Renewable Energy 37 (2012) 213e225

Contents lists available at ScienceDirect

Renewable Energy
journal homepage: www.elsevier.com/locate/renene

Distribution feeder reconfiguration considering fuel cell/wind/photovoltaic power

plants
Taher Niknam a, *, Abdollah Kavousi Fard b, Alireza Seifi b
a
Department of Electronic and Electrical Engineering, Shiraz University of Technology, Shiraz, Iran
b
Department of Electrical and Computer Engineering, Shiraz university, Shiraz, Iran

a r t i c l e i n f o a b s t r a c t

Article history: In this paper a new method based on multi-objective modified Honey Bee Mating Optimization
Received 8 December 2010 (MHBMO) algorithm is presented to investigate the DFR problem with the consideration effect of the
Accepted 13 June 2011 renewable energy sources (RESs). In the proposed algorithm, a new mating process is suggested to
Available online 20 July 2011
overcome the shortcomings which exist in the original HBMO. During the optimization process, the
proposed algorithm finds a set of non-dominated (Pareto) optimal solutions which are stored in an
Keywords:
external memory called repository. Also a fuzzy clustering technique is utilized to handle the size of the
Modified Honey Bee Mating Optimization
repository in the specified limits. Moreover, a fuzzy-based decision maker is adopted to select the ‘best
(MHBMO)
Distribution feeder reconfiguration (DFR)
compromised’ solution among the non-dominated optimal solutions of multi-objective optimization
Multi-objective optimization problem. In order to evaluate the feasibility and effectiveness of the proposed algorithm, two standard
Fuel cell distribution systems are used as case studies.
Wind energy
2011 Elsevier Ltd. All rights reserved.
Photovoltaic (PV)

1. Introduction Generally, the structure of the distribution systems is con-

In recent years, the necessity of concentration on the air
pollution and global warming as two of the greatest threats to the
health of the human beings and animals is conceived more than
before. Energy insecurity and rising prices of conventional energy
sources are also considered as major threats to the economic and
political stability [1]. Therefore in recent years, a growing trend for
finding new environmentally friendly replaceable sources of
energy have been made to reduce the harmful effects of the
conventional energy sources; that is to find alternative non-
traditional/renewable energies. Low emission, high efficiency,
compatibility with other modular subsystems and suitable power
quality and reliability are some benefits of RESs [2]. In addition, in
recent years a wide range of researches have been made to inves-
tigate the use of different kinds of RESs like wind energy, photo-
voltaic cells, biogas energy, fuel cells, combined heat and power
systems (CHP), etc. in the voltage level of distribution [3,4].
Therefore it is important to study the effect of RESs on the distri-
bution networks, in particular on the DFR problem.
Abbreviations: MHBMO, Modified Honey Bee Mating Optimization; DFR,
distribution feeder reconfiguration; PV, photovoltaic; MOP, multi-objective Opti-
mization; MDFR, multi-objective DFR; RES, renewable energy source.
* Corresponding author.
E-mail address: niknam@sutech.ac.ir (T. Niknam).
structed as radial networks so that to have appropriate and suitable
protection coordination. In addition, the radiality of the network
gives this benefit that it is possible to change the structure of the
network by the use of automatic or manual switches in a way that
the goal of supplying all consumers, minimizing total power losses
and improving power quality would be achieved satisfactorily. On
the other hand, the discrete nature of the switches and the radiality
of the network are big barriers to get use of the traditional opti-
mization methods in the multi-objective distribution feeder
reconfiguration (MDFR) problem. In order to simplify the MDFR
problem, traditional methods have proposed converting the multi-
objective optimization problem to a single objective optimization
problem [5,6]. However with this method, much computational
resources are needed to find the multiple optimal solutions.
In recent years, so many attempts have been made by the
researchers to find an appropriate approach to solve the MDFR
problem. In Ref. [7] a new method based on artificial neural network
is proposed to recognize the configuration of the network corre-
sponding to different load levels. In Ref. [8], Taylor and Lubkeman
have suggested a new expert system based on heuristic methods to
reduce the entire search space effectively. In [9], Chiou et al pre-
sented a new method based on variable scaling hybrid differential
evolution (VSHDE) to investigate the DFR problem while the total
power losses has been considered as the objective function.
Kashem et al. have suggested a new method based on dis-
tance measurement to locate the loops in the system and then

0960-1481/$ e see front matter
2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.renene.2011.06.017
214 T. Niknam et al. / Renewable Energy 37 (2012) 213e225

Nomenclature Ri resistance of ith branch (U)

Ii current of ith branch (A)
X state variables vector PFC,i power of the ith FC (kW)
n number of state variables PPV,i power of the ith PV unit (kW)
NFC number of FC power units PWind,i power of the ith wind unit (kW)
NPV number of PV power units a speed reduction factor
NWind number of wind power units g random value in the range of [0,1]
Ii,br current of the ith branch F1 random value in the range of [0,1]
Psub active power production of substation (kW) F2 random value in the range of [0,1]
hi electrical efficiency of the ith FC F3 random value in the range of [0,1]
PLRi part load ratio of the ith FC prob(D) probability of adding the sperm of drone D to the
CFC,i cost of power generated by the ith FC unit ($) queen spermatheca
CPV,i cost of power generated by the ith PV unit ($) Xmut improved brood
CWind,i cost of power generated by the ith wind unit ($) Irand,SP random integer in the range [1,NSP]
NOxFC,i nitrogen oxide pollutants of the ith FC power unit (kg/ fi(X) the ith objective function
kW h) Fi(X) values of the augmented fi(X)
SO2FC,i sulphur oxide pollutants of the ith FC power unit (kg/ Neq number of equality constraints of the DFR problem
kW h) Nueq number of inequality constraints of the DFR problem
NOxPV,i nitrogen oxide pollutants of the ith PV power unit (kg/ L1 penalty factor
kW h) L2 penalty factor
SO2PV,i sulphur oxide pollutants of the ith PV power unit (kg/ Nipop number of the bees
kW h) Squeen queen speed
NOxWind,i nitrogen oxide pollutants of the ith wind power unit fmin
i lowest limit of ith objective function
(kg/kW h) fmax
i highest limit of ith objective function
SO2Wind,i sulphur oxide pollutants of the ith wind power unit D(f) absolute difference between the fitness of the drone
(kg/kW h) and the queen
NOxGrid nitrogen oxide pollutants of the grid (kg) NSP size of the queen spermatheca
SO2Grid sulphur oxide pollutants of the grid (kg) V(I)FC output voltage (current) of fuel cell
Vmax maximum voltage magnitude of ith bus (V) PtotalFC total power output of fuel cell
Vmin minimum voltage magnitude of ith bus (V) KP proportional coefficient
mfi(X) membership function of the ith objective function E0 potential of fuel cell in thermodynamic equilibrium
D drone P(Q)I-G active (reactive) power output of induction generator
Xbrood,j the jth brood VbusI
G voltage of induction generator terminal bus
Sp queen spermatheca matrix s slip of induction generator speed
j a function that converts wind speed to WT power
output Subscripts
w1,2,3,max WT cut-in, rated, cut-off and maximum wind speed, FC fuel cell
respectively PV photovoltaic
Gt solar radiation br branch
Gstd solar radiation in the standard environment set as SP spermatheca
1000 W/m2 max maximum
Rc certain radiation point set as W/m2 min minimum
P(Q)in input active (reactive) power of the synchronous mut mutation
generator(equal to the output active (reactive) power rand random
of RES connected to synchronous generator) eq equality
Eq no-load voltage (constant value) ueq inequality
Xd synchronous reactance ipop initial population
VbusSY voltage of synchronous generator terminal bus Sub substation
Nb number of branches

a switching plan is utilized to improve the load balancing in the
corresponding loop [10]. In Ref. [11], Debaprya has proposed a new
approach based on fuzzy theory to solve the MDFR problem suffi-
ciently. In order to solve the load balancing problem, Baran and Wu
have suggested a Mixed Integer Programming (MIP) based method
while the total power losses as the objective function is reduced
simultaneously [12]. In Ref. [13], Zhou et al. have suggested a new
algorithm to support the load balancing and service restoration in
the two feeder networks simultaneously.
In Refs. [14] and [15] Niknam has proposed a new method based
on evolutionary algorithms and transformation of MDFR problem
to solve a single objective optimization problem. As mentioned
before, in this method, so much computational resources and many
runs are needed to find the optimal solutions and consequently the
effectiveness of the algorithm will be decreased. In Ref. [16],
Niknam has proposed a new hybrid method to solve the MDFR
problem while the proposed method results in a single optimal
solution. Thus here again, several satisfying solutions which can be
considered as Pareto optimal solutions are neglected.
In fact the discrete nature of the tie and sectionalizing switches
along with the consideration effect of the RESs make the MDFR
problem a nonlinear complex multi-objective optimization.
Conventional and classical methods such as linear programming,
mixed integer programming, quadratic programming, etc., can be
 T. Niknam et al. / Renewable Energy 37 (2012) 213e225 215

used to solve this problem; however, these methods reach local 2.2. Objective functions
minimum and some of them cannot handle the integer problems
[16]. Evolutionary methods owing to independence on the type of - Minimization of the power losses (f1): power losses objective
objectives function and constraints can be used to solve the MDFR function is evaluated as:
problem. Nevertheless, some evolutionary methods neither converge
to global minima nor have short convergence time. Recently, a new X
Nbr
f1 ðXÞ ¼ Ploss ðXÞ ¼ Ri  jIi j2 (1)
optimization algorithm based on honey bee mating has been used to
i¼1
solve difficult optimization problems [14]. The original HBMO often
converges to local optima. In order to avoid this shortcoming, where Tiei and Swi are the states of the ith tie switch and sec-
a new method is proposed to improve the mating processing. tionalizing switch respectively. Also in this paper, 0 and 1 are used
Therefore in this paper a new method based on MHBMO algo- to show the states of open and close, respectively.
rithm is proposed to investigate the MDFR problem while the
effects of the RESs are considered simultaneously. In order to - Minimization of the voltage deviation of the buses (f2):
improve the algorithm, the mating process is corrected so that to voltage deviation objective function is evaluated as:
overcome the two main shortcomings which exist in the traditional
HBMO; that is dependency of the algorithm on the initial param- f2 ðXÞ ¼ devðXÞ ¼ max½j1
Vmin j and j1
Vmax j (2)
eters and, the possibility of being trapped in local optima. During
the optimization process, the set of Pareto optimal solutions which - Minimization of the total cost (f3): the cost objective function
are found by the algorithm would be stored in an external memory is the summation of the cost of power production by the grid
called repository. In addition, to find the ‘best compromised’ solu- and the RESs. The grid cost can be evaluated as follows:
tion among the Pareto optimal solutions, a fuzzy-based mechanism
is introduced and applied to the set Pareto solutions set. Therefore
Csub ¼ price  Psub (3)
the main contributions of this paper are as follows: (1) proposing
a new modification process to improve the performance of the The cost of FC power sources is evaluated by [17]:
HBMO algorithm, (2) the use of HBMO algorithm to solve the MDFR
problem for the first time and (3) the use of the idea of the non- PFC;i
dominated solutions in the MDFR optimization problem. Finally, CFC;i ¼ 0:04$=kW h 
hi
to see the feasibility and superiority of the proposed method, two Pgi
test systems are used as case studies. PLRi ¼
Pmaxi
if PLRi < 0:050hi ¼ 0:2716
if PLRi  0:050hi ¼ 0:9033 PLR5i
2:9996 PLR4i þ 3:6503 PLR3i
2. Problem formulation

2:0704 PLR2i þ 0:3747 (4)
2.1. Decision variables The cost of PV and wind units is evaluated similarly by Eq. (5).
The cost of generation of each kW h is a function of three param-

 eters [18]: (I) investment cost, (II) operation and maintenance cost,
X ¼ Tie; Sw; Pg
(III) fuel cost. Therefore the cost value is evaluated as follows [18]:

 CPV;i ¼ a þ b  PPV;i
Sw ¼ Sw1 ; Sw2 ; Sw3 .SwNSw
CWind;i ¼ a þ b  PWind;i
Capital costð$=kWÞ*CapacityðkWÞ*Gr (5)
Tie ¼ ½Tie1 ; Tie2 ; Tie3 ; .; Tietie  a ¼
Life timeðYearÞ*365*24*LF
b ¼ Fuel costð$=kWÞ þ O&M Costð$=kW hÞ
Pg ¼ ½PFC ; PPV ; PWind 
Therefore the total cost is as follows:

 X
NFC X
NPV X
Nwind
PPV ¼ PPV;1 ; PPV;2 ; .; PPV;NPV f3 ðXÞ ¼ Cost ¼ CFC;i þ CPV;i þ Cwind;i þ Csub (6)
i¼1 i¼1 i¼1


PFC ¼ PFC;1 ; PFC;2 ; .; PFC;NFC - Minimizing the total emission produced (f4): the total
emission of the grid and the RESs is as follows [19]:


PWind ¼ PWind;1 ; PWind;2 ; .; PWind;NWind
P
NFC P
NPV NP
Wind
f4 ðXÞ ¼ Emission ¼ EFC; i þ EPV;i þ EWind;i þEGrid

 i¼1 i¼1 i¼1
Xqueen ¼ xq;1 ; xq;2 ; .; xq;N ; N ¼ 2*Ntie þ NFC þ NPV þ NWind  kg MW h
1
EType;i ¼ NOxType;i þSO2Type;i ¼ K1Type;i þK2Type;i PType;i

  lb MW h
1
Spi ¼ si;1 ; si;2 ; .; si;N ; i ¼ 1; 2; .; NSp EGrid ¼ NOxGrid þSO2Grid ¼ K1Grid þK2Grid Psub

 (7)
Xmut ¼ xmut;1 ; xmut;2 ; .; xmut;N
where Type can be: FC, PV or Wind. The related values of these

 parameters are shown in Table 2. It should be noted that the values
Xbrood ¼ xbr;1 ; xbr;2 ; .; xbr;N of K1 and K2 related to Wind and PV units are supposed to be zero in
216 T. Niknam et al. / Renewable Energy 37 (2012) 213e225

this paper. However in some investigations, these two variables (K1 - Bus voltage constraints:
and K2) are considered to get nonzero values. In fact in these
studies, the contamination related to the process of the construc- Vmin  V  Vmax (12)
tion of the RESs is considered in the evaluations.
3. Modeling of the RESs
2.3. Limits and constraints
3.1. Fuel cell
- Limits associated with distribution lines:
Among different kinds of RESs, FC has developed into one of the
most important and influential devices in the current power
Line
Pij;min < PijLine < Pij;max
Line
(8) systems especially in the voltage level of distribution. Simple
structure and operation [8], high efficiency while having the ability
of following load variations [3,4], high reliability and cleanness
- Distribution power flow equations: [20,21] are some of the attractive characteristics of this type of
RESs. In recent years, the knowledge of construction of FCs has
improved profitably in a way that the efficiency of some kinds of
NP
bus   the FCs has become near twice that of the conventional power
Pi ¼ Vi Vj Yij cos qij
di þ dj
i¼1 plants [22].
(9)
P
Nbus  
Qi ¼ Vi Vj Yij sin qij
di þ dj
i¼1 3.2. Photovoltaic

One of the huge sources of energy which can support a big

In fact, this equation is a load flow equation which can play the
role of an equality constraint.
- Keeping the radiality of the network: since the distribution
networks are assumed as radial networks, thus during the
reconfiguration, this quality of the network should be
preserved carefully. Each loop is consisted of a sectionalizing
switch and a tie switch simultaneously. Each time that a loop is
formed in the network; one of the switches should be opened
in a way that the radiality of the network would be preserved.
- Feeder current limitation [16]:
jIf;i j  Ifmax
;i ; i ¼ 1; 2; .; Nf
- RESs constraints on active power production are:
pmin;FC;i  pFC;i  pmax;FC;i ; pmin;PV;i  pPV;i  pmax;PV;i ;
pmin;Wind;i  pWind;i  pmax;Wind;i
part of the required energy of the human beings is solar energy.
In order to get use of this ever lasting source of energy, PV as an
inevitable technology in the area, has attracted the attention of
many researchers in recent years [1]. In fact by the use of arrays
of cells which are constructed of some kinds of semiconductors
that exhibit the photovoltaic effect, the solar radiation can
be converted into direct current (DC) electricity sufficiently
[1,23].
3.3. Wind
With the developments which has happened in the technology
of the wind power generations in recent years, there has been an
(10)
increasing interest to get use of the wind farms which are con-
nected directly to the power system. Indeed, the kinetic energy
which exists in the wind airflows can be changed into electricity
power by the use of wind farms. Especially, the importance of use
of this kind of energy in the regions like high altitude and open
flat areas which have more constant and regular winds is more
ð11Þ
evident.

Fig. 1. Modeling of RESs. a) PQ model with simultaneous three-phase control. b) PQ model with independent three-phase control. c) PV model with simultaneous three-phase
control. d) PV model with independent three-phase control.
T. Niknam et al. / Renewable Energy 37 (2012) 213e225 217

Fig. 2. Block diagram of MHBMO algorithm.

218 T. Niknam et al. / Renewable Energy 37 (2012) 213e225

3.4. Modeling of the RESs
In order to have an appropriate model for the RESs, two subjects
should be considered precisely: (I) the operation of the RESs, (II) the
type of connection of the RES to the grid.
- Fuel cells (FCs): the connection of these chemical sources of
energy into the grid is done via power electronic devices based
on rectifier and inverter. The output power can be modeled
as [24]:
PFC ¼ VFC  IFC
Considering the inner resistance for the FC (R), the total power
produced by the FC is as follows:
2 2
PtotalFC ¼ PFC þ RIFC ¼ VFC IFC þ RIFC
grid directly. In the variable speed type, firstly, by the use of
synchronous or a double-fed induction generator the wind
energy is converted to the electrical energy and then by the use
of power electronic devices, the generated electrical energy is
changed to the grid compatible AC power [25]:
8
< 0   wt  w1 or wt  w3
t
PWT ¼ j wt w1  wt  w2 t ¼ 1; 2; .; T (17)
:
PWT w2  wt  w3
With the above explanation, it can be deduced that the
connection of the RESs to the grid is done via: 1) power electronic
(13) devices and/or 2) synchronous/asynchronous electric machines. In
the first case (via power electronic devices), the RESs should be
modeled with regard to the characteristics of the interface control
circuit. In the second case (direct connection via electric machines),
the RESs should be modeled according to the operation of the
(14) electric machines. These models are explained in the following:

The regulation related to the active power production of a FC is a. Power electronic interface: as mentioned above, in this case,
defined as: the RESs are modeled according to their control circuit. If the
control circuit is designed to control P and V independently, the
PtotalFC ¼ KP ðE0
VFC Þ2 (15) RESs can be modeled as a PV bus. On the other hand, when it is
designed to control P and Q independently, the model is
- Photovoltaic (PV): the connection of this type of RESs to the considered as a PQ bus.
grid is done in a similar way as FCs. The output power and the b. Induction generator model: the active and reactive powers of
solar radiation are related to each other as follows [25]: induction generators are usually supposed to be function of the
slip [26]:
8  2   
>
> Gt PI
G ¼ PI
G VbusI
G ; s 
  >
< t (18)
Psn 0 < G < Rc QI
G ¼ QI
G VbusI
G ; s
PPV Gt ¼ Gstd Rc t ¼ 1; 2; .; T (16)
>
> t
>
: Psn G Gt >Rc If the weak relationship between the slip and the reactive power
Gstd
is neglected and also PI-G is supposed to be constant then Eq. (18) is
Note it that in this equation, the relationship of photovoltaic cell reduced as:
temperature and PPV is neglected. 
PI
G ¼ constant
  (19)
QI
G ¼ f VbusIG
- Wind turbines: This kind of RES is divided into two categories
of fixed and variable speed. In the fixed speed type, the rotor of In the steady state condition which the values of the bus volt-
the squirrel cage induction generator is rotated by the propeller ages can be supposed to be near 1 p.u.; the induction generators
through gear box while the induction motor is connected to the can be modeled as PQ buses.

Fig. 3. Single line diagram of 32-bus test system before DFR.

 T. Niknam et al. / Renewable Energy 37 (2012) 213e225 219

Table 1 where here X is the control variable of making decision. Also n is the
Specifications of RESs (test system 1). number of objective functions.
Capacity (kW) Type Location In fact in MOP, the idea of optimality is replaced with that of
RES 1 350 FC 8 Pareto optimality or non-dominated solution. This characteristic is
RES 2 350 FC 11 due to the ability of MOP in finding the set of Pareto optimal
RES 3 350 PV 18 solutions. Pareto optimal solution is the optimal solution which
RES 4 350 Wind 25
cannot be improved in one of the objective functions unless dete-
riorate the performance of the solution in at least one of the rest.
Thus X* is called a Pareto optimal solution if it is impossible to find
a solution X in U such that X dominates X* ˛ U. U is the set of all the
c. Synchronous generator model: according to the structure of
vectors (X) which observe the constraints and limitations. In defi-
the excitation system, the synchronous machines are divided
nition the solution X1 dominates X2 if the following two conditions
into the following two categories [27]:
are satisfied:
c-1- regulating excitation voltage: this type itself is divided into the
following two categories: 1Þ c j˛f1; 2; .; ng; fj ðX1 Þ  fj ðX2 Þ
 voltage control mode (constant terminal voltage); which the (23)
2Þ d k˛f1; 2; .; ng; fk ðX1 Þ < fk ðX2 Þ
RES can be modeled as PV bus.
 power factor control mode (constant power factor); which the 5. Fuzzy-based clustering
RES can be modeled as PQ bus.
c-2- fixed excitation voltage:, the reactive power for a round rotor 5.1. Keeping the size of the repository
synchronous generator can be evaluated as:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi As mentioned before, the set of Pareto optimal solutions which
u !
u E2 2 2
Vbus are found during the optimization process are stored in an external
t q 2
Qin ¼
Pin
SY
(20) memory (or repository). Since the repository size is constant, the
Xd Xd
number of the Pareto solutions should not exceed a specified
number. Therefore a fuzzy-based clustering technique is utilized
Considering Pin as a constant value, Eq. (20) is simplified to:
here to control the size of the repository. The membership function

Pin ¼ constant assigned to each objective function is as follows:
  (21)
Qin ¼ f VbusSY 8
>
> 1 for fi ðXÞ  f min
>
< 0 for f ðXÞ  fimax
Similarly to Eq. (19), here RES can be supposed as a PQ bus. i
Therefore the RESs can be modeled as PV or PQ buses. mfi ðXÞ ¼ f max
f ðXÞ i (24)
>
> i
On the other hand, since the division of the loads in the distri- > i
: f max
f ðXÞ fi
min
 fi ðXÞ  fimax
i i
bution networks is unbalanced, the operation and control of the RESs
can be implemented in two forms: (1) simultaneous three-phase The values of fkmin and fkmax are evaluated by single optimization
control, (2) independent three-phase control (single phase control). of the kth objective function. Finally, the Pareto solutions are sorted
Therefore according to the model of the RES and the type of control according to the following equation:
which is utilized, four models are defined for the RESs (Fig. 1) [8]: Pn  
i ¼ 1 ui  mfl Xj
Nm ðjÞ ¼ Pm Pn   (25)
i ¼ 1 ui 
PQ model with simultaneous three-phase control. j¼1 mfl Xj
PQ model with independent three-phase control.
PV model with simultaneous three-phase control. where n is the number of the objective functions and m is the
PV model with independent three-phase control. number of the Pareto solutions in the repository.

5.2. Determination of the best compromised solution

4. Multi-objective optimization
After getting Pareto optimal solutions (that is after the termi-
The process of optimization of different conflicting objective nation of the optimization process), the best compromised solution
functions when all the constraints and limitations are observed should be extracted from the repository. In order to achieve this
simultaneously is called multi-objective optimization problem goal, a fuzzy-based mechanism is utilized here. In fact this method
(MOP). The MOP can be described as [16]: offers the best satisfying solution to the decision maker. Therefore,
after the evaluation of Nm for all the Pareto solutions by Eq. (25), the
minF ¼ ½f1 ðXÞ; f2 ðXÞ; .; fn ðXÞT repository is sorted in descending order. The best compromised
s:t: (22) solution is that for which the value of Nm is maximum. Note that
gi ðXÞ < 0 i ¼ 1; 2; .; Nueq here ui is supposed to be unit so that to give equal preferences to all
hi ðXÞ ¼ 0 i ¼ 1; 2; .; Neq the objective functions.

Table 2
Emission factors related to NOx, CO2 and SO2.

Emission type Emission factors (kg/MW h)

Grid Gas turbine Micro turbine Wind PV FC Internal combustion (IC)

NOx 2.2952 0.0136 0.1995 0 0 0.52163 2.1319
CO2 921.25 488.97 723.93 0 0 502.58 6.49.54
SO2 3.5834 0.0027 0.0036 0 0 3628.7 0.2059
220 T. Niknam et al. / Renewable Energy 37 (2012) 213e225

6. Modified Honey Bee Mating Optimization algorithm Table 3

Comparison of active power losses objective functions evaluated by different
methods neglecting RESs (test system 1).
6.1. Original HBMO
Method Power loss Minimum Open switches
Honey bee as a social insect with special behaviors and [KW] voltage

instructions have been the source of inspiration for the human Goswami [31] 143.69 0.93781964 s7,s9,s14,s32,s37
Vanderson Gomes [32] 139.53 0.93781964 s7,s9,s14,s32,s37
beings during the years. The honey bees’ society is consisted of
McDermott et al. [33] 139.53 0.93781964 s7,s9,s14,s32,s37
three groups in general: 1) the queen or female, 2) the drones or PSOeSFLA [34] 139.53 0.93781964 s7,s9,s14,s32,s37
males and 3) the workers. Each of these groups has a special task MSFLA [35] 139.53 0.93781964 s7,s9,s14,s32,s37
which should be implemented in such a way that the total condi- DPSO [14] 139.53 0.93781964 s7,s9,s14,s32,s37
tion of their society improves effectively. HBMO algorithm simu- DPSOeHBMO [14] 139.53 0.93781964 s7,s9,s14,s32,s37
PSOeACO [15] 139.53 0.93781964 s7,s9,s14,s32,s37
lates each of the phases of the natural mating process so that to give DPSOeACO [16] 139.53 0.93781964 s7,s9,s14,s32,s37
a satisfying algorithm which would be profitable in the optimiza- HBMO [29] 139.53 0.93781964 s7,s9,s14,s32,s37
tion applications. The mating process between the queen and each Shirmohammadi [36] 140.26 0.93781964 s7, s10, s14, s32,s37
of the drones is implemented probabilistically with an annealing The proposed 139.53 0.93781964 s7,s9,s14,s32,s37
function as follows [28]: algorithm


Df
probðDÞ ¼ exp
(26)
SðtÞ

xmut;j ; if 41  42
After each mating process, the queen speed decreases. If the xbrood1;j ¼
x ; otherwise
mating process is successful, the corresponding drone sperm is  queen;j
xmut;j ; if 43  42 (30)
added to the queen spermatheca, else it is discarded and the next xbrood2;j ¼
xj ; otherwise
drone is chosen for mating. The speed of the queen after each   
mating process is updated as follows: Xbrood;3 ¼ h  Xqueen þ a  Xqueen
SP Irand;SP

Now by the use of Eq. (23), the non-dominated solutions among

Sðt þ 1Þ ¼ a  SðtÞ
The mating process continues until the time that the speed of
the queen reaches to a specific value or her spermatheca become
full (for more details see Refs. [28,29]). Now the breeding process is
simulated. If the position of any of the new broods is better than
that of the queen, then it will replace the queen. This process of
mating and breeding continues until the time that the best satis-
fying queen (solution) would be achieved.
6.2. Modified HBMO (MHBMO) algorithm
It was mentioned before that the original HBMO suffers from
two main deficiencies; that is the reliance of the HBMO algorithm
on its parameters and the possibility of being trapped in local
optima. These two shortcomings root from the mating process.
Thus in order to improve the algorithm performance, the mating
process should be corrected sufficiently. In the original HBMO, after
that the process of adding the drones’ sperm to the queen sper-
matheca is completed and the queen spermatheca is constructed,
then the breeding process is implemented as follows:
 
Xbrood;j ¼ Xqueen þ g  Xqueen
Spi
(27) Xbrood,1, Xbrood,2, Xbrood,3 and the ith individual in the drones pop-
ulation are evaluated and stored in the repository.
For each individual in the drones’ population (Xi), the modifi-
cation process, as described above, should be repeated. After that
the breeding process for all the individuals is completed, the
repository is updated by Eq. (23).
In order to improve the MHBMO algorithm, the process of
generating drones’ population should be amended too. In the
original HBMO, after that the breeding process for all the drones’
population is finished then the old drones’ population is discarded
and a new generation is produced randomly. In the MHBMO algo-
rithm, this process is corrected as follows:
As mentioned before, for each drone in the population (Xi), three
new modified broods (Xbrood,1, Xbrood,2, Xbrood,3) are generated by Eq.
(30). After selection of the non-dominated solutions among the
three generated modified broods and the ith drone, the individual
who the summation of its membership functions is the most will
replace the corresponding drone (Xi) in the drones’ population. It is
evident that if the summation of the membership functions of Xi is
greater than that of the others, then it will be kept in its position.
Subsequently after a complete breeding process, the old drones’
population is updated and utilized as the new generation of drones
satisfactorily. The complete process of MHBMO algorithm is
depicted in Fig. 2.
(28)

In the proposed MHBMO algorithm, the breeding process is Table 4

Comparison of voltage deviation objective functions evaluated by different methods
modified as follows:
neglecting RESs (test system 1).
After that the queen spermatheca is constructed similar to
original HBMO, then three drones k1, k2 and k3 are chosen from the Method Voltage deviation Minimum Open switches
[p.u.] voltage
queen spermatheca randomly in a way that k1 s k2 s k3 s i where
i is the ith individual in the drones’ population. Thus by the use of DPSO [14] 0.06120031 0.93879681 s7,s9,s14,s32,s37
PSOeACO [15] 0.06120031 0.93879681 s7,s9,s14,s32,s37
the queen spermatheca, a new improved brood is generated as DPSOeHBMO [14] 0.06120031 0.93879681 s7,s9,s14,s32,s37
follows: DPSOeACO [16] 0.06120031 0.93879681 s7,s9,s14,s32,s37
GA 0.06218097 0.93781902 s7,s10,s14,s32,s37
Xmut ¼ Spk1 þ b  ðSpk2
Spk3 Þ (29) PSO 0.06120031 0.93879681 s6,s9,s14,s32,s37
HBMO [29] 0.06120031 0.93879681 s7,s9,s14,s32,s37
Now by the use of Xmut, Xqueen and Xi (the ith drone), three new
The proposed 0.06120031 0.93879681 s7,s9,s14,s32,s37
modified broods would be generated. The modification process is algorithm
implemented as follows:
 T. Niknam et al. / Renewable Energy 37 (2012) 213e225 221

Table 5 one by constructing an augmented objective function as Eq. (31).

Comparison of objective functions evaluated by different methods considering RESs In order to observe all the constraints, the penalty factors L1 and
(test system 1).
L2 are used here to prevent violating the constraints.
Objective function Method Best solution States of the switches
Power losses GA 84.54971 s6,s12,s35,s17,s37
2 3
[kW] PSO 80.74696 s33,s14,s8,s31,s37 F1 ðXÞ
HBMO 82.36062 s6,s12,s11,s31,s37
6 F2 ðXÞ 7
FðXÞ ¼ 4
F ðXÞ 5
The proposed 76.98436 s7,s14,s9,s30,s37
algorithm 3
F4 ðXÞ 41
Voltage deviation GA 0.0490462 s7,s14,s11,s31,s37 3
2 !
[p.u.] PSO 0.0478782 s7,s34,s10,s31,s37
P
Neq NP
ueq
7
HBMO 0.0481048 s33,s34,s8,s31,s37 6 f1 ðXÞ þ L1 ðJi ðXÞÞ2 þL2 ðMax½0;
gi ðXÞÞ2
7
The proposed 0.0442284 s6,s11,s14,s31,s37 6 i¼1 i¼1 7
6 7
algorithm 6 !7
6 P
N NP 7
6 eq ueq
7
Cost [$] GA 154.79529 s6,s11,s17,s34,s37 6 f2 ðXÞ þ L1 ðJi ðXÞÞ2 þL2 ðMax½0;
gi ðXÞÞ2 7
6 ¼ i¼1 7
PSO 154.28952 s7,s9,s12,s32,s37 6 i 1 7
HBMO 154.38659 s6,s10,s34,s36,s37 ¼6 !7
6 P
N NP 7
The proposed 154.16894 s7,s9,s14,s32,s37 6 eq ueq
7
6 f3 ðXÞ þ L1 ðJi ðXÞÞ2 þL2 ðMax½0;
gi ðXÞÞ2 7
algorithm 6 i¼1 i¼1 7
6 7
Emission [kg] GA 14,624.099 s2,s21,s34,s36,s37 6 !7
6 P
N NP 7
PSO 14,206.125 s11,s12,s15,s33,s37 4 eq ueq
7
HBMO 14,399.873 s21,s24,s33,s34,s36
f4 ðXÞ þ L1 ðJi ðXÞÞ2 þL2 ðMax½0;
gi ðXÞÞ2 5
i¼1 i¼1
The proposed 14,074.587 s6,s14,s11,s31,s37
41
algorithm
(31)

7. Application of the MHBMO in MDFR

L1 and L2 are supposed to be 106 in the paper.
Step 1: defining the input data.
Step 2: changing the constrained MOP to an unconstrained one: Step 3: generation of the initial population. The initial pop-
in this step, the constrained MOP is changed to an unconstrained ulation is as follows:

Fig. 4. Single line diagram of 32-bus test system after DFR. a) Single optimization of f1 evaluated by the proposed method considering RESs. b) Single optimization of f2 evaluated by
the proposed method considering RESs. c) Single optimization of f3 evaluated by the proposed method considering RESs. d) Single optimization of f4 evaluated by the proposed
method considering RESs.
222 T. Niknam et al. / Renewable Energy 37 (2012) 213e225
23
X1
6 X2 7
6 7
initial population ¼ 6 7 Step 9: generation of the new drones’ population. Among the ith
4 . 5
XNipop Nipop ðNtie þNsw þNg Þ
Ng ¼ NPV þ NFC þ NWind
Step 4: evaluation of the objective functions. In this step the
values of the objective functions and their corresponding
membership functions are evaluated.
Step 5: formation of the repository. Here by the use of Eq. (23)
and the membership functions evaluated in the last step, all
the Pareto solutions are evaluated and stored in the repository.
Step 6: selection of the queen. The queen is selected from the
repository randomly.
Step 7: formation of the queen spermatheca matrix. Firstly, the
queen flies by her maximum speed far from the nest. Now
a drone is selected from the drones’ population randomly and
mates with the queen. Therefore according to the values of the
objective functions and by the use of Eq. (26), prob(D) would be
evaluated. Now a value in the range of [0,1] is generated
randomly and compared to prob(D). If prob(D) is bigger than the
generated random value, then the sperm of the specified drone
is added to the queen spermatheca, else another drone is chosen
from the population randomly and the mating process is
repeated. The mating process continues until the time that the
queen spermatheca becomes full or her speed reduces to the
specified value.
Fig. 5. 3-D plot of the Pareto optimal solutions considering RESs for 3-objective function (test system 1).
Step 8: breeding process. This process is implemented as
described in Section 6.2.
drone, Xbrood,1, Xbrood,2 and Xbrood,3 the individual who the
summation of its membership functions is the most (so the
fittest individual) will replace the ith drone.
ð32Þ
Step 10: if all the drones are checked go to step 11, else return to
step 8.
Step 11: updating the repository. In this step the repository is
updated so that all solutions in the repository would be Pareto
optimal solutions.
Step 12: updating the queen. A new queen is selected from the
updated repository randomly.
Step 13: generation of the queen speed: The queen speed will be
generated randomly as follows:

Squeen ¼ randð$Þ  ðSmax
Smin Þ þ Smin (33)
Step 14: termination criterion. If the termination criterion is
achieved, finish the algorithm, else return to step 6.
8. Simulation results
8.1. Case study 1
The first test system is a 12.66 kV system which is consisted of
32 buses, 5 sectionalizing switches and 5 tie switches so that to be
 T. Niknam et al. / Renewable Energy 37 (2012) 213e225 223

Table 7
Specifications of the RESs (test system 2).

Capacity (KW) Type Location

RES 1 450 FC 6
RES 2 450 FC 21
RES 3 450 Wind 28
RES 4 450 Wind 39
RES 5 450 PV 44
RES 6 450 PV 51

optimization is done to compare the performance of the proposed

Fig. 6. Single line diagram of 69 bus test system (tie lines are connected from node 10
to 70, 12 to 20, 14 to 90, 38 to 48, and 26 to 54).
able to form 5 branch loops. The network diagram proposed by
Baran and Wu is shown in Fig. 3 [12]. The initial power losses before
the reconfiguration is 202.67 kW. Table 1 shows the location, the
capacity and the type of each RES. Also the emission factors related
to NOx, CO2 and SO2 are shown in Table 2.
Since MHBMO algorithm is used in this paper for the first time
to solve DFR problem, then first of all, a single objective
algorithm with the other works done in the area and also with
some of the famous evolutionary algorithms such as Genetic
Algorithm (GA), Particle Swarm Optimization (PSO) algorithm,
Honey Bee Mating Optimization (HBMO) algorithm, etc. In Tables 3
and 4, the results of the single optimization of the power losses (f1)
and the voltage deviation (f1), neglecting RESs are shown respec-
tively. It can be seen that the proposed algorithm is capable of
finding the best solution as well as the best switching sufficiently. It
should be noted that since just f1 and f2 are investigated in the other
works mentioned in Tables 3 and 4, so the comparison of the other
two objective functions (cost (f1) and emission (f1)) are not shown
in the tables. However the complete comparison of all the objective
functions while the effect of the RESs is considered will be
described in the following. In order to see the appropriate perfor-
mance of the proposed method and also the effect of the RESs on
the network simultaneously, in Table 5, the result of the single
objective optimization by MHBMO and some of the other algo-
rithms mentioned before is shown comparatively. The simulation
results show that the performance of the proposed algorithm is
much more satisfying than the other algorithms shown in Table 5.
In order to have a more precise look on the results, in Fig. 4, the
state of the network after applying single objective optimization by
the proposed method is depicted. It should be noted that, in Fig. 4,
just the states of the switches (tie switches and sectionalizing
switches) are shown and the power production of the RESs are not
revealed here. It is evident from this figure, that the switching
operation has a direct relationship with the optimization of each of
the objective functions individually. Nevertheless, it can be seen
that the states of the switching corresponding to the optimization
of f2 and f4 are completely similar to each other. This result could be

Table 6
Some of the non-dominated solutions found for MDFR problem (test system 1). Table 8
Comparison of objective functions evaluated by different methods considering RESs
Power losses Voltage deviation Cost ($) Emission (test system 2).
(KW) (p.u.) (kg)
Objective function Method Best solution States of the switches
1 076.984 0.052854 156.923 14,081.05
2 090.550 0.049126 155.642 16,402.84 Power losses GA 88.4282 s70,s71,s11,s58,s61
3 088.137 0.048260 156.235 15,955.57 [kW] HBMO 87.2016 s70,s71,s14,s58,s61
4 085.580 0.048357 156.179 15,747.99 PSO 86.2222 s70,s71,s13,s55,s61
5 093.921 0.049159 155.364 17,287.72 The proposed 85.2230 s70,s71,s14,s56,s61
6 077.404 0.044224 156.934 14,110.31 algorithm
7 106.265 0.056111 154.964 18,512.59
Voltage deviation GA 0.0580294 s70,s71,s72,s54,s61
8 099.280 0.051258 155.403 17,260.70
[p.u.] HBMO 0.0572424 s70,s71,s72,s58,s61
9 101.437 0.056263 154.931 18,566.69
PSO 0.0572262 s70,s71,s14,s58,s61
10 101.162 0.051158 155.347 17,722.32
The proposed 0.0572147 s70,s20,s11,s58,s61
11 091.196 0.050928 156.959 15,147.87
algorithm
12 132.036 0.062519 154.405 21,312.95
13 095.416 0.060376 155.035 17,761.61 Cost [$] GA 158.21158 s42,s13,s11,s58,s61
14 134.018 0.062111 154.168 22,016.35 HBMO 157.28998 s70,s13,s11,s58,s74
15 086.276 0.051784 156.909 14,825.49 PSO 157.07320 s70,s71,s11,s58,s74
16 084.314 0.048066 156.978 14,656.03 The proposed 156.05831 s70,s71,s12,s58,s61
17 077.280 0.044228 156.941 14,074.58 algorithm
18 091.875 0.049822 156.083 15,900.95
Emission [kg] GA 7,211.096 s42,s19,s72,s58,s21
19 101.470 0.053645 155.336 17,826.48
HBMO 7,078.044 s35,s13,s72,s58,s68
20 102.815 0.056175 155.364 17,525.11
PSO 7,186.769 s40,s13,s72,s54,s62
Best compromised 084.314 0.048066 156.978 14,656.03 The proposed 7,026.466 s70,s71,s11,s58,s62
solution algorithm
224 T. Niknam et al. / Renewable Energy 37 (2012) 213e225

Table 9 one. This deduction can also be inferred from Fig. 5 which shows
Some of the non-dominated solutions in four-objective DFR problem (test system 2). the 3-D plot of the objective functions with regard to each other.
Power losses Voltage deviation Cost ($) Emission However the little difference which exists in the results of the
(KW) (p.u.) (kg) single optimization of f2 and f4 separately, roots in the difference
1 085.223 0.0572319 159.3069 12,587.71 which exists in the power production of the RESs. Actually, the
2 092.913 0.0572259 161.8965 12,576.07 single optimization of f4 has selected the best operating point for
3 113.798 0.0700551 157.2814 21,293.49
the FC1 (on bus 8), FC2 (on bus 11), PV and Wind plant to be
4 088.460 0.0585779 162.2050 11,629.81
5 091.024 0.0572147 165.6285 07,029.07 350 MW equally, while the single optimization of f2 has preferred to
6 102.611 0.0586040 159.6404 16,570.59 reduce the power production of FC1 by 6MW to 344 MW. The effect
7 103.141 0.0572385 156.7690 21,598.31 of this power production management on the values of the other
8 090.412 0.0572165 164.4271 08,581.18 objective functions can be seen in Table 6. The similarity of the
9 091.337 0.0572149 165.5630 07,199.97
10 099.266 0.0572464 156.1511 22,798.47
behavior of f2 and f4 can be deduced from this table too. In Table 6,
11 093.994 0.0572356 159.7799 16,654.01 a set of the Pareto optimal solutions which are evaluated by the
12 121.537 0.0700697 157.9254 21,278.96 multi-objective optimization of the four objective functions is
13 105.148 0.0586094 156.4227 22,723.45 shown. It can be seen that the best solutions which are evaluated by
14 086.000 0.0572289 162.1507 12,608.01
single optimization in Table 5 are considered as the Pareto solutions
15 099.267 0.0572466 156.0583 22,934.94
16 119.139 0.0736479 159.4350 18,500.87 in the multi-objective optimization in Table 6. It could already be
17 088.853 0.0572158 164.7303 08,006.55 deduced too. In fact since each of these solutions are the best with
18 104.265 0.0572419 156.3078 22,889.30 regard to the corresponding objective function, thus there could
19 090.580 0.0572247 165.6108 07,026.46 not be found any solution that can dominate these solutions and
20 086.276 0.0572269 162.9434 11,393.48
therefore omit them from the repository. Each of the solutions in
Best compromised 090.412 0.0572165 164.4271 08,581.18 Table 6 can be a satisfying solution depending on the preferences of
solution
the decision maker. However in order to offer the best compro-
mised solution to the decision maker, by the use of fuzzy clustering
deduced from Table 5 too. In fact this result shows that these two technique (as explained in Section 5.2), the best compromised
objective functions have a similar behavior in decreasing and solution is evaluated and shown in the last row of Table 6. In Fig. 7,
increasing in a way that in a wide range of variation, the the 3-D plot of the Pareto optimal solutions is depicted while each
improvement of one will result in the improvement of the other figure belongs to the multi-objective optimization of three of the

Fig. 7. 3-D plot of the Pareto optimal solutions considering RESs for 3-objective function (test system 2).
 T. Niknam et al. / Renewable Energy 37 (2012) 213e225
objective functions simultaneously. In this figure, each star denotes
a Pareto optimal solution. The rectangular boxes show the best
optimal solution with regard to each of the objective functions
individually. Also a circle is depicted in each figure to distinguish
the best compromised solution from the other solutions effectively.
8.2. Case study 2
The second case study is a 12.66 kV, 69-bus distribution test
system selected from the PG&E distribution system [30]. The
network diagram is depicted in Fig. 6. The location, capacity and
type of the RESs are shown in Table 7. In Table 8, a complete
comparison between the results of single optimization by the
proposed algorithm, GA, PSO and original HBMO is revealed while
the effect of the RESs is considered. As it can be seen from this table,
the selection of the best optimal solutions for case study 2 depends
on the state of the switching completely. It is evident from Table 7,
that the proposed algorithm has improved the values of the
objective functions effectively. In Table 9, a set of the Pareto optimal
solutions which are selected from the repository set are shown.
With comparison, it can be seen from this table that the behavior of
the f2 and f4 in the optimization process is similar to each other. This
similarity was discussed in case study (1) too. As mentioned before,
each of these solutions can be an appropriate solution according to
the preferences. However, in order to offer the best satisfying
solution to the decision maker, the best compromised solution
evaluated by the proposed method is shown in the last row of
Table 9. In order to see the improvement of each objective function
with regard to the rest, in Fig. 7, the 3-D plot of the Pareto optimal
solutions evaluated by the multi-objective optimization of each
three objective functions individually is depicted. As mentioned
before, the best compromised solution is discriminated from the
other solutions by a circle in the figure.
9. Conclusion
In this paper, a new modified evolutionary algorithm based on
HBMO is proposed to solve the multi-objective DFR problem while
the effect of the RESs is considered. In the proposed method, the
concept of the Pareto optimality is utilized to take advantage of the
non-dominated solutions evaluated during the optimization process
so that to allow the decision maker to apply his/her preferences in
the implementation. In order to control the size of the repository,
a fuzzy-based technique is proposed and utilized in the optimiza-
tion. Also, the concept of fuzzy clustering is used to extract the best
compromised solution from the set of the Pareto optimal solutions
effectively. The simulation results show that the proposed modified
evolutionary algorithm has an appropriate performance in large-
scale optimization problems such as MDFR problem. In addition
the simulation results show that the dependability of the proposed
algorithm on the initial parameters is much less than that of the
other well-known algorithms such as GA, PSO and original HBMO.
References
[1] Jacobson MZ. Review of solutions to global warming, air pollution and energy
security. Energy Environ Sci; 2009:1e55.
[2] Liu T, Xu G, Cai P, Tian L, Huang Q. Development forecast of renewable energy
power generation in China and its influence on the GHG control strategy of
the country. Renewable Energy 2011;36:1284e92.
[3] Braun RJ, Klein SA, Reindl DT. Evaluation of system configurations for solid
oxide fuel cell-based micro-combined heat and power generators in resi-
dential applications. J Power Sources 2006;158:1290e305.
[4] Dalton GJ, Lockington DA, Baldock TE. Feasibility analysis of renewable energy
supply options for a grid-connected large hotel. Renewable Energy 2009;34:
955e64.
225
[5] Chang PC, Chen SH, Liu. Sub-population genetic algorithm with mining gene
structures for multi-objective flowshop scheduling problems. Expert Syst Appl
2007;33(3):762e71.
[6] Lin CM, Gen M. Multi-criteria human resource allocation for solving multi-
stage combinatorial optimization problems using multi-objective hybrid
genetic algorithm. Expert Syst Appl 2008;34(4):2480e90.
[7] Kim H, Ko Y, Jung KH. Artificial neural-network based feeder reconfiguration
for loss reduction in distribution systems. IEEE Trans Power Delivery 1993;
8(3):1356e66.
[8] Taylor T, Lubkeman D. Implementation of heuristic search strategies for
distribution feeder reconfiguration. IEEE Trans Power Delivery 1990;5(3):
239e45.
[9] Chiou J, Chang C. Variable scaling hybrid differential evaluation for solving
network reconfiguration of distribution system. IEEE Trans Power Sys 2005;
20(2):668e74.
[10] Kashem MA, Ganapathy V, Jasmon GB. Network reconfiguration for load
balancing in distribution networks. IET Gener Transm and Distrib 1999;146:
563e7.
[11] Debaprya D. A fuzzy multi-objective approach for network reconfiguration of
distribution systems. IEEE Trans Power Delivery 2006;21(1):202e9.
[12] Baran ME, Wu FF. Network reconfiguration in distribution systems for loss
reduction and load balancing. IEEE Trans Power Delivery 1989;4(2):1401e7.
[13] Zhou Q, Shirmohammadi D, Liu WHE. Distribution feeder reconfiguration for
service restoration and load balancing. IEEE Trans Power Syst 1997;12(2):724e9.
[14] Niknam T. An efficient hybrid evolutionary algorithm based on PSO and
HBMO algorithms for multi-objective distribution feeder reconfiguration.
Energy Convers Manage 2009;50(8):2074e82.
[15] Niknam T. An efficient hybrid evolutionary based on PSO and ACO algorithms for
distribution feeder reconfiguration. Eur Trans Electr Power 2010;20(5):575e90.
[16] Niknam T. A new hybrid algorithm for multi-objective distribution feeder
reconfiguration. Cybernet and Syst: An International J 2009;40(6):508e27.
[17] Niknam T, Meyman HZ, Mojarrad HD. A novel multi-objective fuzzy adaptive
chaotic PSO algorithm for optimal operation management of distribution
network with regard to fuel cell power plants. Eur Trans Electr Power, in press.
[18] Niknam T. A new approach based on ant colony optimization for daily Volt/
Var control in distribution networks considering distributed generators.
Energy Convers Manage 2008;49:3417e24.
[19] Niknam T, Zeinoddini Meymand H, Doagou Mojarrad H, Aghaei J. Multi-
objective daily operation management of distribution network considering
fuel cell power plants. IET Renewable Power Generation, in press.
[20] Niknam T, Meymand HZ, Nayeripour M. A practical algorithm for optimal
operation management of distribution network including fuel cell power
plants. Renewable Energy 2010;35:1696e714.
[21] Ro K, Rahman S. Control of grid-connected fuel cells plants for enhancement
of power system stability. Renewable Energy 2003;28:397e407.
[22] Jurado F, Valverde M, Cano A. Effect of a SOFC plant on distribution system
stability. J Power Sources 2004;129:170e9.
[23] Singh PP, Singh S. Realistic generation cost of solar photovoltaic electricity.
Renewable Energy 2010;35:563e9.
[24] Shen M, Meuleman W, Scott K. The characteristics of power generation of
static state fuel cells’. J Power Sources 2003;115:203e9.
[25] Marwali MKC, Ma H, Shahidehpour SM, Abdul-Rahman KH. Short-term
generation scheduling in photovoltaic-utility grid with battery storage. IEEE
Trans Power Syst 1998;13:1057e62.
[26] Naka S, Genji T, Fukuyama Y. Practical equipment models for fast distribution
power flow considering interconnection for distributed generators. In: IEEE
Power Engineering Society Summer Meeting, vol. 2; 2001. p. 1007e12.
[27] Chen H, Chen J, Shi D, Duan X. Power flow study and voltage stability analysis
for distribution systems with distributed generation. In: IEEE Power Engi-
neering Society General Meeting, vol. 8; 2006. p. 18e22.
[28] Niknam T. Application of honey-mating optimization on estate estimation of
a power distribution system including distributed generators. J Power Zhe-
jiang Univ Sci A 2008;9(12):1753e64.
[29] Niknam T. Application of honey bee mating optimization on distribution state
estimation including distributed generators. J Zhejiang Univ Sci A 2008;9:
1753e64.
[30] Kashem MA, Ganapathy V, Jasmon GB. A geometrical approach for network
reconfiguration based loss minimization in distribution systems. Electrical
Power Energy Syst 2001;23:295e304.
[31] Goswami SK, Basu SK. A new algorithm for the reconfiguration of distribution
feeders for loss minimization. IEEE Trans Power Delivery 1992;7(3):1484e91.
[32] Gomes FV, Carneiro Jr S, Pereira JLR, Vinagre MP, Garcia PAN, Araujo L. A new
heuristic reconfiguration algorithm for large distribution systems. IEEE Trans
Power Syst 2005;20(3):1373e8.
[33] McDermott TE, Drezga I, Broadwater RP. A heuristic nonlinear constructive
method for distribution system reconfiguration. IEEE Trans Power Syst 1999;
14(2):478e83.
[34] Niknam T, Azad Farsani E. A hybrid evolutionary algorithm for distribution
feeder reconfiguration. Sci China Tech Sci 2010;53:950e9.
[35] Niknam T, Azad Far sani E, Nayreipour M. An efficient multi-objective modi-
fied shuffled frog leaping algorithm for distribution feeder reconfiguration
problem. Euro Trans Electr Power 2011;21(1):721e39.
[36] Shirmohammadi D, Hong HW. Reconfiguration of electric distribution networks
for resistive line loss reduction. IEEE Trans Power Syst 1998;4(1):1492e8.