Applied Energy 102 (2013) 943–949

Contents lists available at SciVerse ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Artificial bee swarm optimization algorithm for parameters identification

of solar cell models
Alireza Askarzadeh ⇑, Alireza Rezazadeh
Faculty of Electrical and Computer Engineering, Shahid Beheshti University, GC, Evin 1983963113, Tehran, Iran

h i g h l i g h t s

" Artificial bee swarm optimization (ABSO) is proposed to identify the solar cell parameters.
" The performance of ABSO algorithm is quite promising.
" The results of ABSO outperform those of the other studied algorithms.
" ABSO is a helpful technique for solar cell parameters identification.

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

Article history: An accurate mathematical model is an extremely helpful tool for simulation, evaluation, control, and
Received 2 June 2012 optimization of solar cell systems. Due to the non-linearity of the solar cell models and the inability of
Received in revised form 30 August 2012 traditional optimization methods to accurately identify the unknown parameters, recently, metaheuristic
Accepted 28 September 2012
algorithms have attracted significant attention. Artificial bee swarm optimization (ABSO) is a recently
Available online 9 November 2012
invented algorithm inspired by the intelligent behaviors of honey bees such as collection and processing
of nectar. In this paper, we propose an ABSO-based parameter identification technique based on the sin-
Keywords:
gle and double diode models for a 57 mm diameter commercial (R.T.C. France) silicon solar cell. The
Mathematical modeling
Parameter identification
results obtained by ABSO algorithm are quite promising and outperform those found by the other studied
Artificial bee swarm optimization methods.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction
Thanks to the growing price of fossil fuels and air pollution, the
tendency to use renewable energy sources has noticeably in-
creased over the past years. Among renewable energy sources, so-
lar cells are being used all over the world to produce electricity not
only for the high demand of electrical power, but also for their
promising features like easy installation, little maintenance, no
pollution, and noise-free.
Solar cell accurate modeling has received significant attention
in recent years [1–5]. Solar cell modeling primarily involves the
formulation of the non-linear current vs. voltage (I–V) curve. Sev-
eral models have been developed to represent the behavior of
the system under different operating conditions. They vary from
models with simple assumptions to advanced models accompany-
ing with many physical variables. Nevertheless, two solar cell mod-
els are widely used in practice: the single and double diode models.
⇑ Corresponding author. Tel.: +98 21 29904178; fax: +98 21 22431804.
E-mail addresses: a_askarzadeh@sbu.ac.ir (A. Askarzadeh), a-rezazade@sbu.ac.ir
(A. Rezazadeh).
The single diode model is much more common than its counter-
part. This model contains five unknown parameters while the dou-
ble diode model has seven unknown parameters. Accurate
determination of the parameters plays an important role in solar
cell simulation, performance evaluation, design, optimization,
and control. Therefore, parameters identification with the help of
a capable optimization technique is necessity.
Two main approaches have been used in the literature to solve
the parameter identification problem: traditional [6–9] and meta-
heuristic [10–12] search approaches. In [6] a modified non-linear
least error squares estimation approach based on Newton’s meth-
od has been developed to obtain solar cell parameters. Dependency
on the initial point used in the iterative method is a major defi-
ciency of this approach. Besides, this type of optimization method
is local in nature and may achieve a local optimum rather than a
global one if multiple solutions exist. Similar analytical methods
have been proposed in [7–9]. These traditional optimization tech-
niques needing continuity, convexity and differentiability condi-
tions for being applicable, involve heavy computations, are
sensitive to the initial solution, and most often lead to local optima.
Another reason for the inability of traditional methods to

0306-2619/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.apenergy.2012.09.052
944 A. Askarzadeh, A. Rezazadeh / Applied Energy 102 (2013) 943–949

effectively solve the parameter identification problem is the non- t

linearity of the solar cell characteristic. As a result, in solar cell
t
optimization problems, to increase the probability of obtaining
the global solution at a reasonable time, metaheuristic search ap- s
proaches have been proposed [13–16]. In the literature, solving
the problem of solar cell parameter identification can be addressed
by genetic algorithm (GA) [10], particle swarm optimization (PSO)
[11], and simulated annealing (SA) [12]. Although these metaheu-
d1 d2 sh
ristic algorithms yield better results than traditional ones, they
have their respective limits [17]. As the most popular metaheuris- ph
sh
tic algorithm, GA frequently finds promising regions of search
space quickly, but it has some drawbacks: it has a trend to con-
verge towards local optima rather than the global optimum, unless
the objective function is defined well; suffers from the lack of good
local search ability; and simpler optimization algorithms may find
Fig. 1. The double diode model of solar cell.
better solutions than GA at a same amount of computation time. In
comparison with GA, the advantages of PSO are that it is easy to
implement and there are few parameters to adjust. Nevertheless,
2. Problem formulation
this algorithm has some drawbacks: PSO performance depends
strongly on its parameters; it might easily lose the diversity; and
2.1. Solar cell models
may be influenced by premature convergence, especially when
the best solution is a local one. SA is another metaheuristic trying
Several models have been developed to describe the I–V charac-
to simulate the process of annealing in metallurgy. Due to the fact
teristic of solar cells, but only two models are used in practice.
that SA is a solo-searcher, its performance is sensitive to the start-
These models will be briefly discussed in the following subsections.
ing point of the search. Furthermore, there is no rigorous theoret-
ical foundation for determining SA parameters, particularly the
2.1.1. Double diode model
parameters of the cooling schedule. The selection of these param-
For modeling an ideal solar cell under illumination, a light-gen-
eters is extremely difficult and the designer needs to trial in order
erated current source is shunted with a rectifying diode. Neverthe-
to obtain values that can provide a proper optimization in a rea-
less, in practice to consider the space charge recombination current
sonable amount of computational time.
the current source is also shunted with another diode and a shunt
The non-linearity of the solar cell characteristics expects a high-
leakage resistor to take into account the partial short circuit cur-
performance optimization technique. The success of an optimiza-
rent path near the cell’s edges related to the semiconductor impu-
tion algorithm depends strongly on the ability of providing a good
rities and non-idealities. In addition, due to the solar cell metal
balance between exploration and exploitation. Exploration refers
contacts and the semiconductor material bulk resistance, a series
to generation of new solutions in as yet unseen regions of the
resistor is connected with the cell shunt elements [24]. The equiv-
search space and exploitation means the concentration of the algo-
alent circuit of the double diode model is as Fig. 1.
rithm’s search at the vicinity of current good solutions. The inabil-
The terminal current, It, can be written as follows:
ity of the algorithm to make a good balance between exploration
and exploitation leads to premature convergence, getting trapped It ¼ Iph  Id1  Id2  Ish ð1Þ
in a local optima, and stagnation. Motivated by the swarm behav-
where Iph is the photo-generated current, Id1 denotes the first diode
ior of honey bees, bee algorithms (BAs) are recently invented meta-
current, Id2 is the second diode current, and Ish denotes the shunt
heuristic optimization techniques. BAs employ different kinds of
resistor current.
bees with different updating patterns to probe the search space.
Considering Shockley equation for the diodes currents and
Using distinct updating patterns increases the flexibility of the
substituting the current of the shunt resistor, Eq. (1) is rewritten
algorithm to establish good balance between exploration and
as the following form:
exploitation. Hence, BAs are promising algorithms that have more
chance than the other algorithms to find the optimal parameters of
solar cells which are closer to the real ones. The effectiveness of
BAs has led to their application to optimization problems in differ-
ent areas [18–21]. Different approaches have been developed for t

simulating the intelligent behavior of honey bees [20–23]. Here, t

artificial bee swarm optimization algorithm, ABSO, proposed by
the authors [20] is used to solve the parameter identification prob- s
lem. In ABSO, two kinds of bees are employed: onlooker and scout
bees. Scout bees are those bees that try to provide exploration
while the most focus of onlooker bees is to provide exploitation.
In this paper, ABSO is applied to identify the optimal parameters
of a 57 mm diameter commercial (R.T.C. France) silicon solar cell d sh
[6]. In order to evaluate the ABSO performance, the obtained re-
ph
sults are compared with those of genetic algorithm (GA), chaos sh
particle swarm optimization (CPSO), simulated annealing (SA), pat-
tern search (PS), and harmony search (HS) algorithm.
The rest of this paper is arranged as follows: Section 2 provides
a description of the solar cell models; In Section 3, ABSO algorithm
is explained in detail; Simulation results and discussions are re-
ported in Section 4, and finally, conclusion is stated is Section 5. Fig. 2. The single diode model of solar cell.
 A. Askarzadeh, A. Rezazadeh / Applied Energy 102 (2013) 943–949 945

Table 1 The single diode model can be formulated as follows:

Upper and lower range of the solar cell parameters.    
qðV t þ Rs It Þ V t þ Rs It
Parameter Lower Upper It ¼ Iph  Isd exp 1  ð3Þ
nkT Rsh
Rs (O) 0 0.5
Rsh (O) 0 100 In this model, the parameter identification problem reduces to
Iph (A) 0 1 find five parameters, namely, Rs, Rsh, Iph, Isd, and n.
Isd (lA) 0 1
n 1 2
2.2. Optimization process

The unknown parameters can be identified by the help of an

    optimization technique and a set of experimental I–V data ob-
qðV t þ Rs It Þ
It ¼ Iph  Isd1 exp 1 tained from the real system. In the optimization algorithm, each
n1 kT solution is defined by a vector, x, where x = [Rs Rsh Iph Isd1 Isd2 n1
   
qðV t þ Rs It Þ V t þ Rs I t n2] in the double diode model and x = [Rs Rsh Iph Isd n] in the single
 Isd2 exp 1  ð2Þ
n2 kT Rsh diode model. To define the objective function, Eqs. (2) and (3) are
rewritten in their homogeneous forms as follows:
where Isd1 and Isd2 are the diffusion and saturation currents, respec-    
tively, Vt is the terminal voltage, Rs and Rsh are the series and shunt qðV t þ Rs It Þ
f ðV t ; It ; xÞ ¼ It  Iph þ Isd1 exp 1
resistances, q is the electronic charge, k denotes the Boltzmann con- n1 kT
   
stant, n1 and n2 are the diffusion and recombination diode ideality qðV t þ Rs It Þ V t þ Rs I t
factors, and T (K) is the cell temperature.
þ Isd2 exp 1 þ ð4Þ
n2 kT Rsh
Eq. (2) contains seven unknown parameters, namely, Rs, Rsh, Iph,
   
Isd1, Isd2, n1, and n2. To reflect the solar cell performance as well as qðV t þ Rs It Þ V t þ Rs I t
that of the actual system, the identification of the parameters is f ðV t ; It ; xÞ ¼ It  Iph þ Isd exp 1 þ ð5Þ
nkT Rsh
essential.
The value of f is calculated for each pair of the experimental
data. We use the root mean square error (RMSE) as a criterion to
2.1.2. Single diode model quantify the difference between the model results and the experi-
Single diode model is widely used to represent the solar cell mental data. RMSE is defined by the following equation:
behavior. This model is made by combining together both diodes rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
currents, under the introduction of a non-physical diode ideality 1 XN
RMSE ¼ i¼1 i
ðf ðV t ; It ; xÞÞ2 ð6Þ
factor, n. The equivalent circuit of this model is as Fig. 2. N

Initialization:
Determine swarm size, percentage of scout and onlooker bees, maximum number of iterations
and the other adjustable parameters
Generate an initial swarm in the search space based on Eq. (13)
Do
Compute objective function of the swarm
Update the best achievement for each bee
Sort bees based on their objective function value
Specify scout and onlooker bees
For i = 1 to number of scout bees
For j = 1 to d
Employ Eq. (7)
Next j
Next i
For i = 1 to number of onlooker bees
Select interesting elite bee
For j = 1 to d
Employ Eq. (10)
Next j
Next i
Update adjustable parameters
Until termination criterion is met
Return the best achievement of the population as optimal solution

Fig. 3. Pseudocode of artificial bee swarm optimization algorithm.

946 A. Askarzadeh, A. Rezazadeh / Applied Energy 102 (2013) 943–949

1
Real data
Double diode model
0.8

0.6

Current (A)
0.4

0.2

-0.2

-0.4
-0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6
Voltage (V)

Fig. 5. Comparison between the I–V characteristics resulted from the experimental
data and the double diode model.

0.5
Real data
Double diode model
0.4

0.3
Power (W)

0.2

0.1

-0.1

-0.2
-0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6
Voltage (V)

Fig. 6. Comparison between the P–V characteristics resulted from the experimental
data and the double diode model.

0.18

0.16
Fig. 4. Flowchart of the ABSO algorithm for parameter identification of the solar cell
models. 0.14
Objective function

0.12

where N is the number of the experimental data. 0.1

During the optimization process, the objective function is min-
0.08
imized with respect to the parameters range. The upper and lower
bounds of the parameters, provided by the literature survey [6– 0.06
12], are shown in Table 1. The model parameters (five or seven
0.04

0.02
Table 2
A comparison between the results obtained by the HS-based algorithms and the other 0
ones for the double diode model parameter identification. 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Iteration
Item ABSO PS SA HS
Rs (O) 0.03657 0.0320 0.0345 0.03545 Fig. 7. Convergence process of ABSO algorithm during the identification process of
Rsh (O) 54.6219 81.3008 43.1034 46.8269 the double diode model.
Iph (A) 0.76078 0.7602 0.7623 0.76176
Isd1 (lA) 0.26713 0.9889 0.4767 0.12545
Isd2 (lA) 0.38191 0.0001 0.0100 0.25470 based on the selected model) are successively adjusted by the opti-
n1 1.46512 1.6000 1.5172 1.49439 mization algorithm, until a termination criterion is met. It is clear
n2 1.98152 1.1920 2.0000 1.49989
that the smaller the objective function value, the better the solu-
RMSE 9.8344e4 0.01518 0.01664 0.00126
tion is.
 A. Askarzadeh, A. Rezazadeh / Applied Energy 102 (2013) 943–949 947

Table 3 onlooker bee makes a decision to follow a specific dancer is not

Relative error for each measurement (double diode model). obvious, but is likely in relation with the quality of the nectar
Measurement Vt (V) It measured (A) It calculated (A) Relative source. Consequently, as the quality of a food source increases,
error the probability of its selection increases, too. The value of ne has
1 0.2057 0.764 0.764031 4.02e05 a significant role on the ABSO performance. If it is selected too
2 0.1291 0.762 0.762629 0.00083 small, the algorithm may be influenced by premature convergence,
3 0.0588 0.7605 0.761343 0.00111 while the large values of ne have bad effect on the search ability of
4 0.0057 0.7605 0.760162 0.000445
5 0.0646 0.76 0.75908 0.00121
the ABSO algorithm.
6 0.1185 0.759 0.758081 0.00121 A scout bee is employed to explore randomly the food source
7 0.1678 0.757 0.757139 0.00018 region to find new food sources. It adjusts its position with a ran-
8 0.2132 0.757 0.756193 0.001066 dom walk function, wf. Scout bees are used to maintain the diver-
9 0.2545 0.7555 0.755132 0.000487
sity of the swarm. By this way, the probability of trapping at local
10 0.2924 0.754 0.753694 0.000406
11 0.3269 0.7505 0.751392 0.00119 minima decreases. Each scout bee adjusts its position as
12 0.3585 0.7465 0.747322 0.0011 j
13 0.3873 0.7385 0.740044 0.00209 xjnew ¼ xj þ r s  wf ð7Þ
14 0.4137 0.728 0.727331 0.000919
15 0.4373 0.7065 0.706896 0.00056
where xnew is the new position, x denotes the current position,
16 0.459 0.6755 0.675265 0.000348 j = 1, 2, . . . , d denotes the variable index, rs is a random number be-
17 0.4784 0.632 0.630889 0.001758 tween 1 and 1, and wf, defined by Eq. (8), is a vector depending on
18 0.496 0.573 0.572114 0.001546 the lower bound, l, and upper bound, u, of each. The radius of walk
19 0.5119 0.499 0.499533 0.00107
function, s, decreases with a linear function from a large value, smax,
20 0.5265 0.413 0.413525 0.00127
21 0.5398 0.3165 0.31723 0.00231 to a small one, smin, during the iterations as Eq. (9).
22 0.5521 0.212 0.21209 0.00042 1 2 d
23 0.5633 0.1035 0.102694 0.007788 wf ¼ s  ðju1  l j; ju2  l j; . . . ; jud  l jÞ ð8Þ
24 0.5736 0.01 0.00927 0.07253
25 0.5833 0.123 0.12439 0.01126 s ¼ smax  ðsmax  smin Þiter=itermax ð9Þ
26 0.59 0.21 0.20917 0.003965
where iter is the iteration index and itermax is the maximum number
of iterations. The large value of s in the first iterations makes the
possibility of global search and a small value in the last iterations
3. Artificial bee swarm optimization algorithm (ABSO)
allows a local search.
Each onlooker bee memorizes the food sources that it has found
The collection and processing of nectar are instances of intelli-
so far and their quality. At each iteration, each onlooker bee selects
gent behaviors of honey bees. The main difference between a bee
one of the dancers by tournament selection approach as its own
swarm and other population-based algorithms is that in a bee
interesting elite. It then adjusts its position using its own knowl-
swarm different kinds of bees are employed which use different
edge and that of its interesting elite bee. In tournament selection,
types of trajectories to amend their positions.
weak dancers have a smaller chance of being chosen when the
In ABSO, each food source is considered as a position (feasible
tournament size is large. This approach permits the selection pres-
solution) in a d-dimensional food source region and the objective
sure to be adjusted easily by the tournament size. In this paper,
function of each bee indicates the quality of the food source discov-
tournament size is set to 2. Each onlooker bee adjusts its position
ered by it. The bees leave their hive to find new food sources and
according to
return to share their observations in relation to the quality of the
discovered food sources. They are then partitioned into two xjnew ¼ xj þ wb  r b ðxjb  xj Þ þ we  r e ðxje  xj Þ ð10Þ
groups, onlookers and scouts, based on the quality of the best food
where xe is the interesting elite bee, xb is the best achievement
source that they have found so far. The percentage of each group is
found by the onlooker bee, and rb and re are random numbers from
selected manually. It is better to select a small number of bees as
the interval [0, 1]. To make a trade-off between global and local
scouts. In order to determine the kind of each bee, the bees are
search, wb and we are defined as decreasing linear functions. For
ranked in light of their objective functions. In this case, better bees
enhancing the global search in the beginning of the algorithm the
are those that have smaller values of the objective function. Then, a
maximum values (wbmax and wemax) and converging toward an opti-
predefined number of the bees which have the worst quality are
mal solution at the end of the algorithm minimum values (wbmin
chosen as scouts and the others make onlookers. A random strat-
and wemin) are used. The parameters of wb and we are defined as
egy is employed by the scout bees to fly over the food source area.
follows:
Among the onlooker bees, ne bees with the best quality are chosen
as elites. Each hive includes a dance region in which each elite bee wb ¼ wb max  ðwb max  wb min Þiter=iter max ð11Þ
dances and tries to encourage the onlookers to move toward the
best food source it has discovered so far. The way in which each we ¼ we max  ðwe max  we min Þiter=iter max ð12Þ

Table 4
A comparison between the results obtained by the HS-based algorithms and the other ones for the single diode model parameter identification.

Item ABSO CPSO GA PS SA HS

Rs (O) 0.03659 0.0354 0.0299 0.0313 0.0345 0.03663
Rsh (O) 52.2903 59.012 42.3729 64.1026 43.1034 53.5946
Iph (A) 0.76080 0.7607 0.7619 0.7617 0.7620 0.76070
Isd (lA) 0.30623 0.4000 0.8087 0.9980 0.4798 0.30495
n 1.47583 1.5033 1.5751 1.6000 1.5172 1.47538
RMSE 9.9124e4 0.00139 0.01908 0.01494 0.01900 9.9510e4
948 A. Askarzadeh, A. Rezazadeh / Applied Energy 102 (2013) 943–949

1 0.35
Real data
Single diode model
0.8 0.3

0.6 0.25

Objective function
Current (A)

0.4 0.2

0.2 0.15

0
0.1

-0.2
0.05

-0.4
-0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Voltage (V)
Iteration
Fig. 8. Comparison between the I–V characteristics resulted from the experimental
Fig. 10. Convergence process of ABSO algorithm during the identification process of
data and the single diode model.
the single diode model.

Step 2: The value of the objective function for each bee is com-
0.5 puted based in Eq. (6).
Rea1 data Step 3: The bees are ranked based on their objective functions.
0.4
Single diode model Step 4: Onlooker and scout bees are specified.
Step 5: The position of the onlooker and scout bees is updated
0.3 according to their patterns.
Step 6: If a bee exceeds the search space, it is replaced with the
Power (W)

0.2 previous position.

Step 7: Steps 2 to 6 are repeated until itermax is met.
0.1 Step 8: The best achievement of the swarm is selected as the
optimal solution.
0
Fig. 4 depicts the flowchart of the ABSO algorithm which is used
-0.1 to identify the optimal parameters of the solar cell models.

-0.2
-0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 4. Results and discussions
Voltage (V)
The I–V characteristic of a 57 mm diameter commercial (R.T.C.
Fig. 9. Comparison between the P–V characteristics resulted from the experimental France) silicon solar cell is used to evaluate the efficiency of the
data and the single diode model.
Table 5
Relative error for each measurement (single diode model).
ABSO starts with a swarm of N candidate solutions, called bees. Measurement Vt (V) It measured (A) It calculated (A) Relative
It is usual to specify each bee as a d-dimensional vector, error
xi ¼ ½x1i ; x2i ; . . . ; xdi , where i = 1, 2, . . . , N. Each element of the vector
~
1 0.2057 0.764 0.764201 0.00026
is corresponding to an unknown parameter of the problem under 2 0.1291 0.762 0.762737 0.00097
consideration and is given a value in the allowed range. The bees 3 0.0588 0.7605 0.761393 0.00117
are randomly initialized in the search space. The initial swarm 4 0.0057 0.7605 0.76016 0.000447
5 0.0646 0.76 0.759032 0.001274
should cover the entire search space as much as possible. This is
6 0.1185 0.759 0.757992 0.001328
achieved by uniformly randomizing the bees using the following 7 0.1678 0.757 0.757017 2.21e05
equation: 8 0.2132 0.757 0.756047 0.001259
9 0.2545 0.7555 0.754977 0.000692
j j
xji ¼ l þ a  ðuj  l Þ ð13Þ 10 0.2924 0.754 0.753547 0.0006
11 0.3269 0.7505 0.751277 0.00103
where a is a random number between 0 and 1. 12 0.3585 0.7465 0.74726 0.00102
During the subsequent iterations, the bees update their posi- 13 0.3873 0.7385 0.740051 0.0021
tions according to the related pattern until the maximum number 14 0.4137 0.728 0.727411 0.000809
15 0.4373 0.7065 0.707033 0.00076
of iterations is met. During movement, if a bee exceeds search 16 0.459 0.6755 0.675431 0.000102
space, its position will not be updated. Fig. 3 depicts the pseudo- 17 0.4784 0.632 0.631046 0.001509
code of the ABSO algorithm. 18 0.496 0.573 0.57223 0.001344
The steps of the proposed algorithm used in this study to obtain 19 0.5119 0.499 0.499591 0.00118
20 0.5265 0.413 0.413524 0.00127
the optimal parameters of solar cell models are as follows.
21 0.5398 0.3165 0.317184 0.00216
22 0.5521 0.212 0.212023 0.00011
Step 1: A swarm of bees are randomly initialized in the search 23 0.5633 0.1035 0.10263 0.008404
space. Each solution is defined by a vector, x, where x = [Rs Rsh 24 0.5736 0.01 0.00931 0.068809
Iph Isd1 Isd2 n1 n2] in the double diode model and x = [Rs Rsh Iph 25 0.5833 0.123 0.12438 0.01118
26 0.59 0.21 0.20911 0.00423
Isd n] in the single diode model.
 A. Askarzadeh, A. Rezazadeh / Applied Energy 102 (2013) 943–949
proposed ABSO-based parameter identification method. The exper-
imental data has been adopted from the system under 1 sun
(1000 W/m2) at 33 °C [6]. The adjustable parameters of the ABSO
algorithm, obtained by trial, are as follows: the swarm size is set
to 30 of which 25 bees are onlooker and 5 bees are scout, ne = 5,
wbmax = wemax = 2.5, wbmin = wemin = 1.25, itermax = 5000, smax = 0.2,
and smin = 0.02.
Matlab environment is implemented for programming ABSO
algorithm to identify the solar cell models parameters. Table 2 sum-
marizes the optimal parameters and RMSE value for the double
diode model found by ABSO algorithm in comparison with the re-
sults obtained by pattern search (PS) [25] and simulated annealing
(SA) [12]. The results are also compared with those of harmony
search (HS) algorithm. Motivated by the music improvisation pro-
cess, HS is a metaheuristic algorithm which has received significant
attention during recent years [26,27]. However, we are interested in
studying the performance of HS in the parameter identification
problem. It is evident that ABSO outperforms PS, SA, and HS because
it has found the minimum RMSE value. As a result, the optimal
parameters found by ABSO are closer to the real parameters of
the system. After ABSO the best performance belongs to HS.
To investigate the quality of the identified parameters, they are
put into the double diode model, and the I–V characteristic is
reconstructed. This is simply performed by applying Newton meth-
od when It is unknown while Vt is known. The I–V characteristic re-
sulted from the identified model along with the experimental data
are shown in Fig. 5. As can be observed, the model is in good agree-
ment with the experimental data. The power vs. voltage (P–V)
characteristic of the system is also shown in Fig. 6. The similarity
between the model results and the performance of the real system
proves the superiority of the identification process. The conver-
gence process of ABSO algorithm is illustrated in Fig. 7, indicating
the best value of the objective function during the iterations.
The relative error, e ¼ Imeasured Icalculated
Imeasured
, for each measurement is gi-
ven in Table 3. This table confirms the high accuracy of the identi-
fied model.
The results of ABSO are listed in Table 4 in comparison with
those found by chaos particle swarm optimization algorithm
(CPSO) [11], genetic algorithm (GA) [10], pattern search (PS) [25],
simulated annealing (SA) [12], and harmony search (HS) in param-
eters identification of the solar cell single diode model. The ABSO
yields better results than CPSO, GA, PS, SA, and HS. Based on the
RMSE value found, the search power of the algorithms in the param-
eters identification of the single diode model can be ordered as
ABSO > HS > CPSO > PS > SA > GA. The optimal parameters are put
into the single diode model and Newton method is used to recon-
struct the I–V characteristic. Fig. 8 indicates that the I–V character-
istic found by ABSO is in good agreement with the experimental
data. The P–V characteristic and convergence process of the proposed
algorithm are also represented in Figs. 9 and 10, respectively. The
relative error for each measurement is also given in Table 5.
In order to make a comparison between the precision of the
double and single diode models the results reported in Tables 2
and 4 are considered. From Table 2, the best RMSE value is
9.8344e4 while this value is 9.9124e4 from Table 4, both found
by ABSO algorithm. It can be drawn that the double diode model is
slightly more accurate than the single diode model. However, the
error of 0.79% denotes the suitability and accuracy of the single
diode model.
5. Conclusion
This paper deals with the application of artificial bee swarm
optimization (ABSO) algorithm for obtaining accurate results from
the solar cell double and single diode models. The proposed algo-
949
rithm is used to identify the parameters of a 57 mm diameter com-
mercial (R.T.C. France) silicon solar cell. The obtained results are
quite promising and conquer those found by the other methods.
As a result, ABSO can be a good candidate to solve the optimization
problems of solar cell systems.
References
[1] Ishaque K, Salam Z, Mekhilef S, Shamsudin A. Parameter extraction of solar
photovoltaic modules using penalty-based differential evolution. Appl Energy
2012;99:297–308.
[2] Orioli A, Gangi AD. A procedure to calculate the five-parameter model of
crystalline silicon photovoltaic modules on the basis of the tabular
performance data. Appl Energy 2013;102:1160–77.
[3] Sandrolini L, Artioli M, Reggiani U. Numerical method for the extraction of
photovoltaic module double-diode model parameters through cluster analysis.
Appl Energy 2010;87:442–51.
[4] Amrouche B, Guessoum A, Belhamel M. A simple behavioural model for solar
module electric characteristics based on the first order system step response
for MPPT study and comparison. Appl Energy 2012;91:395–404.
[5] Bonanno F, Capizzi G, Graditi G, Napoli C, Tina GM. A radial basis function
neural network based approach for the electrical characteristics estimation of
a photovoltaic module. Appl Energy 2012;97:956–61.
[6] Easwarakhanthan T, Bottin J, Bouhouch I, Boutrit C. Nonlinear minimization
algorithm for determining the solar cell parameters with microcomputers. Sol
Energy 1986;4:1–12.
[7] Chan DSH, Phillips JR, Phang JCH. A comparative study of extraction methods
for solar cell model parameters. Solid-State Electron 1986;29:329–37.
[8] Jian A, Kapoor A. Exact analytical solutions of the parameters of real solar cells
using Lambert W-function. Sol Energy Mater Sol Cells 2004;81:269–77.
[9] Saleem H, Karmalkar S. An analytical method to extract the physical
parameters of a solar cell from four points on the illuminated J–V curve. IEEE
Electron Device Lett 2009;30:349–52.
[10] AlRashidi MR, AlHajri MF, El-Naggar KM, Al-Othman AK. A new estimation
approach for determining the I–V characteristics of solar cells. Sol Energy
2011;85:1543–50.
[11] Wei H, Cong J, Lingyun X, Deyun S. Extracting solar cell model parameters
based on chaos particle swarm algorithm. In: International conference on
electric information and control engineering (ICEICE); 2011. p. 398–402.
[12] El-Naggar KM, AlRashidi MR, AlHajri MF, Al-Othman AK. Simulated annealing
algorithm for photovoltaic parameters identification. Sol Energy 2012;86:266–74.
[13] Gomez M, Lopez A, Jurado F. Optimal placement and sizing from standpoint of
the investor of photovoltaics grid-connected systems using binary particle
swarm optimization. Appl Energy 2010;87:1911–8.
[14] Reinoso CRS, Cutrera M, Battioni M, Milone DH, Buitrago RH. Photovoltaic
generation model as a function of weather variables using artificial
intelligence techniques. Int J Hyrogen Energy 2012. http://dx.doi.org/
10.1016/j.ijhydene.2011.12.081.
[15] Ekren O, Ekren BY. Size optimization of a PV/wind hybrid energy conversion
system with battery storage using simulated annealing. Appl Energy
2010;87:592–8.
[16] Niknam T, Taheri SI, Aghaei J, Tabatabaei S, Nayeripour M. A modified honey
bee mating optimization algorithm for multiobjective placement of renewable
energy resources. Appl Energy 2011;88:4817–30.
[17] Dai C, Chen W, Zhu Y. Seeker optimization algorithm for digital IIR filter
design. IEEE Trans Ind Electron 2010;57:1710–8.
[18] Samanta S, Chakraborty S. Parametric optimization of some non-traditional
machining processes using artificial bee colony algorithm. Eng Appl Artif Intell
2011;24:946–57.
[19] Lien LC, Cheng MY. A hybrid swarm intelligence based particle-bee algorithm
for construction site layout optimization. Expert Syst Appl 2012;39:9642–50.
[20] Askarzadeh A, Rezazadeh A. A new artificial bee swarm algorithm for
optimization of proton exchange membrane fuel cell model parameters. J
Zhejiang Univ Sci C 2011;12:638–46.
[21] Akbari R, Mohammadi A, Ziarati K. A novel bee swarm optimization algorithm
for numerical function optimization. Commun Nonlinear Sci Numer Simulat
2010;15:3142–55.
[22] Yang XS. Engineering optimizations via nature-inspired virtual bee algorithms.
Lect Notes Comput Sci 2005;3562:317–23.
[23] Karaboga D, Basturk B. A powerful and efficient algorithm for numerical
function optimization: artificial bee colony (ABC) algorithm. J Global Optim
2007;39:459–71.
[24] Wolf M, Noel G, Stirm R. Investigation of the double exponential in the
current-voltage characteristics of silicon solar cells. IEEE Trans Ind Electron
1977;53:1017–26.
[25] AlHajri MF, El-Naggar KM, AlRashidi MR, Al-Othman AK. Optimal extraction of
solar cell parameters using pattern search. Renew Energy 2012;44:238–45.
[26] Askarzadeh A, Rezazadeh A. A grouping-based global harmony search
algorithm for modeling of proton exchange membrane fuel cell. Int J
Hydrogen Energy 2012;36:5047–53.
[27] Askarzadeh A, Rezazadeh A. An Innovative global harmony search algorithm
for parameter identification of a PEM fuel cell model. IEEE Trans Ind Electron
2012;59:3473–80.