Renewable Energy 99 (2016) 270e279

Contents lists available at ScienceDirect

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

Comparison of two PV array models for the simulation of PV systems

using five different algorithms for the parameters identification
Sofiane Kichou a, *, Santiago Silvestre a, Letizia Guglielminotti a, Llanos Mora-Lo

pez b,
~ oz-Cero
Emilio Mun
n c
a
MNT Group, Electronic Engineering Department, UPC-BarcelonaTech. Barcelona, C/ Jordi Girona 1-3, Modul C4 Campus Nord UPC, 08034 Barcelona, Spain
b
Dpto. Lenguajes y Ciencias de la Computacio
laga, Campus de Teatinos, sn, Ma

n, Universidad de Ma
laga 29071, Spain
c
IDEA Research Group, University of Ja

en, Campus de Las Lagunillas, 23071 Ja

en, Spain

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

Article history: Simulation is of primal importance in the prediction of the produced power and automatic fault
Received 19 April 2016 detection in PV grid-connected systems (PVGCS). The accuracy of simulation results depends on the
Received in revised form models used for main components of the PV system, especially for the PV module. The present paper
30 June 2016
compares two PV array models, the five-parameter model (5PM) and the Sandia Array Performance
Accepted 3 July 2016
Available online 12 July 2016
Model (SAPM). Five different algorithms are used for estimating the unknown parameters of both PV
models in order to see how they affect the accuracy of simulations in reproducing the outdoor behavior
of three PVGCS. The arrays of the PVGCS are of three different PV module technologies: Crystalline silicon
Keywords:
PV modeling
(c-Si), amorphous silicon (a-Si:H) and micromorph silicon (a-Si:H/mc-Si:H).
Simulation The accuracy of PV module models based on the five algorithms is evaluated by means of the Route
Parameter extraction Mean Square Error (RMSE) and the Normalized Mean Absolute Error (NMAE), calculated for different
Metaheuristic algorithms weather conditions (clear sky, semi-cloudy and cloudy days). For both models considered in this study,
the best accuracy is obtained from simulations using the estimated values of unknown parameters
delivered by the ABC algorithm. Where, the maximum error values of RMSE and NMAE stay below 6.61%
and 2.66% respectively.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction accurate automatic fault detection procedures is crucial [1e3]. Main

faults in PV systems are caused by short circuits or open circuits in
The photovoltaic (PV) market has grown rapidly in recent years PV modules, inverter disconnections and the presence of shadows
worldwide, especially in developed countries, where this growth on the PV array plane [4e6].
has been exponential. One of the main reasons for the high growth On the other hand, the integration of grid-connected PV systems
of the PV industry is the reduction of the cost of PV generation as also requires the capability of managing the uncertainty related to
well as the improvement of the quality and performance of the the fluctuating energy output inherent to these generation plants.
electronics associated with these generation systems. The moni- For this purpose, it is very important to develop accurate fore-
toring and regular performance supervision on the functioning of casting models in order to achieve an easy integration of PV gen-
grid-connected PV systems is basic to ensure an optimal energy eration plants into traditional power distribution systems [7,8].
harvesting and reliable power production at competitive costs. Simulation plays a crucial role in both outdoor behavior fore-
Detecting faults in PV systems can minimize generation losses by casting and automatic fault detection of grid-connected PV sys-
reducing the time in which the system is working below its point of tems. The precision of simulation results depends on the models
maximum power generation. In this context, the development of used for the main components of the PV system, especially the PV
module models [9,10]. Moreover, the accuracy of the PV module
models is strongly affected by the way of extracting their unknown
parameters. Several research works discussed the topic of PV model
* Corresponding author.
parameters estimation, by applying different methods based on
E-mail addresses: kichousofi[email protected] (S. Kichou), Santiago.silvestre@upc.
edu (S. Silvestre), [email protected] (L. Guglielminotti), llanos@lcc. analytical [11], numerical [12,13] and bio-inspired optimization
uma.es (L. Mora-Lo
pez), [email protected] (E. Mun ~ oz-Cero

n). solution [14e20].

http://dx.doi.org/10.1016/j.renene.2016.07.002
0960-1481/© 2016 Elsevier Ltd. All rights reserved.
 S. Kichou et al. / Renewable Energy 99 (2016) 270e279 271

Previous works investigated the accuracy of PV module models internal data acquisition card of the inverter recorded both
focusing on the I-V curve of the PV module [21e24] or on the I-V parameters.
characteristic of a PV array [25]. The objective of this study is to
n
The monitoring system included in the PV arrays located in Jae
compare two PV array models to analyze the simulation of grid- consists of three SMA Sunny SensorBox devices, installed in the
connected PV systems in real conditions of work. The accuracy of same plane as the PV generators, capable to measure solar radia-
the simulations in reproducing the actual behavior of the PV system tion, module and ambient temperatures together with wind speed.
is evaluated by means of the results obtained from different Two Pt100 RTD were pasted to the rear surface of the modules
parameter extraction techniques based on five algorithms: Lev- under test to measure the cell temperature in each PV array. An
enbergeMarquardt algorithm (LMA), genetic algorithm (GA), par- anemometer and a temperature probe were also available. All
ticle swarm optimization (PSO), differential evolution (DE) and sensors were supplied by SMA and connected to three Sunny
artificial bee colony (ABC) algorithm. SensorBox devices. An additional irradiance sensor, aKipp & Zonen
The two PV array models included in this study are the five- CMP11 pyranometer, was also installed and connected to one of the
parameter model (5PM) [26,27] and the Sandia Array Perfor- latter devices. The three of them were serially connected to the
mance Model (SAPM) developed by Ref. [28]. Three real grid- inverters via a RS-485 bus and then to a Sunny Webbox, from which
connected PV systems are included in the study to validate the environmental and operation could be retrieved.
accuracy of the models. Each one of the PV systems is formed by PV
modules of different technologies: Crystalline silicon (c-Si), amor-
3. PV array models
phous silicon (a-Si:H) and micromorph silicon (a-Si:H/mc-Si:H) in
order to outline differences in the prediction due to solar cell type.
As it has been previously mentioned, the two PV array models
The remainder of the paper is organized as follows: In Section 2,
included in this study are the 5PM [26,27,29] and the SAPM
the PV systems included in the study are described. The PV array
developed by Ref. [28].
models and the parameters extraction techniques used in this
The 5PM, also called one diode model, is one of the most used in
study are summarized in Sections 3 and 4 respectively. Results
simulation of PV modules and arrays. Moreover, root mean square
obtained are shown in Section 5. Finally, conclusions are detailed in
errors (RMSE) of 4.26% [3], 4.39% [30] and 5.12% [31] were reported
Section 6.
in the estimation of the energy produced by grid-connected PV
systems in simulations of dynamic behavior of c-Si PV generators by
2. Description of the PV systems using this model. On the other hand, simulations of a-Si PV arrays by
using the SAPM model have obtained errors below 4.1% on sunny
Three grid connected PV systems formed by PV modules of days [32]. In our approach, the model parameters are calculated by
different technologies were used in this study. means of parameter extraction methods having as main input data
The first PV system is located in San Sebasti
an (Spain). The PV daily actual profiles of module temperature, irradiance on the PV
array is formed by 30 c-Si PV modules with a peak power of array plane and output voltage and current of the PV array.
4.8 kWp connected to a single phase inverter.

n (Spain). Each PV array
The other two PV systems are sited in Jae 3.1. Five-parameter model
is connected to single phase inverter with AC nominal powers of
1.2 kW. One of the PV arrays is formed of 15 a-Si:H PV modules, The 5PM of a solar cell includes a parallel combination of a
rated 60-W peak, and the second PV array consists of 8 micromorph photogenerated controlled current source Iph, a diode, described by
PV modules, rated 110-Wp each. Main characteristics of the PV the well-known single-exponential Shockley equation [33], a shunt
systems and PV modules forming the arrays are given in Table 1 and resistance Rsh and a series resistance Rs modeling the power losses.
Table 2 respectively. The I-V characteristic of a solar cell is given by an implicit and
The following parameters were monitored in the three PV nonlinear equation as follows:
arrays: Current, voltage, power (DC and AC), cosine (4), frequency,  
irradiance and module temperature with a sampling rate of VþRs I
!  
5 min. nVt V þ Rs I
I ¼ Iph  Io e 1  (1)
In the PV system located in San Sebastia
n, the irradiance was Rsh
measured by using a calibrated solar cell installed in the plane of
the modules. The module temperature was measured using a Pt100 where Io and n are the reverse saturation current and ideality factor
sensor fitted to the back of the module, in the middle of a cell. The of the diode respectively and Vt is the thermal voltage.

Table 1
PV systems description.

Main Parameters PV system 1 PV system 2 PV system 3

PV Module c-Si a-Si:H/mc-Si:H a-Si:H

Location
n (Spain)
San Sebastia
n (Spain)
Jae
Latitude: 43 170 9.800 N Latitude: 37 470 14.3500 N
Longitude: 1 590 55.4 00 W Longitude: 3 460 39.73 00 W
Altitude: 41 m. Altitude: 511 m
Nominal power 4.8 kWp 880 Wp 900 Wp
Modules per inverter 30 8 15
Modules in series (Nsg) 15 4 3
Strings in parallel (Npg) 2 2 5
Tilt - Orientation 20 e 9 East 30 e 0 South 35 e 0 South
Inverter Ingecon SUN 5 Sunny Boy SB1200
Single-phase inverter Single-phase inverter
5 kW 1.2 kW
272 S. Kichou et al. / Renewable Energy 99 (2016) 270e279

Table 2
Main parameters of PV modules.

PV module Parameters PV system 1 PV system 2 PV system 3

Isc (A) 9.46 2.5 1.19

Voc (V) 22.2 71 92
Current at Maximum Power Point: Impp (A) 8.65 2.04 0.9
Voltage at Maximum Power Point: Vmpp (V) 18.5 54 67
Temperature Coefficient of Voc bVoc (V/ C) 0.084 0.248 0.280
Temperature Coefficient of Isc aIsc (A/ C) 4.60  103 1.40  103 0.89  103

Eq. (1) can also be written as follows,  

Ego Eg

Vto Vt !3
I ¼ Iph  Id  Ish (2) IscM e Tc
IoM ¼   (11)
VocM ! Tc*
where Id and Ish are the currents across the diode and shunt resis- n Ns Vto

tance respectively. Np e 1
The photogenerated current can be evaluated for any arbitrary
value of irradiance, G, and cell temperature, Tc, by using the where IscM and VocM are the short-circuit current and the open-
following equation: circuit voltage of the PV module respectively, Vto is the thermal
  voltage at STC, Eg the energy bandgap of the semiconductor and Ego
G
Iph ¼ Isc þ ki Tc  Tc* (3) is the energy bandgap at T ¼ 0 K.
G* The value of the energy bandgap of the semiconductor at any
where G* and Tc* are respectively the irradiance and cell temper- cell temperature Tc is given by:
ature at standard test conditions (STC): 1000 W/m2 (AM1.5) and
25  C, ki (A/ ) is the temperature coefficient of the current and Isc
agap Tc2
Eg ¼ Ego  (12)
(A) is the solar cell short circuit current at STC. bgap þ Tc
Some PV modules are formed by parallel strings of solar cells
where agap and bgap are fitting parameters characteristic of the
connected in series. However, most PV modules include one single
semiconductor.
string of solar cells. Therefore, the model of the solar cell can be
Finally, the current IshM, also included in Eq. (9) is given by the
scaled up to the model of the PV module using the following Eqs.
following equation:
(4)e(8):
VM þ IM RsM
IM ¼ Np I (4) IshM ¼ (13)
Np RshM

IscM ¼ Np Isc (5) The same procedure can be applied to scale up the model of the
PV module to the model of a PV array by taking into account the
number of PV modules connected in series by string, Nsg, and the
VM ¼ Ns V (6) number of parallel strings in the PV array, Npg [27].

VocM ¼ Ns Voc (7)

3.2. SAPM model
Ns
RsM ¼ Rs (8) The SAPM model is an empirical model defined by the following
Np
equations [28]. The PV array power at the maximum power point
Where subscript M stands for ‘Module’, Ns is the number of solar (MPP), Pmp (W), is evaluated as follows:
cells connected in series and Np is the number of parallel branches
of solar cells forming the module. Pmpg ¼ Impg  Vmpg (14)
Then, the output current of the PV module, IM, is obtained
rewriting Eq. (2) as follows: where, Impg (A) and Vmpg (V) are the coordinates of the MPP of the
PV array.
  The model uses the normalized irradiance, Ee, defined as
IM ¼ Np Iph  IdM  IshM (9)
follows,
The diode current, IdM, included in Eq. (9) is given by: G
  Ee ¼ (15)
G*
" VM þIM RsM #
n Ns Vt Then, the current and voltage of the MPP of the PV array can be
IdM ¼ IoM e 1 (10)
calculated by using the following equations:
h    i
where VM (V) and IM (A), are the output voltage and current of the Impg ¼ Npg Impo C0 Ee þ C1 Ee2 1 þ aImp Tc  Tc*
PV module respectively.
(16)
The saturation current of the diode IoM (A) depends strongly on
temperature and it is given by:
 S. Kichou et al. / Renewable Energy 99 (2016) 270e279 273

h 4.1. Genetic algorithm

Vmpg ¼ Nsg Vmpo þ C2 Ns dðTc ÞlnðEeÞ þ C3 Ns ðdðTc ÞlnðEeÞÞ2
 i The Genetic Algorithm (GA) developed by John Holland in the
þ bVmp Ee Tc  Tc* 1970s is a technique for solving constrained and unconstrained
optimization problems inspired from the biological evolution.
(17)
The optimization function is encoded as arrays of binary char-
acter strings representing the chromosomes. The fitness of chro-
dðTc Þ ¼ nkðTc þ 273:15Þ=q (18)
mosomes in the population is evaluated by the objective function
for each iteration. Fitter chromosomes are stochastically selected in
where, Impo (A) and Vmpo (V) are the PV module current and
terms of the elitist strategy, which ensures the progeny chromo-
voltage of the MPP at STC, C0 and C1 are empirically determined
somes inherit the best possible combination of the genes of their
coefficients (dimensionless) which relate Imp to the effective irra-
parents. Some of the chromosomes in the population are modified
diance, C0 þ C1 ¼ 1, aImp ( C1) is the normalized temperature co-
via genetic operators like crossover and mutation, forming new
efficient for Imp, C2 (dimensionless) and C3 (V1) are empirical
chromosomes for the next generation. The reason why GA applies
coefficients which relate Vmp to the effective irradiance, d(Tc) is the
crossover and mutation may lie in their capability of avoiding local
thermal voltage per cell at temperature Tc, q is the elementary
optima in the searching process. Several researches applied GA to
charge, 1.60218  1019 (coulomb), k is the Boltzmann’s constant,
extract the parameters of a PV model from measured IeV curves
1.38066  1023 (J/K) and bVmp (V/ C) is the temperature coefficient
[17,35].
for module Vmp at STC.
In this paper, the genetic algorithm available in the Global
The models contain several coefficients and parameters that
Optimization toolbox of MATLAB has been used for minimizing the
must be calculated because are not routinely provided by the PV
objective function Eq. (19) [17].
module’s manufacturer. For this purpose, we used the parameter
extraction techniques described in the following section.
4.2. Differential evolution

4. Parameter extraction techniques

Differential evolution (DE) was proposed by Rainer Storn and
Kenneth Price in 1997 [36]. Similar to other evolutionary algo-
The parameter extraction techniques employed in this study are
rithms, DE is a population based, derivative-free function optimizer.
based on five optimization algorithms that evaluate the model
An advantage of DE over GA is that DE treats possible solutions as
parameters of the two PV array models in real conditions of work,
real-number strings, and thus encoding and decoding are not
using as inputs daily profiles of solar irradiance and cell tempera-
required.
ture together with monitored DC output current and voltage.
The target vector x ¼ [x1, x2, …, xi] where i ¼ 1,2, …, NP repre-
For the five-parameter model of the PV module, the model pa-
sents a population of NP random candidate solutions. The vector of
rameters: Iph, Io, n, Rs, and Rsh are evaluated by using Eqs. (3)e(13)
the ith particle, xi indicates a series of parameters to be extracted,
and actual daily profiles of monitored current and voltage at the
e.g. xi ¼ [Iph,Io,n,Rs,Rsh] for the one-diode model and xi ¼ [C0, C1, C2,
DC output of the three PV arrays included in the study, together
C3, n, aImp, bVmp]. For a D-dimension optimization problem, a
with actual daily profiles of G and Tc at the specific locations
random candidate solution is given by:
detailed in Section 2.
Regarding the SAPM, the same idea is considered for the esti- up
xlow
j  xi;j  xj (20)
mation of the empirical coefficients of the model parameters: C0, C1,
C2, C3, n, aImp and bVmp using Eqs. (15)e(18).
where xlow
j and xup
j are the lower and the upper limits of the jth
The objective function for optimization using metaheuristic al-
vector component respectively, i ¼ 1, 2, …, NP and j ¼ 1, 2, …, D.
gorithms is defined as the RMSE of the error of all data points given
After the initialization DE enters a loop of evolutionary opera-
by Eq. (19) [19,34], where the N represent the number of measured
tions: mutation, crossover and selection considering the maximum
data, Vi and Ii represent the measured voltage and current of the
number of generations tmax, where t ¼ 1, 2, …, tmax.
data point i.
In the mutation step, for each xi at generation t, three vectors xr0,
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi xr1 and xr2 are chosen randomly from the set {1, 2, …,NP}\{i} to
u
u1 X N
SðqÞ ¼ t ½Ii  IðVi; qÞ2 (19) generate a donor vector by:
N i¼1  
vtþ1
i ¼ xtr0 þ F xtr1  xtr2 (21)
where q ¼ f (Iph,Io,n,Rs,Rsh) for the five parameter model and q ¼ f
(C0, C1, C2, C3, n, aImp, bVmp) for the SAPM. where F is a differential weight, known as scaling parameter, usu-
The parameter extraction algorithms implemented in MATLAB/ ally ranges in the interval [0, 1].
Simulink environment are executed until function S(q), given by Eq. The crossover operation is used to decide whether to exchange
(19), is minimized. Figs. 1 and 2 show the Simulink block diagram of with donor vector. By generating a random integer index Jr 2 [1, D]
the 5PM and SAPM used in the parameter extraction procedures. and a randomly distributed number ki 2 [0, 1], the jth dimension of
Thus, the result of the parameter extraction algorithms is a set of PV vi, namely ui,j, is updated according to:
module parameters for the 5PM and a set of empirical parameters (
for the SAPM that allow the best approach to the real daily evolu- vtþ1 ; ki  CR or i ¼ Jr
utþ1 ¼ i;j
(22)
tion of DC output current and voltage of the PV arrays.
i;j
xti;j ; ki > CR and isJr
Two parameter extraction methods are used in this study. The
first method is a numerical solution based on Lev- where CR is a crossover probability in the interval [0, 1]. The
enbergeMarquardt algorithm (LMA) detailed in a previous work crossover scheme formulated by Eq. (22) used in the present work
[12]. The second method is based on different metaheuristic algo- is called binomial strategy.
rithms (GA, DE, PSO and ABC) which are described below. The selection operation, selects the best one from the parent
274 S. Kichou et al. / Renewable Energy 99 (2016) 270e279

Fig. 1. Simulink block diagram for the 5PM.

Fig. 2. Simulink block diagram for the SAPM.

vector xti , and the trial vector utþ1

i solution with the minimum    
objective value, using the following expression: vtþ1
i
¼ wvti þ aε1 xti  gbest t þ bε2 xti  pbestit (25)
(    
utþ1 ; f utþ1  f xti
xtþ1 ¼ i i (23) a ¼ 1.5, b ¼ 2. The random vectors 3 1 and 3 2 are in the range [0,
i
xti ; othewise
1]. The w is the inertia weight, used to balance global and local
search abilities, it is considered constant and set equal to 0.9.
where f(x) is the fitness function to be minimized. Therefore, if a
Finally, lower and upper boundaries are set to ensure that
particular trial vector is found to result in lower fitness value, it will
particles are within the predetermined range. The PSO will
replace the existing target vector; otherwise, the target vector is
continue to search for better solutions until it meets the stopping
retained.
criterion.
4.3. Particle swarm optimization
4.4. Artificial bee colony algorithm
Particle swarm optimization (PSO) is a population based sto-
chastic optimization technique developed by Kennedy and Eber- The artificial bee colony algorithm (ABC) is an optimization
hart [16] and is inspired by the social behavior of bird flocking or algorithm inspired by the natural foraging behavior of honey bees.
fish schooling. It was successfully applied in the parameter extraction of solar cell
PSO search possible solution in a search space by adjusting the models [19,34]. In the ABC, there are food sources representing the
trajectories of particles. The best position encountered of the par- solutions of optimization problems and honey bees (classified into
ticle i is designed by pbesti. In a swarm of particles, there are N local employed bees, onlooker bees and scout bees) representing the
best positions, and the best solution is denoted by gbest. operations to the solutions. The employed bees investigate po-
The velocities and positions of particles, as well as the algorithm tential food sources and share information with onlooker bees. The
parameters (inertia weight w and learning parameters a, b) are food sources of higher quality will have higher possibility to be
firstly initialized. In an iteration t, the fitness of particles is evaluated selected by onlooker bees. If the quality of the employed bees’ food
individually by the objective function. By attracted toward pbesti and sources is relatively low, they will change to scout bees to
gbest, the particle moves according to the following expression: randomly explore new potential food sources. Consequently, the
exploitation is promoted by employed and onlooker bees while the
xtþ1
i ¼ xti þ vtþ1
i (24) exploration is performed by scout bees. The implementation of the
ABC algorithm in MATLAB is carried out by following the same
where vtþ1
i is the velocity, expressed as: steps of given in the previous works [19,34,37].
 S. Kichou et al. / Renewable Energy 99 (2016) 270e279 275

5. Results Table. 4
Lower and upper boundaries selected for each PV module model parameter.

The results of simulation of grid-connected PV systems in real C0 [0e2] Iph [A] [0e10]
conditions of work were obtained under different weather condi- C1 [1e1] Io [A] [107e1011]
tions: clear sky, semi-cloudy, and cloudy weather. The two PV array C2 [10e10] n [1e2]
C3 [10e100] Rs [U] [0e20]
models described above were used for forecasting the output po- aImp [ C1] [104e102] Rsh [U] [50e105]
wer of the three different PV systems using the extracted param- bVmp [V/ C] [1e0]
eters delivered by the five algorithms.
The adjustable parameters chosen for the GA, DE, PSO and ABC
algorithms and the lower and upper boundaries selected for each
calculated for the two PV models given in Tables 7 and 8. The values
parameter are summarized in Tables 3 and 4.
quantify discrepancies between measured data (DC output current,
The optimization algorithms used in the parameter extraction
voltage and power) versus simulated ones predicted by the two PV
techniques evaluate the model parameters of the PV module; Iph, Io,
array models using the five algorithms (LMA, GA, PSO, DE and ABC).
n, Rs, Rsh, in case of the 5PM, and C0, C1, C2, C3, n, aImp, bVmp, in case of
Two metrics were used: The Route Mean Square Error (RMSE) [32]
SAPM.
and the Normalized Mean Absolute Error (NMAE) [10]. For the error
In the case of using the extraction method based on LMA, an
calculation an irradiance filter was applied to the data set. Only the
average number of 10 iterations are needed in order to find a set of
data corresponding to irradiance values above 200 W/m2 were
solar cell model parameters for an input data set corresponding to
considered, since the inverters start working in these conditions.
one day of real operation of the PV array. On the other hand, for the
Below this irradiance value, the PV systems are in an open circuit
extraction method relied on the metaheuristic algorithms (GA, PSO,
configuration, and the resulting values are misleading.
DE and ABC) the average number of iterations is much higher, by
The DC output power of the PV array is obtained as a product of
around 500 iterations are needed.
current and voltage in both real and simulated results.
Moreover, the parameter extraction methods were applied for
As a general trend, the errors obtained in the case of SAPM
each sample day separately, in order to get the optimal set of pa-
model were smaller than in the case of the 5PM for all PV systems
rameters of the two PV models that allows reproducing the real
and weather conditions regardless of the solar cell technology.
behavior of the PV systems with best accuracy. As the extracted
Similarly, for each PV system the error decreases with improving
parameters values obtained by the different algorithms are very
weather conditions: The error for clear sky day was smaller than for
close to each other, it is decided to show the mean value of each
semi-cloudy day, while for cloudy day the largest discrepancy was
extracted parameter. The set of the extracted parameters are listed
always found, as anticipated from the inspection of Figs. 3 e 8.
in Tables 5 and 6.
The maximum values of RMSE and NMAE obtained for the
In order to present the best variety of results, and see the per-
output power using the SAPM model were 6.02% and 2.40%
formance of the two models using real conditions of solar irradi-
respectively. These values were provided by simulations based on
ance and cell temperature, it was chosen to display the DC output
LMA of the PV system 1 with c-Si PV modules in a cloudy day.
current evolution over the course of a clear sky day for PV system 1,
Nevertheless, for the PV systems 2 and 3 based on different PV
a semi-cloudy day for PV system 2 and a cloudy day for PV system 3.
module technologies, the RMSE and NMAE errors obtained for DC
Figs. 3e8 show the measured DC output current of the three PV
output power were below 4% and 1.86 %.
systems, compared with the simulation results obtained with the
On the other hand, in the simulations based on the 5PM the
two PV array models using the extracted set of parameters esti-
maximum values of RMSE and NMAE obtained regarding the DC
mated by the five optimization algorithms considered in this study.
output power were increased up to 13.55% and 5.30% for PV system
As it can be seen in the figures, a good agreement is always
1 based on LMA. However, for the PV systems 2 and 3, even based
found between the measured data and the SAPM simulation curves,
on the LMA, the obtained values of RMSE and NMAE were 6.99%
while the curves obtained with the 5PM are less close to the real
and 3.29 %.
monitored curve. Moreover, it is found that a better agreement
The accuracy of the PV module models in reproducing the
between real and simulated curve is always reached in clear sky
behavior of the PV array under outdoor conditions of solar irradi-
days rather than in cloudy days. It is qualitatively noted that the
ance and cell temperature depends also on the used methods for
worse the weather conditions, the more difficult is for the models
parameters estimation. As it can be seen from Tables 7 and 8, the
to approximate real data as expected.
metaheuristic algorithms provide lower values of RMSE and NMAE
By comparing the optimization algorithms used for the esti-
than the numerical traditional method based on the LMA.
mation of the unknown parameters of the two PV array models, it
Considering the SAPM, the passage from using the LMA to GA as
can be clearly seen that the metaheuristic algorithms provide good
a main algorithm of the parameter extraction, reduces the
results compared to the LMA in all weather conditions and for both
maximum values of RMSE and NMAE of the DC output power to
PV models.
5.84% and 2.35% taking into account all the PV systems and weather
These considerations are confirmed by values of errors
conditions. This passage from LMA to GA also affects the accuracy of
the 5PM, where the maximum values of RMSE and NMAE of the DC
output power were reduced to 11.23% and 4.12% respectively.
Table. 3
The best accuracy of simulations using the SAPM was obtained
Selected parameters of each algorithm
by using the ABC algorithm for the estimation of the unknown
Algorithm parameters GA PSO DE ABC parameters. The greatest RMSE and NMAE values obtained
Population (colony) size, (NP) 100 100 100 100 regarding the DC power of the PV system 1 were 5.78% and 2.26%.
Inertia weight, (w) e 0,9 e e Otherwise for PV system 2 the errors values don’t exceed 3.13% and
a and b e 1.5 and 2 e e 1.61%, and for PV system 3 the best accuracy is achieved, whatever
Crossover probability (CR) e e 0.4 e
Number of onlooker bees e e e 50
the weather condition, the RMSE and NMAE are below 1.43% and
Limit of scout bees e e e 420 1.02% respectively.
Maximum number of iteration 1000 1000 1000 1000 On the other hand, for the 5PM, the best forecasting of the DC
276 S. Kichou et al. / Renewable Energy 99 (2016) 270e279

Table. 5
Mean values of the main PV module parameters obtained from the parameter extraction algorithms for the 5PM.

PV system Day Weather conditions Rs [U] Rsh [U] Io [A] Iph [A] n

1 09/12/2013 Clear sky 0.662 660.011 1.07  108 8.7268 1.191

18/12/2013 Semi cloudy 0.701 651.880 1.14  108 8.7366 1.192
20/12/2013 Cloudy 0.701 651.894 1.14  108 8.7366 1.192
2 05/07/2012 Clear sky 5.771 25.96  103 2.32  107 2.2055 1.223
12/05/2012 Semi cloudy 7.321 20.34  103 4.90  107 2.2462 1.290
12/11/2012 Cloudy 8.010 21.31  103 1.20  107 2.2462 1.289
3 07/08/2011 Clear sky 12.354 3.358  103 8.82  109 1.0751 1.343
12/05/2012 Semi cloudy 17.915 2.365  103 7.92  109 1.0627 1.351
12/11/2012 Cloudy 19.796 2.865  103 1.36  109 1.0686 1.351

Table. 6
Average values of main parameters obtained from the parameter extraction algorithms for the SAPM.

PV System Day Weather conditions C0 C1 C2 C3 n aImp [ C1] bVmp [V/ C]

1 09/12/2013 Clear sky 1.0438 0.2000 2.0686 21.2425 1.1619 4.32  103 0.1067
18/12/2013 Semi cloudy 0.9138 0.0552 1.6104 10.9348 1.1613 4.32  103 0.1168
20/12/2013 Cloudy 0.9762 0.1468 2.0351 12.7702 1.162 4.32  103 0.0554
2 05/07/2012 Clear sky 0.8887 0.0662 2.575 31.7208 1.2177 5.8  104 0.2819
12/05/2012 Semi cloudy 0.9237 0.0500 2.995 43.1182 1.2459 5.8  104 0.2692
12/11/2012 Cloudy 0.9208 0.0608 2.4241 20.0134 1.2466 5.8  104 0.4632
3 07/08/2011 Clear sky 0.8229 0.0500 2.1346 18.999 1.3162 7.52  103 0.2467
12/05/2012 Semi cloudy 0.7973 0.0400 2.7898 27.9781 1.3537 7.52  103 0.3299
12/11/2012 Cloudy 1.0010 0.1086 1.7077 7.8209 1.2941 7.52  103 0.4998

Fig. 3. Evolution of the DC-current of the PV system 1 using SAPM for clear sky day (December 09th, 2013).

Fig. 4. Evolution of the DC-current of the PV system 1 using 5PM for clear sky day (December 09th, 2013).

output power of the PV systems is also obtained from simulations Finally, regarding the DC output current, the highest values of
using the estimated parameters provided by the ABC algorithm. RMSE obtained in clear sky and semi cloudy day, are below 2.91% in
Considering the worst weather condition, the RMSE and NMAE case of SAPM and 3.42% in case of 5PM. In order to make the ob-
values related to DC output power obtained for the PV system 1 are tained results more comprehensive, other machines learning used
6.6% and 2.67%. However, for the PV systems 2 and 3 the errors for modeling the DC output current of PV arrays were considered.
values remain below 3.65% and 2.07%. Ameen et al. [13] reported RMSE of 5.67% in a work based on
 S. Kichou et al. / Renewable Energy 99 (2016) 270e279 277

Fig. 5. Evolution of the DC-current of the PV system 2 using SAPM for semi-cloudy day (May 12th, 2012).

Fig. 6. Evolution of the DC-current of the PV system 2 using 5PM for semi-cloudy day (May 12th, 2012).

Fig. 7. Evolution of the DC-current of the PV system 3 using SAPM for cloudy day (November 12th, 2012).

Fig. 8. Evolution of the DC-current of the PV system 3 using 5PM for cloudy day (November 12th, 2012).
278 S. Kichou et al. / Renewable Energy 99 (2016) 270e279

Table 7
Calculated RMSE (%) and NMAE (%) for the SAPM.

PV system Day Weather Error [%] LMA GA PSO DE ABC

I V P I V P I V P I V P I V P

1 09/12/2013 Clear sky RMSE 0.64 2.09 1.72 0.64 1.26 1.18 0.64 0.84 1.00 0.65 0.84 0.99 0.65 0.71 0.63
NMAE 0.27 1.43 0.77 0.25 0.97 0.58 0.26 0.62 0.45 0.26 0.62 0.45 0.27 0.48 0.25
18/12/2013 Semi cloudy RMSE 2.91 4.09 2.87 2.51 2.98 2.68 2.50 2.98 2.63 2.50 2.90 2.59 2.50 2.89 2.59
NMAE 1.29 2.11 1.12 0.86 1.83 0.97 0.83 1.84 0.94 0.83 1.70 0.89 0.83 1.69 0.91
20/12/2013 Cloudy RMSE 6.37 5.06 6.02 6.41 4.90 5.84 6.36 4.91 5.77 6.35 4.87 5.79 6.37 4.91 5.78
NMAE 2.43 3.51 2.40 2.54 3.34 2.35 2.44 3.34 2.26 2.44 3.32 2.27 2.44 3.35 2.26
2 05/07/2012 Clear sky RMSE 1.33 1.42 1.55 1.29 0.82 1.14 1.31 0.81 1.14 1.29 1.02 1.06 1.27 0.84 1.03
NMAE 0.46 1.48 0.78 0.53 1.23 0.70 0.47 1.29 0.58 0.51 1.73 0.55 0.53 1.47 0.52
12/05/2012 Semi cloudy RMSE 1.54 1.13 1.55 1.52 0.98 1.53 1.52 1.11 1.41 1.75 1.49 1.36 1.53 1.11 1.32
NMAE 0.62 1.67 0.88 0.59 1.50 0.88 0.59 1.90 0.87 0.75 2.68 0.85 0.61 1.89 0.83
12/11/2012 Cloudy RMSE 2.75 3.50 3.51 2.78 3.32 3.17 2.76 3.22 3.15 2.76 3.22 3.15 2.76 3.31 3.13
NMAE 0.70 5.91 1.84 0.68 4.59 1.65 0.69 4.32 1.62 0.68 4.31 1.61 0.69 4.57 1.61
3 07/08/2011 Clear sky RMSE 1.37 0.92 1.43 1.04 0.95 1.17 1.04 0.88 1.10 1.04 0.77 0.99 1.04 0.76 0.98
NMAE 1.25 0.56 0.78 0.90 0.64 0.66 0.90 0.56 0.59 0.91 0.64 0.51 0.90 0.61 0.48
12/05/2012 Semi cloudy RMSE 1.91 0.89 2.20 1.23 0.81 1.10 1.24 0.90 0.93 1.24 0.82 1.07 1.23 0.89 0.91
NMAE 1.70 0.81 1.07 1.05 0.68 0.49 1.08 0.82 0.43 1.07 0.68 0.48 1.07 0.81 0.41
12/11/2012 Cloudy RMSE 2.67 2.39 4.00 2.40 1.87 2.16 2.42 1.62 1.98 2.42 1.68 2.07 2.25 1.62 1.42
NMAE 2.12 3.27 1.86 1.75 2.34 1.09 1.79 2.04 0.66 1.75 2.08 1.06 1.75 2.04 1.01

Table 8
Calculated RMSE (%) and NMAE (%) for the 5PM.

PV system Day Weather Error [%] LMA GA PSO DE ABC

I V P I V P I V P I V P I V P

1 09/12/2013 Clear sky RMSE 1.78 1.39 2.29 1.76 1.39 2.23 1.75 1.39 2.22 1.75 1.38 2.21 1.75 1.38 2.21
NMAE 0.89 0.98 1.05 0.88 0.98 1.05 0.88 0.98 1.05 0.87 0.97 1.04 0.87 0.96 1.04
18/12/2013 Semi cloudy RMSE 3.42 3.93 4.96 3.37 3.84 4.88 3.37 3.80 4.05 2.84 3.82 3.72 2.55 4.84 3.69
NMAE 1.38 2.48 2.19 1.35 2.48 2.13 1.34 2.45 1.94 1.28 2.46 1.80 0.97 3.08 1.74
20/12/2013 Cloudy RMSE 10.34 4.92 13.55 9.34 5.80 11.23 7.73 4.87 6.96 6.41 6.29 7.79 5.60 4.91 6.60
NMAE 4.37 3.63 5.30 4.30 3.51 4.12 3.63 3.32 2.91 3.17 4.76 2.99 2.14 3.62 2.67
2 05/07/2012 Clear sky RMSE 1.35 2.07 2.43 1.34 2.07 2.42 1.34 2.06 2.41 1.34 2.06 2.40 1.34 1.38 2.09
NMAE 0.48 3.03 1.59 0.48 3.02 1.59 0.48 3.03 1.59 0.47 3.01 1.57 0.47 2.47 1.45
12/05/2012 Semi cloudy RMSE 1.60 2.98 3.51 1.60 2.92 3.41 1.60 2.28 3.13 1.60 2.27 3.13 1.61 2.12 3.07
NMAE 0.64 5.40 2.50 0.65 5.24 2.42 0.65 3.71 2.10 0.65 3.70 2.10 0.64 3.72 2.08
12/11/2012 Cloudy RMSE 4.13 3.24 5.01 3.16 3.25 4.86 2.44 2.98 3.98 3.70 3.24 4.60 3.50 3.14 3.64
NMAE 1.53 5.83 3.87 1.15 5.83 3.17 0.87 5.09 2.54 1.27 5.83 2.72 1.16 5.29 2.06
3 07/08/2011 Clear sky RMSE 1.91 2.44 3.32 1.90 2.43 3.31 1.91 2.16 1.57 1.83 1.92 2.12 0.85 2.31 1.28
NMAE 1.61 1.77 1.71 1.60 1.75 1.73 1.61 1.59 1.69 1.09 0.89 1.01 0.79 1.88 0.67
12/05/2012 Semi cloudy RMSE 1.66 2.68 3.53 1.72 2.09 3.36 1.67 1.97 3.34 1.65 1.95 3.17 1.66 1.95 3.02
NMAE 1.51 2.49 1.78 1.52 1.74 1.67 1.52 1.76 1.66 1.51 1.74 1.60 1.51 1.75 1.53
12/11/2012 Cloudy RMSE 5.36 5.10 6.99 3.44 5.10 4.84 2.53 2.36 2.63 2.12 2.52 1.89 2.09 2.53 1.78
NMAE 4.25 3.22 3.29 2.76 3.21 2.44 1.89 2.18 1.42 1.60 2.24 0.91 1.51 2.26 0.80

artificial neural networks for forecasting the output current of a PV for applications of supervision and forecasting of energy produc-
array. Ibrahim et al. [38] published a novel machine learning con- tion. The RMSE obtained in the comparison of the daily evolution of
sisting in using random forests technique for modeling the output main electrical parameters of the PV systems is below 8% in all
current of a PV array, the RMSE provided is of 2.74%. cases except the case of using LMA and GA algorithms to simulate
the c-Si PV module working in cloudy conditions. This effect can be
6. Conclusions explained taking into account that the values of series, Rs, and
shunt, Rsh, resistances forming part of the model parameter set vary
Two PV array models have been compared in this work for with the irradiance, whereas both parameters have been assumed
simulation purposes: The 5PM and the SAPM. These models were constant in the performed simulations. An advantage of the 5PM
applied to reproduce the behavior of three grid connected PV sys- lies in the physical meaning of the set of model parameters that
tems with different topologies and solar cell technologies. The provides relevant information about the PV array and allows an
models parameters were obtained from daily monitored profiles of easy comparison between different PV modules.
G, Tc, and output DC current and voltage of the PV arrays using five On the other hand, the SAPM model is an empirical model
different optimization algorithms (LMA, GA, PSO, DE and ABC). including a set of model parameters in which some of them have
The metaheuristic algorithms are more efficient than the little physical meaning. Nevertheless, the SAPM model showed a
traditional LMA algorithm in estimating the unknown parameters high accuracy degree in the simulation of the PV systems behavior
of both PV module models, essentially in bad weather conditions. independently of the solar cell technology. The RMSE values ob-
The GA provides high values of RMSE compared to the other bio- tained for the DC output power of the PV arrays in the simulations
inspired algorithms. The ABC algorithm is slightly more accurate stayed below 6.05% for the PV system 1 even in cloudy days. For the
than the DE and PSO algorithms. PV system 2 this error dropped below 3.52%. However, for the PV
The 5PM allowed simulating the dynamic behavior of the PV system 3 the RMSE values are below 4% even in cloudy days and
systems included in this study with an acceptable accuracy degree case of using LMA. The SAPM model demonstrated best potential
 S. Kichou et al. / Renewable Energy 99 (2016) 270e279 279

for the simulation of PV systems in real operating conditions; this [17] M.S. Ismail, M. Moghavvemi, T.M.I. Mahlia, Characterization of PV panel and
global optimization of its model parameters using genetic algorithm, Energy
holds even when using thin film technologies of PV modules.
Convers. Manag. 73 (2013) 10e25, http://dx.doi.org/10.1016/
j.enconman.2013.03.033.
Acknowledgments [18] J. Ma, Z. Bi, T.O. Ting, S. Hao, W. Hao, Comparative performance on photo-
voltaic model parameter identification via bio-inspired algorithms, Sol. En-
ergy 132 (2016) 606e616, http://dx.doi.org/10.1016/j.solener.2016.03.033.
This work was partly supported by the Spanish Science and [19] E. Garoudja, K. Kara, A. Chouder, S. Silvestre, Parameters extraction of
Innovation Ministry and the ERDF within the frame of the project photovoltaic module for long-term prediction using artificial bee colony

n de la energía generada por mo
‘Estimacio
dulos fotovoltaicos de optimization, in: 3rd Int. Conf. Control. Eng. Inf. Technol., 2015, pp. 1e6,
http://dx.doi.org/10.1109/CEIT.2015.7232993.
capa delgada: influencia del espectro’ under expedient code [20] V. Jack, Z. Salam, An improved method to estimate the parameters of the
ENE2008-05098/ALT. single diode model of photovoltaic module using differential evolution, in:
Electric Power and Energy Conversion Systems (EPECS), 4th International
Conference, IEEE, 2015, pp. 1e6.
References [21] M. de Blas, J. Torres, E. Prieto, A. Garcıa, Selecting a suitable model for char-
acterizing photovoltaic devices, Renew. Energy 25 (2002) 371e380, http://
[1] C. Ventura, G.M. Tina, Development of models for on-line diagnostic and en- dx.doi.org/10.1016/S0960-1481(01)00056-8.
ergy assessment analysis of PV power plants: the study case of 1 MW Sicilian [22] G. Ciulla, V. Lo Brano, V. Di Dio, G. Cipriani, A comparison of different one-
PV plant, Energy Procedia 83 (2015) 248e257, http://dx.doi.org/10.1016/ diode models for the representation of I-V characteristic of a PV cell,
j.egypro.2015.12.179. Renew. Sustain. Energy Rev. 32 (2014) 684e696, http://dx.doi.org/10.1016/
[2] S. Silvestre, M.A. Da Silva, A. Chouder, D. Guasch, E. Karatepe, New procedure j.rser.2014.01.027.
for fault detection in grid connected PV systems based on the evaluation of [23] L. Hontoria, J. Aguilera, F. Almonacid, G. Nofuentes, P. Zufiria, Artificial neural
current and voltage indicators, Energy Convers. Manag. 86 (2014) 241e249, networks applied in PV systems and solar radiation, Artif. Intell. Energy
http://dx.doi.org/10.1016/j.enconman.2014.05.008. Renew. Energy Syst. (2006) 163e200.
[3] a. Chouder, S. Silvestre, Automatic supervision and fault detection of PV sys- [24] S. Lineykin, M. Averbukh, A. Kuperman, An improved approach to extract the
tems based on power losses analysis, Energy Convers. Manag. 51 (2010) single-diode equivalent circuit parameters of a photovoltaic cell/panel,
1929e1937, http://dx.doi.org/10.1016/j.enconman.2010.02.025. Renew. Sustain. Energy Rev. 30 (2014) 282e289, http://dx.doi.org/10.1016/
[4] S. Silvestre, A. Chouder, E. Karatepe, Automatic fault detection in grid con- j.rser.2013.10.015.
nected PV systems, Sol. Energy 94 (2013) 119e127, http://dx.doi.org/10.1016/ [25] T. Ma, H. Yang, L. Lu, Development of a model to simulate the performance
j.solener.2013.05.001. characteristics of crystalline silicon photovoltaic modules/strings/arrays, Sol.
[5] S. Silvestre, S. Kichou, A. Chouder, G. Nofuentes, E. Karatepe, Analysis of cur- Energy 100 (2014) 31e41, http://dx.doi.org/10.1016/j.solener.2013.12.003.
rent and voltage indicators in grid connected PV (photovoltaic) systems [26] R. Overstraeten, R. Mertens, Characterisation and Testing of Solar Cells and
working in faulty and partial shading conditions, Energy 86 (2015) 42e50, Modules, Hilger, Bristol, Engl, 1986.
http://dx.doi.org/10.1016/j.energy.2015.03.123. [27] L. Castan~ er, S. Silvestre, Modelling Photovoltaic Systems Using PCspice, 2002,
[6] W. Chine, A. Mellit, V. Lughi, A. Malek, G. Sulligoi, A. Massi Pavan, A novel fault http://dx.doi.org/10.1002/0470855541.
diagnosis technique for photovoltaic systems based on artificial neural net- [28] D.L. King, J.a Kratochvil, W.E. Boyson, Photovoltaic array performance model,
works, Renew. Energy 90 (2016) 501e512, http://dx.doi.org/10.1016/ Online 8 (2004) 1e19, http://dx.doi.org/10.2172/919131.
j.renene.2016.01.036. [29] W. De Soto, S.A. Klein, W.A. Beckman, Improvement and validation of a model
[7] A. Dolara, F. Grimaccia, S. Leva, M. Mussetta, E. Ogliari, A physical hybrid for photovoltaic array performance, Sol. Energy 80 (2006) 78e88, http://
artificial neural network for short term forecasting of PV plant power output, dx.doi.org/10.1016/j.solener.2005.06.010.
Energies 8 (2015) 1138e1153, http://dx.doi.org/10.3390/en8021138. [30] A.N. Celik, N. Acikgoz, Modelling and́experimental verification of the oper-
[8] Y.M. Saint-Drenan, S. Bofinger, R. Fritz, S. Vogt, G.H. Good, J. Dobschinski, An ating current of mono-crystalline photovoltaic modules using four- and five-
empirical approach to parameterizing photovoltaic plants for power fore- parameter models, Appl. Energy 84 (2007) 1e15, http://dx.doi.org/10.1016/
casting and simulation, Sol. Energy 120 (2015) 479e493, http://dx.doi.org/ j.apenergy.2006.04.007.
10.1016/j.solener.2015.07.024. [31] A. Chouder, S. Silvestre, N. Sadaoui, L. Rahmani, Modeling and simulation of a
[9] A.K. Tossa, Y.M. Soro, Y. Azoumah, D. Yamegueu, A new approach to estimate grid connected PV system based on the evaluation of main PV module pa-
the performance and energy productivity of photovoltaic modules in real rameters, Simul. Model. Pract. Theory 20 (2012) 46e58, http://dx.doi.org/
operating conditions, Sol. Energy 110 (2014) 543e560, http://dx.doi.org/ 10.1016/j.simpat.2011.08.011.
10.1016/j.solener.2014.09.043. [32] J. Peng, L. Lu, H. Yang, T. Ma, Validation of the Sandia model with indoor and
[10] A. Dolara, S. Leva, G. Manzolini, Comparison of different physical models for outdoor measurements for semi-transparent amorphous silicon PV modules,
PV power output prediction, Sol. Energy 119 (2015) 83e99. Renew. Energy 80 (2015) 316e323, http://dx.doi.org/10.1016/
[11] M. Hejri, H. Mokhtari, M.R. Azizian, M. Ghandhari, L. So €der, On the parameter j.renene.2015.02.017.
extraction of a five-parameter double-diode model of photovoltaic cells and [33] C. Sah, R.N. Noyce, W. Shockley, Carrier generation and recombination in PeN
modules, IEEE J. Photovolt. 4 (2014) 915e923, http://dx.doi.org/10.1109/ junctions and PeN junction characteristics, Proc IRE 45 (1957) 1228e1243.
JPHOTOV.2014.2307161. [34] D. Oliva, E. Cuevas, G. Pajares, Parameter identification of solar cells using
[12] S. Kichou, S. Silvestre, G. Nofuentes, M. Torres-Ramírez, A. Chouder, D. Guasch, artificial bee colony optimization, Energy 72 (2014) 93e102, http://dx.doi.org/
Characterization of degradation and evaluation of model parameters of 10.1016/j.energy.2014.05.011.
amorphous silicon photovoltaic modules under outdoor long term exposure, [35] J.D. Bastidas-Rodriguez, G. Petrone, C.A. Ramos-Paja, G. Spagnuolo, A genetic
Energy 96 (2016) 231e241, http://dx.doi.org/10.1016/j.energy.2015.12.054. algorithm for identifying the single diode model parameters of a photovoltaic
[13] A.M. Ameen, J. Pasupuleti, T. Khatib, Modeling of photovoltaic array output panel, Math. Comput. Simul. (2015), http://dx.doi.org/10.1016/
current based on actual performance using artificial neural networks, j.matcom.2015.10.008.
J. Renew. Sustain. Energy 7 (5) (2015) 1e11, 053107. [36] R. Storn, K. Price, Differential evolution e a simple and efficient heuristic for
[14] M.F. AlHajri, K.M. El-Naggar, M.R. AlRashidi, A.K. Al-Othman, Optimal global optimization over continuous spaces, J. Glob. Optim. 11 (1997)
extraction of solar cell parameters using pattern search, Renew. Energy 44 341e359, http://dx.doi.org/10.1023/A:1008202821328.
(2012) 238e245. [37] D. Karaboga, B. Akay, A comparative study of Artificial Bee Colony algorithm,
[15] D.H. Muhsen, A.B. Ghazali, T. Khatib, I.A. Abed, A comparative study of Appl. Math. Comput. 214 (2009) 108e132, http://dx.doi.org/10.1016/
evolutionary algorithms and adapting control parameters for estimating the j.amc.2009.03.090.
parameters of a single-diode photovoltaic module’s model, Renew. Energy 96 [38] I.A. Ibrahim, A. Mohamed, T. Khatib, Modeling of photovoltaic array using
(2016) 377e389. random forests technique, in: Conference on Energy Conversion (CENCON),
[16] R. Eberhart, J. Kennedy, A new optimizer using particle swarm theory, IEEE, 2015, October, pp. 390e393.
MHS’95, Proc. Sixth Int. Symp. Micro Mach. Hum. Sci. (1995) 39e43, http://
dx.doi.org/10.1109/MHS.1995.494215.