Renewable and Sustainable Energy Reviews 66 (2016) 415–434

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews

journal homepage: www.elsevier.com/locate/rser

Assessment of optimization algorithms capability in distribution

network planning: Review, comparison and modification techniques
Mahdi Sedghi n, Ali Ahmadian, Masoud Aliakbar-Golkar
Faculty of Electrical Engineering, K. N. Toosi University of Technology, P.O. Box 16315-1355, Tehran, Iran

art ic l e i nf o a b s t r a c t

Article history: Optimal expansion of medium-voltage power networks because of load growth is a combinatorial pro-
Received 11 September 2015 blem which is important from technical and economic points of view. The planning solutions consist of
Received in revised form installation and/or reinforcement of high voltage/medium voltage (HV/MV) substations, feeder sections,
4 March 2016
distributed generation (DG) and storage units to expand the capacity of the network. The cost objective
Accepted 13 August 2016
Available online 23 August 2016
function of the system should be minimized subject to the technical constraints. Due to the complicacy
and the complexity of the problem, it should be solved by modern optimization algorithms. In this paper,
Keywords: the most famous optimization algorithms for solving the distribution network planning problem are
Distribution network planning reviewed and compared, and some points are proposed to improve the performance of the algorithms. In
Distributed generation
order to compare the algorithms in practice, and verify the proposed improvement points, the numerical
Storage
studies on three test distribution networks are presented. The results show that every algorithm has its
Optimization algorithms
own advantages and disadvantages in specific conditions. However, in general manner, the hybrid Tabu
search/genetic algorithm (TS/GA) and the improved particle swarm optimization (PSO) algorithm pro-
posed in this paper are the best choices for optimal distribution network planning.
& 2016 Elsevier Ltd. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
2. Metaheuristic glossary and related definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418
3. Optimization algorithms: classification, review and
modification techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418
3.1. Tabu search (TS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
3.2. Simulated annealing (SA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
3.3. Genetic algorithm (GA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
3.4. Evolutionary strategy (ES) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
3.5. Artificial immune system (AIS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
3.6. Ant colony optimization (ACO) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
3.7. Ant colony system (ACS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
3.8. Particle swarm optimization (PSO) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
3.9. Hybrid TS/GA (Memetic) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
4. Problem statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
4.1. Objective function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
4.2. Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
4.3. Operation strategies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
5. Numerical study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
5.1. Cases description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
5.2. Numerical results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427

n
Corresponding author.
E-mail address: meh.sedghi@gmail.com (M. Sedghi).

http://dx.doi.org/10.1016/j.rser.2016.08.027
1364-0321/& 2016 Elsevier Ltd. All rights reserved.
416 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434

5.3. Comparison and discussion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427

6. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432

Nomenclature to the e-th failure event (h)

CAP
SDG ,d capacity of the d-th DG source (MVA)
1) Problem parameters CAP
SST ,k power rating of the k-th storage unit (MVA)
SdDG
,t operation power of d-th DG unit in t-th load level (MVA)
ADG set of all selected DG sources SkST, t operation power of k-th storage unit during t-th load
AEV set of all failure events level (MVA)
AFD set of all selected feeder sections a
SFD power flow of the i-th feeder of size a during y-th year
, i, y
ALD set of all load nodes (MVA)
ASFD set of all feeder sections directly connected to new CAP , a
SFD capacity of the i-th feeder section of size a (MVA)
,i
candidate HV/MV substations SsSS dispatched apparent power from the s-th HV/MV
,t
ASS set of all selected HV/MV substations substation during t-th load level (MVA)
AST set of all selected storage units Tt time duration of the t-th load level (h/yr)
AT set of all load levels Vl, t calculated voltage magnitude in l-th load point at t-th
INS
CDG ,d cost of installing the d-th DG source ($) load level (p.u.)
INS
CST ,k installing cost of the k-th storage unit ($) Vmin minimum allowed operation voltage (p.u.)
OM
CST ,k fixed operation and maintenance cost of the k-th Vmax maximum allowed operation voltage (p.u.)
storage unit ($) xo(l, e ) binary variable associated to the l-th load point outage
REP
CST ,k replacement cost of the k-th storage unit ($) due to e-th failure event
INS
CFD ,i fixed cost of installation feeder i ($) R standard ramp function
CO cost function of customer outage ($) λl, e average failure rate affected the l-th load node in case
INS
CSS ,s cost function of installing/upgrading the s-th HV/MV of the e-th failure event
substation ($)
CEsSS
,t electricity market price at the s-th HV/MV substation
2) Algorithms parameters
during t-th load level ($/MW h)
INS
CF installation cost of the system ($)
μ number of selected parents in ES algorithm
CF OPR operation cost of the system ($)
θ number of generated offsprings in ES algorithm
CF RLB reliability cost of the system ($)
CAP iter iteration counter
EST ,k capacity of the k-th storage unit (MW h)
c1, c2 positive constants as learning factors
EkST, t variable energy stored in k-th storage unit during t-th
mDG mutation rate for DG units
load level (MW h)
ST mST mutation rate for storage units
Emin ,k minimum allowed state-of-charge (SOC) of the k-th
mFD mutation rate for MV feeder sections
storage unit (MW h)
ST mSS mutation rate for HV/MV substations
Emax maximum allowed SOC of the k-th storage unit
q real variable that determines the relative importance
(MW h)
of the exploitation over the exploration in ACS
F objective function ($)
q DG controller parameter for intensification the DG unit
hl type of the l-th load node
solutions
Infr inflation rate
q FD controller parameter for intensification the MV feeder
Intr interest rate
solutions
NT life of the project (yr)
q0FD controller parameter for MV feeder solutions to re-
OCdDG,t operation cost of the d-th DG source including main-
move candidate HV/MV substations in IPSO
tenance cost at t-th load level ($)
q SS controller parameter for intensification the HV/MV
PdDG
,t power generation of the d-th DG unit at t-th load level
substation solutions
(MW)
vijk the j-th component of the i-th velocity vector in k-th
PlLD
, y (t ) demanded power of the l-th load node at t-th hour of
iteration of PSO
day in y-th year (MW)
w inertia weight in PSO
PlLD
,t demanded power of the l-th load node in t-th load
x iDG binary decision variable associated to the installation
level (MW)
of DG in i-th node
PtLOSS total power loss of the distribution network in t-th
x ijk the j-th component of the i-th particle's position in k-
load level (MW)
RES th iteration of PSO
PDG ,l restored power of the l-th load point by selected DG k
xgbest ,j the j-th component of the global best position vector
units (MW)
RES in k-th iteration of PSO
PST ,l restored power of the l-th load point by selected sto- k
xpbest , ij the j-th component of previous best position vector of
rage units (MW)
the i-th particle in k-th iteration of PSO
PsSS
,t dispatched real power from the s-th HV/MV substa-
x iST binary decision variable associated to the installation
tion in t-th load level including network losses (MW)
of storage unit at i-th node
PkST, t power of the k-th storage unit at t-th load level (MW)
α maximum number of iterations in which the HV/MV
Pf (t ) probability of failure event at t-th hour
substations are removed
re repair/replacement time of the equipment failed due
 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434
1. Introduction
Expansion of existing distribution networks, because of the load
growth, is a common and combinatorial problem which is im-
portant from economic and technological points of view [1]. The
problem, in its general form, consists of optimal placement and
sizing of the HV/MV substations, distributed generation (DG)
sources and storage units as well as medium-voltage feeders rout-
ing and design while minimizing complex cost objective function
under technical constraints [2–5]. However, sometimes only limited
types of facilities are considered for installation and optimization
e.g. only feeder routing. The objective function is generally com-
posed of network investment cost, operation cost and reliability
cost. The optimal solution must satisfy the capacity constraint of
facilities. The operational voltage magnitude should be kept in al-
lowable interval. In addition, the distribution network should have a
radial configuration because of protection system constraints.
Sometimes, it is necessary to design a system to meet load demand
in new regions without existing facilities (i.e. greenfield planning).
But the most common issue in distribution planning deals with the
reinforcement and expansion of an existing system. The existing
devices have construction cost only for reinforcement, and they
have no construction costs associated with them for expansion,
however, they have associated operation costs [3–5]. The distribu-
tion network may be expanded over a single stage or several stages.
Unlike the single stage expansion planning, the date of installing the
facilities should be determined optimally in multistage distribution
network planning as well. [1,2].
In recent years, a lot of mathematical models and algorithms
have been developed for solving distribution network planning
problem. A comprehensive review of classical models and issues
can be found in [6]. Transportation algorithm [7], Branch and Bound
procedure [8,9], and Mixed Integer Programming [10] are the most
important classical methods used for distribution network plan-
ning. However, the new modern optimization algorithms, as me-
taheuristics, are more useful for non-convex problems with com-
plex combinatorial structures [11]. The metaheuristics are iterative-
based methods. In order to reach the global optimum, they require
infinite number of iterations which is not feasible in practice. In a
single run with limited iteration number, there is no guarantee to
obtain the global optimal solution. Hence, the metaheuristics
should be run several times to achieve an approximate global op-
timum which is practically acceptable [5]. Because of their more
flexibility and controlled random manner, the modern algorithms
are preferable compared with the classical methods to be used in
combinatorial problems such as distribution network planning.
Comparison of these algorithms is an interesting and useful
study in practice. Some researchers have compared the capability of
the optimization algorithms generally in power system applications
[12–18]. In [12,13], some metaheuristic techniques (e.g. GA and PSO)
are compared for reactive power planning in power system. Simu-
lated annealing (SA), TS, GA and PSO algorithms are evaluated for
comparison in [14] for optimal placement of biomass power plants.
A modified binary PSO algorithm is proposed in this reference for
optimal placement of biomass power plants. The results of case
studies show the proposed PSO algorithm in [14] outperforms the
other algorithms for this problem. Several metaheuristics such as
TS, SA, GA, PSO, ACO, evolutionary strategy (ES) and hybrid ap-
proach are considered for comparison and application to generation
expansion problem (GEP) in [15], where some modifications are
suggested to improve the methods. The virtual mapping procedure
(VMP) and penalty factor approach (PFA) are proposed to improve
the efficiency of metaheuristic techniques for solving GEP. Further
an intelligent initial population generation (IIPG) is introduced to
reduce the computation time. A similar study is represented in [16],
where TS, SA, PSO, some evolutionary algorithms such as GA and ES,
417
and hybrid approach based on two-phase algorithm (e.g. combi-
nation of an evolutionary algorithm and SA) are compared with
each other. The numerical results show that the PSO, evolutionary
algorithms and hybrid approaches perform better than SA and TS to
deal with GEP. In [17,18], optimization algorithms have been com-
pared in renewable-based hybrid system studies. The authors have
proposed the renewable-based resources modeling methods and
have reviewed the optimization application in operation and
planning of hybrid systems. The application of computational op-
timization techniques in microgrid planning is reviewed in [19]
where microgrid planning problem has been defined firstly, and
then, the optimization algorithm applications have been reviewed
in the next section. In [20], the PSO algorithm is compared against
GA to optimize geothermal power plants. It is concluded that the
PSO converges to better objective function values thanks to more
inherent diversification. In addition, some metaheuristic methods
are compared in [21] for electricity demand estimation. The results
show that the PSO algorithm is preferred for demand estimation
problem when the velocity of the particles is controlled. A com-
parative study is performed in [22] to investigate the GA and PSO
algorithms when they are employed for DG planning in a dis-
tribution network. Unlike the results of [20,21] it is concluded in
[22] that GA outperforms the PSO algorithm in case studies. How-
ever, many important algorithms such as SA, ant colony system
(ACS) and artificial immune system (AIS) are neglected in [20–22].
Whereas many methods and algorithms such as TS, SA, GA, ES,
AIS, ACO, ACS, PSO, and hybrid TS/GA are recently developed for
distribution network planning, comparison of optimization algo-
rithms is needed to let the planners select the best algorithm for a
typical distribution network which is under study. Some re-
searchers have reviewed the optimization algorithms application
in distribution network planning. A review on distribution system
expansion models has been presented in 1993 [23]. However, it is
an outdated work that has not considered the most modern al-
gorithms. In [24,25], two valuable reviews on models and methods
used in distribution system planning have been provided. More-
over the literature review shows that whenever metaheuristics are
applied to a specific problem, they need particular modifications
related to the structure of that problem. Hence, modified algo-
rithms are more efficient than standard ones. As a result, this
paper not only considers the review and comparison of the opti-
mization algorithms, but also proposes some new modifications
for them. This paper reviews the application of the most important
metaheuristics i.e. TS, SA, GA, ES, AIS, ant colony optimization
(ACO), ACS and PSO algorithms and some hybrid approaches to
distribution network planning. A comparative study is developed
to specify the advantages and disadvantages of the algorithms as
well as the proposed modifications. Classical methods are not
considered in this paper.
On the other hand, unlike the conventional distribution net-
work expansion planning, a general model for distribution net-
work planning is used here. In conventional distribution network
planning, only HV/MV substations and/or feeder sections are
considered. However, in this paper, dispatchable DG and battery
energy storage units are taken for consideration as well. As a re-
sult, a general optimization problem of distribution network
planning is used here to compare the algorithms experimentally.
This study determines the effectiveness of the algorithms when
they are employed for planning of modern power distribution
systems namely active distribution networks (ADNs). The com-
parison approach is based on numerical studies. The uncertain
conditions are neglected in the current work. However, some op-
eration strategies for DG and storage units are used considering
the variable load.
The contribution of this work is summarized as follows that
distinguished it from the previous works:
418 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434
– Numerical studies are provided in this paper for eight most
popular heuristic algorithms.
– In addition to review the algorithms, a simulation-based com-
parative study has been performed to discover the advantages
and disadvantages of every method.
– Some modification techniques are proposed to improve the
performance of algorithms considering especial features of dis-
tribution network planning.
– In addition to the conventional components, the modern tech-
nologies such as energy storage units are considered in reviews
and simulations too.
– A clear classification among greenfield, single stage and multi-
stage planning is provided in literature review.
The remainder of the paper is as follows: Firstly, the algorithms
glossary and related definitions are presented. The metaheuristics
review, analysis and modification techniques are included in Sec-
tion 3. In Section 4, distribution network expansion planning
problem is described and formulated. Section 5 consists of nu-
merical studies, comparisons and discussions. Concluding remarks
are explained in Section 6.
2. Metaheuristic glossary and related definitions
In this section, important terms and ideas are presented to
effectively convey the concepts involved with optimization algo-
rithms. These terms and ideas offer a basic understanding of the
concepts that are to be covered in detail in the following sections.
1) Greedy search: It is a local search strategy concerned in opti-
mization algorithms. According to this strategy, the first
neighbor solution which is better than the current solution is
selected to continue local search.
Fig. 1. Classification of distribution network planning methods.
2) Hill-climbing: Unlike greedy search, in hill-climbing algorithm,
a neighbor solution with lower quality may be accepted to
continue the local search along a trajectory.
3) Population: A group of candidate solutions (called individuals)
to an optimization problem.
4) Offspring: A new candidate solution which is produced by
previous candidate solutions called parents.
5) Mutation: It is an operator in evolutionary algorithms to gen-
erate offspring. It is analogous to biological mutation. Mutation
changes one or more gene values in a chromosome from its
initial state.
6) Crossover: It is an operator in evolutionary algorithms to
generate two offsprings regarding two parents. It is analogous
to biological crossover. Crossover is a process of taking more
than one parent solutions and producing child solutions from
them.
7) Pheromone: A chemical substance produced by ants. They use
pheromone to communicate with their family.
3. Optimization algorithms: classification, review and
modification techniques
Branch Exchange (BE), Tabu search, simulated annealing, ge-
netic algorithm, evolutionary strategy, artificial immune system,
ant colony optimization, ant colony system, particle swarm opti-
mization and hybrid approaches such as memetic are the famous
metaheuristics which have been used for distribution network
planning. A classification of the methods and algorithms is shown
in Fig. 1, where the metaheuristics are divided to two categories:
trajectory-based and population-based algorithms.
Recently a lot of papers have been presented for distribution
network planning using optimization algorithms. For example,
some previous pertinent works in which metaheuristic techniques
 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434 419

Table 1
Properties of the papers which have used metaheuristics for distribution network planning.

Optimization algorithm Refs. Type of planning Equipment for installation, expansion or reinforcement

Greenfield planning Single stage expansion Multistage expansion HV/MV substation MV feeder DG Storage

BE [26] * * *
[27] * * *
[28] * *

TS [29] * * *
[30] * * *
[31] * * *
[32] * * *

SA [33] * *
[34] * *
[35] * *

GA [36] * *
[37] * * *
[38] * * *
[39] * * *
[40] * *
[41] * *
[42] * *
[43] * * *
[44] * *
[45] * * *
[46] * * *
[47] * *
[48] * * *
[49] * * * *
[50] * *
[51] * * * *
[52] * * * *
[53] * * *
[54] * *
[55] * * *
[56] * *
[57] * * *
[58] * * *

ES [59] * * *
[60] * *
[61] * *

AIS [62] * *
[63] * *
[64] * *
[65] * *

ACO [66] * *
[67] * * *
[68] * *
[69] * *
[70] * *

ACS [71] * * *
[72] * * * *

PSO [73] * *
[74] * * *
[75] * *
[76] * * * *
[77] * * *
[78] * *
[79] * * *
[80] * *
[81] * *
[82] * *
[83] * * * *
[84] * * *
[85] * *
[86] * *
[87] * *
[88] * * * * *

TS/GA [89] * * *
[90] * * * *
[91] * * * *
420 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434
have been used for distribution network planning are shown in
Table 1. The algorithms used in these papers are presented in
Table 1 as well. Moreover, this table represents the properties of
the planning approach and also the equipment that have been
considered for installation, expansion or reinforcement.
As can be seen, most researchers have considered the single-
stage expansion rather than greenfield and multistage planning.
GA and PSO are the most popular algorithms, and ACO is arranged
in the third grade. GA has been usually used for MV feeder plan-
ning, while PSO is mostly employed in problems which contain DG
siting and sizing. Many papers have considered the optimal
planning of MV feeders beside DG sources. However, the storage
units are not investigated in lots of papers. In few papers which
include storage planning, either GA or PSO is utilized for
optimization.
Metaheuristics are iteration-based techniques which search in
solutions space of the combinatorial problems to find an approx-
imate global optimal solution within a reasonable computation
time. Detailed information about optimization algorithms can be
found in [92,93]. These algorithms have been extremely applied to
distribution network planning problem and some various mod-
ifications were proposed. In this section, the most important al-
gorithms are reviewed and some new modifications and hy-
bridization approaches are proposed.
Fig. 2. The flowchart of proposed local search method.
3.1. Tabu search (TS)
TS was introduced by Glover in [11]. TS is a local-search-based
algorithm that starts off with a feasible current solution. During
iterations, it searches locally along a trajectory. In each single
iteration, it moves to the best neighbor solution of the current
solution, even if the neighbor solution is worse than the current
solution regarding the objective function. As a result, TS is a hill-
climbing algorithm. Also, a Tabu memory is used in TS to prevent
cycling. Several papers have applied TS to distribution network
planning problem e.g. [29–32].
In conventional distribution network planning, branch ex-
change (BE) method is used for local search [27,94]. However, BE
method can be applied only to feeder sections. In practice, TS can
not consider all neighbors of the current solution in limited
computation time, so usually a probabilistic local search is used in
TS [92]. Therefore, the result of TS depends on the initial solution.
It is casual to obtain the global optimum, and TS needs to be in-
corporated with auxiliary algorithms.
In this paper, a novel local search method is proposed for dis-
tribution network planning which includes feeders, DG sources
and storage units expansion, as shown in Fig. 2.
According to the proposed local search, only the state of one
type of facility (i.e. either feeders or DGs or storage units) is
 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434
changed to produce a neighbor solution. A DG or storage unit can
be added to/removed from the network. However, it is also fea-
sible to remove/add another DG or storage, if the generated ran-
dom number within (0,1) is less than 0.5. As a result, the location
of a DG or storage unit may be changed immediately. This sto-
chastic strategy let the optimization algorithm pass the local op-
timums. In this algorithm, mFD and mDG are controller parameters
that 0 < mFD < 1 and 0 < mDG < 1. For sizing DG and storage units,
in the proposed method, several candidate capacities can be
considered and modeled by separate binary values. Therefore, the
dimension of the problem increases. Finally, the candidate HV/MV
substations to which no feeder is connected, are removed from the
candidate solution.
As the main drawback, performance of the TS algorithm is
dependent on initial solution in practice. So it is not suitable for
large-scale distribution network planning, where an appropriate
initial solution can not be estimated easily. There is no guarantee
to obtain a global optimum when TS is used for optimization.
However, the advantage of TS is its simplicity as it has the least
number of controller parameters i.e. only number of neighbors and
maximum iteration. So it does not need serious effort to tune the
parameters that is a challenge in metaheuristics application.
3.2. Simulated annealing (SA)
SA was first presented by Kirkpatrick in [95]. It is a hill-climbing
algorithm similar to TS. However, it considers only one neighbor in
each single iteration. The trend of SA to hill-climbing is reduced
during iterations. The tendency of SA from exploration to ex-
ploitation is controlled by two parameters (i.e. temperature and
cooling rate). As a result, performance of SA will be independent of
initial solution, if the temperature is taken high enough. However,
with a very high temperature, the algorithm may diverge. Unlike
TS, selecting appropriate values for controller parameters is a
challenge in SA implementation. It is proved mathematically in
[92] that the algorithm converges asymptotically to the global
optimum with probability one. So, infinite iterations are needed to
achieve the global optimal solution. In practice it is infeasible be-
cause of limited computation time. Usually the algorithm stops
after predetermined number of unsuccessful iterations. Therefore
the mentioned probability is less than one. However, the algo-
rithm is run several times to improve the results [93].
SA algorithm has been applied to distribution network plan-
ning problem in some papers such as [33–35]. When DG and
storage units are contemplated in distribution network planning,
the local search method proposed in Section 3.1 for TS can be used
for SA algorithm as well.
3.3. Genetic algorithm (GA)
The basic idea of GA was first introduced in [96]. GA is an
evolutionary algorithm that is based on the population of some
solutions. During successive iterations, the population is promoted
by applying genetic operators (i.e. crossover, mutation and selec-
tion). In Ref. [97], it is proved that if the current best individual of
the population is not eliminated by GA operators, the canonical GA
(with infinite number of iterations) converges to the global opti-
mum with probability one. Unfortunately, GA execution is very
time-consuming in practice, even with finite number of iterations
[36].
GA is widely used for distribution network planning, for in-
stance in [36–58]. High computation cost is the main flaw of GA,
when it is used for distribution network planning. In order to
modify the standard binary GA, some techniques have been pro-
posed in previous works. Literature review shows that the most
effective modification (to reduce the computation cost) is
421
proposed in [36,37] where GA is specifically adjusted for dis-
tribution network planning. In the improved GA, the substations
and feeders are coded directly, and modified crossover and mu-
tation operators are used. To produce an offspring, the crossover
operator initially selects the feeder sections which are common
between the parents, and then, in order to generate a radial con-
nective graph, it selects different feeders randomly from one of the
parents. Moreover, mutation operator of the modified GA is a BE
process. More details about the modifications can be found in
[36,37].
Regarding the especial crossover and mutation operators, the
offsprings are more similar to their parents in modified GA. As a
result, diversification of the modified algorithm is less than the
conventional GA, however, the intensification increases. The nu-
merical results of [36,37] show that using the modified GA effi-
ciently improves the computation cost. In addition, tuning para-
meters of the modified GA is not difficult compared with the other
algorithms such as SA.
Here, the modified crossover operator can be easily extended to
use for DG and storage planning. While the proposed local search
in Section 3.1 is utilized for mutation. In some cases, HV/MV
substations are expanded as well. So, the mutation operator
should be applied to the candidate HV/MV substations as shown in
literature [39]. This operator eliminates randomly a HV/MV sub-
station and removes all feeder sections which are directly con-
nected to it. The operator is helpful when the planning problem
includes several candidate HV/MV substations. However, the mu-
tation rate of HV/MV substations is reduced during iteration to
prevent diverging.
3.4. Evolutionary strategy (ES)
Evolutionary strategy algorithm is similar to GA. After gen-
erating the initial population, μ parents are randomly selected to
produce θ new offsprings using mutation operator. Then the next
population is chosen among parents and offsprings by the selec-
tion operator. ES is used in some papers for conventional dis-
tribution network planning [59–61], however, DG and storage
units should be concerned in ADNs planning as well. In this way,
the proposed local search in Section 3.1 should be used for mu-
tation in ES. Moreover, if the planning problem includes some new
candidate HV/MV substations, the mutation operator is applied to
the HV/MV substations as described in Section 3.3.
3.5. Artificial immune system (AIS)
AIS algorithm is based on human body's defense process
against viruses [92]. Like GA and ES, AIS is a population-based
algorithm, but it utilizes the following three operators: duplica-
tion, mutation and selection. Starting from a randomly generated
population, the solutions are reproduced with different rates.
Considering the objective function, more suitable solutions are
duplicated with a more rate. Then the solutions are mutated with
different rates. The less suitable solutions are changed with a
higher mutation rate. Finally the selection operator is applied to
the whole population to produce a more appropriate group
of solutions. The algorithm is well described in [98], and it is
used for distribution network planning in some papers such as
[62–65].
AIS is more intelligent than GA, thanks to guided mutation and
duplication operators. However, tuning the mutation and dupli-
cation rates is a challenge for AIS application in practice. Similar to
GA, the proposed local search strategy can be employed for mu-
tation operator in AIS.
422 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434
3.6. Ant colony optimization (ACO)
Ant colony optimization is a swarm intelligent algorithm in
which the ants use some probabilistic strategies to move on the
path and use the pheromone details for communicating with
other ants. So, the distribution network expansion problem
should be modeled as a graph firstly. In this graph, each node
denotes a device e.g. a candidate substation or feeder section.
Ants are placed first randomly on the start node. Then, they are
allowed to move through the feasible routes. Each ant generates a
candidate solution and then the objective function is calculated.
The pheromone level is updated considering the objective func-
tion and the selected branches by ants. In the next iteration, the
ants are enabled to select the path probabilistically considering
the pheromone level. The procedure is iterated until the termi-
nation criterion is satisfied. The details of ACO can be found
in [99].
ACO is applied to distribution network planning in some papers
such as [66–70]. As a modification technique, an improved ACO
namely Min-Max ACO (minimum-maximum ant colony optimi-
zation) can be used for distribution network expansion planning.
Min-Max ACO algorithm, in which the pheromone level is boun-
ded, has been presented in [99]. It is shown that Min-Max ACO is
more efficient than conventional ACO. However, the Min-Max ACO
needs more computation time for convergence [99].
3.7. Ant colony system (ACS)
Ant colony system algorithm is introduced to improve perfor-
mance of ACO for large-scale problems. ACS includes two mod-
ifications [93]:
1. Using local update of pheromone.
2. Using a rule of transition depending on parameter q ( 0 < q < 1)
which defines a balance between diversification and
intensification.
When ants are moving on the graph, if q > rand, then only
branches which have the most pheromone level are selected and
probabilistic movements are neglected. When q is high, the in-
formation gathered in the previous iterations is utilized more, and
a non explored path can not be chosen. The details of ACS can be
found in [93]. A Min-Max ACS can be defined similar to the Min-
Max ACO algorithm with constrained pheromone level. In practice,
the major challenge of ACS is to set q to an appropriate value.
Some papers have implemented the ACS algorithm for dis-
tribution network planning [71,72]. However, in ADN planning,
several candidate HV/MV substations, feeder sections, DG sources
and storage units are considered simultaneously. When the pro-
blem includes some new candidate HV/MV substations, they are
modeled as separate nodes in the graph of ACS. An example of this
model is presented in Fig. 3, where substations no. 1, 4 and 6 are
selected for installation. As a result, feeder sections which are di-
rectly connected to substations no. 2, 3 and 5, are moved to a Tabu
Fig. 3. Modeling of candidate HV/MV substations and the other components in ACS graph.
list. Then ants are allowed to select only the other facilities (i.e.
admissible feeders, DG and storage units). These facilities are se-
lected within a similar graph shown in Fig. 3, after the HV/MV
substations are selected.
In this case, to prevent premature converging to a local opti-
mum for substations, some mutations are needed seriously in
ACS. However, permanent mutations in HV/MV substation states
can disrupt the convergence. In this paper, a novel ACS-based
hybrid algorithm namely GA/ACS is proposed for distribution
expansion planning when it includes some candidate HV/MV
substations and other equipment, simultaneously. The proposed
hybrid algorithm consists of two phases. In phase one, ACS al-
gorithm is executed for all facilities while the mutation operator
is applied only to the HV/MV substation solutions. In this phase,
local update of pheromone is abandoned in order to increase
diversification for the other components (i.e. feeders, DG and
storage units). The intensification parameter q for these compo-
nents should not be very high, but it must be high enough for
substations to prevent diverging. After sufficient iteration, mu-
tation rate is decreased to zero instantaneously and the algorithm
is switched to the second phase. In the switching step, all the
pheromone evaporates (for all the components). In the second
phase, q is set to 1 for HV/MV substations and it is increased
appropriately for feeders, DG and storage units. As a result, the
selected HV/MV substations are kept fixed in the second phase.
The second phase will be interrupted after predetermined num-
ber of unsuccessful iterations. The flowchart of the proposed GA/
ACS algorithm is represented in Fig. 4.
Although the proposed GA/ACS algorithm is more intelligent
than the conventional ACS, it needs more effort for tuning the
parameters appropriately. The hybrid GA/ACO algorithm can be
applied as well, using the ACO algorithm instead of ACS.
3.8. Particle swarm optimization (PSO)
PSO is a swarm intelligent algorithm. It is based on the move-
ment of some particles. These particles, which denote feasible
solutions, share their explorations among themselves. Every par-
ticle flies in the solutions space with a corresponding velocity
which is regulated iteratively considering its own previous ex-
perience and the other particles' experience. The details of PSO can
be found in [92]. In discrete space, the particles are manipulated
according to the following equations:
(
vijk + 1 = w × vijk + c1 × rand × x pbest
k k
, ij − xij )
k
(
+ c2 × rand × xgbest k
, j − xij ) (1)
⎧
⎪ 1,
xijk + 1 = ⎨
(
if rand < sigmoid vijk + 1 )
⎪
⎩ 0, otherwise (2)
where
 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434 423

Fig. 4. Flowchart of the proposed hybrid GA/ACS algorithm.

1 HV/MV substations. In the proposed improved PSO (IPSO), Eq. (2)

( )
sigmoid vijk + 1 =
1 + exp ( )
−vijk + 1 (3)
is modified for DG and storage units as follows:

⎧
PSO is extremely applied to distribution network planning ⎪ 1,
⎪
if { rand < sigmoid( v )}
k+1
ij
problem, e.g. in [73–88]. Literature review shows that PSO algo-
xij = ⎨
k+1
AND { ⎡⎣ x = 1⎤⎦OR⎡⎣ q < rand⎤⎦
rithm is not robust when the dimension of the problem increases. ⎪
k
gbest, j
u
}
In this paper, a modified PSO algorithm based on [88] is proposed ⎪
⎩ 0, otherwise (4)
to eliminate this flaw. Moreover, the proposed algorithm is pow-
erful if the expansion planning problem includes several candidate where u ∈ {DG, ST}. For feeder sections the equation is adjusted as
424 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434

follows: 4. Problem statement

⎧ In order to apply the metaheuristics to distribution network

⎪ 1, if
⎪
{ (
rand < sigmoid vijk + 1 )}
⎪ AND ⎡⎣ xgbest
k
{ ⎤ ⎡ FD < rand⎤
, j = 1⎦OR⎣ q ⎦ }
⎪ tion expansion planning specifies the optimal route and size of
xij = ⎨
k+1
⎪ AND
⎪ ⎡ x k + 1 ∉ A ⎤OR⎡ q FD < rand⎤OR⎡⎣ iter > α⎤⎦
⎪ ⎣ ij{ SFD⎦ ⎣ 0 ⎦ }
⎪
⎩ 0, otherwise
In Eqs. (4) and (5), the new controller parameters, i.e.
0 < q DG < 1, 0 < q ST < 1 and 0 < q FD < 1, are embedded to increase
the intensification. The controller parameter q0FD is used to remove
the feeder sections which are directly connected to the new can-
didate HV/MV substations. This process i.e. removing the new
candidate HV/MV substations, is terminated after α iterations. q0FD
and α are new controller parameters that prevent installing the
excessive HV/MV substations. If q DG = 0, q ST = 0, q FD = 0 and
q0FD = 0, then the IPSO is converted to the conventional PSO. The
controller parameters should be set appropriately to make the
IPSO effective. In other words, if these fixed parameters are too
small, the IPSO will be similar to the conventional PSO. On the
other hand, if they are set to very high values, the premature
convergence to a local optimum will be happened. The suitable
settings can be found by trial and error.
3.9. Hybrid TS/GA (Memetic)
Incorporating the local search of TS algorithm into the muta-
tion operator of GA is the most common strategy in metaheur-
istics hybridization [92]. In comparison with the GA, the muta-
tion operator of the hybrid TS/GA is more intelligent, and so the
hybrid algorithm will be more suitable. Sometimes, hybrid TS/GA
is called memetic algorithm which was first proposed by Moscato
in [100]. TS/GA is applied to the distribution network planning
problem in some papers such as [89–91]. The modified GAs
proposed in literature can be used to couple with TS for dis-
tribution network planning. Using TS algorithm increases the
computation cost of hybrid approach, however, tuning the para-
meters is not more complicated than GA, thanks to the simplicity
of TS.
expansion planning and compare the algorithms experimentally,
the planning problem is formulated in this section. The distribu-
new feeders beside the location and capacity of new DG sources,
storage units and new HV/MV substations, and/or the capacity
increment of existing HV/MV substations to minimize the cost
(5) objective function considering technical constraints. New HV/MV
substations and new feeders are selected among several candidate
substations and feeders, while the new DG and storage units can
be installed in every load node. The conductor type of new feeder
sections as well as the size of DG units and energy storage devices
can be selected among several candidate ones. For example, a
simple test network is shown in Fig. 5.
It is assumed that the distribution company (DISCO) is the
owner and operator of the distribution network. In addition, it is
the institute that invests for installing/reinforcement of equipment
to meet the load growth in distribution network. These assump-
tions have been usually used in the previous works [86].
The objective function contains three terms: investment cost,
operation cost and reliability cost. Operation cost consists of DG
operation cost and the cost of purchased electricity at HV/MV
substations. The operation cost of storage units can be modeled by
fixed operation and maintenance cost which should be calculated
considering the capacity of storage units [101]. The interest and
inflation rates are used to estimate the objective function in the
base year of study.
In this paper, the energy not supplied (ENS) is considered as the
reliability cost of the distribution network. The ENS term appears
due to the failure events in HV/MV substations or feeder sections.
The outage cost is determined considering the interrupted load,
type of the customer and outage duration. However, DG and sto-
rage units can restore the lost loads partly, if it is feasible from
technical points of view. The availability of restoration sources
should be evaluated for every failure event.
In an acceptable solution, the power of the feeders, substations,
DG sources and storage units should not exceed the maximum
capacity, the required load should be supplied in normal condi-
tions, and the voltage magnitude of all nodes should be limited to
the allowed boundary. In addition the configuration of the net-
work must be radial [94]. As a result, the mathematical formula-
tion can be written as follows:
4.1. Objective function

F = CF INS + CF OPR + CF RLB (6)

where

F INS = ∑ INS
CFD ,i + ∑ INS
CSS ,s + ∑ INS
CDG ,d
i ∈ AFD s ∈ ASS d ∈ ADG

∑ ⎡ C INS + C OM + C REP ⎤
+ ⎣ ST , k ST , k ST , k⎦
k ∈ AST (7)

F OPR
NT ⎡ ⎤
= ∑ ∑ Tt × ⎢ ∑ CEsSS ( SS
, t × Ps, t + ) ∑ OCdDG ⎥
y = 1 t ∈ AT
⎢⎣ s ∈ A d ∈ ADG
,t
⎥⎦
SS

⎛ 1 + Infr ⎞ y − 1
×⎜ ⎟
⎝ 1 + Intr ⎠ (8)
Fig. 5. A simple example of distribution network for expansion planning.
 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434 425

NT 4.3. Operation strategies

F RLB = ∑ ∑ ∑ xo(l, e) × λ e × CO(hl , re )
y = 1 l ∈ ALD e ∈ AEV
Distribution network planning procedure should be evaluated
⎧ 24 re ⎫ considering operation modes of the system. In this section, op-
∑ ∑ Pf (t ) × R⎡⎣ PlLD ⎤⎪
⎪
×⎨ , t (t + m) − PDG, l − PST , l (t , m, re )⎦⎬
RES RES
⎪
⎩ t=1 m=1
⎪
⎭ eration strategies for DG and storage units are described, assuming
(9)
a three-level load model of the variable loads. Distributed re-
where Eqs. (7)–(9) represent the installation cost, operation cost sources such as DG and storage units can be used in distribution
and reliability cost, respectively. networks for several purposes. Here, DG and storage units are
allocated for both peak shaving and reliability enhancement [88].
4.2. Constraints Consequently, DG sources are used to inject the electrical power to
the network only when the load level is high or normal, assuming
three-level load model [88]. As a result, DG units are stayed in
a CAP, a standby mode in light-load times. Here, it is supposed that dis-
0 ≤ SFD , i, t ≤ SFD, i , ∀ i ∈ AFD , ∀ t ∈ AT (10)
patchable DG units have operation cost as shown for example in
[49,58,86,88]. In other cases, it is not necessary to keep DG units
on standby mode when there is low demand of energy. The sto-
0 ≤ SsSS CAP rage units are charged when the price of electrical energy is low,
, t ≤ SSS, s , ∀ s ∈ ASS , ∀ t ∈ AT (11)
and then they are discharged in high-price periods. As a result, the
storage units are kept in standby mode when the price level is
normal.
0 ≤ SdDG CAP Moreover, DG and storage units can be used to reduce the ENS
, t ≤ SDG, d, ∀ d ∈ ADG , ∀ t ∈ AT (12)
appeared due to failure events. When a failure event happens, the
failed section is cut off from the other devices using normally
closed switches. Then DG and storage units restore the lost load in
0 ≤ SkST, t ≤ SST
CAP
∀ k ∈ AST , ∀ t ∈ AT islanding mode operation. At first, the loads with the higher
, k, (13)
priority are restored. Then the loads which have the lower priority
are joined, if the capacity constraint of DG and storage units is not
violated.
ST ST ST The amount of energy which is stored in storage units i.e. the
Emin , k ≤ Ek, t ≤ Emax , k, ∀ k ∈ AST , ∀ t ∈ AT (14)
SOC, is variable versus time. Unlike dispatchable DG sources, the
restored load power by storage units is dependent on both repair/
replacement duration and the time at which the failure event
occurs. For every probable failure event, the SOC should be esti-
∑ PsSS
,t + ∑ PdDG
,t = ∑ PkST, t + ∑ PlLD
, t + Pt
LOSS
, ∀ t ∈ AT
s ∈ ASS d ∈ ADG k ∈ AST l ∈ ALD (15) mated from periodic charge/discharge curve of the storage units to
determine the ENS index of reliability. Similar to storage units, the
restored load power by non-dispatchable DG sources e.g. solar and
wind generation, depends on the time at which the failure event
Vmin ≤ Vl, t ≤ Vmax, ∀ l ∈ ALD , ∀ t ∈ AT (16) occurs. However, non-dispatchable lonely DG sources cannot be
used in islanding mode of operation.

NLD = NFD (17)

where Eqs. (10)–(14) represent the constraints of capacity, Eq. Table 2

Economic and technical data used for experimental test [88,101].
(15) shows the power balance, Eq. (16) shows the voltage con-
straint, and Eq. (17) indicates the constraint of radial network. Economic/technical parameters Value
INS
CSS , s is a discrete function of installing/upgrading capacity of the
s-th substation. The capacity of the substations is determined to DG investment cost ($/kW) 2000
DG operation cost ($/MW h) 50
satisfy Eq. (15) for the power balance constraint, in which PtLOSS is
Fixed O&M cost of DG ($/kW h) 0.0153
computed by executing the backward/forward sweep load flow DG normal operation power factor 1
considering distributed energy resources [102]. Storage investment cost ($/kW h) 225
The replacement cost of the storage units is included in the cost Unit cost of power rating for storage ($/kW) 175
objective function, because of limited lifetime of battery energy Fixed O&M cost for storage ($/kW) 20
Storage replacement cost ($/kW h) 225
storage units. In other words, the number of charge/discharge Number of charge/discharge cycles in storage life 10,000
cycles of the batteries is restricted. Whenever the number of cycles Interest rate (%) 14
exceeds the maximum value, the replacement cost is added to the Inflation rate (%) 7
investment cost in Eq. (7). In addition, battery energy storage Failure rate of feeders (f/km-yr) 0.2174
Failure rate of HV/MV transformers (f/yr) 0.0650
should not be discharged completely because of the technical Repair/replacement time of feeders (h) 4
constraints [103]. Here, state-of-charge (SOC) of the battery units Repair/replacement time of HV/MV transformers (h) 8
is limited as a constraint in Eq. (14).
426 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434

Table 3 industrial loads. The second case is a 48-bus network with one
Information of the three test distribution networks used for case studies. existing HV/MV substation, for expanding only the feeders, DG
Parameters of typical networks Case 1 Case 2 Case 3 sources and storage units. The third case is a 71-bus network with
one existing and five candidate HV/MV substations. It includes the
Voltage level (kV) 33 20 20
Allowed voltage deviation (%) 5 5 5
Number of the existing load nodes 8 16 24
Number of the new load nodes 14 31 41 80.044
Worst
Existing HV/MV substations capacity (MVA) 5 15 15
Best
Number of the candidate HV/MV substations 0 0 5
80.042
Number of the existing feeder sections 8 16 24
Number of the candidate feeder sections 41 119 178
Average peak load per node (kW) 320 315 400 80.04

Objective function (M$)

Power factor of loads 0.8 (lag) 0.9 (lag) 0.9 (lag)
Annual increment load rate (%) 3 3 3
Life time of the project (yr) 25 25 25 80.038

80.036

Table 4 80.034
Sector interruption cost for reliability cost evaluation [49].

80.032
User sector Priority no. Interruption duration (min) & Cost ($/kW)

60 min 240 min 480 min 80.03

TS SA GA ES AIS ACO ACS PSO IPSO TS/GA
Residential 3 0.482 4.914 15.69
Industrial 2 9.085 25.16 55.81 Fig. 7. Comparison of optimization algorithms performance in case 1.
Commercial 1 8.552 31.32 83.01

167
Worst
Best
166.8
Table 5
Loading levels and market price data.
166.6
Load level Percentage of peak load Time duration (h/ Market price
Objective function (M$)

(%) yr) ($/MW h)

166.4

High 100 2190 75

Normal 70 4380 50 166.2
Low 50 2190 35
166

165.8

165.6

165.4
TS SA GA ES AIS ACO ACS PSO IPSO TS/GA

Fig. 8. Comparison of optimization algorithms performance in case 2.

282
Worst
Fig. 6. Daily charge/discharge curve of storage units. Best
281

280

5. Numerical study
Objective function (M$)

279

5.1. Cases description

278

In order to experimentally analyze and compare the capability 277

of optimization algorithms for distribution network expansion
planning, and to verify the proposed improvement points, three 276

different distribution networks as test cases are studied. The first

275
case is a 23-bus network with one existing HV/MV substation, for
expanding only the feeders, DG sources and storage units. The 274
properties of the network are shown in [71]. These properties are TS SA GA ES AIS ACS PSO IPSO TS/GA GA/ACO GA/ACS

modified for an expansion planning. In this case, all load nodes are Fig. 9. Comparison of optimization algorithms performance in case 3.
 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434 427

250 data are shown in Table 5.

objective functions (1000$)

200
The daily charge/discharge curve of storage units is shown in
Fig. 6. This curve is used to estimate the SOC of the storage units
Difference of

150 during a day. The active power of battery is a negative value when
it is discharged.
100

50 5.2. Numerical results

0
TS SA GA ES AIS ACO ACS PSO IPSO TS/GA The tests are implemented using MATLAB 7.4. Each algorithm is
run several times and the results are presented in Figs. 7–9, which
Fig. 10. Difference between the best and the worst objective functions in case 2.
show the best and the worst cost functions obtained from opti-
mization algorithms in three different cases. Among these, the
IPSO and hybrid GA/ACS are the new algorithms proposed in this
1800
paper. The conventional ACS and proposed GA/ACS algorithms
objective functions (1000$)

1500
achieve the same results in case 1 and 2, because there is no new
1200
Difference of

candidate HV/MV substation in these cases. However, they are

900 different in the 3rd case, hence, the performance of GA/ACS is
600 shown only in case 3.
300 Figs. 7–9 represent that there is no difference between the best
0 and the worst solutions in case 1 which is a simple problem, while
TS SA GA ES AIS ACS PSO IPSO TS/GA GA/ACO GA/ACS
the difference between the best and the worst objective functions,
Fig. 11. Difference between the best and the worst objective functions in case 3. in case 2 and 3, are considerable and shown respectively in Figs. 10
and 11. The convergence time of the algorithms is given in Table 6.
Figs. 12–14 show the objective function information during
iterations of the optimization algorithms. These data are presented
Table 6
as lines for trajectory-based algorithms, and as box plots for po-
Convergence time of the optimization algorithms in numerical studies.
pulation-based algorithms.
Optimization algorithm Convergence time (s)
5.3. Comparison and discussion
Case 1 Case 2 Case 3

TS 8 1181 1535
According to Table 2, the test network in case 1 has the least
SA 16.5 2067 1553 dimension and it is the simplest test case, however, the 3rd case is
GA 41.5 1590 2297 the most complex one. Fig. 7 shows that all the algorithms except
ES 43.06 1872 3892 ACO obtain the same optimal solution in case 1. The ACO algorithm
AIS 77.15 1706 4139
results in the worst solution in this case as well as the other cases.
ACO 11.8 2429 –
ACS 29 824 1413 So, ACS algorithm, as a modified version of ACO is preferred to be
PSO 126.5 1216 1428 used for optimal distribution planning.
IPSO 36.7 997 3163 Among trajectory-based algorithms, SA is more appropriate
TS/GA 28.9 1995 4347 than TS, especially in the third case. The reason is that the hill-
GA/ACO – – 427.5
GA/ACS – – 2911
climbing strategy is controlled by a dynamic temperature in SA, as
it is visible in Figs. 12–14(a), (b). As a result, uncontrolled hill-
climbing is not suitable for the distribution network planning.
However, the main flaw of SA is that it can not find the optimal
expansion of all the HV/MV substations, feeder sections, DG location of HV/MV substations correctly in case 3, because it
sources and storage units. Properties of the cases 2 and 3 are considers only one neighbor solution in every single iteration. The
advantage of TS is its fast convergence (see Table 6). It is the best
presented in [104]. In these cases, the load types consist of re-
choice for uncritical optimization problems when the computation
sidential and commercial loads.
cost is an important index. For example, it would be a suitable
Two candidate types of overhead MV feeders are considered for
method if the problem should be solved several times for lots of
numerical studies. Properties of the feeders are adopted from
scenarios. Another advantage of TS is its simplicity. It has the least
[49,105]. Three candidate sizes of DG sources (400, 800 and
number of controller parameters for tuning that is a challenge for
1000 kVA) and three candidate capacities of storage units (1.8,
application of the heuristic methods.
2 and 2.1 MW h) are taken for consideration in these studies. Among population-based algorithms, GA is widely applied to
Every model of renewable/nonrenewable dispatchable DG distribution planning problem. It provided acceptable results in all
sources and storage units can be used in the proposed metho- three cases, however, combination of TS to GA i.e. hybrid TS/GA,
dology. However, in this paper, fuel cell DGs and zinc-bromine (Zn/ can improve the results. Hybrid TS/GA utilizes not only the ad-
Br) battery units are used for the case studies. Economic and vantages of population search in several areas, but also the ben-
technical data are given in Table 2 and more details of the cases efits of local searches in every area. It seeks probable optimal areas
are shown in Table 3. The parameters of the batteries are obtained in more details through the local search of TS. So, the results
from [101]. would be more suitable. This improvement, which is obvious in
Table 4 gives the costumer sector interruption cost for different Figs. 8 and 9, is considerable when the problem is more complex.
load types used in the study. The loading levels and market price Figs. 12–14(c), (d) show the impact of TS algorithm on the popu-
428 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434

Fig. 12. Objective function data during iteration of algorithms in case 1.

M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434 429

Fig. 13. Objective function data during iteration of algorithms in case 2.

430 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434

Fig. 14. Objective function data during iteration of algorithms in case 3.

 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434
Table 7
List of advantages and disadvantages of all the algorithms.
Algorithm Advantages
TS It convergences fast.
Tuning the controller
parameters is easy.
It is suitable for simple and un-
critical problems.
SA It does not depend on the initial
solution.
It is capable to obtain the
global optimum in large-scale
problems.
GA It does not depend on the initial
solution.
It is capable to obtain the global
optimum in complex problems.
Tuning the controller para-
meters is not critical.
ES It does not depend on the initial
solution.
It is capable to obtain the global
optimum in complex problems.
Tuning the controller parameters
is not critical.
It is simpler than GA for im-
plementation (without crossover
operator).
AIS It does not depend on the initial
solution.
It is capable to obtain the global
optimum in complex problems.
It obtain more accurate results
than GA.
ACO It needs less computation time in
simple planning.
ACS It is more suitable than ACO.
It convergences fast.
PSO It has less controller parameters
than IPSO.
IPSO It is appropriate for large-scale
networks.
It is suitable when several HV/
MV substations are considered
beside the feeders, DG and sto-
rage units.
The convergence time is less
than TS/GA.
Hybrid It is helpful when several candi-
GA/ACS date HV/MV substations are taken
into account in addition to the
other facilities.
It is more useful than GA and ACO.
It converges faster than IPSO and
hybrid TS/GA.
Hybrid It is more precise than GA, ES and
TS/GA AIS.
Tuning the controller parameters
is not critical.
Disadvantages
It needs an appropriate initial
solution.
There is no guarantee to obtain
an optimal solution near the
global optimum in complex and
high-dimension problems.
It needs more computation time
than TS for convergence.
It needs to set the controller
parameters appropriately.
It is not suitable when several
candidate HV/MV substations
are taken into account beside the
other facilities.
It needs much more computa-
tion time compared with trajec-
tory-based methods and PSO.
It is not reliable for large-scale
networks.
It needs much more computa-
tion time compared with trajec-
tory-based methods and PSO.
It is not reliable for large-scale
networks.
It is less reliable than GA.
It needs much more computation
time compared with trajectory-
based methods, GA, ES and PSO.
Tuning the controller para-
meters is a challenge.
It can not achieve the global
optimum in distribution plan-
ning even for small networks.
Tuning the controller para-
meters is a challenge.
It has several controller para-
meters for tuning.
It is not helpful for complex
problems.
It is not suitable for large-scale
networks.
Tuning the controller para-
meters is a challenge.
It has many controller para-
meters for tuning.
It has many controller para-
meters for tuning.
It is less accurate than IPSO.
It needs more computation time,
especially for large-scale
networks.
It is not reliable when several
candidate HV/MV substations
are taken into account beside the
other facilities.
431
lation of GA. The main influence is visible in the variance of dis-
tributions. As a result, the intensification of GA has been increased
using TS algorithm.
ES and AIS algorithms approximately obtain the same results
which are provided by GA. However, using the hybrid TS/GA is
preferred compared to GA, ES and AIS. Figs. 12–14 show that un-
like the hybrid TS/GA, the solutions of GA, ES and AIS are more
normally distributed around the median solution, while the so-
lutions of TS/GA has a more congestion around the best solution.
However, the hybrid TS/GA usually needs more computation time
for convergence, as shown in Table 6.
Fig. 9 displays that the proposed hybrid GA/ACS in this paper is
more efficient than conventional ACS algorithm. Also, the pro-
posed GA/ACS is more suitable than SA, GA, ES, AIS and hybrid TS/
GA in the third case. The mutations of GA help the ACS to find the
optimal state of HV/MV substations and then the dimension of
search space decreases in the second phase. As a result, ACS can
find an optimal solution easier with a balance between diversifi-
cation and intensification.
Fig. 7 shows that conventional PSO and the proposed IPSO
obtain the same result in case 1 where the problem is not very
complex. However, the conventional PSO can not achieve the
global optimum in case 2 and 3. The proposed modification for
standard PSO algorithm is useful and applicable, as the proposed
IPSO reduces objective function noticeably, compared with the
conventional PSO in the third case. Figs. 13 and 14 show that the
variance of objective functions in IPSO is significantly less than
that of the conventional PSO. This is because of using the new
controller parameters i.e. q DG , q ST and q FD in IPSO algorithm. These
controller parameters have to be used in complex distribution
planning which needs more intensification. The IPSO algorithm
achieves the best optimal solution in case 2 and 3 among all the
algorithms, as shown in Figs. 8 and 9. On the other hand, the other
new controller parameters i.e. q0FD and α , help the IPSO find the
optimal location of the new HV/MV substation in case 3, where
the conventional PSO fails to find the optimal location of the
substation. Therefore the proposed algorithm is much more ap-
propriate when the distribution planning includes the expansion
of HV/MV substations, MV feeders, DG sources and storage units,
simultaneously. Another advantage of the IPSO is that it converges
faster than the other algorithms such as ES, AIS and TS/GA in case
2 and 3, as shown in Table 6.
Figs. 10 and 11 indicate that the difference between the best
and the worst objective functions, through several runs, are con-
siderable in case 2 and 3. As a result, every algorithm should be
run several times to obtain a reliable optimal solution. Among all,
ACO and SA algorithms result in the largest difference between the
best and the worst functions in case 2 and case 3, respectively.
Hence, they have the worst confidence in a single run. In addition,
it can be seen that the confidence of GA increases when it is
coupled to the TS algorithm.
Regarding the numerical results, it is concluded that the in-
tensified algorithms outperform the other ones. So the in-
tensification is more important than the diversification when the
optimization algorithms are employed for distribution network
planning. This may not be valid for the other applications e.g.
geothermal power plants optimization in [20], where the di-
versification is more vital. The reason is that distribution network
planning is usually a high-dimension and discrete optimization
problem where the decision variables have more critical
situations.
The advantages and disadvantages of all the algorithms are
listed in Table 7, and properties of the algorithms are compared in
Table 8. These tables are helpful to select the most appropriate
algorithm.
432 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434
Table 8
Properties of the algorithms in comparison with each other ( : excellent,
Optimization Optimization
capability in capability in
Algorithm
simple complex
problem problem
TS
SA
GA
ES
AIS
ACO
ACS
PSO
IPSO
Hybrid
GA/ACS
-
Hybrid
TS/GA
6. Conclusions
Recently, a large number of modern optimization algorithms
have been proposed to solve distribution networks expansion
planning problem. Selection and application of the best algorithm
among all algorithms for a typical problem is a challenge in dis-
tribution network planning. As a result, a comparative analysis is
useful and essential to find the best algorithm. In this paper, the
most famous and popular optimization algorithms for distribution
expansion planning are briefly reviewed and analyzed. Then a
general form of distribution network expansion planning problem
is formulated, and consequently the algorithms are compared
experimentally. In addition, some new modifications such as a
new hybrid GA/ACS algorithm and an improved PSO algorithm are
proposed, regarding the especial conditions of distribution net-
work planning.
Numerical studies are performed on three test distribution
networks. The HV/MV substations, MV feeder sections, dispatch-
able DG sources and storage units are considered for distribution
expansion planning, while the objective function comprises the
investment, operation and reliability costs. Some optimal strate-
gies are used for DG and storage units, regarding the three-level
variable load. DG sources and storage units are used not only for
peak cutting but also for reliability enhancement.
The numerical results show that the capability of the meta-
heuristics depends on the dimension and complexity of the pro-
blem, because the number of local optimums increases as the size
of network raises. In particular, the problem becomes more com-
plex when the problem consists of new candidate HV/MV sub-
stations placement and sizing beside the other equipment ex-
pansion planning. Case studies display the ability and efficiency of
the proposed modifications. As a result, every algorithm has some
advantages and disadvantages which are addressed in this paper.
However, the hybrid TS/GA (including the mentioned modifica-
tions in literature) and the Improved PSO algorithm proposed in
this paper are generally the most powerful algorithms for dis-
tribution network expansion planning. The comparison can be
extended to very large-scale distribution network planning as a
future work.
: good, : poor).
Independency Confidence
Computation Tuning
on initial of a single
cost complexity
solution run
References
[1] Bagheri A, Monsef H, Lesani H. Renewable power generation employed in an
integrated dynamic distribution network expansion planning. Electr Power
Syst Res 2015;127:280–96.
[2] Esmaeeli M, Kazemi A, Shayanfar HA, Haghifam MR. Multistage distribution
substations planning considering reliability and growth of energy demand.
Energy 2015;84:357–64.
[3] Viral R, Khatod DK. Optimal planning of distributed generation systems in
distribution system: a review. Renew Sustain Energy Rev 2012;16:5146–65.
[4] Paliwal P, Patidar NP, Nema RK. Planning of grid integrated distributed
generators: a review of technology, objectives and techniques. Renew Sus-
tain Energy Rev 2014;40:557–70.
[5] Sedghi M, Ahmadian A, Golkar MA. Optimal storage planning in active dis-
tribution network considering uncertainty of wind power distributed gen-
eration. IEEE Trans Power Syst 2016;31:304–16.
[6] Khator SK, Leung LC. Power distribution planning: a review of models and
issues. IEEE Trans Power Syst 1997;12:1151–9.
[7] Crawford DM, Holt Jr SB. A mathematical optimization technique for locating
and sizing distribution substations, and deriving their optimal service areas.
IEEE Trans Power Appar Syst 1975;94:230–5.
[8] Wall DL, Thompson GL, Northcote-Green JED. An optimization model for
planning radial distribution networks. IEEE Trans Power Appar Syst
1979;98:1061–8.
[9] Thompson GL, Wall DL. A branch and bound model for choosing optimal
substation locations. IEEE Trans Power Appar Syst 1981;100:2683–8.
[10] Gonen T, Foote BL. Distribution-system planning using mixed-integer pro-
gramming. In: Proceeding of the IEEE generation, transmission and dis-
tribution. vol. 128; 1981. p. 70–79.
[11] Glover F. Future paths for integer programming and links to artificial in-
telligence. Comput Oper Res 1986;13:533–49.
[12] Keypour R, Sharifzadeh H. A comparative study on performance of meta-
heuristics optimization methods for optimal var sizing and allocation. In:
Proceeding of IEEE region 8 international conference on computational
technologies in electrical and electronics engineering (SIBIRCON). IEEE; 2010.
p. 777–782.
[13] Eghbal M, Yorino N, Zoka Y, ElAraby EE. Application and comparison of
metaheuristic techniques to reactive power planning problem. IEEJ Trans
Electr Electron Eng 2008;3:721–30.
[14] Reche-López P, Ruiz-Reyes N, García Galán S, Jurado F. Comparison of me-
taheuristic techniques to determine optimal placement of biomass power
plants. Energy Convers Manag 2009;50:2020–8.
[15] Kannan S, Slochanal SMR, Padhy NP. Application and comparison of meta-
heuristic techniques to generation expansion planning problem. IEEE Trans
Power Syst 2005;20:466–75.
[16] Kannan SMRSS, Slochanal SMR, Baskar S, Murugan P. Application and com-
parison of metaheuristic techniques to generation expansion planning in the
partially deregulated environment. IET Gener Transm Distrib 2007;1:111–8.
[17] Fathima AH, Palanisamy K. Optimization in microgrids with hybrid energy
systems – a review. Renew Sustain Energy Rev 2015;45:431–46.
[18] Sinha S, Chandel SS. Review of recent trends in optimization techniques for
solar photovoltaic-wind based hybrid energy systems. Renew Sustain Energy
 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434
Rev 2015;50:55–769.
[19] Gamarra C, Guerrero JM. Computational optimization techniques applied to
microgrids planning: a review. Renew Sustain Energy Rev 2015;48:413–24.
[20] Clarke J, McLay L, McLeskey Jr JT. Comparison of genetic algorithm to particle
swarm for constrained simulation-based optimization of a geothermal
power plant. Adv Eng Inform 2014;28:81–90.
[21] Askarzadeh A. Comparison of particle swarm optimization and other meta-
heuristics on electricity demand estimation: a case study of Iran. Energy
2014;72:484–91.
[22] Shi R, Cui C, Su K, Zain Z. Comparison study of two meta-heuristic algorithms
with their applications to distributed generation planning. Energy Procedia
2011;12:245–52.
[23] Temraz HK, Quintana VH. Distribution system expansion planning models:
an overview. Electr Power Syst Res 1993;26:61–70.
[24] Georgilakis PS, Hatziargyrious ND. A review of power distribution planning
in the modern power systems era: models, methods and future research.
Electr Power Syst Res 2015;121:89–100.
[25] Rezaee-Jordehi A. Optimization of electric distribution systems: a review.
Renew Sustain Energy Rev 2015;51:1088–100.
[26] Aoki K, Nara K, Satoh T, Kitagawa M, Yamanaka K. New approximate opti-
mization method for distribution system planning. IEEE Trans Power Syst
1990;5:126–32.
[27] Nara K, Satoh T, Kuwabara H, Aoki K, Kitagawa M, Ishihara T. Distribution
systems expansion planning by multi-stage branch exchange. IEEE Trans
Power Syst 1992;7:208–14.
[28] Miguez E, Cidras J, Diaz-Dorado E, Garcia-Dornelas JL. An improved branch-
exchange algorithm for large-scale distribution network planning. IEEE Trans
Power Syst 2002;17:931–6.
[29] Parada V, Ferland JA, Arias M, Schwarzenberg P, Vargas LS. Heuristic de-
termination of distribution trees. IEEE Trans Power Deliv 2010;25:861–9.
[30] Ramirez-Rosado IJ, Dominguez-Navarro JA. New multiobjective Tabu search
algorithm for fuzzy optimal planning of power distribution systems. IEEE
Trans Power Syst 2006;21:224–33.
[31] Junior BRP, Cossi AM, Contreras J, Mantovani JRS. Multiobjective multistage
distribution system planning using Tabu search. IET Gener Transm Distrib
2014;8:35–45.
[32] Golshan MH, Arefifar SA. Distributed generation, reactive sources and net-
work-configuration planning for power and energy-loss reduction. IET Gener
Transm Distrib 2006;153:127–36.
[33] Parada V, Ferland JA, Arias M, Daniels K. Optimization of electrical dis-
tribution feeders using simulated annealing. IEEE Trans Power Deliv
2004;19:1135–41.
[34] Nahman JM, Peric DM. Optimal planning of radial distribution networks by
simulated annealing technique. IEEE Trans Power Syst 2008;23:790–5.
[35] Popovic ZN, Kerleta VD, Popovic DS. Hybrid simulated annealing and mixed
integer linear programming algorithm for optimal planning of radial dis-
tribution networks with distributed generation. Electr Power Syst Res
2014;108:211–22.
[36] Rivas-Davalos F, Irving MR. An efficient genetic algorithm for optimal large-
scale power distribution network planning. In: Proceeding of the IEEE power
technology conference. vol. 3; 2003. p. 1–5.
[37] Rivas-Dávalos F, Irving MR. The edge-set encoding in evolutionary algo-
rithms for power distribution network planning problem (part I): single-
objective optimization planning. In: Proceeding of the IEEE electronics, ro-
botics and automotive mechanics conference. vol. 1; 2006. p. 203–208.
[38] Taroco C, Carrano EG, Neto OM, Takahashi RHC. A faster genetic algorithm for
substation location and network design of power distribution systems. In:
Proceeding of the IEEE evolutionary computation congress (CEC). IEEE; 2012.
p. 1–6.
[39] Haghifam MR, Shahabi M. Optimal location and sizing of HV/MV substations
in uncertainty load environment using genetic algorithm. Electr Power Syst
Res 2002;63:37–50.
[40] Jimenez-Estevez GA, Vargas LS, Marianov V. Determination of feeder areas
for the design of large distribution networks. IEEE Trans Power Deliv
2010;25:1912–22.
[41] Carrano EG, Soares LA, Takahashi RH, Saldanha RR, Neto OM. Electric dis-
tribution network multiobjective design using a problem-specific genetic
algorithm. IEEE Trans Power Deliv 2006;21:995–1005.
[42] Kong T, Cheng H, Hu Z, Yao L. Multiobjective planning of open-loop MV
distribution networks using ComGIS network analysis and MOGA. Electr
Power Syst Res 2009;79:390–8.
[43] Ouyang W, Cheng H, Zhang X, Yao L, Bazargan M. Distribution network
planning considering distributed generation by genetic algorithm combined
with graph theory. Electr Power Compon Syst 2010;38:325–39.
[44] Carrano EG, Cardoso RTN, Takahashi RHC, Fonseca CM, Neto OM. Power
distribution network expansion scheduling using dynamic programming
genetic algorithm. IET Gener Transm Distrib 2008;2:444–55.
[45] Nazar MS, Haghifam MR, Nazar M. A scenario driven multiobjective pri-
mary–secondary distribution system expansion planning algorithm in the
presence of wholesale–retail market. Int J Electr Power Syst 2012;40:29–45.
[46] Cossi AM, da Silva LGW, Lazaro RAR, Mantovani JRS. Primary power dis-
tribution systems planning taking into account reliability, operation and
expansion costs. IET Gener Transm Distrib 2012;6:274–84.
[47] Celli G, Mocci S, Pilo F, Loddo M. Optimal integration of energy storage in
distribution networks. In: Proceeding of power technology conference. IEEE;
2009. p. 1–7.
433
[48] Naderi E, Kiaei I, Haghifam MR. NaS technology allocation for improving
reliability of DG-enhanced distribution networks. In: Proceeding of the IEEE
11th international conference on probabilistic methods applied to power
systems (PMAPS). IEEE; 2010. p. 148–153.
[49] Falaghi H, Singh C, Haghifam MR, Ramezani M. DG integrated multistage
distribution system expansion planning. Int J Electr Power Syst
2011;33:1489–97.
[50] Carvalho PMS, Ferreira LAFM. Distribution quality of service and reliability
optimal design: individual standards and regulation effectiveness. IEEE Trans
Power Syst 2005;20:2086–92.
[51] Pilo F, Celli G, Mocci S, Soma GG. Multi-objective programming for optimal
DG integration in active distribution systems. In: Proceeding of the IEEE
power and energy society general meeting. IEEE; 2010. p. 1–7.
[52] Naderi E, Seifi H, Sepasian MS. A dynamic approach for distribution system
planning considering distributed generation. IEEE Trans Power Deliv
2012;27:1313–22.
[53] Borges CLT, Martins VF. Multistage expansion planning for active distribution
networks under demand and distributed generation uncertainties. Int J
Electr Power Syst 2012;36:107–16.
[54] Borges CL, Falcao DM. Optimal distributed generation allocation for relia-
bility, losses, and voltage improvement. Int J Electr Power Syst 2006;28:413–
20.
[55] Najafi S, Hosseinian SH, Abedi M, Vahidnia A, Abachezadeh S. A framework
for optimal planning in large distribution networks. IEEE Trans Power Syst
2009;24:1019–28.
[56] Sattianadan D, Sudhakaran M, Dash SS, Vijayakumar K, Ravindran P. Optimal
placement of DG in distribution system using genetic algorithm. Swarm Evol
Memetic Comput 2013;8298:639–47.
[57] Diaz-Dorado E, Cidrás J, Miguez E. Application of evolutionary algorithms for
the planning of urban distribution networks of medium voltage. IEEE Trans
Power Syst 2002;17:879–84.
[58] Ouyang W, Cheng H, Zhang B, Yaoc L. Distribution network planning method
considering distributed generation for peak cutting. Energy Convers Manag
2010;51:2394–401.
[59] Bouchard DE, Salama MMA, Chikhani AY. Optimal distribution feeder routing
and optimal substation sizing and placement using evolutionary strategies.
In: Proceeding of electrical and computer engineering conference. IEEE;
1994. p. 661–664.
[60] Diaz-Dorado E, Pidre JC, García EM. Planning of large rural low-voltage
networks using evolution strategies. IEEE Trans Power Syst 2003;18:1594–
600.
[61] Jimenez-Estevez G, Vargas LS, Palma-Behnke R. An evolutionary approach for
the greenfield planning problem in distribution networks. In: Proceeding of
the IEEE international joint conference on neural networks. IEEE; 2007. p.
1744–1749.
[62] Souza BB, Carrano EG, Neto OM, Takahashi RHC. Immune system memetic
algorithm for power distribution network design with load evolution un-
certainty. Electr Power Syst Res 2011;81:527–37.
[63] Carrano EG, Guimarães FG, Takahashi RH, Neto OM, Campelo F. Electric
distribution network expansion under load-evolution uncertainty using an
immune system inspired algorithm. IEEE Trans Power Syst 2007;22:851–61.
[64] Muhtazaruddin MN, Jamian JJ, Fujita G, Baharudin MA, Wazir MW, Mokhlis
H. Distribution network loss minimization via simultaneous distributed
generation coordination with network reconfiguration. Arab J Sci Eng
2014;39:1–11.
[65] Junjie MA, Yulong W, Yang L. Size and location of distributed generation in
distribution system based on immune algorithm. Syst Eng Procedia
2012;4:124–32.
[66] Alvarado JM, Alvarado EV, Arevalo MA, Quituisaca SP, Gomez JF, Oliveira-De
Jesus D. Ant colony systems application for electric distribution network
planning. In: Proceeding of the 15th international conference on intelligent
system applications to power systems. IEEE; 2009. p. 1–6.
[67] Jia-mei Z, Ren HE. A new idea for power distribution network planning using
ant colony optimization. In: Proceeding of the 5th international conference
on wireless communications, networking and mobile computing. IEEE; 2009.
p. 1–4.
[68] Jia-mei Z, Hong-e R. Improved ant colony algorithm for power distribution
network planning. In: Proceeding of the 2nd international conference on
advanced computer control (ICACC). IEEE. vol. 1; 2010. p. 357–360.
[69] Falaghi H, Haghifam MR. ACO based algorithm for distributed generation
sources allocation and sizing in distribution systems. In: Proceeding of IEEE
lausanne power technology. IEEE; 2007. p. 555–560.
[70] Sun W, Shang W. Research on distribution network optimization using im-
proved ant colony algorithm. In: Proceeding of the sixth world congress on
intelligent control and automation. vol. 1; 2006 p. 2599–2603.
[71] Gomez JF, Khodr HM, De Oliveira PM, Ocque L, Yusta JM, Villasana R, et al.
Ant colony system algorithm for the planning of primary distribution cir-
cuits. IEEE Trans Power Syst 2004;19:996–1004.
[72] Favuzza S, Graditi G, Ippolito MG, Sanseverino ER. Optimal electrical dis-
tribution systems reinforcement planning using gas micro turbines by dy-
namic ant colony search algorithm. IEEE Trans Power Syst 2007;22:580–7.
[73] Cheng S, Chen MY, Wai RJ, Wang FZ. Optimal placement of distributed gen-
eration units in distribution systems via an enhanced multi-objective particle
swarm optimization algorithm. J Zhejiang Univ Sci C 2014;15:300–11.
[74] Ziari I, Ledwich G, Ghosh A. Optimal integrated planning of MV–LV dis-
tribution systems using DPSO. Electr Power Syst Res 2011;81:1905–14.
434 M. Sedghi et al. / Renewable and Sustainable Energy Reviews 66 (2016) 415–434
[75] Ameli A, Bahrami S, Khazaeli S, Haghifam MR. A multiobjective particle
swarm optimization for sizing and placement of DGs from DG owner's and
distribution company's viewpoints. IEEE Trans Power Deliv 2014;29:1831–
40.
[76] Samper ME, Vargas A. Investment decisions in distribution networks under
uncertainty with distributed generation – Part II: implementation and re-
sults. IEEE Trans Power Syst 2013;28:2341–51.
[77] Ganguly S, Sahoo NC, Das D. Mono-and multi-objective planning of electrical
distribution networks using particle swarm optimization. Appl Soft Comput
2011;11:2391–405.
[78] Bhumkittipich K, Phuangpornpitak W. Optimal placement and sizing of
distributed generation for power loss reduction using particle swarm opti-
mization. Energy Procedia 2013;34:307–17.
[79] Ziari I, Ledwich G, Ghosh A, Platt G. Optimal distribution network re-
inforcement considering load growth, line loss, and reliability. IEEE Trans
Power Syst 2013;28:587–97.
[80] AlRashidi MR, AlHajri MF. Proper planning of multiple distributed generation
sources using heuristic approach. In: Proceeding of 4th international con-
ference on modeling, simulation and applied optimization (ICMSAO). IEEE;
2011. p. 1–5.
[81] El-Zonkoly AM. Optimal placement of multi-distributed generation units
including different load models using particle swarm optimization. Swarm
Evol Comput 2011;1:50–9.
[82] Kansal S, Kumar V, Tyagi B. Optimal placement of different type of DG
sources in distribution networks. Int J Electr Power Syst 2013;53:752–60.
[83] Ziari I, Ledwich G, Ghosh A, Platt G. Integrated distribution systems planning
to improve reliability under load growth. IEEE Trans Power Deliv
2012;27:757–65.
[84] Sahoo NC, Ganguly S, Das D. Simple heuristics-based selection of guides for
multi-objective PSO with an application to electrical distribution system
planning. Eng Appl Artif Intell 2011;24:567–85.
[85] Aman MM, Jasmon GB, Bakar AHA, Mokhlis H. A new approach for optimum
DG placement and sizing based on voltage stability maximization and
minimization of power losses. Energy Convers Manag 2013;70:202–10.
[86] Ahmadigorji M, Amjady N. A new evolutionary solution method for dynamic
expansion planning of DG-integrated primary distribution networks. Energy
Convers Manag 2014;82:61–70.
[87] AlRashidi MR, AlHajri MF. Optimal planning of multiple distributed gen-
eration sources in distribution networks: a new approach. Energy Convers
Manag 2011;52:3301–8.
[88] Sedghi M, Aliakbar-Golkar M, Haghifam MR. Distribution network expansion
considering distributed generation and storage units using modified PSO
algorithm. Int J Electr Power Syst 2013;52:221–30.
[89] Wang C, Wang S. The automatic routing system of urban mid-voltage dis-
tribution network based on spatial GIS. In: Proceeding of international
conference on power system technology. IEEE. vol. 2; 2004. p. 1827–1832.
[90] Mori H, Yoshida T. An efficient multi-objective memetic algorithm for un-
certainties in distribution network expansion planning. In: Proceeding of
IEEE power & energy society general meeting. IEEE; 2009. p. 1–6.
[91] Mori H, Yoshida T. Probabilistic distribution network expansion planning
with multi-objective memetic algorithm. In: Proceeding of the electric
power conference (EPEC). IEEE; 2008. p. 1–6.
[92] Burke EK, Kendall G. Search methodologies: Introductory Tutorials in Opti-
mization and Decision Support Techniques. New York: Springer-Verlag;
2005.
[93] Dréo J. Metaheuristics for Hard Optimization: Methods and Case Studies.
Berlin, Germany: Springer-Verlag; 2006.
[94] Neimane V. On development planning of electricity distribution networks
[Ph.D. thesis].Stockholm, Sweden: Royal Institute of Technology; 2001.
[95] Kirkpatrick S, Gelatt CD, Vecchi MP. Optimization by simulated annealing.
Science 1983;220:671–80.
[96] Holland JH. Adaptation in Natural and Artificial Systems. USA: The University
of Michigan Press; 1975.
[97] Rudolph G. Convergence analysis of canonical genetic algorithms. IEEE Trans
Neural Netw 1994;5:96–101.
[98] Kilic O, Nguyen QM. Application of artificial immune system algorithm to
electromagnetics problems. Prog Electro Res 2010;20:1–17.
[99] Dorigo M, Stutzle T. Ant colony optimization. USA, Cambridge: MIT Press;
2004.
[100] Moscato P. On evolution, search, optimization, genetic algorithms and mar-
tial arts: towards memetic algorithms, Technical Report No. C3P 826. Caltech
Concurrent Computation Program. Pasadena, CA: California Institute of
Technology;1989.
[101] Poonpun P, Jewell WT. Analysis of the cost per kilowatt hour to store elec-
tricity. IEEE Trans Energy Convers 2008;23:529–34.
[102] Sedghi M, Golkar MA. Analysis and comparison of load flow methods for
distribution networks considering distributed generation. Int J Smart Electr
Eng 2012;1:27–32.
[103] Li Q, Choi SS, Yuan Y, Yao DL. On the determination of battery energy storage
capacity and short-term power dispatch of a wind farm. IEEE Trans Sustain
Energy 2011;2:148–58.
[104] Sedghi M. Distribution network expansion using modern optimization al-
gorithms [M.Sc. thesis]. Tehran, Iran: K. N. Toosi University of Technology;
2009.
[105] Siahkali H, Roshanfekr R. Distribution network planning using supplying area
and PSO algorithms. J Iran Assoc Electr Electron Eng 2005;2:17–30.