Renewable and Sustainable Energy Reviews 77 (2017) 955–969

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews

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

Identification of relevant input variables for prediction of 1-minute time- MARK

step photovoltaic module power using Artificial Neural Network and
Multiple Linear Regression Models
⁎
Amit Kumar Yadava,b, , S.S. Chandela
a
Centre for Energy and Environmental Engineering, National Institute of Technology, Hamirpur, Himachal Pradesh 177005, India
b
Electrical and Electronics Engineering Department, National Institute of Technology Sikkim, Ravangla, Barfung Block, South Sikkim 737139, India

A R T I C L E I N F O A BS T RAC T

Keywords: In photovoltaic (PV) modules manufacturer provides rating under standard test conditions (STC). But STC
Photovoltaic module hardly occur under outdoor conditions so it is important to investigate PV power by experimental analysis. In
Power prediction this study extensive literature survey of PV module electrical characteristics by conventional methods and ANN
Multi-crystalline silicon techniques are carried out. It is found that experimental analysis of PV modules maximum power under outdoor
Maximum power
conditions remains a major research area. For this measurement of 75 Wp PV module are performed under
Artificial Neural Network
outdoor conditions at Centre for Energy and Environmental Engineering, National Institute of Technology,
Multiple Linear Regression
Hamirpur, India. To find most influencing variables for PV power prediction, five different sets of parameters
are served as inputs to establish five Artificial Neural Network (ANN) models and Multiple Linear Regression
(MLR) Models which is novelty of this paper. The results shows that solar radiation and air temperature are
found to be most influencing input variables for ANN based prediction of maximum power produced by PV
module with mean absolute percentage (MAPE) of 2.15 %. The mean absolute percentage (MAPE) errors for
ANN models are found to vary between 2.15 % to 2.55 % whereas for MLR models it varies from 13.04 % to
19.34 %, showing better prediction of ANN models.

1. Introduction model of solar cell is shown in Fig. 1 with parallel resistor denoting
leakage current, series resistor denoting internal resistance for current
The photovoltaic (PV) module is a reliable technology directly flow, diode and photo current. The I − V equation of solar cell is
converts solar radiation (SR) into power. The PV installed capacity is expressed as:
increasing globally with more than 40% since 2009 and its growth ⎧ ⎛ q ⎫ V + IRs
⎞
increases all over the world [1]. The PV installed capacity will reach I = IL − Io⎨exp⎜ (V + IRs )⎟ − 1⎬ −
3000 GW in 2050, providing 4 500 TWh per year (covers 11 % of global ⎩ ⎝ akTc ⎠ ⎭ Rsh (1)
electricity production) and it is beneficial for socio-economic develop- where IL is photon current, Io is diode reverse saturation current, a is
ment by reducing carbon dioxide emission. In first decade the PV ideality factor, q charge on electron (1.6 × 10−19C ), k is Boltzmann
generation cost is expected to reduce by more than 50 %. The PV constant (1.38 × 10−23J/ K ),Tc is cell temperature, Rs is series resistance
commercial and residential systems will achieve the first level of grid and Rsh is shunt resistance , Rsh has large value and Rs is small so it can
parity by 2020 in many regions. be neglected in the analysis.
A PV module contain solar cells which convert SR into electricity by Solar cells are connected in parallel and series combination in a PV
photovoltaic effect [2]. When sunlight is incident on solar cells, module. The I − V characteristics [3] of a PV module are given by
photons energy more than band-gap energy of a semiconductor are equation:
absorbed and generate electron-hole pairs. Due to influence of p-n
⎡ ⎛ qV ⎞⎤
junction electric field charge carriers (electron-hole pairs) are acceler- I = npIL − npIo⎢exp⎜ − 1⎟⎥
ated and create photocurrent which is directly proportional to SR. A ⎣ ⎝ kTcans ⎠⎦ (2)
solar cell has nonlinear electrical characteristics which vary with SR
where np is number of module connected in parallel, ns is number of
intensity and cell temperature. The equivalent circuit of single diode

⁎
Corresponding author at: Centre for Energy and Environmental Engineering, National Institute of Technology, Hamirpur, Himachal Pradesh 177005, India.
E-mail address: amit1986.529@rediffmail.com (A.K. Yadav).

http://dx.doi.org/10.1016/j.rser.2016.12.029
Received 19 November 2015; Received in revised form 24 November 2016; Accepted 6 December 2016
Available online 31 January 2017
1364-0321/ © 2016 Elsevier Ltd. All rights reserved.
A.K. Yadav, S.S. Chandel Renewable and Sustainable Energy Reviews 77 (2017) 955–969

Nomenclature Pmax Power at maximum power point (W)

R Correlation coefficient
ANN Artificial Neural Network RMSE Root Mean Square Error
BSMT Back Surface Module Temperature (⁰C) SR Solar Radiation (W / m2 )
I Current (A) T Air Temperature (⁰C)
Imax Current at maximum power point (A) V Voltage (V)
Isc Short Circuit Current (A) Vmax Voltage at maximum power point (V)
MAPE Mean Absolute Percentage Error (%) Voc Open Circuit Voltage (V)
MLR Multiple Linear Regression WS Wind Speed (m/s)

module in series, a is ideality factor which varies from 1 to 2 and PV modules maximum power in STC with SR of 1000 W/m2 and
indicates solar cell deviation from the ideal p-n junction characteristics temperature 25 ⁰C. The STC rarely occurs under outdoor conditions so
[4]. The cell reverse saturation current Io varies with temperature more accurate models are required to estimate maximum power by PV
according to equation [5]: modules under outdoor conditions.
The main objective of this study is to identify most influencing
⎡ T ⎤3 ⎡⎛ eg ⎞⎛ 1 1 ⎞⎤
Io = Ir ⎢ c ⎥ exp⎢⎜q ⎟⎜ − ⎟⎥ variables for ANN based prediction of PV maximum power under
⎣ Tref ⎦ ⎢⎣⎝ ka ⎠⎝ Tref Tc ⎠⎥⎦ (3) outdoor conditions. For this purpose five ANN models (ANN-1, ANN-2,
ANN-3, ANN-4 and ANN-5) are developed to predict maximum power
where Ir is the reverse saturation current at Tref , Tref is the cell reference by a 75 Wp Si- multi crystalline PV module. The input variables to ANN
temperature and eg is the band gap energy of semiconductor used in the models are five different combinations of irradiance, air temperature,
cell. The photo current IL depends on SR and the cell temperature is back surface module temperature, maximum voltage and maximum
given by: current. To evaluate the prediction accuracy of ANN models a
⎡ SR ⎤ comparison with five Multiple Linear Regression (MLR) Models is also
IL = [Iscref + Kisc(Tc − Tref )]⎢ ⎥ carried out.
⎣ 1000 ⎦ (4)
The paper is organized as follows: a brief literature survey on PV
where Iscref is the cell short circuit current at reference temperature and module electrical characteristics is presented in Section 2. Methodology
radiation, Kisc is the short circuit current temperature coefficient and is described in Section 3. Results and discussion are presented in
SR in W/m2. The PV array power can be calculated by [6]: Section 4 and conclusions are given in Section 5.
P=I×V (5)
2. Literature review of PV module electrical characteristics
⎡ ⎛ qV ⎞⎤
P = npILV − npIoV ⎢exp⎜ − 1⎟ ⎥ parameters evaluation
⎣ ⎝ kTcans ⎠⎦ (6)
The performance of PV system is influenced by SR and temperature
Maximum power point voltage [7] Vmax can be calculated by setting
dP therefore to design an accurate PV system electrical characteristic
= 0 and Vmax is calculated using equation:
( dV )
model under real operating condition is necessary. This section reviews
⎛ qV ⎞⎡⎛ qV ⎞ ⎤ (I + Io ) various methodologies for determining electrical characteristics of PV
exp⎜ max ⎟⎢⎜ max ⎟ + 1⎥ = L modules using conventional and Artificial Neural Network based
⎝ kTcans ⎠⎣⎝ kTcans ⎠ ⎦ Io (7)
methods.
PV module output voltage is a function of photo current which is
calculated by load current and is given by equation: 2.1. Conventional models
⎛ akTc ⎞ ⎡ IL + Io − I ⎤
V=⎜ ⎟ln⎢ ⎥ − RsI Conventional models compute I − V characteristics to determine
⎝ q ⎠ ⎣ Io ⎦ (8)
maximum power using Eqs. 1 to 8. A MATLAB simulink model is used
Therefore, maximum power of PV module is influenced by ambient for calculating current (I ) – voltage (V ) and power (P )-voltage (V )
temperature, SR and back surface module temperature so modeling of characteristics of PV module using 60Wp and 125Wp solar modules to
PV cells and modules becomes an important research area in photo- find maximum power [9,10]. Tsai [11] also used simulink model for
voltaics. PV modeling mainly incorporates explicit I = f (V ) and implicit obtaining electrical characteristics of a Siemens SM46 PV module and
I = f (I , V ) models for estimating maximum power [8]. The explicit studied the effect of SR on PV power and cell temperature. The
model is simple analytical equations based on assumptions and simulation and experimental results are found close to each other.
requires less computational time. The implicit model is more accurate Marion et al. [12] used bilinear interpolation method to translate
than explicit model and incorporates several parameters such as diode I − V curve of PV module to required value of SR and T by using four
ideality factor, diode reverse saturation current, photon current, shunt reference I − V curve. This method is applied to PV module of seven
resistance and series resistance which are difficult to obtain from PV different technologies and root mean square error is 1.4% for max-
cells manufacturers. In addition the manufacturers provide rating of imum power,1.3% for maximum voltage, 0.8% for maximum current,
0.5% for open circuit voltage, 0.3% for short circuit current, and 2% for
fill factor.
Ishaque et al. [13] proposed a MATLAB simulink model using a
double diode model of solar cell. The model is freely downloadable
from web site (http://sites.google com/site/drkishaque/Downloads)
gives better accuracy at low irradiance and is useful for performance of
PV module under partial shading conditions. The accuracy is verified
for multi-crystalline, mono-crystalline and thin-film PV modules. It is
found that less computational time is required for calculating power.
Fig. 1. Solar cell equivalent circuit. Celsa and Tina [14] used PLC technique to simulate non uniform

956
A.K. Yadav, S.S. Chandel
distribution of SR and T for PV modules connected in parallel and
series.
Chouder et al. [15] presented a graphical user interface of
LABVIEW for simulation, monitoring and modeling of PV module for
I − V characteristics. The root mean square error (RMSE) between
simulations and measured results is found to be less than 3.81 %.
Bencherif and Chermitti [16] proposed a method for calculating
reverse saturation current (Io ), series resistance (Rs ), shunt resistance
(Rsh ), photocurrent (IL ) and ideality factor (a ) using Voc, Isc, maximum
value of current, voltage and power as given in manufacturer datasheet.
The MAPE between measured and simulated values of maximum
current, voltage and power of PV module is found to be not more than
0.1%.
Ibrahim et al. [17] determined electrical characteristics of 50 W,
100 W mono crystalline silicon and 17 W triple junction (amorphous
silicon) PV modules. The electrical characteristics of PV modules are
calculated for sun tracking and south facing orientation of PV modules.
The electrical characteristic with sun tracking system is found to be
better than fixed orientation of PV system. The efficiencies of 100 W,
50 W and 17 W are found to be 23% (spring season), 10.55% (autumn
season) and 4.82% (autumn season) respectively.
Fuentes et al. [18] used algebraic methods to model PV module
maximum power for Mediterranean and Continental-Mediterranean
climates in the city of Jaén. The deviation in open circuit voltage, peak
power, short circuit current due to module mismatch is observed so
good characterization of module under STC is recommended.
Hernandez et al. [19] developed methods for maximum power
calculation of PV modules using manufacturer parameters, average
value of T and SR. The method is verified with 3.6 kWp building
integrated PV system and found that calculated values of energy is
more than measured values for January, February and March months
due to circuit model approximation, calculation of cell temperature
using monthly average ambient temperature, SR measurement in
cloudy season and decrease of inverter efficiency with low value of SR.
Ibrahim et al. [20] studied the effect of tilt angle on electrical
characteristics of a 50 Wp mono-crystalline silicon PV module. The
short circuit current increases up to tilt angle equal to latitude and then
it decreases with increase in tilt angle. The open circuit voltage
decreases with increase in tilt angle and module temperature. The PV
power increases with intensity of SR.
Dolara et al. [21] analyzed diagonal, vertical and horizontal shading
on electrical characteristics of poly-crystalline and mono-crystalline PV
modules. The PV current decreases as shading is varied from 0% to
100%. Due to shading two maximum powers points occur on power
voltage curve and output power is reduced by more than 30% with 50%
shading of PV cell. It is suggested that bypass diodes in PV module are
important for proper functioning.
Patel and Agarwal [22] presents MATLAB/SIMULINK model of
P − V and I − V characteristics for PV array in partial shading condi-
tion. The multiple peaks are found on P − V curve under partial
shading. The model is freely available and downloads from web site
(http://www.ee.iitb.ac.in/uma/∼phiren/).
Chen et al. [23] developed 300 W, 60 V emulator for finding
electrical characteristics under uniform and shading conditions of
SR. The emulator generated voltage-power curve of a single PV module
and series, parallel connected PV module under different solar radia-
tion uniformity conditions.
Wang et al. [24] proposed methodology of calculating dynamic
resistance (ratio of change in voltage to change in current) of multi
crystalline Si PV module with different values of SR and T. This method
is useful to find maximum power and matching of impedance under
different operating condition. Alessandro et al. [25] calculated shunt
resistance of PV cell connected in string of module using slope
measurement of I-V characteristics of modules without considering
intrinsic diode parameter. The shunt resistance is found to be 2 to 20 k
Ω cm2.
957
Renewable and Sustainable Energy Reviews 77 (2017) 955–969
Table 1
Temperature difference for predicting back surface module temperature.
Types of Module Installation Δt (°C)
Glass/cell/glass open rack 3
Glass/cell/glass close roof 1
Glass/cell/polymer sheet open rack 3
Glass/cell/polymer sheet insulated back 0
Polymer/thin film/steel open rack 3
22X Linear Concentrator tracker 13
Skoplaki and Palyvos [26] presented various correlations of cell
temperature (Tc ) for performance of PV module and its importance in
installation of PV array, PV thermal collector and building integrated
photovoltaic system. The cell temperature is more than back surface
module temperature (BSMT ) due to solar radiation (SR ) and substrate
materials of PV module. The correlations of Tc in terms of BSMT is
given by following equation.
SR
Tc = BSMT + Δt
SRref (9)
2
where SRref is reference solar radiation (1000 W/m ) incident on PV
module, Δt is temperature difference between Tc and Tm which is 2 to
3 °C for a flat plate module installed on an open-rack. The temperature
difference is zero for flat plate with back surface thermally insulated.
The empirical calculated value of Δt is shown in Table 1 [27]. The back
surface module temperature (BSMT ) in terms of air temperature (T )
and wind speed (ws ) is given by following correlation [28] where a, b
are constant.
BSMT = SR(e a + bws ) + T (10)
In another study Skoplaki and Palyvos [29] also presented a review
of correlation equations for PV module power, efficiency, electrical
characteristics in terms of Tc . The PV module output power decreases
with operating temperature. King et al. [30] developed methodology for
finding T coefficient of cells, modules and array. The error source in
measurement of PV parameters are presented and highlighted that
temperature coefficient should be considered for sizing and designing
of PV system.
Hernandez et al. [31] developed correlations of SR and T in terms
of random number for calculating temperature for calculating perfor-
mance of grid connected photovoltaic system. The correlation coeffi-
cient (R) is more than 98% showing predicted power is close to
measured power.
Kaplani and Kaplanis [32] studied effect of WS, wind incidence
angle and PV inclination angle through f coefficient on the temperature
of the PV module. The f coefficient correlates PV module temperature
in terms of incident SR and temperature on PV module. It was
experimentally found that f coefficient lies between 0.037 and
0.012 m2 K/W.
Jiang et al. [33] determined PV module junction temperature with
p-n junction semiconductor theory. The junction temperature increases
with increase in radiation and ambient temperature. The ambient
temperature and junction temperature difference increases linearly
with increase in solar radiation.
Kamkird et al. [34] investigated the effect of temperature on
current, voltage and power of 3.76 kWp amorphous silicon solar cell,
3.60 kWp poly crystalline silicon solar cell and 2.88 kWp hetero
junction intrinsic thin layer. It is found that amorphous silicon solar
cell has lowest negative temperature coefficient and is good for current,
voltage and power generation in hot humid climate of Thailand.
Bahaidarah et al. [35] investigated the effect of water cooling
technique at back surface of PV module for Dhahran, Saudi Arabia.
Due to cooling of PV module the maximum power increases and PV
panel efficiency is increases by 9%. It is found that temperature rise of
PV module decreases with increase in rate of water flow and at same
A.K. Yadav, S.S. Chandel
rate of water flow temperature increases with increase in SR.
Garcia and Balenzategui [36] applied International Standards (EN
61215, EN-61646) for determining normal operating cell temperature
of PV device. It is concluded that temperature rise in PV module is high
in afternoon than morning at same value of SR and for accurate
calculation of module temperature different structures (glass–glass,
glass–tedlar, window thermally isolated modules) should be consid-
ered.
Torres-Ramírez et al. [37] presented Osterwald's and constant fill
factor for outdoor conditions to model different types of thin film PV
modules located in Madrid and Jaén. The relative error between
measured and estimated value is found to be -0.5% to 4%.
Gaglia et al. [38] performed experimental analysis of PV module
under outdoor condition in Athens, Greece. It was found that annual
PV efficiency was 8.7% and PV efficiency was 18% lower than efficiency
reported under STC.
Chantana et al. [39] studied influence of environmental parameters
on performance ratio (PR) of Si- PV module using multiple regression
analysis. It is demonstrated that PR is significantly affected by angle of
Incidence.
Peng et al. [40] developed Sandia Array Performance Model
(SAPM) empirical coefficient for amorphous silicon (a-Si) PV module.
It was found that SAPM give accurate results for sunny days and in
accurate result for over cast days due to instrument error. The error is
less than 4 % for simulated and measure energy output.
Sadok and Mehdaoui [41] presented electrical characteristics of PV
module under outdoor condition in Sahara region. The experimental
values of I-V curves are fitted with least square method and found close
to Rauschenbach’s model.
Adinoyi and Said [42] discussed effect of dust on output of PV
module in Saudi Arabia. It was illustrated that dust accumulation on
PV surface decreases PV output power by 50% and single dust storm
decreases power by 20%.
Ozden et al. [43] performed outdoor testing of mono-Si, amorphous
Si thin film and a CdTe thin film arrays for 48 months under same
climatic condition of central Anatolia. The average efficiencies for these
modules are found to be 11.86%, 6.40% and 5.30%. The degradation
rates are 0.40%, 1.88% and 10.60% respectively.
Visa et al. [44] analyzed electrical characteristics of mono-, poly-
crystalline silicon, CdTe, CIS and CIGS PV modules in temperate
mountain climate for 14 months. It is found that poly-crystalline
silicon perform best in snowy winters rather than warm summer and
high relative efficiency losses are viewed for the CIS and CdTe thin film
modules.
Limmanee et al. [45] used Si wafer and thin film PV module for
performing electrical characteristics under tropical climate. It is found
that temperature coefficient below -0.31%/°C beneficial for energy
yield. The degradation rate of 10–50 % is found in PV module.
Elibol et al. [46] performed experimental analysis of 2.35 kW
mono-crystalline, 2.64 kW polycrystalline and 2.40 kW amorphous
silicon PV panels under outdoor condition of Turkey. The performance
ratio (PR) is 73%, 81% and 91% and panel efficiency is found to be
4.79%, 11.36% and 13.26%. The statiscal analysis shows that 1°
increase in temperature increases efficiency of a-Si, polycrystalline by
0.029%, 0.033% and decreases the efficiency of mono-crystalline
panels by 0.084%.
Başoğlu, et al. [47] used crystalline (c-Si), multi crystalline (mc-Si)
and cadmium–telluride (Cd–Te) modules for calculating PR and
capacity factor (CF) in Northwest of Turkey. PRs are 83.8%, 82.05%
and 89.76% for mc-Si, c-Si and Cd–Te arrays, respectively. CF of Cd–
Te array is highest during all month so it is reliable PV module under
İzmit climatic condition. Sharma and Chandel [48] studied degradation
rate in electrical characteristics of multi-crystalline-silicon PV system
in 1 kWp system after 2.5 year in outdoor condition of western
Himalayan region in India. It was calculated that three modules shows
50% and 7 modules shows 0.6–2.5% degradation in power. Rajput
958
Renewable and Sustainable Energy Reviews 77 (2017) 955–969
Table 2
Proposed combination of input variables for ANN based models for maximum PV power
prediction.
ANN Models Input Variables Combinations
ANN-1 solar radiation, air temperature, back surface module
temperature, open circuit voltage and wind speed
ANN-2 solar radiation, air temperature, back surface module
temperature, open circuit voltage
ANN-3 solar radiation, air temperature, back surface module
temperature
ANN-4 solar radiation, air temperature
ANN-5 solar radiation, back surface module temperature
et al. [49] performed I-V analysis of 90 mono-crystalline silicon PV
modules installed at rooftop of National Institute of Solar Energy
(NISE), Gurgaon. The modules is under outdoor condition since 22
years. The maximum and minimum power is degraded by 4.1% and
0.3% per year.
2.2. ANN models for power prediction under outdoor conditions
In power generation fields ANN have been used for various
applications such as SR prediction [50–54], PV tracking system [55–
59], electrical energy consumption [60] and electrical characteristic
estimation [61].
Bahgat et al. [62] used ANN to predict PVmax and normal operating
power of UNI-POWER US-64 solar PV module for clear and cloudy
days. The network utilized SR and PV module surface temperature as
inputs for prediction.
Almonacid et al. [63] used ANN technique for predicting electrical
characteristics of Si-crystalline PV modules. It is found that ANN
method generates I-V curve of PV module for low value of SR. It is
suggested that ANN is a useful tool to determine PV module power
under outdoor conditions proving useful for PV system installations. In
other study Almonacid et al. [64] build ANN model for predicting Pmax
of High Concentrator Photovoltaic (HCPV) modules at south of Spain.
The input parameters to model is air mass, direct normal irradiance
(DNI), precipitable water, T and wind speed (WS). The ANN model
gives better prediction except for DNI value below 400 W/m2 and WS
above 3 m/s. Almonacid et al. [65] used MLP to estimate I − V
characteristics of PV CIS module. The model incorporates SR and
module cell temperature as input for estimation. This method is useful
for PV installer from installation point of view.
Sulaiman et al. [66] incorporate partial evolutionary programming
(EP) ANN for power prediction of 42 kWp grid connected PV system
installed at roof of Quadrangle Building, University of New South
Wales, Australia. The network used SR, WS and T as inputs and PV
power as output. The optimum momentum, hidden layer neurons and
learning rate for training are calculated by EP. The ANN model with EP
gives better prediction than only ANN. Bonanno et al. [67] used radial
basis function neural network (RBFNN) to predict electrical character-
istics of PV module. The input variables to RBFNN are SR and T. The
back propagation algorithm is used for training the RBFNN.
Mellit and Pavan [68] used two ANN models (ANN-1, ANN-2) for
predicting power of 20 kWp grid-connected photovoltaic at Trieste
(Italy). The SR, T is used as input variables for ANN-1 model and for
ANN-2 model, SR is used as the input variable. The RMSE in ANN-1
model varies from 0.00375 to 0.00398 and RMSE for ANN-2 model
varies from 0.00442 to 0.00478, showing T affect has less effect on
prediction accuracy of model.
Mellit et al. [69] utilized ANN models for estimating power of
50 Wp Si-polycrystalline PV module in cloudy and sunny days. The
comparative analysis shows that ANN model estimate power better
than one-diode models, polynomial regression, analytical and multiple
linear regression. Brano et al. [70] used T, solar cell temperature, SR,
A.K. Yadav, S.S. Chandel Renewable and Sustainable Energy Reviews 77 (2017) 955–969

Fig. 2. Class-A Sun Simulator installed at the Photovoltaic Energy Research laboratory, CEEE, NIT-H.

Fig. 3. Electrical Characteristics of 75 Wp multicrystalline Silicon PV module measured at STC.

Table 3 Table 4
Specifications of Class-A Sun Simulator. Electrical Characteristics of a 75 Wp PV module as per manufacturer data and measured
sun simulator at STC (1000 W/m2, A.M. 1.5 spectrum, 25 °C cell temperature) at CEEE,
Type Spectral Match Irradiance Spatial Temporal NIT-H.
(each interval) Non-Uniformity Instability
PV module Manufacturer Data Measured data using
Class A 0.75–1.25 2% 2% characteristics Class-A Sun Simulator
at CEEE

WS, open circuit voltage, short circuit current as input variables to Maximum Power: Pm (W) 79.9 77
Short Circuit Current: Isc 4.816 5.075
different ANN models for predicting Pmax of PV modules.
(A)
Piliougine et al. [71] used instantaneous clearness index (CI), angle Open Circuit Voltage: Voc 23.03 21.8
of incidence, SR and module temperature to model PV module. Due to (V)
incorporation of angle of incidence and CI better result is obtained. Maximum Power 4.406 4.594
Current: Imp (A)
Veerachary and Yadaiah [72] utilized ANN for operating point
Maximum Power Voltage: 18.13 16.8
selection of dc motor run by PV. The SR is taken as input and converter Vmp (V)
chopping ratio as output in ANN. The predicted and calculated value of Cell Efficiency (% ) 13.7 13.71
converter chopping ratio is closely matched with each other. Module Efficiency (%) 11.9 11.29
Graditi et al. [73] developed two ANN model for estimating power
produced by a 1 kWp PV plant situated at ENEA Portici Research
Centre, in Italy South region. The inputs to first ANN model is global prediction of PV module Pmax. ANN model give better results than
SR, T and input to second model is clear sky SR. The R value for first conventional model.
model is 98.51 % and for second is 99.36 %. Fernández et al. [75] used ANN model for estimating Pmax of low
Hiyama and Kitabayashi [74] used WS, SR and T for ANN concentration PV module for building integration under real outdoor

959
A.K. Yadav, S.S. Chandel Renewable and Sustainable Energy Reviews 77 (2017) 955–969

Fig. 4. Schematic of outdoor experimental setup for 75 Wp PV module at CEEE, NIT-H.

Table 5 of hidden layer, learning rate, activation function, learning algorithm

Specifications of Solar Analyzer. and momentum rate. The prediction error of less value is obtained by
hybrid EP and ANN model.
Model Measured Range Resolution Accuracy
value Routh et al. [82] used WS, SR and wind direction as input to ANN
models for estimation of solar cell temperature which is important for
Solar Analyzer DC Voltage 0–10 V 0.001 V +1%+(1% of performance of solar cell. The temperature trend of sixty years over
(SOLAR- Vopen+0.1V)
Bangladesh is considered and ANN models are trained and tested using
600) 10–60 V 0.01 V +1%+(1% of
Vopen+0.1V) ten years data from Dhaka station from Bangladesh Meteorological
DC Current 0.01–10 A 1 mA +1%+(1% of Department.
Ishort+9 mA) Ramawan et al. [83] presented Bat Algorithm-ANN for prediction
10–12 A 10 mA +1%+(1% of of output power of grid-connected PV system using inputs as SR, air
Ishort+0.09 mA)
and module temperature. Bat Algorithm is used to decide learning rate,
DC Current 0.01–10 A 1 mA +1 %+9 mA
10–12 A 10 mA +1%+0.09 A momentum rate and hidden layer neurons. This method give better
results than evolutionary programming based ANN model.
Singh et al. [84] proposed fuzzy theory, adaptive neuro-fuzzy
conditions. The mode incorporate direct and diffuse SR, module interface system, ANN and generalized neural network for forecasting
temperature, incidence angle for Pmax prediction. The RMSE in ANN of solar power from 101 kW Rooftop PV plant installed at IIT Jodhpur.
model is found to be 2.32% and R is 99 %. The proposed method forecast better than conventional models.
Askarzadeh [76] used I, SR and T for ANN based prediction of PV Singh et al. [85] used ANN and generalized neural network (GRNN)
module voltage. The back propagation algorithm with one hidden layer for solar power forecasting. The utilized input variables are WS, SR and
provides better prediction for an amorphous silicon PV module. T. It is found that GRNN model predict better than ANN model.
Saberian et al. [77] developed feed forward back propagation and Huang et al. [86] introduced solar azimuth and zenith angle in
regression neural network to model PV output. Both networks utilized addition with module temperature and SR to improve the accuracy of
SR, mean T, Tmin and Tmax as inputs and power as target variables. The ANN models for estimating PV outdoor power. The input variables
performance of both networks is good but feed forward back propaga- solar azimuth and zenith angle are also important for Support Vector
tion give better results than regression neural network. and Gaussian process Regression for estimating PV outdoor power.
Niu et al. [78] studied effect of season, light intensity, temperature,
day and seasons on PV power production. Adaptive neural network and 3. Research gap
scene simulation knowledge mining technique are used for prediction.
The Guangdong PV power generation is selected for validation and it is Based on literature review the identified research gaps are as
found that the proposed approach give better accuracy. follows:
Jiang et al. [79] used Chebyshev Functional Link Neural Network
(CFLNN) for prediction of current for PV modules. The CFLNN 1. Maximum power prediction under outdoor conditions of different
incorporate SR, T, V as input variables and current as output variable. rating PV modules needs to be carried out.
It is found that CFLNN predict PV module current better than 2. For prediction of PV modules electrical characteristics the input
analytical model. variables used in ANN models are direct normal irradiance, air mass,
Hussain et al. [80] use SR, T and module temperature as inputs in precipitable water, wind speed, solar radiation, PV module surface
ANN for prediction of output power from grid-connected photovoltaic temperature, ambient temperature, open circuit voltage, short
system. The particle swarm optimization is used for selecting hidden circuit current, instantaneous clearness index, diffuse solar radiation
layers neurons for ANN. This method has least error in comparison to and incidence angle. Therefore effect of these input variables
Fast Evolutionary Programming-ANN. combinations needs to be considered for maximum power prediction
Sulaiman et al. [81] presented hybridization of Evolutionary using ANN models.
Programming (EP) and ANN for prediction of energy from Grid 3. Comparison of different ANN models such as multilayer perceptron
connected PV system. The ANN model incorporates SR and PV module model developed with nftool, radial basis function neural network
temperature as inputs for prediction. The EP is hybridize with Multi- and generalized regression neural network, etc in prediction of PV
Layer Feed forward Neural network for finding out optimum neurons modules maximum power to be done in order to identify the best

960
A.K. Yadav, S.S. Chandel Renewable and Sustainable Energy Reviews 77 (2017) 955–969

Fig. 5. Measured values of data (a) solar radiation, (b) wind speed, (c) air temperature, (d) back surface module temperature, (e) open circuit voltage, (f) maximum voltage, (g)
maximum current and (h) maximum PV power.

ANN prediction models.
4. Comparison of PV power prediction with regression and ANN
models needs to be considered more to find out accuracy errors.
In this study research gaps numbers 4 and few parts of 2 are
considered and remaining will be carried out in future research. The
mostly input variables used in ANN models for prediction of power are
air temperature, cell temperature, solar irradiance, wind speed, open
circuit voltage, short circuit current, back surface module temperature.
However, influence of input variables on prediction accuracy of PV

961
A.K. Yadav, S.S. Chandel Renewable and Sustainable Energy Reviews 77 (2017) 955–969

Table 9
Statistical Analysis for ANN-4 Model.

Hidden R for MAPE RMSE ANN architecture

layer Training for for
neurons Testing Testing

10 95.61 3.83 0.96

11 94.25 4.24 1.10
12 93.30 4.83 1.19
13 98.33 2.15 0.59 in = 2, Hn = 13, on = 1
14 92.12 5.10 1.29
15 88.74 5.23 1.55
16 87.30 5.76 1.64
17 96.32 3.37 0.88
Fig. 6. ANN Architecture. 18 89.87 5.49 1.47
19 85.93 6.35 1.73
20 94.09 4.31 1.12
Table 6
Statistical Analysis for ANN-1 Model.

Table 10
Hidden R for MAPE RMSE ANN architecture
Statistical Analysis for ANN-5 Model.
layer Training for for
neurons Testing Testing
Hidden R for MAPE RMSE ANN architecture
12 96.01 3.68 0.92 layer Training for for
13 93.56 3.89 1.17 neurons Testing Testing
14 94.40 4.34 1.09
10 93.59 3.72 1.17
15 93.91 4.35 1.14
11 97.42 2.55 0.74
16 97.56 2.38 0.72 in = 5, Hn = 16, on = 1
12 82.53 5.41 1.93
17 91.56 5.48 1.34
13 88 4.79 1.60
18 96.80 2.59 0.82
14 83.10 7.26 1.90 in = 2, Hn = 11, on = 1
19 82.60 7.07 1.93
15 85.01 5.97 1.79
20 91.57 5.37 1.34
16 95.98 3.45 0.92
21 83.62 6.97 1.87
17 90.19 5.63 1.44
22 83.53 7.01 1.88
18 84.97 6.10 1.79
19 87.79 6.33 1.61
20 85.94 6.98 1.73
Table 7
Statistical Analysis for ANN-2 Model.

Hidden R for MAPE RMSE ANN architecture

module maximum power prediction using ANN techniques is not
layer Training for for clearly reported. Therefore in present study the novelty is first STC of
neurons Testing Testing 75 Wp module is taken using Class-A Sun Simulator and outdoor
maximum power is measured using Solar Analyzer in CEEE labs. For
11 96.29 3.37 0.89
analysis five ANN models are developed with different input combina-
12 92.11 5.01 1.29
13 91.86 4.91 1.32 tions to evaluate prediction accuracy for maximum PV power predic-
14 96.94 2.83 0.80 tion as shown in Table 2.
15 92.56 4.01 1.26
16 96.40 2.99 0.87 in = 4, Hn = 21, on = 1
17 91.25 5.07 1.36 4. Methodology
18 96.41 3.65 0.87
19 89.24 4.57 1.51 4.1. Measurement of PV module electrical characteristics using Class-
20 88.29 5.87 1.58
21 97.98 2.26 0.65
A sun simulator

A 75 Wp Si multi crystalline PV module is used in this study. The

electrical characteristics of PV module at STC are measured using a
Table 8 Class-A sun simulator [Fig. 2] installed at the Photovoltaic Energy
Statistical Analysis for ANN-3 Model. Research Laboratory of Centre for Energy and Environmental
Engineering (CEEE), National Institute of Technology, Hamirpur
Hidden R for MAPE RMSE ANN architecture
layer Training for for
(NIT-H) Himachal Pradesh, India are shown in Fig. 3. The sun
neurons Testing Testing simulator is also called artificial sun which is used to provide
illumination approximately equal to natural sun for indoor test facility.
11 81.77 8.08 1.97 The sun simulator light is controlled in three dimensions namely
12 94.36 4.03 1.09
13 84.85 7.27 1.80
spectral content, spatial uniformity and temporal stability. The speci-
14 83.61 6.39 1.87 fication of class A sun simulator is given in Table 3.
15 94.83 3.95 1.05 in = 3, Hn = 20, on = 1 The electrical characteristics of 75 Wp PV modules under STC as
16 83.41 7.50 1.88 provided by the manufacturer Maharishi Solar, India and as measured
17 92 5.39 1.30
during the experiment using Sun simulator are given in Table 4.
18 86.42 6.49 1.70
19 92.70 4.86 1.25
20 97.70 2.54 0.69
4.2. Experimental set up and analysis
21 92.48 4.93 1.26

The schematic of outdoor experimental setup for carrying out the

outdoor measurements of a 75 Wp PV module is shown in Fig. 4. A

962
A.K. Yadav, S.S. Chandel
Fig. 7. Performance Plot of ANN-1 Model.
Fig. 8. Performance Plot of ANN-2 Model.
RTD temperature sensor is fixed on the back surface of the 75 Wp
module to measure back surface module temperature. The meteorolo-
gical data are continuously measured at one minute time step at the
outdoor experimental PV test facility setup on the roof top of CEEE. SR
is measured by LP02 pyranometer having spectral range 350–
2800 nm, sensitivity 15μV /Wm −2 , temperature range -40 °C to 80 °C
and measurement range 0 to 2000 W/m2. Wind speed is measured by a
3 cup type anemometer sensor with range 0 to 67 m/s, resolution
0.1 m/s and accuracy better than 0.5 m/s. Air temperature is measured
by a RTD100 temperature sensor. The data are recorded by a multi-
channel data logger connected to a computer installed in the PV
laboratory. The electrical characteristics parameters are measured by
Solar Module Analyzer equipped with computer. The specifications of
SOLAR-600 used for outdoor PV characteristics measurements are
given in Table 5.
The experiment was carried out from 9.52 a.m. to 4.16 p.m. on
963
Renewable and Sustainable Energy Reviews 77 (2017) 955–969
October 22, 2013 for developing and validation of models. The 1
minute measured values of various parameters namely global solar
radiation, wind speed, air temperature, back surface module tempera-
ture, open circuit voltage, maximum voltage, maximum current and
maximum power of module are shown in are shown in Fig. 5(a, b, c, d,
e, f, g, h) respectively.
4.3. Artificial Neural Network Models for PV power prediction
As an alternative to conventional methods, ANN methods have
been used in solar radiation prediction [87,88] and photovoltaic
applications [89]. The multilayer perceptron (MLP) neural network is
mostly used for prediction which consists of an input layer, hidden
layers and output layer. The input layer contains input vector x and
output layer give output vector y [Fig. 6]. The number of hidden layer
neurons is calculated [90,91] by following equation:
A.K. Yadav, S.S. Chandel Renewable and Sustainable Energy Reviews 77 (2017) 955–969

Fig. 9. Performance Plot of ANN-3 Model.

Fig. 10. Performance Plot of ANN-4 Model.

in + on PV power under outdoor conditions. The output layer incorporates

Hn = + sn
2 (11) linear activation function and hidden layer contains different nonlinear
where Hn and sn are number of hidden layer neurons and number of function or neurons. The nonlinear function is also called as activation
data samples used in ANN model, in and on indicate number of input function and it is usually sigmoid function f (x ).
and output variables. The sensitivity test is used for validating the 1
number of hidden layer neurons by calculating change in mean f (x ) =
1 + e−x (13)
absolute percentage error (MAPE) when number of hidden layer
neurons is changed by ± 5 from calculated hidden layer neurons using Thus the ANN model with in inputs, Hn hidden layer neurons and a
Eq. 11. The MAPE is shown in Eq. (12) single output layer has nonlinear mapping from an input x to an output
y given by following equation.
⎛ n
PV max . poweri(ANN )− PVmax.poweri(measured ) ⎞
1
MAPE=⎜⎜ ∑ ⎟ × 100
⎟ Hn ⎡ ⎛ in ⎞⎤
⎝n i =1
PV max . poweri(measured ) ⎠ y= ∑ ⎢⎢wjf ⎜⎜∑ wji. xi⎟⎟⎥⎥
j= 0 ⎣ ⎝ i =0 ⎠⎦ (14)
(12)
where PV max . poweri(ANN ) is the maximum power predicted by ANN The parameters of ANN model are given by weights and biases that
and PV max . poweri (measured ) is the measured value of module maximum are connected between layers. The biases are denoted by the subscripts
power. The ANN architecture with least MAPE is used for prediction of i = 0 and j = 0 in Eq. (14). The weights (wj ,wji ) govern non linear

964
A.K. Yadav, S.S. Chandel Renewable and Sustainable Energy Reviews 77 (2017) 955–969

Fig. 11. Performance Plot of ANN-5 Model.

Table 11
ANN Models Prediction Accuracy of testing data sets.

Models Input Variables MAPE RMSE Highlights

ANN-1 SR, T, BSMT, Voc, WS 2.38 0.72 High prediction accuracy

ANN-2 SR, T, BSMT, Voc 2.26 0.65
ANN-3 SR, T, BSMT 2.54 0.69
ANN-4 SR, T 2.15 0.59
ANN-5 SR, BSMT 2.55 0.74

Fig. 13. Correlation between measured and predicted Pmax of the employed PV module
by ANN-2 for testing data point.

n
1
RMSE = ∑ (PV max . poweri(ANN )− PVmax.poweri(measured ))2
n i =1 (15)

4.4. Implementation of ANN model

As discussed earlier, the PV module maximum power is influenced

Fig. 12. Correlation between measured and predicted Pmax of the employed PV module
by ANN-1 for testing data point.
mapping of variables. The training algorithm in ANN model is
Levenberg–Marquardt (LM) algorithm which is an approximation to
the Newton method. During training the network weights are adjusted
to minimize root mean square error (RMSE) given by following
equation.
by change in solar radiation (SR ), air temperature (T ), back surface
module temperature (BSMT ), open circuit voltage (Voc ) and wind speed
(WS), so the maximum power of PV module should be a function of
these parameters:
PV max . power = f (SR, T, BSMT, Voc, WS) (16)
Therefore, the research problem is computation of the function f
where the relationship among the variables is non linear and cannot be

965
A.K. Yadav, S.S. Chandel
Fig. 14. Correlation between measured and predicted Pmax of the employed PV module
by ANN-3 for testing data point.
Fig. 15. Correlation between measured and predicted Pmax of the employed PV module
by ANN-4 for testing data point.
solved by conventional methods. So, in order to consider the effect of
each input variable on PV module maximum power prediction five
ANN models (ANN-1, ANN-2, ANN-3, ANN-4 and ANN-5) are
developed in the present study using different combinations of input
variables. ANN-1 model utilized SR, T, BSMT, Voc , WS as input vari-
ables. The input variables to ANN-2 model areSR, T, BSMT, Voc . The
ANN-3 model incorporates SR, T, BSMT as input variables. ANN-4
model uses SR, T as inputs and ANN-5 model utilizes SR, BSMT as
input variables .The output variable of all five ANN models is maximum
PV power (PV max . power ).
966
Renewable and Sustainable Energy Reviews 77 (2017) 955–969
Fig. 16. Correlation between measured and predicted Pmax of the employed PV module
by ANN-5 for testing data point.
5. Results and discussion
5.1. ANN models accuracy evaluation
The five ANN models (ANN-1, ANN-2, ANN-3, ANN-4 and ANN-5)
are developed using MATLAB software (R 2011a). For training ANN
models 200 data points are used and for testing 45 data points are
utilized from Fig. 5. The number of hidden layer neurons of five ANN
models is calculated using Eq. 11 and sensitivity test is performed as
shown in Tables 6–10.
The performance plot of five ANN models shows that mean square
error becomes minimum as the number of epochs is increasing
(Figs. 7–11). The epochs are one complete sweep of training, validation
and testing data set. The test set error and validation set error has
comparable characteristics and no major over fitting has happened
near epoch 4, 5 and 6 (where best validation performance has taken
place). The performance curve is a plot of mean square error (MSE)
with number of epochs. The MSE plot in training data has lower curve
and has an upper curve in validation data set. The network with least
MSE in validation curve is called as trained ANN model. The training of
ANN models stops automatically when validation error stops improv-
ing as indicated by an increase in MSE of validation data samples.
The prediction accuracy of ANN models are calculated with MAPE
given by Lewis [92]. The MAPE ≤ 10% indicates high prediction
accuracy, 10% ≤ MAPE ≤ 20% indicates good prediction, 20% ≤ MAPE
≤ 50% indicates reasonable prediction, MAPE≥ 50% indicates inaccu-
rate forecasting. The maximum MAPE of testing data for ANN-1, ANN-
2, ANN-3, ANN-4 and ANN-5 models are shown in Table 11, showing
high prediction accuracy of ANN models. The obtained results show
that ANN-4 model with input variable as solar radiation and tempera-
ture has least MAPE, so it can be used for prediction of maximum
power of 75 Wp PV module at any site as solar radiation and
temperature are most easily accessible variables. The MAPE of ANN-
1 model is more than ANN-2 showing wind speed as less influencing
variable in prediction of maximum power. The MAPE of ANN-3 model
is more than ANN-2 model showing open circuit voltage affects
prediction accuracy of network. The ANN-5 model with solar radiation
and back surface module temperature has high MAPE. The correlation
between measured and predicted maximum power of PV module by
A.K. Yadav, S.S. Chandel
Table 12
MLR Models.
Models Correlations
MLR-1 PVmax.power = − 0.007SR + 0.0304T + 0.1003BSMT + 1.5623Voc + 0.2268WS−12.6485
MLR-2 PVmax.power = − 0.0013SR + 0.0360T + 0.0933BSMT + 1.5520Voc − 12.3582
MLR-3 PVmax.power = 0.0037SR + 0.0486T−0.0167BSMT + 20.0173
MLR-4 PVmax.power = 0.0033SR + 0.0500T + 19.410
MLR-5 PVmax.power = 0.0032SR − 0.0197BSMT + 21.6679
ANN models for testing data point are shown in Figs. 12–16.
5.2. Multiple Linear Regression Models
In Multiple Linear Regression (MLR) model the output Y is related
to input X for each observation. The critical assumption of the MLR
model is that the conditional mean function is linear as follows:
E (Y / X ) = α + β1X1 + ....... + βpXp (17)
where α is intercept and βj are coefficients or slopes.
In this study five MLR models (MLR-1, MLR-2, MLR-3, MLR-4 and
MLR-5) are developed using least squares approach. The same
database (Fig. 3) is used to develop five MLR Models are shown in
Table 12. The five MLR models are validated with 1 minute experi-
mental values of PV module maximum current, maximum voltage,
maximum power, open circuit voltage, irradiance, air temperature,
back surface module temperature and wind speed taken on 5 December
2013. It is found that MAPE value varies from 14.76% to 20.34% which
lie in the range of good prediction as per Lewis prediction accuracy
evaluation.
6. Conclusions
In this study, an overview of conventional and ANN methods for PV
module power prediction is provided. Based on research gap experi-
mental analysis by Solar Analyzer for 75Wp PV module under outdoor
conditions is presented. In addition five ANN and MLR models are
developed with different input variable combinations to find out most
influencing variables for the prediction of maximum power. For this
ANN-1 model utilized SR, T, BSMT, Voc , WS as input variables. The
input variables to ANN-2 model are SR, T, BSMT, Voc . The ANN-3
model incorporate input variables as SR, T, BSMT . The ANN-4 model
uses SR, T as inputs and ANN-5 model utilizes SR, BSMT as input
variables. The MAPE for ANN-1, ANN-2, ANN-3, ANN-4 and ANN-5
models are found to be 2.38 %, 2.26 %, 2.54%, 2.15% and 2.55 %,
showing high prediction accuracy of all ANN models.
It is found that wind speed is less influencing variable for the
prediction of PV module maximum power. The ANN-4 model has least
MAPE (2.15%) showing solar radiation and air temperature are most
important variables in PV module maximum power prediction using
ANN model. The ANN-4 model can be applied to predict PV maximum
power based on predicted solar radiation and air temperature.
Therefore ANN models remove the complexity of solving explicit and
implicit equations for maximum power of PV module.
The MAPE for MLR models varies from 13.04 % to 19.34 %
showing good prediction accuracy as per Lewis evaluation criteria. It is
found that MLR-4 model has least MAPE (13.04%), showing solar
radiation and air temperature are important variables for prediction of
maximum power using MLR models.
The MAPE of ANN models are less than MLR models proving ANN
models give better results than MLR models.
In future work, different ANN techniques will be applied to PV
modules of different rating and technologies for maximum power
prediction under outdoor conditions.
967
Renewable and Sustainable Energy Reviews 77 (2017) 955–969
MAPE Highlights
14.76
14.80 Good
19.34 prediction
13.04 accuracy
18.44
References
[1] IEA . Technology roadmap, solar photovoltaic energy. Paris: OECD/IED; 2010.
[2] D'aiello RV. Photovoltaic device performance: an overview. Solar Cells
1986;18:243–9.
[3] Hussein KH, Muta I, Hoshino T, Osakada . Maximum photovoltaic power tracking:
an algorithm for rapidly changing atmospheric conditions. IEEE Proc Gener,
Transm, Distrib 1995;142:59–64.
[4] Khallat MA, Rahman S. A probabilistic approach to photovoltaic generator
performance prediction. IEEE Trans Energy Convers 1986;3:34–40.
[5] Vachtsevanos G, Kalaitzakis K. A hybrid photovoltaic simulator for utility inter-
active studies. IEEE Trans Energy Convers 1987;2:227–31.
[6] Araujo GL, Sánchez E. Analytical expressions for the determination of the
maximum power point and the fill factor of a solar cell. Solar Cells 1982;5:377–86.
[7] Khallat MA, Rahman S. A probabilistic approach to photovoltaic generator
performance prediction. IEEE Trans Energy Convers 1986;3:34–40.
[8] Chin VJ, Salam Z, Ishaque K. Cell modelling and model parameters estimation
techniques for photovoltaic simulator application: a review. Appl Energy
2015;154:500–19.
[9] Qi C, Ming Z. Photovoltaic module simulink model for a stand-alone PV system.
Phys Procedia 2012;24:94–100.
[10] Krismadinata Rahim NA, Ping HW, Selvaraj J. Photovoltaic module modeling using
simulink/matlab. Procedia Environ Sci 2013;17:537–46.
[11] Tsai H. Insolation-oriented model of photovoltaic module using Matlab/Simulink.
Solar Energy 2010;84:1318–26.
[12] Marion B, Rummel S, Anderberg A. Current–voltage curve translation by bilinear
interpolation. Prog Photovolt: Res Appl 2004;12:593–607.
[13] Ishaque K, Salam Z, Syafaruddin . A comprehensive MATLAB Simulink PV system
simulator with partial shading capability based on two-diode model. Solar Energy
2011;85:2217–27.
[14] Celsa D, Tina GM. Matlab/Simulink Model of Photovoltaic Modules/Strings under
Uneven Distribution of Irradiance and Temperature. In: Proceedings of the 6th
International Renewable Energy Congress (IREC); 2015. p. 1-6
[15] Chouder A, Silvestre S, Taghezouit B, Karatepe E. Monitoring, modelling and
simulation of PV systems using LabVIEW. Sol Energy 2013;91:337–49.
[16] Bencherif M, Chermitti A. New method to assess the loss parameters of the
photovoltaic modules. J Renew Sustain Energy 2012;4:1–19.
[17] Ibrahim A, Ramadan MRI, El-Sebaii AA, El-Broullesy SM. Annual performance of
solar modules with tilting angle facing south and sun tracking in Tanta, Egypt. Int J
Renew Energy Res 2011;1(1):26–40.
[18] Fuentes M, Nofuentes G, Aguilera A, Talavera DL, Castro M. Application and
validation of algebraic methods to predict the behaviour of crystalline silicon PV
modules in Mediterranean climates. Solar Energy 2007;81:1396–408.
[19] Hernandez 1J, Vallejo W, Gordillo G. Practical method for estimating the power
and energy delivered by photovoltaic modules operating under non-standard
conditions. Prog Photovolt: Res Appl 2013;21:867–75.
[20] Ibrahim A, Ramadan MRI, El-Sebaii AA, El-Broullesy SM. Outdoor and indoor
measurements of the electrical characteristics of a mono crystalline silicon solar
module. Int J Sustain Energy 2013;32(3):137–52.
[21] Dolara A, Lazaroiu GC, Leva S, Manzolini G. Experimental investigation of partial
shading scenarios on PV (photovoltaic) modules. Energy 2013;55:466–75.
[22] Patel H, Agarwal V. MATLAB-Based Modeling to Study the Effects of Partial
Shading on PV Array Characteristics. IEEE Trans Energy Convers
2008;23(1):302–10.
[23] Chen C, Chang H, Kuo C, Lin C. Programmable energy source emulator for
photovoltaic panels considering partial shadow effect. Energy 2013;54:174–83.
[24] Wang J, Shieh J, Su Y, Kuo K, Chang Y, Liang Y, Chou J, Liao K, Jiang J. A novel
method for the determination of dynamic resistance for photovoltaic modules.
Energy 2011;36:5968–74.
[25] Alessandro V, Guerriero P, Daliento S, Gargiulo M. A straightforward method to
extract the shunt resistance of photovoltaic cells from current–voltage character-
istics of mounted arrays. Solid-State Electron 2011;63:130–6.
[26] Skoplaki E, Palyvos JA. Operating temperature of photovoltaic modules: a survey of
pertinent correlations. Renew Energy 2009;34:23–9.
[27] King DL, Boyson WE, Kratochvil JA. Photovoltaic array performance model. Report
SAND2004-3535. Available from: 〈http://www.sandia.gov/pv/docs/PDF/King
%20SAND.pdf〉; 2004.
[28] King DL, Boyson WE, Kratochvill JA. Photovoltaic Array Performance Model. New
Mexico: Sandia National Laboratories; 2004.
[29] Skoplaki E, Palyvos JA. On the temperature dependence of photovoltaic module
electrical performance: A review of efficiency/power correlations. Sol Energy
A.K. Yadav, S.S. Chandel
2009;83:614–24.
[30] King DL, Kratochvil JA, Boyson WE. Temperature coefficients for PV modules and
arrays: measurement methods, difficulties, and results. In: Proceedings of IEEE
26th PVSC; 1997. p. 1183-6
[31] Hernandez J, Gordillo G, Vallejo W. Predicting the behavior of a grid-connected
photovoltaic system from measurements of solar radiation and ambient tempera-
ture. Appl Energy 2013;104:527–37.
[32] Kaplani E, Kaplanis S. Thermal modelling and experimental assessment of the
dependence of PV module temperature on wind velocity and direction, module
orientation and inclination. Sol Energy 2014;107:443–60.
[33] Jiang J, Wang J, Kuo K, Su Y, Shieh J, Chou J. Analysis of the junction temperature
and thermal characteristics of photovoltaic modules under various operation
conditions. Energy 2012;44:292–301.
[34] Kamkird P, Ketjoy N, Rakwichian W, Sukchai S. Investigation on temperature
coefficients of three types photovoltaic module technologies under thailand
operating condition. Proced Eng 2012;32:376–83.
[35] Bahaidarah H, Subhan A, Gandhidasan P, Rehman S. Performance evaluation of a
PV (photovoltaic) module by back surface water cooling for hot climatic conditions.
Energy 2013;59:445–53.
[36] García MCA, Balenzategui JL. Estimation of photovoltaic module yearly tempera-
ture and performance based on Nominal Operation Cell Temperature calculations.
Renew Energy 2004;29:1997–2010.
[37] Torres-Ramírez M, Nofuentes G, Silva JP, Silvestre S, Muńoz JV. Study on
analytical modelling approaches to the performance of thin film PV modules in
sunny inland climates. Energy 2014;73:731–40.
[38] Gaglia AG, Lykoudis S, Argiriou AA, Balaras CA, Dialynas E. Energy efficiency of PV
panels under real outdoor conditions - an experimental assessment in Athens,
Greece. Renew Energy 2017;101:236–43.
[39] Chantana J, Kamei A, Minemoto T. Influences of environmental factors on Si-based
photovoltaic modules after longtime outdoor exposure by multiple regression
analysis. Renew Energy 2017;101:10–5.
[40] Peng J, Lu L, Yang H, Ma T. Validation of the Sandia model with indoor and
outdoor measurements for semi-transparent amorphous silicon PV modules.
Renew Energy 2015;80:316–23.
[41] Sadok M, Mehdaoui A. Outdoor testing of photovoltaic arrays in the Saharan
region. Renew Energy 2008;33:2516–24.
[42] Adinoyi MJ, Said SAM. Effect of dust accumulation on the power outputs of solar
photovoltaic modules. Renew Energy 2013;60:633–6.
[43] Ozden T, Akinoglu BG, Turan R. Long term outdoor performances of three different
on-grid PV arrays in central Anatolia-An extended analysis. Renew Energy
2017;101:182–95.
[44] Visa I, Burduhos B, Neagoe M, Moldovan M, Duta A. Comparative analysis of the
infield response of five types of photovoltaic modules. Renew Energy
2016;95:178–90.
[45] Limmanee A, Udomdachanut N, Songtrai S, Kaewniyompanit S, Sato Y, Nakaishi
M, Kittisontirak S, Sriprapha K, Sakamoto Y. Field performance and degradation
rates of different types of photovoltaic modules: a case study in Thailand. Renew
Energy 2016;89:12–7.
[46] Elibol E ÖT, Tutkun N, Köysal O. Outdoor performance analysis of different PV
panel types. Renew Sustain Energy Rev 2017;67:651–61.
[47] Başoğlu ME, Kazdaloğlu A, Erfidan T, Bilgin MZ, Çakır B. Performance analyzes of
different photovoltaic module technologies under İzmit, Kocaeli climatic condi-
tions. Renew Sustain Energy Rev 2017;67:651–61.
[48] Sharma V, Chandel SS. A novel study for determining early life degradation of
multi-crystalline-silicon photovoltaic modules observed in western Himalayan
Indian climatic conditions. Sol Energy 2016;134:32–44.
[49] Rajput P, Tiwari GN, Sastry OS, Bora B, Sharma V. Degradation of mono-
crystalline photovoltaic modules after 22 years of outdoor exposure in the
composite climate of India. Sol Energy 2016;135:786–95.
[50] Qazi A, Fayaz H, Wadi A, Raj RG, Rahim NA. The artificial neural network for solar
radiation prediction and designing solar systems: a systematic literature review. J
Clean Prod 2015. http://dx.doi.org/10.1016/j.jclepro.2015.04.041.
[51] Yadav AK, Malik H, Chandel SS. Application of rapid miner in ANN based
prediction of solar radiation for assessment of solar energy resource potential of 76
sites in Northwestern India. Renew Sustain Energy Rev 2015;52:1093–106.
[52] Yadav AK, Malik H, Chandel SS. Solar radiation prediction using Artificial Neural
Network techniques: a review. Renew Sustain Energy Rev 2014;33:772–81.
[53] Yadav AK, Malik H, Chandel SS. Selection of most relevant input parameters using
WEKA for artificial neural network based solar radiation prediction models. Renew
Sustain Energy Rev 2014;31:509–19.
[54] Dahmani K, Dizene R, Notton G, Paoli C, Voyant C, Nivet ML. Estimation of 5-min
time-step data of tilted solar global irradiation using ANN (Artificial Neural
Network) model. Energy 2014;70:374–81.
[55] Makhloufi MT, Khireddine MS, Abdessemed Y, Boutarfa A. Tracking power
photovoltaic system using artificial neural network control strategy. I.J. Intell Syst
Appl 2014;12:17–26.
[56] Hiyama T, Kouzuma S, Imakubo T. Evaluation of neural network based real time
maximum power tracking controller for PV system. lEEE Trans Energy Convers
1995;10(3):543–8.
[57] Hiyama T, Kouzuma S, Imakubo T. Identification of optimal operating point of PV
modules using neural network for real time maximum power tracking control. IEEE
Trans Energy Convers 1995;10(2):360–7.
[58] Bahgata ABG, Helwab NH, Ahmadb GE, Shenawy ETE. Maximum power point
tracking controller for PV systems using neural networks. Renew Energy
2005;30:1257–68.
[59] Veligorskyi O, Chakirov, R., Vagapov, Y. Artificial neural network-based maximum
968
Renewable and Sustainable Energy Reviews 77 (2017) 955–969
power point tracker for the photovoltaic application. 1st International Conference
on Industrial Networks and Intelligent Systems; 2015. p.133–8.
[60] GKF , Yau KKW. Predicting electricity energy consumption: a comparison of
regression analysis, decision tree and neural networks. Energy 2007;32:1761–8.
[61] Zarate LE, Soares DA, Bittencout FR, Machado L. Artificial neural networks applied
for representation of curves current-voltage of photovoltaic modules. In
:Proceedings of the 6th IEEE International Conference on Industrial Informatics;
2008.
[62] Bahgat ABG, Helwa NH, Ahamd GE, Shenawy ETE. Estimation of the maximum
power and normal operating power of a photovoltaic module by neural networks.
Renew Energy 2004;29:443–57.
[63] Almonacid F, Rus C, Hontoria L, Fuentes M, Nofuentes G. Characterisation of Si-
crystalline PV modules by artificial neural networks. Renew Energy
2009;34:941–9.
[64] Almonacid F, Fernández EF, Rodrigo P, Pérez-Higueras PJ, Rus-Casas C.
Estimating the maximum power of a High Concentrator Photovoltaic (HCPV)
module using an Artificial Neural Network. Energy 2013;53:165–72.
[65] Almonacid F, Rus C, Hontoria L, Muńoz FJ. Characterisation of PV CIS module by
artificial neural networks. A comparative study with other methods. Renew Energy
2010;35:973–80.
[66] Sulaiman SI, Rahman TKA, Musirin I. Partial evolutionary ANN for output
prediction of a grid-connected photovoltaic system. Int J Comput Electr Eng
2009;1(1):40–5.
[67] 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 photo-
voltaic module. Appl Energy 2012;97:956–61.
[68] Mellit A, Pavan AM. Performance prediction of 20 kWp grid-connected photo-
voltaic plant at Trieste (Italy) using artificial neural network. Energy Convers
Manag 2010;51:2431–41.
[69] Mellit A, Sağlam S, Kalogirou SA. Artificial neural network-based model for
estimating the produced power of a photovoltaic module. Renew Energy
2013;60:71–8.
[70] Brano VL, Giuseppina Ciulla G, Falco MD. Artificial neural networks to predict the
power output of a PV panel. Int J Photo Energy 2014;1:12.
[71] Piliougine M, Elizondo D, Mora-López L, Sidrach-de-Cardona M. Modelling
photovoltaic modules with neural networks using angle of incidence and clearness
index. Prog Photovolt: Res Appl 2015;23:513–23.
[72] Veerachary M, Yadaiah N. ANN based peak power tracking for PV supplied dc
motors. Sol Energy 2000;69(4):343–50.
[73] Graditi G, Ferlito S, Adinolfi G, Tina GM, Ventura C. Performance estimation of a
thin-film photovoltaic plant based on an Artificial Neural Network model. In:
Proceedings of the fifth International Renewable Energy Congress IREC; 2014
[74] Hiyama T, Kitabayashi K. Neural network based estimation of maximum power
generation from PV module using environmental information. IEEE Trans Energy
Convers 1997;12(3):241–7.
[75] Fernández EF, Almonacid F, Sarmah N, Rodrigo P, Mallick TK, Pé rez-Higueras P.
A model based on artificial neuronal network for the prediction of the maximum
power of a low concentration photovoltaic module for building integration. Sol
Energy 2014;100:148–58.
[76] Askarzadeh A. Voltage prediction of a photovoltaic module using artificial neural
networks. Int Trans Electr Energy Syst 2014;24(12):1715–25.
[77] Saberian A, Hizam H, Radzi MAM, Kadir MZAA, Mirzaei M. Modelling and
prediction of photovoltaic power output using artificial neural networks. Int J
Photoenergy 2014;2014:1–10.
[78] Niu D, Wei Y, Chen Y. Photovoltaic power prediction based on scene simulation
knowledge mining and adaptive neural network. Math Prob Eng 2013;2013:1–6.
[79] Jiang LL, Maskell DL, Patra JC. Chebyshev Functional Link Neural Network-based
modeling and experimental verification for photovoltaic arrays. In: Proceedings of
the 2012 international joint conference on neural networks (IJCNN); 2012. p. 1-8.
[80] Hussain TN, Sulaiman SI, Musirin I. A hybrid artificial neural network for grid-
connected photovoltaic system output prediction. In: Proceedings of IEEE sym-
posium on computers & informatics (ISCI); 2013. p. 108-111.
[81] Sulaiman SI, Muhammad KS, Musirin I, Shaari S. Hybridization of Meta-
Evolutionary Programming and Artificial Neural Network for predicting grid-
connected photovoltaic system output. In: Proceedings of IEEE TENCON spring
conference; 2013. 445–9.
[82] Routh RK, Bin YAH, Hossain MN, Asasduzzaman MM, Hossain MI, Husnaeen U,
Mubarak M. Artificial neural network based temperature prediction and its impact
on solar cell. In: Proceedings of international conference on informatics, electro-
nics & vision (ICIEV); 2012. p. 897–902.
[83] Ramawan MK, Othman Z, Sulaiman SI, Musirin I, Othman N. A hybrid bat
Algorithm Artificial Neural Network for grid-connected photovoltaic system output
prediction. In: Proceedings of 8th IEEE International Power Engineering and
Optimization Conference (PEOCO); 2014. p. 619–23.
[84] Singh VP, Vijay V, Bhatt MS, Chaturvedi DK. Generalized neural network
methodology for short term solar power forecasting. In: Proceedings of the 13th
international conference on environmental and electrical engineering; 2013. p. 58–
62.
[85] Singh VK, Vaibhav K, Chaturvedi DK. Solar power forecasting modeling using soft
computing approach. In: Proceedings of Nirma University international conference
on engineering (NUICONE) 2012. p. 1–5.
[86] Huang C, Bensoussan A, Edesess M, Tsui KL. Improvement in artificial neural
network-based estimation of grid connected photovoltaic power output. Renew
Energy 2016;97:838–48.
[87] Inman RH, Pedro HTC, Coimbra CFM. Solar forecasting methods for renewable
energy integration. Prog Energy Combust Sci 2013;39:535–76.
A.K. Yadav, S.S. Chandel
[88] Diagne M, David M, Lauret P, Boland J, Schmutz N. Review of solar irradiance
forecasting methods and a proposition for small-scale insular grids. Renew Sustain
Energy Rev 2013;27:65–76.
[89] Mellit A, Kalogirou SA. Artificial intelligence techniques for photovoltaic applica-
tions: a review. Prog Energy Combust Sci 2008;34:574–632.
969
Renewable and Sustainable Energy Reviews 77 (2017) 955–969
[90] Chow SKH, Lee EWM, Li DHW. Short-term prediction of photovoltaic energy
generation by intelligent approach. Energy Build 2012;55:660–7.
[91] Frederick M. Neuroshell 2 Manual, Ward Systems Group Inc., 1996.
[92] Lewis CD. International and business forecasting methods. London: Butter-worths;
1982.