You are on page 1of 8

2978 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 37, NO.

4, AUGUST 2022

A Multiobjective and Multicriteria Approach for


Optimal Placement of Protective Devices and
Switches in Distribution Networks
Miguel Monteiro Costa , Michel Bessani , and Lucas S. Batista

Abstract—The optimal allocation of protective devices and CCx Capital investment cost of device x,
switches in distribution systems concerns the analysis of conflict- x ∈ {f b (fuse-blow), f s (fuse-save), r (recloser),
ing attributes, such as capital investment, SAIDI, ENS, ECOST, s (sectionalizing switch)}
MAIFI, among others. Regarding this multiattribute context,
ICx Installation cost of device x, x ∈ {f b, f s, r, s}
this paper proposes a multicriteria strategy to support decision-
making. From this strategy, an assertive decision can be made M Cx Maintenance cost of device x, x ∈ {f b, f s, r, s}
under different scenarios when defining an appropriate solution T Life time of devices
to a real-world problem. A multiobjective optimization approach Sf Set of possible places for allocation of fuses
is applied to approximate an adequate set of potential solutions for Ssr Set of possible places for allocation of switches or
allocating protective devices and switches. The capital investment reclosers
and the SAIDI index are assumed as objective functions of the Ss Set of locations where only switches can be allo-
problem, and the well-known NSGA-II method is employed as the
cated
optimizer. Experiments are performed using the RBTS - Bus 4; the
obtained trade-off solutions are evaluated in the previously stated Variables
attributes, and assisted decisions are made to illustrate the effect vk , wk , xk Binary variables for the allocation of a fuse-blow
of considering different scenarios of preferences from the decision fuse, fuse-save fuse, and recloser in the k-th place,
unit. The results highlight both the advantages of a multiobjective respectively
approach for allocating protective devices and the usefulness of y k , zk Binary variables for the allocation of a switch in the
a multicriteria strategy to aid assertive decision-making under k-th place
different scenarios.
Index Terms—Power distribution system reliability, protection
devices, decision-making, multiobjective optimization.
I. INTRODUCTION
INIMIZING interruptions, as well as their duration, is
Parameters
λi
NOMENCLATURE

i-th branch permanent failure rate


M a recurring problem for distribution utilities. In addition
to nuisance and damage to customers, supply interruptions can
lead to penalties imposed on distribution utilities by regulatory
γi i-th branch temporary failure rate
agencies. It is estimated that failures in primary distribution
rij Outage duration of j-th load point due to i-th con-
systems are related to approximately 70% of power supply inter-
tingency
ruptions [1]. Therefore, protection devices like circuit breakers
N Iij Number of momentary interruptions of the j-th load
(CBs), reclosers, and fuses, as well as sectionalizing switches
point due to the i-th contingency
(SSs), have an essential role in distribution systems reliability.
Pj Number of customers at the j-th load point
In this sense, it is fundamental to determine the optimal allo-
Lj Load of the j-th load point
cation of these devices to simultaneously maximize reliability
CDFijm Customers’ damage function of the m-th customer
benefits and minimize capital investment [2]. As presented next,
type concerning outage duration rij
however, most of the works in this research area deal with
only a single criterion, aiming at a devices allocation that either
Manuscript received April 12, 2021; revised August 31, 2021; accepted improves the network reliability or reduces the costs.
October 6, 2021. Date of publication October 19, 2021; date of current version
July 25, 2022. This work was supported in part by Brazilian agencies CNPq, The former studies that proposed optimization methods for the
FAPEMIG, and CAPES. Paper no. TPWRD-00580-2021.R1. (Corresponding allocation of such devices, aiming at improving the reliability of
author: Miguel Monteiro Costa.) distribution networks, date back to the 90 s. For instance, in [3]
Miguel Monteiro Costa is with the Graduate Program in Electrical Engineer- it was formulated a combinatorial optimization problem for SSs
ing, Universidade Federal de Minas Gerais, Belo Horizonte, MG 31270-901,
Brazil (e-mail: miguelmonteirorp@yahoo.com.br). allocation regarding costs related to SSs and Expected Customer
Michel Bessani and Lucas S. Batista are with the Department of Electri- Interruption Cost (ECOST). In [4], binary programming together
cal Engineering, Universidade Federal de Minas Gerais, Belo Horizonte, MG with a heuristic method were applied for a predefined quantity of
31270-901, Brazil (e-mail: mbessani@ufmg.br; lusoba@ufmg.br).
Color versions of one or more figures in this article are available at
reclosers and SSs allocation aiming to minimize the System Av-
https://doi.org/10.1109/TPWRD.2021.3120968. erage Interruption Frequency Index (SAIFI). An exact method
Digital Object Identifier 10.1109/TPWRD.2021.3120968 was presented in [5] for allocation of automatic SSs in radial and
0885-8977 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on July 27,2022 at 11:31:59 UTC from IEEE Xplore. Restrictions apply.
COSTA et al.: MULTIOBJECTIVE AND MULTICRITERIA APPROACH FOR OPTIMAL PLACEMENT 2979

meshed networks, considering the costs of these devices and of TABLE I


MAIN FEATURES OF THE PROPOSED APPROACH
the Energy Not Supplied (ENS).
Recent studies on the broader problem of simultaneous allo-
cation of SSs and protective devices used Mixed Integer Linear
Programming (MILP), Mixed Integer Non-linear Programming
(MINLP), and (meta)heuristic algorithms. Concerning MILP
approaches, in [6] it was analyzed the insertion and replacement
of CBs, reclosers, switches, and fault passage indicators. Its
objective function encompasses the costs related to new devices
and relocation of existing ones, outages, and crew. In [7], the
costs related to devices and permanent and temporary faults were
minimized. Besides CBs, reclosers, switches, and sectionalizers, must be adaptable to different needs imposed by the utility’s
two coordination schemes between fuses and reclosers were ana- priorities.
lyzed: fuse-blow (FB) and fuse-save (FS); this last reduces fuses Decision-making was adopted only in [21], which minimizes
blowing due to temporary faults by the acting of an upstream cost, SAIFI and SAIDI, and in [22], which in addition to cost,
recloser. SAIDI and SAIFI, includes distributed generation unavailability
Among the MINLP approaches, the allocation of reclosers and as a fourth objective function. They both choose a final solu-
fuses was investigated in [8] with an objective function derived tion by a min-max approach [23], using the defined objective
from the SAIFI index. Loads’ uncertainties were considered functions, without additional criteria. Besides, the min-max
in [9] for allocation of switches and reclosers regarding the approach was designed assuming the same importance for all
costs related to devices and interruptions. CBs, switches, and the attributes, which is not appropriate to a decision-making
fuses were handled in [10], [11] concerning the minimization process on different decision scenarios.
of costs of devices, interruptions, and ECOST; the presence In this sense, this paper’s main contribution is the proposal
of pre-existing SSs was analyzed by the same authors in [11]. of a multicriteria decision-making approach for the protection
In the group of (meta)heuristics, Particle Swarm Optimization devices allocation problem. This approach can provide a suitable
was employed in [12] for reclosers and automatic switches final solution that is chosen over a set of potential solutions
allocation, considering the costs associated with the ENS and considering the preferences of the decision unit (DU), thus being
devices. A greedy algorithm was designed in [13] for automatic adaptable to multiple decision scenarios. The optimization prob-
switches allocation, aiming to minimize the number of switches. lem was formulated considering the allocation of CBs, reclosers,
Besides, an exchange market algorithm was applied in [14] switches, and two types of (schemes to) fuse: FB and FS, being
for the placement of protection relays in microgrids, aiming to this last treated only in recent studies [7], [19], [24]. The potential
minimize SAIFI. solutions are obtained through the simultaneous minimization
The minimization of the ENS was investigated in [15] for opti- of capital investment and SAIDI. It is worth noting that, in
mal placement of protection devices and distributed generation addition to these two objective functions, the following criteria
in smart grids; the authors took into account the uncertainties are employed in the decision process: ENS, cost of energy not
of renewable power generation together with the possibility of supplied, and momentary interruptions index. As already known,
protective devices’ unavailability. Failures of protective devices the analysis of this set of attributes can be crucial for a suitable
were also considered in [16] for allocation of SSs and reclosers, decision-making in this research field [25].
aiming to optimize a function that considers the costs of devices, The main aspects covered by the proposed approach are listed
interruptions, and ENS. below and faced against the literature in Table I.
Despite these previous efforts, the applicability of such ap- F1) Permanent and temporary faults impact consideration;
proaches in a real-world context is limited, as not one but F2) Incorporation of both FB and FS schemes;
several criteria of interest would be relevant to assist the devices’ F3) Simultaneous optimization of financial and technical
allocation by a distribution utility. In this sense, although in aspects, allowing trade-off analysis between themselves;
smaller number, multiobjective approaches are addressed in F4) Proposition of a suitable decision-making approach;
the literature aiming to present design solutions with a trade- F5) Definition of additional performance criteria to evaluate
off between some relevant quality indicators for supporting the potential solutions;
decision-making. For instance, both the number of customers F6) Incorporation of the DU’s preferences into the decision-
not supplied and installed SSs are optimized in [17]. Allocation making process.
of SSs and reclosers is handled in [18] regarding the optimization In the remainder of this paper, a conceptual background is pre-
of System Average Interruption Duration Index (SAIDI), SAIFI, sented in Section II; Section III details the proposed approach,
and the costs of these devices. SAIDI, ENS, and costs of devices including the problem modeling, the employed methods for
are analyzed in [19] for allocation of control and protective optimization and decision-making, and the analyzed distribution
devices. system; Section IV is devoted to results and discussion; finally,
However, given a hopeful set of trade-off solutions, it is the study is concluded in Section V.
needed a decision aid approach to support the selection of a solu-
tion that best suits the utility preferences. In general, distribution II. BACKGROUND
companies face specific challenges related to their particular
A. Protection and Reliability of Distribution Systems
scenarios [20]. Some of these particularities concern the utility
sector (public, private, mixed), government regulations, among Distribution networks are subjected to temporary and per-
others. In this manner, a decision aid approach for this problem manent faults, which can cause either momentary or sustained

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on July 27,2022 at 11:31:59 UTC from IEEE Xplore. Restrictions apply.
2980 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 37, NO. 4, AUGUST 2022

interruptions. These momentary interruptions can be handled


by devices with an automatic reclosing capability, as CBs and
reclosers. Normally, CBs are allocated at each feeder beginning
and can clear momentary faults or open for permanent ones.
Fuses can assume FS and FB coordination schemes [26].
In the FS scheme, either an upstream recloser or CB with a
reclosing capability operates before the fuse trips to isolate Fig. 1. Flowchart of the proposed methodology.
a downstream fault, i.e., the fuse acts only after automatic
reclosing of the recloser (permanent faults). In the FB scheme,
the fuse acts to both permanent and temporary faults, resulting A. Problem Modeling
in sustained interruptions. The following binary decision variables are used:
Another device considered is the SS, which can isolate the 
1, if there is a FB fuse on k-th place
affected branch. SS also allows restoration of unaffected loads, vk = (2)
either by downstream restoration through alternative feeders or 0, otherwise
distributed resources, or by upstream one, by opening the switch 
and reclosing the protective device if there is a SS between a 1, if there is a FS fuse on k-th place
wk = (3)
circuit breaker or recloser and the affected branch. This last one 0, otherwise
was considered in this work.
where k ∈ Sf , Sf is the set of possible places for allocation of
The usual manner to assess network reliability is by reliability
fuses, and N sf is the cardinality of Sf ;
indexes, which consider the number of interruptions with their 
respective duration and the number of affected loads or cus- 1, if there is a recloser on k-th place
xk = (4)
tomers. SAIDI, SAIFI, ENS, and MAIFI (Momentary Average 0, otherwise
Interruption Frequency Index) indexes (described in [27]) were 
employed in this paper, either as objective functions or decision- 1, if there is a switch on k-th place
making attributes. The ECOST [11] was also considered, which yk = (5)
0, otherwise
measures the interruption cost for different customer types over
a predefined time interval. where k ∈ Ssr, Ssr is the set of possible places for allocation
of switches or reclosers, and N ssr is the cardinality of Ssr;

B. Multiobjective Optimization Problem (MOP) 1, if there is a switch on k-th place
zk = (6)
In a general way, a multiobjective optimization problem 0, otherwise
(MOP) can be defined as a minimization problem [28]: where k ∈ Ss, Ss is the set of locations where only switches
min f (a) = [f1 (a), . . ., fnf (a)]T can be allocated, and N ss is the cardinality of Ss.
a
To validate the proposed approach, a two-objective opti-
subjectto : a ∈ Ω (1) mization problem is stated, min {SAIDI, Total Cost (TC)}, in
T
in which a = (a1 , . . ., anv ) is a candidate solution, f (·) : which SAIDI is a well-known index and TC considers the costs
Rnv → Rnf represents nf conflicting objective functions and associated with the allocated devices and also the ECOST index,
Ω ⊂ Rnv is the feasible search space such that Ω = {a ∈ as follows:
T
Rnv | gi (a) ≤ 0, hj (a) = 0; ∀ i = 1, . . ., ng , j = 1, . . ., nh },
in which f (Ω) defines the reachable objective set and h(·) : min T C = F bC + F sC + RC + SC + ECOSTt (7)
t=1
Rnv → Rnh and g(·) : Rnv → Rng represent the nh equality
and ng inequality constraints, respectively. N
 sf T N
  sf
A viable solution a ∈ Ω dominates another one a ∈ Ω (indi- F bC = (CCf b + ICf b)vk + M Cf bt vk (8)
cated as f (a) ≺ f (a )) if, and only if, fk (a) ≤ fk (a ) ∀ k ∈ k=1 t=1 k=1
{1, . . ., nf } and ∃ k such that fk (a) < fk (a ). Therefore, a N sf T N sf
solution a∗ ∈ Ω is Pareto-optimal for (1) if there is no solution   
F sC = (CCf s + ICf s)wk + M Cf st wk (9)
a ∈ Ω such that f (a) ≺ f (a∗ ). The set of all efficient solutions
k=1 t=1 k=1
is called Pareto-optimal set and the image of this set is called
Pareto-optimal front. N
 ssr T N
  ssr
RC = (CCr + ICr)xk + M Crt xk (10)
k=1 t=1 k=1
III. METHODOLOGY
N
 ssr N
 ss
A flowchart illustrating the methodology is presented in Fig 1.
SC = (CCs + ICs)yk + (CCs + ICs)zk
As shown in this figure, the first step concerns problem mod-
k=1 k=1
eling. Afterward, non-dominated solutions are approximated
by NSGA-II. At last, the MCDM strategy is applied to pro- T N
  ssr T N
  ss
vide the most appropriate solution according to the decision + M Cst yk + M Cst zk (11)
unit preferences. The objectives and constraints that should be t=1 k=1 t=1 k=1
addressed with their practical aspects must be handled in the where CCf b, CCf s, CCr and CCs are the capital investment
problem modeling stage. This first stage is essential to the correct costs, ICf b, ICf s, ICr and ICs are the installation costs,
execution of the proposed methodology. and M Cf b, M Cf s, M Cr and M Cs are, respectively, the

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on July 27,2022 at 11:31:59 UTC from IEEE Xplore. Restrictions apply.
COSTA et al.: MULTIOBJECTIVE AND MULTICRITERIA APPROACH FOR OPTIMAL PLACEMENT 2981

maintenance costs for FB fuse, FS fuse, recloser and SS, and


T is the lifetime of the devices.
The problem constraints concern the possible locations of the
different types of devices, as stated in (2)–(6) [11]: i) protective
devices can be placed on the beginning of branches, ii) fuses are
not allowed on the main feeder, only on laterals, iii) switches
can be placed on either in the beginning or end of branches, and
iv) at most one device can be allocated in each candidate place
as imposed by (12):

vk + wk ≤ 1, ∀ k ∈ Sf and xk + yk ≤ 1 ∀ k ∈ Ssr. (12)

B. Employed Methods
1) Non-Dominated Sorting Genetic Algorithm II (NSGA-II):
NSGA-II [29] was applied as the solution algorithm for the
proposed nonlinear two-objective optimization problem. In this
method, the candidate solutions are ranked concerning two
mechanisms: the fast non-dominated sorting (for classification
in non-dominated fronts) and the crowding distance assignment
(for solutions density estimation).
In order to evolve the alternatives of the MOP, binary tour-
nament selection without replacement and variation operators
such as (i) uniform and two-point crossovers, with the same
probability, and (ii) bit-flip and swap mutations, also with the
same probability were used [30].
2) Decision-Making Approach: Since a set of solutions is
achieved for a MOP, a final action (alternative) should be chosen
concerning some design criteria of interest. To help in this pro-
cess, a Multiple-Criteria Decision-Making (MCDM) strategy
Fig. 2. RBTS - Bus 4. All the possible places to allocate devices are presented.
can be used to aid the selection of the most suitable solution Circuit breakers (in blue) are considered to exist on the initial network state.
based on some judgments, experiences, and expert opinions of
the decision-maker [31].
In this paper, a two-fold simple approach is proposed. First, The TOPSIS method [32] performs a ranking of the alterna-
the widely-known Analytic Hierarchy Process (AHP) [31] is tives from the definition of two utopian solutions: Positive Ideal
used to estimate the priority or weight wi , i = 1, . . . , nc , for Solution (PIS) and Negative Ideal Solution (NIS). The PIS is a
the nc performance criteria. At last, these weight values are vector stated with the best values achieved by the alternatives,
processed by the Technique for Order of Preference by Similarity and the NIS aims at the opposite, having as reference the worst
to Ideal Solution (TOPSIS) [32] to attain a final action that best values achieved. This method is based on the idea that the most
suits the decision-maker’s preferences. preferred alternative should be the most distant to NIS and the
The proposed MCDM strategy follows the next steps: closest one from PIS in some geometrical sense. This method
r Step 1: Definition of a set of potential solutions for the
considers that the criteria have a tendency of monotonically
problem. This set is preferably not so big, having no more decreasing or increasing utility. In this way, it is simple to define
than twenty solutions, chosen by the decision unit. the PIS and NIS solutions. Additional details of the working
r Step 2: Definition of criteria of interest. Beyond the objec-
principle of AHP and TOPSIS are presented afterward.
tive functions, complementary criteria are used to evaluate
the candidate solutions regarding additional aspects not
considered in the optimization phase. C. Test System
r Step 3: Use of the AHP method to define the criteria The proposed strategy was applied in the radial grid composed
weights that reflect the designers’ preferences. of the feeders connected to bus 4 of the Roy Billinton Test System
r Step 4: Use of the TOPSIS method, which receives the pre- (RBTS - Bus 4) [34]. This distribution network, as considered
selected alternatives, the weights of criteria, and indicates in [10], has 7 feeders with 7 CBs and 38 load points. Fig. 2
a final solution. illustrates this test system and emphasizes the possible places
The AHP method consists of a measurement (or classification) for device allocation. Although simple, this network presents
theory based on pairwise comparisons between the alternatives. a reasonable number of possibilities for allocating devices and
This method is based on judgments of experts in the problem thus will be explored in different decision-making scenarios.
area, which are defined from the AHP Fundamental Scale of The initial network state presents SAIDI0 = 4, 36 hr/yr,
Comparisons [33]. Besides, the AHP method allows evaluating SAIFI0 = 0, 76 int/yr and ENS0 = 80, 56 MWh. Its reliability
the consistency of the judgments by using a Consistency Index data, such as number and type of customers, average and peak
(CI) [33]. load at each load point, branch and equipment failure rates,

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on July 27,2022 at 11:31:59 UTC from IEEE Xplore. Restrictions apply.
2982 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 37, NO. 4, AUGUST 2022

Fig. 3. Obtained non-dominated fronts and hypervolumes (HV). Fig. 4. Estimated Pareto-front and analyzed solutions.

among others, were obtained from [34]. The temporary failure


rate was considered four times the permanent failure rate [7],
[24]. The customer damage functions for the different customer
types were extracted from [35]. The capital investment and
acquisition costs of a recloser, remote-controlled switch, and
fuse (both FB and FS) are assumed as $6000, $4700, and $500,
respectively [24]. It was considered a lifetime of 15 years for the
protective devices and switches, and the annual maintenance
cost was assumed as 2% of the acquisition and installation
costs [10]. The switching time was defined as 10 min [11], and
the repair time of affected branches and fuses as 5 hrs and 1,1 hrs,
respectively [7].
In this study, following [36], it was assumed that momentary
interruptions have a duration of five minutes and outages longer Fig. 5. Quantities and types of equipment allocated to each solution with their
than that value are sustained interruptions. It was also assumed respective SAIDI index.
that the CBs are already placed at the beginning of each feeder
and have the capability of automatic reclosing for clearing TABLE II
temporary faults, as used in [7], [9], [12]. EXTREME SOLUTIONS OF THE ESTIMATED PARETO-FRONT

IV. RESULTS AND DISCUSSION


For the NSGA-II, the following parameters were used: pop-
ulation size of 100 solutions, crossover probability of 80%,
mutation probability of 10%, and 5 × 104 objective function
evaluations. To observe the convergence capability of the op- reclosers and switches, are placed in the grid. The greater the
timization method, the NSGA-II was executed five times, as TC value, the lower the SAIDI index is.
presented in Fig. 3, and the similarity among the obtained fronts Different scenarios are evaluated in the following subsections
is evident. A metric known as Hypervolume (HV), which calcu- to illustrate how the most appropriate solution returned by the
lates the region enclosed by an estimated front and a reference proposed approach is affected by the DU’s preferences. The
point dominated by all solutions in this front, can also be used first case study analyzes the extreme solutions of the estimated
for such an evaluation. The reference point is stated, in this Pareto-front. Afterward, the proposed MCDM approach is em-
paper, with the worst values achieved regarding all the estimated ployed for three other case studies to evaluate some different
fronts. This metric is detailed in [37]. The closeness between the contexts of preferences.
obtained non-dominated fronts and their respective HV values
show the capability of the method to reproduce similar results
A. Case Study 1
from different runs.
A final Pareto-front was estimated using the set of non- This case study analyzes the extreme solutions on the esti-
dominated solutions obtained from the five executions, and this mated Pareto-front (Fig. 4), i.e., the best solution according to
front is shown in Fig. 4. This front provides several potential TC (solution 1) and the best one concerning SAIDI (solution
solutions for the proposed problem, allowing a trade-off analysis 161). The ENS, ECOST, and MAIFI indexes for these solutions
between investment and reliability, which is often not possible were also evaluated, as shown in Table II.
in the case of single-objective approaches. It is important to highlight these two solutions because they
This estimated Pareto-front is composed of 161 solutions. are the obtained ones if it is performed a single-objective
Fig. 5 presents the quantities and types of allocated devices approach, considering as objective function either the TC or
besides the SAIDI index of each solution sorted by increasing the SAIDI index. It is notorious that the improved value of a
values of TC. As shown in this figure, with the increase of TC, single-objective function came at the cost of the quality of the
either more protection equipment or more expensive devices, as other attributes. This fact shows the advantage of multiobjective

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on July 27,2022 at 11:31:59 UTC from IEEE Xplore. Restrictions apply.
COSTA et al.: MULTIOBJECTIVE AND MULTICRITERIA APPROACH FOR OPTIMAL PLACEMENT 2983

TABLE III TABLE IV


NON-DOMINATED SOLUTIONS OF INTEREST JUDGMENT BETWEEN CRITERIA IN THE AHP METHOD

indicate the criteria weights. To do this, the relative importance


of each criterion is estimated from a peer comparison among the
criteria.
The comparison matrix among the criteria is shown in Ta-
ble IV. According to the experience of this DU and the AHP
Fundamental Scale of Comparisons, it was considered a pref-
erence between similar and moderate of TC to SAIDI and a
strong preference of TC to the other criteria. Besides, it was
assumed a preference between moderate and strong of SAIDI
to ENS, ECOST, and MAIFI, and a similar preference be-
tween these last ones. For these judgments, the AHP indicated
a Consistency Index of CI = 0, 006 and a weight vector of
optimization, allowing the analysis of a set of solutions with w = [0, 454 0, 300 0, 082 0, 082 0, 082] for TC, SAIDI, ENS,
different levels of investment and reliability. ECOST, and MAIFI, respectively.
In the fourth step of the proposed decision approach, the
B. Selecting a Reduced Set of Potential Solutions criteria weights are used by the TOPSIS method, which analyses
the selected solutions and then indicates solution 125 as the final
The estimated front is composed of 161 design alternatives. one (see Fig. 4 and Table III).
Since a large set of possibilities does not necessarily contribute to Due to the higher weight defined for TC, the selected solution
the decision-making process, this set may be reduced according is not very expensive. If it is compared against the immediately
to the first step of the proposed MCDM. In this way, a decision cheaper solution (i.e., solution 117 of Table III), it is better in
unit may discard solutions that do not improve the SAIDI, SAIFI, all of the other criteria. Besides, if it is compared against the
and ENS indexes in at least 20% with respect to the initial immediately more expensive one (solution 129 of Table III),
network state. solution 125 is better in three of the five used criteria. These
Additionally, a financial decision may be applied to discard remarks indeed highlight the ability of the proposed strategy
the most expensive solutions, that is, solutions that have, for to reflect the preferences of the DU and to capture some better
instance, TC values greater than 80% of the most expensive features of the alternatives.
obtained non-dominated solution, that has T C = $104850, 00.
This process results in 90 non-dominated solutions, which are
the points between the red lines in Fig. 4. Furthermore, since D. Case Study 3
making a decision on a set of 90 solutions can be a challenge In Case Study 3, suppose that the DU intends to give greater
for the designer and even for any decision aid approach, a set of importance to the SAIDI criterion. In the proposed approach,
20 solutions of interest is selected considering uniformly spaced this preference can be stated by modifying the judgments
TC values. in the AHP comparison matrix, for example, by attributing
Based on the second step of the proposed MCDM strategy, a preference between similar and moderate of SAIDI to TC
the selected solutions should be evaluated in complementary and a strong preference of SAIDI to the other criteria. Also,
criteria. As previously mentioned, the additional considered it is assumed a preference between moderate and strong of
criteria are the ENS, ECOST, and MAIFI indexes. Table III TC to ENS, ECOST, and MAIFI, and a similar preference
shows the 20 selected solutions together with the two objective between these last ones. This process results in the weight
values and the three additional criteria. vector w = [0, 300 0, 454 0, 082 0, 082 0, 082], which reflects
The proposed approach allows assisted decisions that reflect a higher weight to SAIDI.
different DU’s preferences. Then, for illustrating the effect of The TOPSIS method pointed out to solution 138 of Table III,
steps 3 and 4 of the proposed approach, three scenarios are also shown in Fig. 4. The solution 138 is better than solutions
performed next considering the set of solutions of Table III. 137 and 140 in four of the five considered criteria, highlight-
ing that the proposed approach can identify the solution that
C. Case Study 2 better meets the DU preferences among the set of potential
solutions.
In Case Study 2, according to the DU’s preferences, higher
importance is attributed to the TC criterion against the others.
In real situations, this can be understood as the improvement of E. Case Study 4
the network reliability spending as little as possible. Concerning In Case Study 4, the MAIFI index was considered the most
the third step of the decision approach, the AHP was applied to important criterion. This context may represent a scenario in

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on July 27,2022 at 11:31:59 UTC from IEEE Xplore. Restrictions apply.
2984 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 37, NO. 4, AUGUST 2022

TABLE V not have a switch in branch 63 and has two more fuses. Thus, it
DEVICES ALLOCATION OF ANALYZED SOLUTIONS
is cheaper than solution 140, and it is better in all criteria, except
in the SAIDI index.
The solution 86, the chosen one when MAIFI has the higher
importance (Case Study 4), allocates fuses for the branches
serving the higher amount of customers. Only one recloser was
allocated at the beginning of feeder 7, which is the second most
loaded feeder and the second one with the greatest number of
customers. Moreover, this solution allocates 22 FB fuses and
only 5 FS fuses, which leads to less temporary interruptions,
resulting in a better value of MAIFI in comparison to the
previously chosen solutions.
At last, it is worth noting the strategic places identified by the
optimizer for the allocation of reclosers. In general, they are allo-
cated at the beginning of the feeders, in the subsequent branches
after CBs, thus protecting a greater number of customers and
allowing a greater number of FS fuses.
For comparison purposes, the min-max approach (as used
in [21], [22]) was applied over the obtained set of potential
solutions. This approach indicates the solution numbered by 104,
as presented in Fig. 4 and Table III. Regardless of the particular
needs of companies, note that this methodology always returns
the same solution, as the same importance is assigned to each
one of the attributes. On the other hand, the proposed MCDM
strategy deals with multiple criteria in an adaptable manner,
*
Allocation of switches in the beginning (B) or end (E) of the branch.
allowing the different practical scenarios reflected by the DU’s
preferences to guide the selection of the most appropriate final
solution. In this way, concerning the case studies analyzed,
which it is strongly recommended to reduce temporary faults’ it is possible to see that alternative actions can be preferable
impact. For that, a high importance level can be assigned to to solution 104. For example, contrasting this solution against
the MAIFI index in the AHP comparison matrix. This can be solution 125 (Case Study 2, which assigns greater importance
done by specifying for MAIFI a preference between similar and to TC), the solution 125 is more expensive but is better in
moderate to TC, a moderate preference to SAIDI, and a strong all other criteria. When solution 104 is compared against so-
preference to the other criteria; the remaining judgments can lution 138 (Case Study 3, which assigns greater importance
be maintained as in Table IV. This procedure results in the to SAIDI), the solution 138 is better in three of five criteria,
weight vector w = [0, 283 0, 183 0, 059 0, 059 0, 415] for TC, including SAIDI.
SAIDI, ENS, ECOST, and MAIFI, respectively, which defines
a greater weight to the MAIFI index. In this case, the TOPSIS V. CONCLUSION
method indicates the solution 86, shown in Fig. 4 and Table III,
which indeed presents the smallest value of MAIFI among all This paper presented a multicriteria decision-making ap-
the analyzed solutions. proach to the optimal placement of switches and protective
devices in distribution networks. First, a multiobjective opti-
mization problem was addressed by NSGA-II, which resulted
F. Discussion in a set of non-dominated solutions that includes a valuable range
Table V shows the branches where each device should be of trade-off alternatives with different investment and reliability
placed according to solutions from each case study. Solutions levels. These solutions were used by the proposed MCDM
117 and 140 are also shown for comparison purposes. strategy to provide the most appropriate solution according to
For Case Study 1, the solution that minimizes the Total Cost the DU’s preferences.
(solution 1) does not allocate devices in the grid; in contrast, In the analyzed scenarios, each one with different DU’s
the solution that minimizes the SAIDI index (solution 161), preferences, the proposed MCDM strategy was able to indicate
i.e., the most expensive one, allocates many devices, including appropriate solutions that reflected these preferences in the
expensive ones as reclosers and switches. In this solution, the decision-making process. This capability was highlighted by
reclosers and switches are placed on the feeders with more comparing the proposed approach against the min-max tech-
customers, allocating two reclosers in each of the feeders 1, nique applied in previous studies. The results indicate that the
3, and 4 (see Fig. 2 and [34]). presented approach can handle different DU’s preferences that
In Case Study 2, it is important to highlight the similarity reflect practical needs that are scenario-dependent.
between solutions 117 and 125; the difference concerns the It is also important to highlight additional relevant system
allocation of five more fuses on solution 125, being 7, 7% more protection issues that have not been addressed in this paper, such
expensive, but enhancing all the reliability indexes. The solution as fault detection and isolation, service restoration, types of loads
obtained in Case Study 3 (solution 138) is similar to 140, but does and customers, government regulations, and the many different

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on July 27,2022 at 11:31:59 UTC from IEEE Xplore. Restrictions apply.
COSTA et al.: MULTIOBJECTIVE AND MULTICRITERIA APPROACH FOR OPTIMAL PLACEMENT 2985

scenarios faced by distribution systems operators. The incorpo- [16] M. Safari, M.-R. Haghifam, and M. Zangiabadi, “A hybrid method for
ration of new practical aspects as the aforementioned ones will recloser and sectionalizer placement in distribution networks consider-
ing protection coordination, fault type and equipment malfunction,” IET
require adequate procedures to quantify how they are affected Gener. Transmiss. Distrib., vol. 15, no. 15, pp. 2176–2190, 2021.
by protection allocation. Moreover, this incorporation must be [17] J. R. Bezerra, G. C. Barroso, R. P. S. Leão, and R. F. Sampaio, “Multiob-
accompanied by updates in the problem modeling, as new ob- jective optimization algorithm for switch placement in radial power dis-
jective functions, constraints, or criteria for the decision-making tribution networks,” IEEE Trans. Power Del., vol. 30, no. 2, pp. 545–552,
Apr. 2015.
methodology. This flexibility emphasizes the applicability of the [18] A. V. Pombo, J. Murta-Pina, and V. F. Pires, “Multiobjective planning of
proposed strategy for both the optimal allocation of traditional distribution networks incorporating switches and protective devices using
types of devices and emerging ones. a memetic optimization,” Rel. Eng. Syst. Saf., vol. 136, pp. 101–108, 2015.
Future works can evaluate both the impact of distributed [19] W. R. Faria, D. de B. Martins, C. A. Nametala, and B. R. Pereira,
“Protection system planning for distribution networks: A probabilistic
resources and island operation in the optimal allocation of approach,” Electric Power Syst. Res., vol. 189, 2020, Art. no. 106612.
devices and network reliability. Additionally, it can be relevant [20] S. Kufeoglu, M. Pollitt, and K. Anaya, “Electric power distribution in the
to treat the optimal allocation of devices and the load restoration world: Today and tomorrow,” 2018. [Online]. Available: https://doi.org/
10.17863/CAM.27667
problem together within a single approach, aiming to increment [21] W. Tippachon and D. Rerkpreedapong, “Multiobjective optimal placement
the service quality and reliability. Future studies also can en- of switches and protective devices in electric power distribution systems
compass protection coordination, fault indicator devices, electric using ant colony optimization,” Electric Power Syst. Res., vol. 79, no. 7,
vehicles, among others. pp. 1171–1178, 2009.
[22] A. V. Pombo, J. Murta-Pina, and V. F. Pires, “A multiobjective placement
of switching devices in distribution networks incorporating distributed
energy resources,” Electric Power Syst. Res., vol. 130, pp. 34–45, 2016.
REFERENCES [23] I. J. Ramirez-Rosado and J. A. Dominguez-Navarro, “Possibilistic model
based on fuzzy sets for the multiobjective optimal planning of electric
[1] L. S. de Assis, J. F. V. González, F. L. Usberti, C. Lyra, C. Cavellucci,
power distribution networks,” IEEE Trans. Power Syst., vol. 19, no. 4,
and F. J. Von Zuben, “Switch allocation problems in power distribution
pp. 1801–1810, Nov. 2004.
systems,” IEEE Trans. Power Syst., vol. 30, no. 1, pp. 246–253, Jan. 2015.
[24] M. Izadi, A. Safdarian, M. Moeini-Aghtaie, and M. Lehtonen, “Optimal
[2] J.-M. Sohn, S.-R. Nam, and J.-K. Park, “Value-based radial distribution
placement of protective and controlling devices in electric power distribu-
system reliability optimization,” IEEE Trans. Power Syst., vol. 21, no. 2,
tion systems: A MIP model,” IEEE Access, vol. 7, pp. 122 827–122 837,
pp. 941–947, May 2006.
2019.
[3] R. Billinton and S. Jonnavithula, “Optimal switching device placement
[25] M. F. McGranaghan, “Quantifying reliability and service quality for dis-
in radial distribution systems,” IEEE Trans. Power Del., vol. 11, no. 3,
tribution systems,” IEEE Trans. Ind. Appl., vol. 43, no. 1, pp. 188–195,
pp. 1646–1651, Jul. 1996.
Jan. 2007.
[4] F. Soudi and K. Tomsovic, “Optimized distribution protection using binary
[26] A. Alam, M. N. Alam, V. Pant, and B. Das, “Placement of protective
programming,” IEEE Trans. Power Del., vol. 13, no. 1, pp. 218–224,
devices in distribution system considering uncertainties in loads, tempo-
Jan. 1998.
rary and permanent failure rates and repair rates,” IET Gener. Transmiss.
[5] G. Celli and F. Pilo, “Optimal sectionalizing switches allocation in distri-
Distrib., vol. 12, no. 7, pp. 1474–1485, 2018.
bution networks,” IEEE Trans. Power Del., vol. 14, no. 3, pp. 1167–1172,
[27] M. Cepin, Assessment of Power System Reliability: Methods and Applica-
Jul. 1999.
tions, 1st ed. London, U.K.: Springer, 2011.
[6] Z. Popovic, B. Brbaklic, and S. Knezevic, “A mixed integer linear pro-
[28] K. Miettinen, Nonlinear Multiobjective Optimization. Boston, MA, USA:
gramming based approach for optimal placement of different types of
Kluwer Academic Publishers, 1999.
automation devices in distribution networks,” Electric Power Syst. Res.,
[29] K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A fast and elitist
vol. 148, pp. 136–146, 2017.
multiobjective genetic algorithm: NSGA-II,” IEEE Trans. Evol. Comput.,
[7] M. Lwin, J. Guo, N. Dimitrov, and S. Santoso, “Protective device and
vol. 6, no. 2, pp. 182–197, Apr. 2002.
switch allocation for reliability optimization with distributed generators,”
[30] A. E. Eiben and J. E. Smith, Introduction to Evolutionary Computing, 2nd
IEEE Trans. Sustain. Energy, vol. 10, no. 1, pp. 449–458, Jan. 2019.
ed. Berlin, Germany: Springer Publishing Company, Incorporated, 2015.
[8] E. Zambon, D. Z. Bossois, B. B. Garcia, and E. F. Azeredo, “A novel
[31] N. Bhushan and K. Rai, Strategic Decision Making: Applying the Ana-
nonlinear programming model for distribution protection optimization,”
lytic Hierarchy Process, (Series Decision Engineering). London, U.K.:
IEEE Trans. Power Del., vol. 24, no. 4, pp. 1951–1958, Oct. 2009.
Springer, 2004.
[9] A. Alam, V. Pant, and B. Das, “Switch and recloser placement in distri-
[32] E. Triantaphyllou, Multi-Criteria Decision Making Methods: A Compara-
bution system considering uncertainties in loads, failure rates and repair
tive Study. New York, NY, USA: Springer-Science Business Media B. V.,
rates,” Electric Power Syst. Res., vol. 140, pp. 619–630, 2016.
2000.
[10] A. Heidari et al., “Reliability optimization of automated distribution
[33] T. Saaty, “Decision making with the analytic hierarchy process,” Int. J.
networks with probability customer interruption cost model in the pres-
Serv. Sci., vol. 1, pp. 83–98, 2008.
ence of DG units,” IEEE Trans. Smart Grid, vol. 8, no. 1, pp. 305–315,
[34] R. N. Allan, R. Billinton, I. Sjarief, L. Goel, and K. S. So, “A re-
Jan. 2017.
liability test system for educational purposes-basic distribution system
[11] A. Heidari, Z. Y. Dong, D. Zhang, P. Siano, and J. Aghaei, “Mixed-integer
data and results,” IEEE Trans. Power Syst., vol. 6, no. 2, pp. 813–820,
nonlinear programming formulation for distribution networks reliability
May 1991.
optimization,” IEEE Trans. Ind. Informat., vol. 14, no. 5, pp. 1952–1961,
[35] A. Heidari, V. G. Agelidis, and H. Zayandehroodi, “Reliability worth
May 2018.
analysis of distributed generation enhanced distribution system consid-
[12] S. Abdi, K. Afshar, S. Ahmadi, N. Bigdeli, and M. Abdi, “Optimal
ering the customer cost model based on optimal radial basis function
recloser and autosectionalizer allocation in distribution networks using
neural network,” in Proc. IEEE 7th Int. Power Eng. Optim. Conf., 2013,
IPSO-Monte Carlo approach,” Int. J. Elect. Power Energy Syst., vol. 55,
pp. 641–646.
pp. 602–611, 2014.
[36] IEEE Guide for Electric Power Distribution Reliability Indices, IEEE
[13] Y. Xu, C. Liu, K. P. Schneider, and D. T. Ton, “Placement of remote-
Standard 1366-2012 - Redline, 2012, pp. 1–92.
controlled switches to enhance distribution system restoration capability,”
[37] E. Zitzler and L. Thiele, “Multiobjective evolutionary algorithms: A
IEEE Trans. Power Syst., vol. 31, no. 2, pp. 1139–1150, Mar. 2016.
comparative case study and the strength Pareto approach,” IEEE Trans.
[14] B. Reimer, T. Khalili, A. Bidram, M. J. Reno, and R. C. Matthews,
Evol. Comput., vol. 3, no. 4, pp. 257–271, Nov. 1999.
“Optimal protection relay placement in microgrids,” in Proc. IEEE Kansas
Power Energy Conf., 2020, pp. 1–6.
[15] M.-R. Yaghoubi-Nia, H. Hashemi-Dezaki, and A. Halvaei Niasar, “Op-
timal stochastic scenario-based allocation of smart grids’ renewable and
non-renewable distributed generation units and protective devices,” Sus-
tain. Energy Technol. Assessments, vol. 44, 2021, Art. no. 101033.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on July 27,2022 at 11:31:59 UTC from IEEE Xplore. Restrictions apply.

You might also like