from the standpoint of objective 2, although
based problems. EA procedures are based on the behavior of is better from the objective 1 perspective. Therefore, there is no
birds, insects, or flocks transforming the whole population into advantage to select instead of . Solution is considered a
search agents for solutions. Each population agent evaluates a poor quality solution. It is said to be dominated by solution
solution for the problem. Thus, in a single simulation run, many that, in turn, is not dominated by any other solution. Thus, is
solutions are available [12]. This feature of EA makes them very called a nondominated solution.
suitable for MOO problems to achieve a set of solutions and not The nondominated set of solutions from a specific problem
only a unique solution. forms the pareto-optimal set, defined as follows [11].
1) Definition 1: Among a set of solutions , the nondomi-
A. Pareto-Optimal Front nated sets of solutions are those that are not dominated by
Despite the multiobjective evolutionary algorithms any member of the set .
(MOEAs) giving a set of solutions in one simulation run, most 2) Definition 2: The nondominated set of the entire feasible
of the solutions are useless for solving the problem. Only a search space is the globally pareto-optimal set.
Authorized licensed use limited to: Sharda University. Downloaded on December 20,2024 at 10:00:05 UTC from IEEE Xplore. Restrictions apply.
BEZERRA et al.: MULTIOBJECTIVE OPTIMIZATION ALGORITHM FOR SWITCH PLACEMENT 547

From the aforementioned definitions, it can be noted that none

of the solutions , , and , from Fig. 1, dominates each other
and no other solution dominates any of them. Thus, they form
a pareto-optimal set of solutions and the line compounded by
the entire set of the nondominated solutions is called the pareto-
optimal front.

III. PROBLEM FORMULATION

The MOSP algorithm proposed in this paper is based on the
discrete binary version of the particle swarm algorithm [13] that
has its continuous version originally proposed in [14]. The flow-
chart of the MOSP algorithm describing each of its steps is pre-
sented in Fig. 2.
The algorithm is developed to obtain a configuration of
switches that minimizes the number of CNS in radial distri-
bution networks in order to reduce the impact of faults for
the customers. The only parameters available are the network
topology and the number of customers per load point. Failure
rates are not available.
The network topology is represented by Graph Theory. An in-
cidence matrix containing the nodes, edges, weights, and the
relationship among them is defined for the analyzed network.
Details about the topology representation can be found in [3].
From matrix , the number of particles, or dimension ,
for each swarm are determined. The dimension corresponds to
the maximum possible number of switches in the distribution
networks. Dimension is given by

(1)

where is the number of branches in the distribution network,

is the number of normally open switches, and is the
number of circuit breakers (CBs). The swarm size is set to
.
Each solution is assigned as a candidate solution
which is a vector of dimension . Each position of corresponds
to a branch in the network that can assume 1 or 0 values. Value
1 means that a switch must be placed in the respective branch,
and value 0 means that there is not a switch in that branch.
At the end of iteration steps, the solutions are classified into
dominated or nondominated in order to form the pareto-optimal
front.

IV. ALGORITHM DESCRIPTION

The MOSP algorithm is divided in three parts: 1) initial
values; 2) searching; and 3) solution classification. The first
one needs the network topology as input for dimension evalu-
ation. The learning factors ( and ) and the weights ( and
) are also set. The second part is the PSO searching which Fig. 2. Flowchart of the MOSP algorithm.
evaluates the solutions within a limited number of iterations.
The last part is the classification of each obtained solution as
dominated or nondominated. The learning factors, also known as acceleration coefficients,
control the searching algorithm to improve solutions' diversity
A. Initial Values or faster convergence. Making , the convergence is
The learning factors and weights need to be set before the reached faster. Making the opposite , the solutions
simulation run. The assignments for the reliability weight achieved are more diverse by reducing the risk of local min-
and cost weight are discussed in Section IV-D and the imum. For building a pareto-optimal front, the diversity is pri-
learning factors are discussed ahead. oritary and, consequently, the second assumption is pursued.
Authorized licensed use limited to: Sharda University. Downloaded on December 20,2024 at 10:00:05 UTC from IEEE Xplore. Restrictions apply.
548 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 30, NO. 2, APRIL 2015

Most studies set each of the learning factors ( and ) at is obtain a reference value for the number of switches. For the
2 [13], [15]. Thus, for this paper, are adopted second stage, a nonzero value is assigned to . For the study
as the initial values. After several simulations for tuning, the cases presented ahead, is adopted considering
learning factors are set to 4 and 2 for the study that reliability and cost are equally important. For , the
cases presented. pareto-optimal front tends to present more reliable and expen-
sive solutions, while for the opposite , the pareto-op-
B. Searching timal front tends to present cheaper and less reliable solutions.
The searching is split in two stages in order to avoid local The member function is given by
minimum. This division allows the adjustment of the weights
( and ) and the number of iterations along the simulation.
At the first stage, a reference value for the reliability (i.e., the (5)
best configuration of switches regardless of the cost) is found.
This reference leads to the second stage for reducing the number
where is the number of branches in the distribution network,
of switches.
is the number of customers switched off by a fault in the th
The value is the best solution from the entire swarm
branch, and is the total number of customers normally sup-
evaluated at the end of each iteration. The value corre-
plied by the distribution network. Thus, the evaluates the
sponds to the best solution among all values ever eval-
number of CNS that a specific set of switches performs consid-
uated. The and solutions are randomly initialized,
ering all of the contingencies.
as shown in Fig. 2, and they are compared at the end of each
The member function is evaluated by
iteration. If the currently proposed solution at iteration
is better than the and/or solutions, assumes the (6)
new and/or solutions, respectively. The velocity
and are updated at the end of each iteration. where is the number of switches in solution and is
the maximum number of switches for any given configuration.
C. Solution Classification
Each solution is stored at the end of each iteration as
V. CASE STUDIES
can be noted from Fig. 2. Then, the set of stored solutions is
classified as dominated or nondominated. This is the basis to The effectiveness of the proposed MOSP algorithm is
achieve the pareto-optimal curves presented ahead for each case demonstrated by the application of two reference distribu-
study. tion networks—the IEEE 123-node test system and the Roy
Billinton test system—as the testbed. Both test systems are
D. Objective Function commonly found in the literature in switch placement works.
The objective function is the central part of the algorithm. It The results obtained by the application of the MOSP algorithm
assesses the proposed solutions from each particle swarm. Each are compared to some of these works.
solution is a set of switches that minimizes the number of The following subsections present the results achieved by
customers affected (unsupplied) by the outages. each of the aforementioned test systems. All simulations have
The objective function is evaluated by only using the network been developed using Matlab software.
topology and the number of customers per load point without
any reliability data. This procedure reduces the objective func- A. IEEE 123-Node Test System
tion design complexity. Accordingly, the objective function The IEEE 123-node test system (IEEE123) [16] is a well-
written for the MOSP algorithm has two member func- known radial distribution test system used as a reference system
tions: the and the . The evaluates the number in many papers. It is a single feeder distribution network with
of CNS by the outages and the evaluates the number of 85 load points as shown in Fig. 3. Two normally open switches
switches present in the solution . are added in the branches and as highlighted in Fig. 3.
The objective function is given by The entire estimated number of customers is 5410, by taking
into consideration that each customer consumes, on average, a
(2)
kW. Thus, for the load point LP3 at node 3, the three-phase load
where and are the reliability and cost weights, respec- is 60 kW, and supplies 60 customers. The load data used in the
tively. The weights represent the importance or priority that simulation can be found in [16].
each objective has in the current simulation. It is important to The results obtained from the application of the MOSP algo-
note that the weight values are assigned accordingly to (3) and rithm to IEEE123 are presented in Fig. 4. The PSO parameters
(4) and the weights used are shown in Table I. The learning fac-
tors are kept between the two stages 4 and 2 for
(3) improving the diversity over the premature convergence as ex-
(4) plained before in Section IV-A. Their values are adjusted em-
pirically during the algorithm development.
At the first stage, 0 and 1, which means that Fig. 4(a) shows a CNS of 593 customers and a number of
the reliability has priority over the cost. The goal for this stage 84 switches at the end of the 30th iteration (first stage). The
Authorized licensed use limited to: Sharda University. Downloaded on December 20,2024 at 10:00:05 UTC from IEEE Xplore. Restrictions apply.
BEZERRA et al.: MULTIOBJECTIVE OPTIMIZATION ALGORITHM FOR SWITCH PLACEMENT 549

TABLE I
PSO PARAMETERS AND WEIGHTS FOR THE IEEE123 SIMULATION

Fig. 3. IEEE 123-node test system.

Fig. 5. Pareto-optimal front for the IEEE 123-node test system.

1035 customers, who are not supplied, is achieved as shown in

Fig. 4(b) at the 110th and last iteration.
From Fig. 4(a), it can also be noted that the CNS, at the first
stage (from the 1st iteration to the 30th), tends to decrease its
value. At same time, the number of switches [from Fig. 4(b)]
tends to increase. The second stage starts at the 31st iteration
and ends at the 110th one. The trends are inverted because the
cost is now considered . This behavior of the CNS and
the number of switches along the two stages is expected since
they conflict.
The solutions achieved are then classified as dominated
or nondominated to compound the pareto-optimal front as
explained in Section II. Fig. 5 shows the pareto-optimal front
obtained from the 15 nondominated solutions achieved. It also
shows the best dominated solutions obtained for the IEEE123.
Each of the nondominated solutions is a possible solution that
can be selected by a utility in accordance with the trade-off
between cost and reliability issues. It is important to highlight
that the configuration of switches of each solution is stored at
each step, and they are available at the end of the simulation.
The results obtained by the MOSP algorithm applied to
IEEE123 are compared to the one presented in [7], TPSO.
In order to properly compare the results, the same objective
function used in the MOSP algorithm is applied to the set of
switches achieved by the TPSO algorithm. The closest solution
among the nondominated ones, presented in Fig. 5, is selected
for comparison to the TPSO algorithm. The results for compar-
Fig. 4. Simulation results for the IEEE 123-node test system: (a) CNS and ison are shown in Table II and the locations as for the MOSP
(b) number of switches. algorithm are presented in Table III. The solution achieved by
the MOSP algorithm is better than that given by the TPSO
algorithm, chiefly for the CNS.
simulation proceeds to the second stage to search a configura- Furthermore, the MOSP algorithm figures out a set of pos-
tion with fewer switches. A configuration with 12 switches for sible solutions and not only a single solution as given in [7]. For
Authorized licensed use limited to: Sharda University. Downloaded on December 20,2024 at 10:00:05 UTC from IEEE Xplore. Restrictions apply.
550 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 30, NO. 2, APRIL 2015

TABLE II
COMPARISON BETWEEN TPSO AND MOSP ALGORITHMS

TABLE III
SWITCH LOCATIONS

Fig. 7. Simulation results for the RBTS bus 4: (a) CNS. (b) Number of
switches.

TABLE IV
PSO PARAMETERS AND WEIGHTS FOR THE RBTS4 SIMULATION

Fig. 6. RBTS bus 4.

(RBTS4). The MOSP algorithm is applied to the RBTS4 system

the IEEE123, the MOSP algorithm presents solutions that range that has seven feeders and 4779 customers, presented in Fig. 6.
from 11 to 87 switches, as shown in Fig. 5 from the nondomi- The simulation results are presented in Fig. 7. The first stage
nated solutions, although the solution proposed by the TPSO ap- of the MOSP algorithm searches for the most reliable set of
points a single number of 30 switches. A number of options aids switches regardless of the number of switches needed. This is
the decision makers with more alternatives in terms of cost and accomplished by making 0.
reliability. In addition, very often, a large number of switches is The values assigned for the PSO parameters and the weights
not practical for most utilities. are shown in Table IV. Again, the values of , , , and
assume the same values as the previous case study. The is
B. Roy Billinton Test System adjusted empirically.
The RBTS is widely applied as a testbed for reliability studies The same trends for CNS and the number of switches can be
in distribution networks. Parameters as load description, feeder observed in Fig. 7: the CNS decreases at first stage (from the 1st
length, number of customers, failure rates of equipment, and the to the 20th iteration) and increases at the second one (from the
network topology are available in [17], where two distribution 21st to the 80th iteration); and the number of switches increases
networks are presented: the RBTS Bus 2 and the RBTS Bus 4 in the first stage and it is strongly reduced at the second stage.
Authorized licensed use limited to: Sharda University. Downloaded on December 20,2024 at 10:00:05 UTC from IEEE Xplore. Restrictions apply.
BEZERRA et al.: MULTIOBJECTIVE OPTIMIZATION ALGORITHM FOR SWITCH PLACEMENT 551

TABLE VI
SWITCH LOCATIONS COMPARISON

represented: the nondominated (good quality solutions) and the

dominated (useless solutions).
The results obtained from the MOSP algorithm are compared
to the ones presented in [9] that applies the mixed-integer linear
programming (MILP) formulation. The MILP results in a set
of 22 switches as the optimal configuration for the case that
loop switches are considered. For a more trustworthy compar-
ison, the CNS of the set of switches achieved in [9] has been
Fig. 8. Pareto-optimal front for the RBTS bus 4. evaluated by the MOSP objective function, which resulted in
531 customers. The closest solution among the 11 presented in
Table V is solution #7, as shown in Fig. 8. The MOSP algo-
TABLE V
NONDOMINATED SOLUTIONS FOR THE RBTS4 rithm yielded a number of switches that were 18% lower than
the MILP formulation, while the CNS from the MOSP algo-
rithm is 3% greater than that from the MILP formulation. The
switch location for the related solutions given by the MOSP and
MILP approaches is shown in Table VI.
The MILP formulation proposed in [9] has been proven to
reach the global optimum in a finite number of steps. In addi-
tion, it does not suffer the problem of local minimum. Albeit,
it is not possible to pinpoint the optimal solution by the MOSP
algorithm, the proximity of the solutions shows its efficiency in
finding suitable solutions.
The MOSP algorithm also gives information on the number
of not supplied customers of great relevance to the utilities. Fur-
thermore, the MOSP algorithm gives alternate solutions for the
planners as for the number and location of power switches, in-
This occurs because weight cost assumes a nonzero value stead of a unique solution.
0.5) at the second stage and zero for the first one.
Although the MOSP algorithm is capable of reducing the C. Remarks
number of switches to a practical value, it is not possible to en- Two remarks are noteworthy about the results obtained by the
sure that one specific set of switches is the optimal solution, as MOSP algorithm. In the IEEE 123-node test system, the MOSP
explained in Section II. Additional higher level information is algorithm presented a far better solution than the TPSO algo-
necessary to aid the decision making according to the needs of rithm, mainly due to the distinct design strategies for the objec-
the utility, that is, a more reliable solution or a cheaper one. tive functions applied in each proposal. The MOSP objective
Therefore, the pareto-optimal front is useful in choosing one function is designed to achieve the minimum number of CNS
single solution that complies with the expectations of a utility while the TPSO algorithm has the minimization of the ECOST
about reliability and cost among a set of nondominated solu- as the objective function.
tions. The pareto-optimal front for the RBTS4, extracted from The ECOST is given by
the simulation results presented in Fig. 7, is shown in Fig. 8.
Numerical results are shown in Table V.
(7)
On the one hand, solution #1 is the cheapest and the less reli-
able one, but, on the other hand, solution #11 is the most expen-
sive and the most reliable one. All of the others at the pareto-op- In [7], where is the number of isolated points due to con-
timal front are variations between the cheapest solution (#1) and tingency , is the number of contingencies, is the cur-
the most reliable (#11). tailed load at load point due to contingency , is the average
The pareto-optimal front given by the proposed algorithm outage time due to the contingency , is the average failure
eases the decision making, because it straightly shows the rate of contingency , and is the outage cost of load
relationship between reliability (CNS) and cost (number of point due to outage with outage duration .
switches) for the switch placement task, reducing it to a Although using different objective functions, the comparison
trade-off analysis. It can be noted that two sort of solutions are is pertinent. It can noted from (7) that fewer affected customers,
Authorized licensed use limited to: Sharda University. Downloaded on December 20,2024 at 10:00:05 UTC from IEEE Xplore. Restrictions apply.
552 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 30, NO. 2, APRIL 2015

as proposed by the MOSP algorithm, implies less loads affected [8] D. Horvath, L. Brillet, S. Roy, S. Conilleau, A. Tantar, J.-C. Boisson,
and minor values of . Thus, the value of ECOST is reduced N. Melab, and E. Talbi, “Local vs. global search strategies in evo-
lutionary grid-based conformational sampling and docking,” in Proc.
when the CNS is reduced. IEEE Congr. Evol. Comput., 2009, pp. 247–254.
In the RBTS4, the results given by the MOSP algorithm [9] A. Abiri-Jahromi, M. Fotuhi-Firuzabad, M. Parvania, and M. Mosleh,
and the MILP formulation are similar. The proximity from the “Optimized sectionalizing switch placement strategy in distribution
systems,” IEEE Trans. Power Del., vol. 27, no. 1, pp. 362–370, Jan.
pareto-optimal front achieved by the MOSP algorithm validates 2012.
the good quality of the solution and its performance to find [10] D. Kondo, C. Almeida, H. Kagan, A. Cunha, M. Gouvea, L. Felber,
solutions close to the global pareto-optimal front as presented M. Braga, and J. Nascimento, “A methodology for reclosers allocation
in distribution networks,” in Proc. IEEE Power Energy Soc. Conf. In-
in Fig. 8. novat. Smart Grid Technol. Latin Amer., 2013, pp. 1–8.
[11] K. Deb, Multi-Objective Optimization Using Evolutionary Algorithms
(Wiley Paperback). Hoboken, NJ, USA: Wiley, 2009.
VI. CONCLUSION [12] C. C. Coello, G. B. Lamont, and D. A. van Veldhuizen, Evolutionary
Algorithms for Solving Multi-Objective Problems (Genetic and Evolu-
tionary Computation). New York, USA: Springer, 2007.
This paper presented an MOSP algorithm based on PSO. The [13] J. Kennedy and R. Eberhart, “Particle swarm optimization,” in Proc.
approach has presented good performance when evaluating con- IEEE Int. Conf. Neural Netw., Nov/Dec. 1995, vol. 4, pp. 1942–1948.
flicting targets, although it is not possible to state the optimal [14] J. Kennedy and R. Eberhart, “A discrete binary version of the par-
ticle swarm algorithm,” in Proc. IEEE Int. Conf. Syst., Man, Cybern. ,
solution. The proposed algorithm presents a set of solutions in- Comput. Cybern. Simul., Oct. 1997, vol. 5, pp. 4104–4108.
stead of a unique solution. It does not require the failure rates [15] Y. Shi and R. Eberhart, “Empirical study of particle swarm optimiza-
of the evaluated system, and information is very often unavail- tion,” in Proc. Congr. Evol. Comput., 1999, vol. 3, pp. 1936–1950.
[16] W. Kersting, “Radial distribution test feeders,” in Proc. IEEE Power
able. Thus, even new distribution networks can be simulated by Eng. Soc. Winter Meeting, 2001, vol. 2, pp. 908–912.
the MOSP algorithm since only the topology and the estimated [17] R. Allan, R. Billinton, I. Sjarief, L. Goel, and K. So, “A reliability
number of customers per load point are required. test system for educational purposes-basic distribution system data and
results,” IEEE Trans. Power Syst., vol. 6, no. 2, pp. 813–820, May
Representative test systems, such as RBTS bus 4 and IEEE 1991.
123-node test systems, have demonstrated that the MOSP al-
gorithm is flexible and applicable to different distribution net- José Roberto Bezerra (S’12) received the M.Sc. degree in electrical engi-
works. The multiobjective approach has shown that a single so- neering from the Federal University of Ceará, Ceará, Brazil, in 2004.
He is actually devoted to his doctorate studies in power systems automation
lution cannot be considered optimal due to the intrinsic relia- and self-healing. His interest areas also encompass cybersecurity for smart grids
bility/cost trade-off in switch placement problems. Thus, the so- and power system protection.
lutions presented at the pareto-optimal front give options to the
power utility personnel for choosing the most suitable configu-
ration of switches that meet the expectations of reliability and Giovanni Cordeiro Barroso received the M.Sc. degree in electrical engi-
investment of the company. neering from Pontifical Catholic University of Rio de Janeiro, Rio de Janeiro,
Brazil, in 1986 and the Ph.D. degree in electrical engineering from the Federal
University of Paraíba, Paraiba, Brazil, in 1996.
He became a Postdoctoral Fellow with the Institute National des Télécommu-
REFERENCES nications—INT, Evry, France, in 2003. Currently, he is an Associate Professor
at the Federal University of Ceará, Fortaleza, Ceará, Brazil. He has experience
[1] A. Moradi, M. Fotuhi-Firuzabad, and M. Rashidi-Nejad, “A reliability in electrical engineering with an emphasis on supervisory control, acting on
cost/worth approach to determine optimum switching placement in dis- the following topics: petri nets, modeling and analysis, colored petri nets, dis-
tribution systems,” presented at the IEEE Transm. Distrib. Conf. Ex- tributed systems, and discrete event systems.
hibit.: Asia Pacific, Washington, DC, USA, 2005.
[2] L. Yiqing, G. Houlei, W. Xin, T. Zhaohong, W. Peng, L. Naiyong,
and X. Minjiang, “Study on new type of ied with integrated functions
in intelligent substation,” in Proc. Int. Conf. Adv. Power Syst. Autom.
Protect., 2011, vol. 1, pp. 161–165. Ruth Pastôra Saraiva Leão received the B.Sc. degree in electrical engineering
[3] J. R. Bezerra, G. C. Barroso, and R. P. Leao, “Switch placement al- in power systems from the Federal University of Ceará, Fortaleza, Ceará, Brazi.
gorithm for reducing customers outage impacts on radial distribution and the Ph.D. degree in electrical engineering from Loughborough University
networks,” in Proc. TENCON 2012–2012 IEEE Region 10 Conf., Nov. of Technology, Loughborough, U.K., and the postdoctoral degree at Kassel
2012, pp. 1–6. University and Institut fr Solare Energieversorgunstech-nik e.V.-ISET, Kassel,
[4] C.-S. Chen, C.-H. Lin, H.-J. Chuang, C.-S. Li, M.-Y. Huang, and C.-W. Germany.
Huang, “Optimal placement of line switches for distribution automa- She has a degree in business administration from Universidade Estadual do
tion systems using immune algorithm,” IEEE Trans. Power Syst., vol. Ceará, Fortaleza. She is Associate Professor in the Department of Electrical En-
21, no. 3, pp. 1209–1217, Aug. 2006. gineering, Federal University of Ceará. Her areas of interest are electric power
[5] H. Falaghi, M.-R. Haghifam, and C. Singh, “Ant colony optimization- quality, renewable and distributed generation, automation of distribution net-
based method for placement of sectionalizing switches in distribution works, as well as smart metering and smart grids.
networks using a fuzzy multiobjective approach,” IEEE Trans. Power
Del., vol. 24, no. 1, pp. 268–276, Jan. 2009.
[6] H. N. Alves, “A hybrid algorithm for optimal placement of switches
devices in electric distribution systems,” IEEE Trans. Latin Amer., vol. Raimundo Furtado Sampaio received the M.Sc. degree in electrical engi-
10, no. 6, pp. 2218–2223, 2012. neering from the Federal University of Ceará, Fortaleza, Ceará, Brazil, in 2002.
[7] A. Moradi and M. Fotuhi-Firuzabad, “Optimal switch placement in dis- Currently, he is Assistant Professor at Federal University of Ceará. His
tribution systems using trinary particle swarm optimization algorithm,” research interest includes multiagent and power system automation and
IEEE Trans. Power Del., vol. 23, no. 1, pp. 271–279, Jan. 2008. protection.

Authorized licensed use limited to: Sharda University. Downloaded on December 20,2024 at 10:00:05 UTC from IEEE Xplore. Restrictions apply.