Solar Energy 173 (2018) 139–151

Contents lists available at ScienceDirect

Solar Energy
journal homepage: www.elsevier.com/locate/solener

Modeling of PV system based on experimental data for fault detection using T

kNN method
⁎
Siva Ramakrishna Madeti , S.N. Singh
Alternate Hydro Energy Centre, Indian Institute of Technology Roorkee, Uttarakhand 247667, India

A R T I C LE I N FO A B S T R A C T

Keywords: In this paper, a string level fault detection and diagnosis technique for photovoltaic (PV ) systems based on k-
Photovoltaic (PV) modeling nearest neighbors (kNN ) rule is proposed. It detects and classifies open circuit faults, line-line (L−L ) faults,
Fault detection techniques partial shading with and with-out bypass diode faults and partial shading with inverted bypass diode faults in
k-Nearest neighbors (kNN) real time. A detailed modeling of the PV systems based on experimental data is presented that only requires
available data from the manufacturer’s datasheet reported under standard test conditions (STC ) and normal
operating cell temperature(NOCT ) . This model considers the temperature dependent variables such as junction
thermal voltage (Vt ), diode quality factor (A) and series resistance(Rs ) . Simulations of the developed model have
been carried out using Matlab/Simulink. A PV analyzer (Solar I-V) of HT instruments is used to measure the
I (V ) characteristics of PV module. The developed model precisely traces theI (V ) characteristics of PV systems at
different irradiance and temperature levels. The simulation results indicate that the error between the measured
data and developed model is less than the models available in the literature. The absolute error is confined in the
range 0.61 to 6.5%. Finally, the data generated from proposed model and experimental setup are used to validate
and test the performance of the proposed fault detection and classification(FDC ) technique. It is observed from
the results that the average of fault classification gives a high accuracy of 98.70%.

1. Introduction classification technique is required. This would enable operator to take

Photovoltaic (PV) systems provide a promising solution to directly
utilizing solar energy and are currently gaining popularity as the
technologies are mature and as the material costs are driven down
(Mercure and Salas, 2012). Global installed PV capacity at the end of
2016 was reported as 310 GWp (PVPS, IEA, 2015). However, as they are
installed in outdoor environment, continuous exposure to harsh cli-
matic conditions (sun beam, rainfall, etc.) may affect the system per-
formance. A monitoring study of PV system is reported in (Firth et al.,
2010) and it was reported that the annual power loss due to various
faults is about 18.9%. Hence, continuous monitoring along with fault
diagnosis techniques is essential to detect the causes affecting the
performance of the PV system (Madeti and Singh, 2017). In order to
detect and clear the faults present in the system, proper fault
corrective measures, which improve the performance of PV system by
minimizing the power losses caused by the faults.
Conventional fault detection and protection methods usually add
fuses and circuit breakers within PV components to prevent PV com-
ponents from experiencing large fault current. These devices are also
not able to detect various faults unique to the PV system (Zhao et al.,
2013). Currently, thermal cameras (Boztepe et al., 2014), earth capa-
citance measurements (ECMs) (Ding et al., 2012), and time-domain
reflectometry (TDR) (Petrone et al., 2007) are the three popular
methods for PV fault diagnosis. Thermal cameras are employed to de-
tect the temperature characteristics of a PV array under fault condi-
tions. Thermal images can be also linked to the maximum power point
tracking (MPPT) algorithm of a PV controller. In practice, a gradual
change in the thermal image of a PV module (e.g., due to device aging)

Abbreviations: AC, alternating current; AI, artificial intelligence; ABC, artificial bee colony; AFSA, artificial fish swarm algorithm; ACMT, adjacent string comparison
measurement techniques; ANFIS, adaptive-neuron fuzzy inference systems; CMM, comparison between measured and modeled; CSO, cat swarm optimization; DC,
direct current; ECM, earth capacitance measurement; EIM, external injection methods; FDC, fault detection and classification; GCPV, grid connected photovoltaic;
HETB, heat exchange and temperature based models; kNN, k-nearest neighbors; L-L, line-line fault; MLT, machine learning techniques; MPPT, maximum power point
tracking; NOCT, normal operating cell temperature; NIWE, national institute of wind energy; PC, personal computer; PSO, particle swarm optimization; PLA, power
loss analysis; PV, photovoltaic; RES, renewable energy sources; SBDF, shading with faulted bypass diode; SBDI, shading with inverted bypass diode; SBDN, shading
with bypass diode normal; STC, standard test conditions; TDR, time-domain reflectometry
⁎
Corresponding author.
E-mail address: ramakrishna.iitroorkee@gmail.com (S.R. Madeti).

https://doi.org/10.1016/j.solener.2018.07.038
Received 3 March 2018; Received in revised form 29 June 2018; Accepted 13 July 2018
Available online 27 July 2018
0038-092X/ © 2018 Elsevier Ltd. All rights reserved.
S.R. Madeti, S.N. Singh Solar Energy 173 (2018) 139–151

Nomenclature ns number of cells connected in series, –

Pac output ac power, W
Symbol Description, Units Parr array power, W
A diode quality factor, – Pmod module power, W
G Measured irradiance, W/m2 Pmp maximum power, W
Gd diffuse solar radiation, kWh/m2 Pstr string power, W
Gg global solar radiation, kWh/m2 q electronic charge, C
Gref reference irradiance, W/m2 Rs series resistance, Ω
Gt tilted solar radiation, kWh/m2 Rsh shunt resistance, Ω
I current of module, A Sw wind speed, m/s
I0 reverse saturation current, A T cell temperature, °C
Iarr array current, A T cell temperature, K
Ibat battery current, A Ta ambient temperature, °C
ID1 diode current, A Tbat battery temperature, °C
Iexp experimental model output current, A TL load temperature, °C
IL load current, A Tm module temperature, °C
Il-l line to line current, A V voltage of module, V
Imod module current, A Varr array voltage, V
Imodel mathematical model output current, A Vbat battery voltage, V
Imp maximum current, A Vg grid voltage, V
Impp current at maximum power point, A VL load voltage, V
Iph photo-generated current, A Vmod module voltage, V
IRsh leakage current flowing through the shunt resistor, A Vmp maximum voltage, V
Isc short circuit current, A Vmpp voltage at maximum power point, V
Isc short-circuit current, A Voc open circuit voltage, V
Istr string current, A Voc open-circuit voltage, V
k Boltzmann’s constant, J/K Vstr string voltage, V
Ki temperature co-efficient for short-circuit current, mA/°C Vt junction thermal voltage, V
Kv temperature co-efficient for open-circuit voltage, mV/°C

poses a technical challenge, and high system costs also limit the wide
application of thermal cameras. The ECM can locate the disconnection
of PV strings, whereas the TDR technology can predict the degradation
of a PV array. Nonetheless, both the ECM and the TDR can only operate
offline. In practice, online diagnosis methods are highly desired, which
can take measurements while the tested device is in operation.
In (Solórzano and Egido, 2013), to monitor the performance of each
PV module the power optimizer is developed, which is connected to
each PV module by replacing the traditional solar junction box. The
power optimizers increase energy output from PV systems by constantly
tracking the maximum power point (MPPT) of each module in-
dividually. However, more pieces of equipment must be used causes
increase in maintenance and cost of the system. Automatic systems for
anomaly detection in the PV systems are thus needed for remote sig-
naling for the detection and classification of the fault. Various methods
have been proposed in the past and their characteristics are discussed in
Table 1. A comprehensive comparison and analysis of various fault
detection techniques has been presented in (Madeti and Singh, 2017).
Due to the non-linear nature of the PV modules, machine learning
techniques have been driven more attention over the recent years. Since
threshold values are required for the other fault detection techniques
which are hard to define for the PV system. From Table 1, it is observed
that the machine learning techniques may be the more feasible option
for accurate fault detection.
Most of the proposed methods for fault classification categorize the
faults in either of the four detectable fault types which are module open
circuit, short circuit faults such as Line-Line or Line-Ground faults. The
open circuit fault can be related to two subclasses: open circuit PV
string and open circuit PV module. The open circuit PV string can be
caused by bad connections or aged power cables, while the open circuit
PV module can be generated by bad connections or broken cells in hot
spot areas. The short circuit fault happens when one or several PV
modules in the PV string are short circuited due to bad connections.
Some methods however only propose methods to detect the faults and
does not propose classification algorithm. And there are others which
make an analysis of the classification features of these faults and does
not propose or implement a detection means.
Some of the methods (Firth et al., 2010; Drews et al., 2007; Chouder
and Silvestre, 2010; Polo et al., 2010), classify the faults in terms of
energy losses. The classification is based on the rate of energy loss and
not whether the four detectable fault types used for categorization.
Their detectable features in addition include component failures, in-
verter defects, MPP tracking failure etc, but are not able to classify the
various faults in DC side of PV systems. A procedure for fault diagnosis
in PV systems with distributed MPPT at module level, power optimizers
DC-DC or micro-inverters DC-AC, is proposed in (Solórzano and Egido,
2013). It has been shown that the designed procedure can diagnose a
large scope of failures including: fixed object shading, localized dirt
generalized dirt, hot spots, module degradation and excessive losses in
DC cables.
A reliable and accurate PV model is essential for fault detection and
classification. It is used for predicting the energy that can be harvested
for a PV plant in a specific location. It is also needed to verify the ef-
fectiveness of developed fault classification technique, MPPT algo-
rithms and control structures. Moreover, various fault diagnosis tech-
niques are based in the analysis of power losses in the PV system. The
losses are calculated by means of comparison between the monitored
data with simulation results. When the simulation model of a PV system
is not accurate which results in an over/under estimation of the output
power in the simulation results. It causes false fault detection.
Various methods have been proposed before for modeling of the PV
module from manufacturer’s data sheet, but they have certain limita-
tions. The model proposed in this paper builds on these limitations and
proposes an improved model. Some of the limitations of the previous
models are discussed below.
Di Piazza and Vitale (2013) proposed a new method for PV mod-
eling; however, it is able to reproduce the behaviour of a PV module and
its respective I (V ) curve, only under STC conditions. This method is

140
 S.R. Madeti, S.N. Singh

Table 1
Comparison of characteristics of various fault detection techniques for PV Plants.
Characteristics CMM ACMT PLA EIM HETB MLT

Threshold values Need to set threshold value. No threshold value needs to be set. Need to set threshold value. No threshold value needs to No threshold value needs to be set. No threshold value
set be set. needs to be set.
Classify the type of Able to classify the type of fault. Not able to classify the type of fault. Able to classify the faults like inverter This method is able to Not able to classify the type of fault. Able to classify the
fault shutdowns, snow cover, shading, locate the faulted module type of fault.
defective inverters, snow cover, but this in the PV string, but not
method is not able to distinguish able to classify the type of
between Line-Line faults, open circuit fault.
faults.
Time taken Relatively less Less Less High High Less
On/Off line mode On line On line On line Off line On line On line
of operation

141
Computa-tion Relatively high Easy Relatively high High High Less
Accurate results Since the expected yield is a This method assumes that all healthy This method requires meteorological Since the classification is Since the method relies on difference Classification
calculated value, system losses such as strings have the same output data, and the whole efficiency of the based on graphical signal measure of heat exchange and PV Accuracy can reach
cable losses, joint losses, etc, is characteristics, however in the case of system depends on the accuracy of the response, the results can be module temperatures, the methods do up to 100 percent.
difficult to model thus different large grid tied PV systems, string meteorological data. This gives very inaccurate. not take into account wind speeds,
methods of manipulating the expected currents can significantly vary during inaccuracies in the fault detection. dust distribution which cause non-
yields, which fail to give accurate normal condition because location of uniform temperature distribution.
results. the strings with respect to the exposed This causes inaccurate results.
sunlight, Cable impedances etc, which
gives erroneous results.
Sensors required Measurement sensors at every module No Irradiation and temperature sensors Required to measure Electrical and External injection devices Does not require Irradiation sensors Available sensors in
are required for increasing detection are required. Meteorological data. are required, such as pulse PV systems can be
accuracy. generators and used.
oscilloscopes
Cost High Less cost High Very High Less cost Less
Solar Energy 173 (2018) 139–151
S.R. Madeti, S.N. Singh
not able to simulate the PV module behaviour on its own under dif-
ferent solar radiation and temperature levels. Method presented in
(ALQahtani et al., 2012) characterizes the PV performance, but it per-
forms well under fixed STC conditions. In case if both irradiance and
temperature are outside the STC condition, it fails to determine the
response. In Ortiz-Rivera and Peng (2005), the proposed method is able
to accurately simulate the I-V and P-V characteristics of a module by
using the manufacturer’s data sheet. However, this method relies on a
parameter Vmin , which is the open-circuit voltage of the PV module at
200 W/m2. In most of the cases Vmin is not specified accurately, since its
value has to be obtained from the I −V characteristics of the data sheet.
The model proposed in (Chenni et al., 2007) gives a good representa-
tion of I (V ) and P (V ) curves of the PV module, but the limitation of
this method is that A and Rs are taken as a fixed values. The fixed values
of A and Rs can limit the accuracy of the model, since these values vary
with temperature. The value of Rs can be determined by matching the
obtained curve with equivalent data obtained through experimental
test. This method of finding Rs is impossible by using the manufacturer’s
data sheet alone (Xiao et al., 2004). An iterative differential evaluation
technique is proposed in da Costa et al. (2010) to derive a relationship
between Rs and temperature. However, it is computationally complex to
find the constant temperature coefficient by using this method. The
iterative technique proposed in Villalva et al. (2009) is the very pro-
mising; but, accuracy of the model is limited due to fixed value of Rs at
different temperatures. Many methods consider A as a constant value
for different PV cell technologies (Villalva et al., 2009; Salmi et al.,
2012; Walker, 2001), but it has been seen that it does vary with tem-
perature.
Conventional methods are incapable of estimating the parameters of
PV modules with high accuracy. This led many scientists to search for
Artificial Intelligence (AI) techniques as alternate approaches to con-
ventional techniques for PV modeling. A neural network modeling
method for single diode equivalent model is presented in Karatepe et al.
(2006) while In Askarzadeh and Rezazadeh (2012) a parameter iden-
tification using harmony search algorithm is proposed. The results in
the paper show that the superior performance of harmony search al-
gorithm over simulated annealing and genetic algorithms. The problem
of parameters identification is formulated as a multi-objective function
and Artificial Bee Colony (ABC) algorithm is proposed for solving the
multi-objective problem (Oliva et al., 2014; Askarzadeh and Rezazadeh,
2013). In Han et al. (2014) a parameter identification based on Artifi-
cial Fish Swarm Algorithm (AFSA) is presented. Simulated annealing
algorithm is also used as an optimization algorithm for finding the
model parameters of the solar cell in El-Naggar et al. (2012). Bird
mating optimizer is developed by Askarzadeh and dos Santos (2015), to
solve the parameter identification problem. The results show the su-
perior performance of the algorithm over Particle Swarm Optimization
(PSO) algorithm. In Saravanan et al. (2013) hybrid GA-PSO algorithm is
proposed for extracting the model parameters of PV cell. In Bastidas-
Rodriguez et al. (2017), Zagrouba et al. (2010). A genetic algorithm for
identifying the parameters for single diode is proposed. a parameter
identification using Cat Swarm Optimization (CSO). The algorithm is
used to estimate unknown parameters of single and double diodes
models (Guo et al., 2016). Also parameter identification based on
Adaptive-Neuron Fuzzy Inference Systems (ANFIS) is presented com-
pared to neural network (NN) (Salem and Awadallah, 2014). The re-
sults show the superior performance of NN over ANFIS. In Jacob et al.
(2015) Artificial Immune System (AIS) is proposed as a solution and
compared to particle swarm optimization is presented. The research
results show the superior performance of AIS over PSO and GA.
This paper uses for the first time the k-Nearest Neighbor (kNN ) rule
based method to detect and classify the fault as well as locate the
faulted string of the PV array in a typical grid tied distributed inverter
PV system. Unlike the existing methods, the proposed fault detection
method is independent of threshold values, accurately classify the type
of fault, operate in online mode, less computational time and lower
142
Solar Energy 173 (2018) 139–151
cost. Moreover, this method utilizes the sensors already available to the
PV system such as voltage, current, irradiation and temperature to
detect and classify the faults. A method for modeling of a PV system is
developed to validate the performance of the proposed method. The
above mentioned modeling methods only use STC points of the man-
ufacturer’s datasheet for PV system modeling. However, the proposed
approach utilizes both NOCT and STC remarkable points to model the
PV system. The points (0, ISC ), (Vmp, Imp) and (VOC , 0) on the V-I char-
acteristic are referred to as the remarkable points. By using these re-
markable points the relationship between 'A' and 'Rs ' with PV module
temperature is developed. This model correctly tracks the behaviour of
PV module under different irradiance and temperature levels. Matlab/
Simulink is used to implement the models and obtain the corresponding
results. The data generated from this simulation model and from an
experimental setup is used for validation and implementation of the
proposed kNN based FDC technique.
The paper is organized as follows: Section 2 shows the developed
method for PV modeling. Section 3 discusses the proposed FDC using
kNN . Fault detection algorithm is described in Section 4. Simulation
results are given in Section 5. Finally, Section 6 concludes the paper.
2. Detailed modeling of the PV system
In this paper, a method is implemented to develop relation between
A and Rs with temperature. This model uses the single diode model in
shown in Fig. 1, for the electrical stimulation of the PV cell. It consists
of current source, a diode, and series and parallel resistances.
The voltage current I (V ) relationship of PV can be given using
Kirchhoff’s current law (Chatterjee et al., 2011), as
I = Iph−ID1−IRsh (1)
where Iph is the photo-generated current, ID1 is the diode current, IRsh is
the shunt resistance current.
The diode current ID1 signifies diffusion and recombination current
in quasi steady state regions of emitter and excess concentration regions
of PN junction. This diode current is represented by Shockley equation
as:
q (V + IR s ) ⎞
⎧ ⎛ ⎜
KTA ⎠ −1⎫
⎟
ID1 = Is e⎝
⎨ ⎬ (2)
⎩ ⎭
−19
where q is the charge of an electron (1.6 × 10 C), K is the
Boltzmann constant (1.3805 × 10−23 J/K), T is temperature, Is is
leakage current, V and I are the module voltage and current, respec-
tively; Rs is the series resistance and A is the diode quality factor.
and IRsh is the leakage current flowing through the shunt resistor Rsh
expressed as
V + IRs
IRsh =
Rsh (3)
Substituting Eqs. (2) and (3) for ID1 and Ish in Eq. (1), then final
expression can be written as
q (V + IR s ) ⎞
⎧ ⎛ ⎜
KTA ⎠ −1⎫−
⎟
V + IRs
I = Iph−Is e⎝
⎨ ⎬ Rsh (4)
⎩ ⎭
Fig. 1. Single-diode model of a PV source (Chatterjee et al., 2011).
S.R. Madeti, S.N. Singh Solar Energy 173 (2018) 139–151

The presence of the Rsh in the circuit model is mainly due to the T = 0.943Tamb + 0.028 × G−1.528 × wind speed + 4.3 (17)
leakage current of the p-n junction depending on the fabrication process
G
of the PV cell. Considering PV modules or fields as a part of an electric T = Tamb + (NOCT −20°C)
800 (18)
generation system, the Rsh has a stronger influence in an uncommon
region of operation, i.e., the constant current region and for low solar In (Xiao et al., 2004) and (Zdravkovic et al., 2009), effect of tem-
irradiance values. Therefore, the simplification introduced by ne- perature on PV characteristics are studied experimentally and it is
glecting Rsh does not affect significantly the validity of the model and its found that the values of Rs and A vary linearly with temperature. Eqs.
use for the scope of emulating the PV source, since the power genera- (19) and (20) provide the improvements in the existing models to give
tion occurs at high irradiance values and in the neighborhood of the the relation of A and Rs under variable temperature.
MPP.
It is a common practice to neglect the term ‘–1′ in Eq. (4) because, in Rstc −Rnoct ⎞
Rs (T ) = Rstc ⎡1 + ⎛ ⎜ (T −Tstc ) ⎤ ⎟
⎢ ⎝ (T −Tnoct ) × ns ⎠ ⎥ (19)
silicon devices, the dark saturation current is much smaller than the ⎣ stc ⎦
exponential term. Then, Eq. (4) becomes as follows:
Astc −Anoct ⎞
q (V + IR s ) ⎞ A (T ) = Astc ⎡1 + ⎛ ⎜ (T −Tstc ) ⎤ ⎟

⎧ ⎛ ⎜

KTA ⎠ ⎫
⎟ ⎢ ⎝ (Tstc −Tnoct ) × ns ⎠ ⎥ (20)
I = Iph−Is e⎝ ⎣ ⎦
⎨ ⎬ (5)
⎩ ⎭ The temperature dependency of thermal voltage of the diode and
q
Considering K1 = AKT
and Is = e K2 , the final PV cell model can be energy band gap is given by
given as:
kT
Vt (T ) =
I = Iph−e[(V + IRs ) K1+ K2] (6) q (21)
By using the equation (6), the PV cell electrical behaviour can be Finally, the output current of PV module at different irradiance and
studied once the four model parameters K1, K2, Rs and Iph are known. temperature is given as
The four parameters can be obtained by the knowledge of the short- V + IRs (T )
circuit current (Isc ), open circuit voltage (Voc ), and the maximum power I = Iph (G, T )−Is (T ) × ⎛e Vt (T ) A (T ) ⎞
⎝ ⎠ (22)
point voltage and current (Vmp and Imp ). The system of four equations
that have to be solved in order to obtain the four unknowns are: where the reverse saturation current as formulated in Jiang et al.
Isc = Iph−e[(Isc Rs ) K1+ K2] (2013) is given by
(7)
Isc, stc + Ki × (T −Tstc )
0 = Iph−e[Voc K1+ K2] (8) Is (T ) = Voc, stc + Kv × (T − Tstc )
e Vt (T ) A (T ) (23)
Imp = Iph−e[(Vmp+ Imp Rs ) K1+ K2] (9) Finally, PV module can be modeled precisely for a variety of cli-
matic conditions by using above equations and NOCT , STC information
dP V = Vmp
| =0 from the manufacturer’s datasheet. The proposed model is simulated in
dV I = Imp (10)
Matlab/Simulink to validate its accuracy and effectiveness. The fol-
And finally the solutions are given by: lowing information about the module is provided in the data sheet
under NOCT and STC conditions: open-circuit voltage Voc , short-circuit
Imp
K1 = current Isc , current Impp , voltage Vmpp at maximum power point (MPP ),
Vmp (Iph−Imp) (11) number of cells connected in series ns , temperature co-efficient for
open-circuit voltage Kv , and short-circuit current Ki . A case study was

Rs =
log ( Iph − Imp
Iph )−V mp K1 + Voc K1 conducted on polycrystalline type, PV module produced by JJ Solar,
Imp K1 PV-JP60F230, India. Technical specifications of JJP60F230 Wp PV
(12)
module available in laboratory is given in Table 2.
K2 = log (Iph )−Voc × K1 (13) The proposed model is experimentally validated at Solar
Laboratory, Alternate Hydro Energy Centre (AHEC), IIT Roorkee. It
Iph = Isc + e (Isc Rs ) K1+ K2 (14) includes multi-crystalline silicon PV module of 230 Wp, a silicon solar
cell based irradiance sensor (Atersa calibrated cell), and a temperature
The information provided in the manufacturer’s data is used to es-
sensor (PT300N). The irradiance sensor (fixed at optimum tilt angle)
timate these parameters for both NOCT and STC conditions. The value
and temperature sensor are used to measure the solar radiation and
of A for both the NOCT and STC conditions can then be found from the
q temperature of PV module. Solar I-V of HT instruments is used to
assumption K1 = AKT using
measure the current-voltage characteristics of PV module. The data
q
Astc = measured by the instrument is transferred to a personal computer (PC)
K1stc × Tstc × K (15)
q Table 2
Anoct = JJP60F230 Wp PV1module data-sheet2electrical1parameters1at
K1noct × Tnoct × K (16)
STC.
When the value of Tnoct is provided as ambient temperature in the Electrical1parameter Value
data sheet, it is necessary to convert the Tnoct into module or cell tem-
perature. The expression for predicting the cell temperature as a func- Rated Power (Pmax ) 230 W
tion of irradiance, ambient temperature and wind speed is given by Volt at Max. power (Vmp) 29.3 V
Chenni et al. (2007). The wind speed data for other sites with height of Open circuit voltage (VOC ) 36.8 V
Current at Max. Power (Imp) 7.85 A
50 and 75 high meters can be provided by National Institute of Wind
Short circuit current (Isc ) 8.4 A
Energy (NIWE) (http://niwe.res.in:8080/NIWE_WRA_DATA/
Tolerance of rated power (%) ±3%
DataTable_D4.jsf). If windspeed parameter is not available, Eq. (18) can Kv −2.13 mV / K
be used to obtain the cell temperature from the irradiance (G ) and the Ki 4.4 6mA/ K
NOCT.

143
S.R. Madeti, S.N. Singh
using Universal Serial Bus (USB-RS232). The technical specifications of
Solar I-V curve tracer, electrical characteristics of irradiance sensor and
temperature sensor are given in Tables 3–5, respectively.
Outdoor experiment readings were taken from 10:00 to 17:00 hrs at
an interval of one hour for a single day (16th Nov 2017). Fig. 2 shows
the electrical characteristics of the PV module, measured by a PV
analyzer ‘Solar I-V’ of HT instruments and simulated model for the PV
module manufactured by JJ solar (JP60F230), under various irradiance
and temperature levels. The results show that the proposed simulation
model is able to reproduce the manufacturer specified characteristics of
the PV module. The values of Rs and A vary linearly with temperature
as 2.934 × 10−3 ohm/°C and 5.57 × 10−3/°C respectively.
In order to evaluate the accuracy of the model, root mean square
error (RMSE) was used (Di Piazza and Vitale, 2013). The RMSE is given
as follow:
n 0.5
2
⎡ ∑ (Isimi−Imesi ) ⎤
RMSE = ⎢ i = 0 ⎥
⎣ n ⎦ (24)
where Isimi and Imesi is the ith
simulation current and the ithmeasured
current, respectively and n is number of measured points. Comparing
the measured values with simulated values, it is noted that the results
are very similar. The RMSE between the measured and simulated values
is 0.0282.
To show the impact of modeling the temperature dependency of Rs
and A , a quantitative error analysis is performed by calculating the
percentage error between the results obtained from the proposed model
with the model developed in Di Piazza and Vitale (2013) (where the
values of Rs and A are taken as fixed), is given in Table 6. The estimated
error is defined as follows:
Isimi−Imesi
Error % = × 100
Isimi (25)
ith
where Isimi is the simulation current from the mathematical model at
voltage V , irradiance G and temperature T , and Imesi is the ith measured
current at voltage V , irradiance G and temperature T .
The value of Rs = 0.2312 Ω and A = 1.3079 are obtained through
the proposed model at 28.1 °C temperature. Fig. 3 shows the compar-
ison results under different irradiance and temperature levels.
From the comparison results, it is found that the error between the
proposed model and measured data is less than the model developed in
Di Piazza and Vitale (2013), which assumes Rs and A values are con-
stant values.
3. Fault detection and classification using kNN
In this paper, the traditional kNN rule is adapted for fault classifi-
cation. In pattern classification, the kNN rule is a method to classify a
new object by examining its distance to the nearest neighboring
training samples in the feature space (Duda et al., 2001; Schmidt et al.,
2004). In Fig. 4, the squares and triangles are training samples from
classes A and B respectively; X , shown as the circle, is the sample to be
classified. In this example, Euclidean distance metric is used and dashed
circles with different radii are distance references for finding nearest
neighbors. For pattern classification, kNN algorithm only requires an
integer k, a set of labeled samples (training data), and a metric to
measure distances. Due to its simplicity, the implementation is
straightforward.
Fig. 5 shows the type of faults which are detected and classified by
this method. The faults detected and classified are: open circuit, L-L
fault, shading with bypass diode normal (SBDN), shading with faulted
bypass diode (SBDF) and shading with inverted bypass diode (SBDI)
fault. These faults may not be identified properly by the conventional
protection devices (Zhao et al., 2013, 2015).
The open circuit fault takes place in PV modules or in PV string due
to several reasons, like: broken cells, physical breakdown of cable
144
Solar Energy 173 (2018) 139–151
joints, loose connections and aged power cables near terminal leads.
During this fault the bypass diode maintains the circuit continuity and
allows current to flow into the inverter. However, this fault causes re-
duction in faulted string voltage and current remains the same, which
leads to major reduction in output power.
L-L fault occurs as a result of accidental connection between two
nodes of the PV array. This fault happens when short circuit occurs
between the current carrying conductors, cable insulation failures, da-
mage caused by mechanical forces and water ingress or corrosion of the
DC junction box. This fault leads to reduction in voltage across the
faulted string, which consequently draws a significant back-feeding
current from the other strings. The conventional over current protection
devices can interrupt the back-feeding current and isolate the faulted
strings. However, under low irradiance below 500 W/m2, the faulty
string draws lower back-feeding currents from the other strings. Under
these circumstances, conventional protection devices may fail to detect
the faults (Zhao et al., 2013).
The partial shading fault can be caused by non-uniform solar ra-
diation on modules due to passing clouds, trees and building near the
PV plant, mismatch in modules. Within a PV string if some of the PV
cells/modules are shaded then those cells/modules get reverse biased
and act as a resistive load to dissipate power. This increases the tem-
perature of a PV module/cell, which leads to permanent destruction of
whole PV system due to fire hazards. By using bypass diodes the losses
due to partial shading can be reduced. Finally, the inverted bypass
diode fault occurs when bypass diodes across PV modules are connected
in reverse direction can be caused by bad connections made by the
operator. This fault causes reduction in open circuit voltage while short
circuit current always decreases due to shading effect. It leads to re-
duction in maximum output power from the module. The PV output
characteristics during these partial shading is similar to that of open
circuit. It varies however in respect to the time the circuit remains open.
The proposed fault detection algorithm uses the time factor of the
shading occurrence for shade detection. During the above mentioned
fault conditions the PV system will operate at lower MPP than normal.
Therefore, to restore the optimal power from the PV system the fault
detection and diagnosis is essential.
In this paper, the mentioned faults are applied at 20%, 40%, 60%
and 80% in the system. For example, 20% short circuit fault refers the
scenarios where 20% of the PV modules in the string are short circuit.
Likewise 40%, 60% and 80% are defined as scenarios where 40%, 60%
and 80% of the PV modules in the string are short circuit, respectively.
Faults occurring at 20%, 40%, 60% and 80% have been collectively
considered under the same fault category. For example, whether an
open circuit fault is occurring at 20%, 40%, 60% or 80%, it will be
characterized as an open circuit fault. The proposed fault detection
technique is applied to the string based PV system. Since the reliability
and observability of string based system is better as compared to cen-
tralized system. Unlike existing fault detection methods, the proposed
approach can accurately estimate the type of fault exists. A detailed
description of kNN rule is given in He and Wang (2007). Some of the
Table 3
Technical specification of ‘Solar I-V’ curve tracer.
Manufacturer name HT instruments
Measuring range of I-V curve/VOC-ISC 1000 V/15 A
Measuring range for photovoltaic testing 1000VDC/265VAC (single-phase)
Internal memory capacity 200 curves I-V 8 days@
PI = 10 min
Recording with selectable integration 5 s – 60 m
period
Detection of I-V curve on PV modules and 128 points
strings
Size in L * W * H 235 * 165 * 75 (mm)
Weight in grams (batteries included) 1200
S.R. Madeti, S.N. Singh Solar Energy 173 (2018) 139–151

Table 4
14
Electrical characteristics of Atersa calibrated cell. G=115.13 W/m2 & T=28.1oC Fixed Rs & A
G=115.13 W/m2 & T=28.1oC Varying Rs & A
Electrical features Configuration 1 Configuration 2 12
G=864.50 W/m2 & T=46.0oC Fixed Rs & A
Voltage – radiation 65 mV dc 100 mV dc 10 G=864.50 W/m2 & T=46.0oC Varying Rs & A
correspondence Per output = 1000 W/m2 Per output = 1000 W/m2
Measurement intrinsic ± 0.1% ±0.2%
8

Error %
error
Measurement error of the ± 2.0% ±2.0%
Reference Pattern 6
Impedance connected to ≥10 M ohms
the output 4

2
Table 5
Specifications of temperature sensor. 0
0 5 10 15 20 25 30 35 40
Feature PT300N
Voltage (V)
Sensor Element PT300N Class A as per EN 60751 Fig. 3. Comparative results between the proposed method and the one pre-
Sensor Housing Self adhesive Aluminum Block,
sented in (Schmidt et al., 2004) at G = 864.50 W/m2 and T = 46.0 °C;
35 mm × 12 mm × 6 mm
Sensor Cable Length: 3 m, PUR coated, shielded (LiHC11Y,
G = 115.13 W/m2 and T = 28.1 °C.
2 × 0.25 mm2)
Operating Temperature −40 to +90 °C
Accuracy ± 0.03 °C to ± 0.3 °C
Uncertainty ± 0.340 °C

8
G=864.50 W/m 2, T=46.0oC
Isim
Imea

6 G=623.40 W/m 2, T=41.8oC

Current (A)

4 G=424.97 W/m 2, T=36.7oC

G=244.32 W/m 2, T=31.6oC

2

G=115.13 W/m 2, T=28.1oC

0
0 5 10 15 20 25 30 35 40

Voltage (V)
Fig. 4. Features from two different classes are shown. One represented with a
Fig. 2. Simulated and measured I (V ) characteristics under different test con-
square and other with a triangle. A test data ‘x’ symbolized by a circle is to be
ditions for JJ solar PV module JP60F230.
classified (Zhao et al., 2015).

reasons to choose a kNN rule for FDC are stated as follows:

intelligent kNN rule is able to reach the global optimum efficiently.
(3) The output power from the strings can be severely affected by
(1) In PV systems, since the output power is highly dependent on cli-
shading; especially shading that occurs due to clouds, surrounding
matic conditions, it is difficult to set threshold values for protective
trees, buildings, rains, etc. In that case, the illumination on the PV
devices. Thus, a learning technique is required to set their own
panels is less than the readings of sensors (located in the inverters)
threshold values by training the system based on observed data. The
that measure irradiance. To improve this, classification with less
training data will vary depending on the electrical properties of PV
outliers is desired. The kNN rule has the advantage of being rela-
panel and climatic conditions of the installed PV system.
tively insensitive to these outliers. The readings of irradiance and
(2) k -means clustering and fuzzy C-means are likely to be trapped in a
temperature for the unilluminated PV panels are not considered in
local optimum when dealing with large and complex data. The

Table 6
Calculated Error (%) between the for model developed in Schmidt et al. (2004) and the proposed model.
Experimental irradiance and temperature values Error between model developed in Schmidt et al. (2004) and measured Error between proposed model and measured data
data (%) (%)

G = 864.50 W/m2; T = 46.0 °C 1.853 0.613

G = 623.40 W/m2; T = 41.8 °C 2.156 1.522
G = 424.97 W/m2; T = 36.7 °C 5.894 3.425
G = 244.32 W/m2; T = 31.6 °C 8.775 5.845
G = 115.13 W/m2; T = 28.1 °C 11.884 6.413

145
S.R. Madeti, S.N. Singh Solar Energy 173 (2018) 139–151

Irradiance sensor

Temperature sensor
5 modules/string Vmpp
Impp

Normal operation

Open circuit with bypass diode

Step up
Utility Grid
Transformer
L-L fault inter string

SBDN: Partial shading with normal bypass diode

SBDI: Partial shading with inverted bypass diode

PV module
Blocking diode
SBDF: Partial shading with faulted bypass diode
Bypass diode
Inverted bypass diode

Inverter

Fig. 5. Modeled PV system and types of faults in dc side of the string inverter system PV array.

training data set.
This technique uses irradiance (W/m2), module temperature (°C),
Vmpp , Impp and Pmpp data signals that convey the information about the
type of fault in order to take the appropriate action.
The proposed kNN rule based FDC technique relies on the trajectory
of a normal sample, which is similar to the trajectories of normal
samples in the training data. Alternatively, the trajectory of a faulty
sample generally exhibits some deviations from the trajectories of
normal training samples. In other words, the distance of a faulty sample
with respect to neighboring training samples must be greater than the
distance of a normal sample to its nearest neighboring training samples.
If we can determine the distances of normal training samples to their
nearest neighboring training samples, we can define a threshold with
certain confidence level. The unclassified samples are considered as
normal if its distances to its nearest neighboring training samples is
below the defined threshold value. Otherwise, the sample is considered
as fault. The classifier has following features:
(1) Model input parameters: Irradiance (W/m2), module temperature

Instantaneous IMPP, VMPP,

Irradiation and Temperature
Data from Metering units at
each string

If first No
energization

Yes kNN based

Classification Start
Data logging for predefined
period of time (10 minutes
interval)
Open 20%, 40%,
60% or 80%

Data Processing
1. Simulate fault values using Start ON DELAY
TIMER SET tshade
measured values.
2. Train the kNN with the measured
normal and simulated Fault values If fault expires No
within tshade

Yes
START Fault Fault Alarm Fault Alarm Open Fault Alarm L-L Fault Alarm SBDN Fault Alarm SBDI Fault Alarm SBDF
Normal Momentary Shading circuit fault fault fault fault fault
Detection

Fig. 6. Flow chart of proposed fault detection method.

146
S.R. Madeti, S.N. Singh Solar Energy 173 (2018) 139–151

(°C), voltage (Vmpp ), current (Impp ) and power (Pmpp ) and SBDF ground faults.
(2) kNN model used: Five input vector (G, T , Vmpp, Impp, Pmpp ), six class
categories (Normal, open, L-L, SBDN, SBDI, SBDF). (a) Simulation setup

4. Fault detection and algorithm description
Fig. 6 shows the flow chart for the fault detection and classification
method. The steps for real time application are stated below.
(i) Step 1 Initiate: Upon energization of the newly erected PV plant,
the data logger starts storing the kNN input vectors in 15 min in-
terval.
(ii) Step 2 Data Preprocess: Simulator calculates the fault values for
the logged data. Then trains the kNN network using these values.
(iii) Step 3 Fault Detection Start: Initiates Fault detection using in-
stantaneous Measurement Data.
(iv) Step 4 Classification Stage: The data is classified into following
categories: Normal, Open, LL, SBDN, SBDI, SBDF. The timer starts
if an open circuit is detected. Timer is set to tshade which is the
average time a shade caused by cloud can occur in that area. If the
open circuit persists upon expiration of timer, fault is classified as
Open. If the open circuit extinguishes within period of tshade , mo-
mentary partial shade is classified.
5. Simulation results
This section presents the performance of the proposed method for
FDC in a PV system, i.e., open circuit faults, line-line faults, SBDN, SBDI
Different types of faults are simulated in a PV system constructed in
MATLAB by applying the toolbox SimPowerSystems. As illustrated in
Fig. 7 with three parallel PV arrays, the irradiance and temperature can
be adjusted on the input side, and inside dashed boxes are the optional
circuits for simulating different faults. The remaining part of Fig. 7
includes the common essential components to simulate a normal PV
system. It includes of 15 multicrystalline silicon PV modules with a
nominal power of 230 W, a DC-DC boost converter, inverter, and grid as
the output. The PV modules are organized in 3 strings and each string is
made up of 5 series connected PV modules to form the entire PV array.
Each string is connected to the inverter input stage.
(b) Fault classification
Normal and five faults are simulated in the string level GCPV
system, at different irradiance and temperature levels. The irradiance
levels are taken at intervals of 25 W/m2 from 100 to 1000 W/m2. At
each irradiance level, the temperature varies between 25 °C and 55 °C,
with a step size of 5 °C. Sample electrical characteristics of the six faults
at G = 700 W/m2, T = 50 °C are shown in Table 7. Similarly electrical
characteristics at other irradiance and temperature have also been used
and the faults are classified based on these characteristics.
The training data distribution is shown in Fig. 8. These data are the
labeled data in the testing stage. Testing and validation of classification

L-L
FAULTS

Look under mask

1000 1000 1000 1000

Irradiance By-Pass diode Irradiance By-Pass diode Irradiance By-Pass diode Irradiance By-Pass diode
25 25 25 25
Temperature Temperature Temperature Temperature
PV Cells PV Cells PV Cells PV Cells
1-20 1-20 1-20 1-20

1000 600 600 600

Irradiance1 By-Pass diode Irradiance1 By-Pass diode Irradiance1 By-Pass diode Irradiance1 By-Pass diode
25 25 25 25
Temperature1 Temperature1 Temperature1 Temperature1
PV Cells PV Cells PV Cells PV Cells
21-40 21-40 21-40 21-40

1000 300 300 300

Irradiance2 Irradiance2 Irradiance2 Irradiance2
25
By-Pass diode By-Pass diode By-Pass diode By-Pass diode
25 25 25
Temperature2 Temperature2 Temperature2 Temperature2
PV Cells PV Cells PV Cells PV Cells
41-60 41-60 41-60 41-60

Fig. 7. An illustrative simulation system for faults induced by (a) L-L, (b) open circuit, (c) SBDN, (d) SBDI, (e) SBDF.

147
S.R. Madeti, S.N. Singh Solar Energy 173 (2018) 139–151

Table 7 demonstrates the one of the groups (open circuit fault) in a 2D plot,
Electrical difference between these six faults (operating conditions: G = 700 W/ where the fault is clearly distinguishable.
m2, T = 50 °C). Tables 8 and 9 shows the classification results of proposed model
Fault type Voltage (V) Current (A) Power (W) and existing model (Di Piazza and Vitale, 2013) in the form of a con-
fusion matrix. The kNN model was developed using the MATLAB
Normal 128.09 5.51 705.60 classification learner app, which classifies a test set based on pre trained
Open decrease (102.47) no change (5.51) decrease (564.63)
data and generates a confusion matrix. Thereafter, Tables 8 and 9 re-
L-L decrease (101.41) decrease (3.51) decrease (355.95)
SBDN decrease (85.41) decrease (4.51) decrease (385.20) presents the confusion matrices as obtained from the classification
SBDI decrease (107.39) decrease (3.45) decrease (370.22) learner app for the respective test cases. The numbers 1–6 represents
SBDF increase (136.12) decrease (3.28) decrease (446.28) different classes, such that: Class 1: Normal, Class 2: Open circuit, Class
3: L-L, Class 4: SBDN, Class 5: SBDI, Class 6: SBDF.
For example, the green box in the second row shows that 63 out of
Normal operation 64 test data in class 2 were correctly classified. The red boxes represent
Open circuit Fault the misclassified data or the classification error, i.e. the number of
L-L Fault
SBDN Fault output class that ended up in wrong class during classification. The
2000 SBDI Fault fourth red box from the left in the second row shows that in 0.26% of
SBDF Fault
cases, class 4 is misclassified as class 2. The percentage of absolute
1500 number of successful classification and misclassification of each class is
represented by gray box. For example, in the top right gray box, 100%
means that the all data pertaining to class 1 has been classified cor-
PMPP (W)

1000
rectly. The bottom gray box in the bottom row of second column in-
dicates that 98.44% of total target class 2 were classified correctly and
500 1.56% were misclassified. The blue box represents the mean correct
classification accuracy and it is 98.70% for this study. Only 1.3% of
1200 tested samples were misclassified.
0 1000
While in case of model proposed in Ding et al. (2012) the mean
)
/m 2

800
600 correct classification accuracy is 89.84% and 10.16% of the tested data
(W

50 400
45 were misclassified. The comparison between the results shows that the
e
nc

40
35 200
ia

30 proposed model network is more efficient for faults classification with

ad

25
0
Irr

Tempe 20 respect to the one presented in Di Piazza and Vitale (2013).

r 15
ature 0
( C)

Fig. 8. Training data distribution for varying irradiance, temperature with the 6. Conclusion
MPP power.

In this paper, a FDC technique for photovoltaic (PV ) systems based

method has been done for unlabeled data. Irradiation data of
180–930 W/m2 with increment of 50 W/m2 and temperatures of 25 to
61 °C with increments of 6 °C have been used for testing. The dis-
tribution of training, validation and testing data for varying Vmpp, Impp
and Pmpp has been shown in Fig. 9. For testing the FDC , experimental
data have been collected by conducting specific tests. The measure-
ments were recorded for every 20 min covering the period of five days;
that is, January 24 to 28, 2018 has been considered. Fig. 10
on kNN was developed. Unlike existing techniques the proposed
learning technique does not require any predefined threshold values for
fault detection, which are difficult to choose for PV systems, since
output power is highly dependent on climatic conditions. A detailed
modeling and simulation of the PV system based on experimental data
was developed which utilizes STC and NOCT remarkable points from
the manufacturer’s datasheet and builds a relationship between 'A' and
'Rs ' with module temperature. The proposed model precisely traces the

1400

1200

1000
P m p String (W)

800

600

400
200
200 180
160
0 140
)

120
(V

100
g
rin

10 80
60
St

8
6 40
p
m

4 20
V

Im p S 2 0
tring 0
(A)

Fig. 9. Training, validation and testing data distribution.

148
S.R. Madeti, S.N. Singh Solar Energy 173 (2018) 139–151

Open circuit fault training data

Open circuit fault validation data
Open circuit fault testing data
Open circuit fault experimental data

8
1000

Open circuit fault training data

Open circuit fault validation data
6 800 Open circuit fault testing data
Open circuit fault experimental data
Im p String (A)

Pm p String (W)
600
4

400

2
200

0
0
20 40 60 80 100 120
20 40 60 80 100 120
Vm p String (V)
Vm p String (V)

Fig. 10. Training, validation and testing data distribution of Open Circuit fault in 2D (a) Vmp vs. Imp, (b) Vmp vs. Pmp.

Table 8
Confusion matrix for kNN test results of proposed model, showing the correct recognition rates and mis-
classification rates. The green boxes represent the number of correctly classified data of each class. The red boxes
represent the misclassified data. The blue box represents the mean correct classification accuracy. The percentage
of absolute number of successful classification and misclassification of each class is represented by gray box.

149
S.R. Madeti, S.N. Singh
Table 9
Confusion matrix for kNN test results of model developed in Schmidt et al. (2004).
I (V ) or P (V ) characteristics of the PV system at different levels of ir-
radiance and temperature. The performance of the proposed FDC
technique was tested by using the data generated from proposed PV
model and experimental setup. The simulation results yield high clas-
sification accuracy for the tested faults, which justifies the fitness of the
proposed FDC technique.
Acknowledgement
One of the authors (SRK) acknowledges the financial support of
Ministry of New and Renewable Energy (MNRE), Government of India,
New Delhi, India in the form of scholarship. The authors also like to
acknowledge Prof. Jan Kleissl and reviewers whose comments greatly
improved this work.
References
ALQahtani, A.H., Abuhamdeh, M.S., Alsmadi, Y.M., 2012. A simplified and comprehen-
sive approach to characterize photovoltaic system performance. Energy Tech, IEEE
29, 1–6.
Askarzadeh, A., dos Santos, Coelho L., 2015. Determination of photovoltaic modules
parameters at different operating conditions using a novel bird mating optimizer
approach. Energy Convers. Manag. 89, 608–614.
Askarzadeh, A., Rezazadeh, A., 2012. Parameter identification for solar cell models using
harmony search-based algorithms. Sol. Energy 86 (11), 3241–3249.
Askarzadeh, A., Rezazadeh, A., 2013. Artificial bee swarm optimization algorithm for
parameters identification of solar cell models. Appl. Energy 102, 943–949.
Bastidas-Rodriguez, J., Petrone, G., Ramos-Paja, C., Spagnuolo, G., 2017. A genetic al-
gorithm for identifying the single diode model parameters of a photovoltaic panel.
Math. Comput. Simul.
Boztepe, M., et al., 2014. Global MPPT scheme for photovoltaic string inverters based on
restricted voltage window search algorithm. IEEE Trans. Ind. Electron. 61 (7),
3302–3312.
Chatterjee, A., Keyhani, A., Kapoor, D., 2011. Identification of photovoltaic source
models. IEEE Trans. Energy Convers. 26 (3), 883–889.
Chenni, R., Makhlouf, M., Kerbache, T., Bouzid, A., 2007. A detailed modeling method for
photovoltaic cells. Energy 32 (9), 1724–1730.
Chouder, A., Silvestre, S., 2010. Automatic supervision and fault detection of PV systems
based on power losses analysis. Energy Convers. Manage. 51 (10), 1929–1937.
da Costa, W.T., Fardin, J.F., Simonetti, D.S., de VBM Neto, L., 2010. Identification of
photovoltaic model parameters by differential evolution. Ind. Technol. (ICIT)
931–936.
Di Piazza, M.C., Vitale, G., 2013. Parameter identification for photovoltaic source models.
Springer Photovoltaic Sources 83–129.
Ding, K., Bian, X., Liu, H., Peng, T., 2012. A MATLAB-Simulink-based PV module model
150
Solar Energy 173 (2018) 139–151
and its application under conditions of nonuniform irradiance. IEEE Trans. Energy
Convers. 27 (4), 864–872.
Drews, A., De Keizer, A.C., Beyer, H.G., Lorenz, E., Betcke, J., Van Sark, W.G.,
Heydenreich, W., Wiemken, E., Stettler, S., Toggweiler, P., Bofinger, S., 2007.
Monitoring and remote failure detection of grid-connected PV systems based on sa-
tellite observations. Sol. Energy 81 (4), 548–564.
Duda, R.O., Hart, P.E., Stork, D.G., 2001. Pattern Classification, second ed. Wiley, New
York.
El-Naggar, K., AlRashidi, M., AlHajri, M., Al-Othman, A., 2012. Simulated annealing al-
gorithm for photovoltaic parameters identification. Sol. Energy 86 (1), 266–274.
Firth, Steven K., Lomas, Kevin J., Rees, Simon J., 2010. A simple model of PV system
performance and its use in fault detection. Sol. Energy 84 (4), 624–635.
Guo, L., Meng, Z., Sun, Y., Wang, L., 2016. Parameter identification and sensitivity
analysis of solar cell models with cat swarm optimization algorithm. Energy Convers.
Manag. 108, 520–528.
Han, W., Wang, H.-H., Chen, L., 2014. Parameters identification for photovoltaic module
based on an improved artificial fish swarm algorithm. Sci. World J.
He, Q.P., Wang, J., 2007. Fault detection using the k-nearest neighbor rule for semi-
conductor manufacturing processes. IEEE Trans. Semicond. Manuf. 20 (4), 345–354.
http://niwe.res.in:8080/NIWE_WRA_DATA/DataTable_D4.jsf.
Jacob, B., Balasubramanian, K., Azharuddin, S.M., Rajasekar, N., 2015. Solar pv model-
ling and parameter extraction using artificial immune system. Energy Proc. 75,
331–336.
Jiang, L.L., Maskell, D.L., Patra, J.C., 2013. A novel ant colony optimization-based
maximum power point tracking for photovoltaic systems under partially shaded
conditions. Energy Build. 58, 227–236.
Karatepe, E., Boztepe, M., Colak, M., 2006. Neural network based solar cell model. Energy
Convers. Manag. 47 (9), 1159–1178.
Madeti, S.R., Singh, S.N., 2017. Monitoring system for photovoltaic plants: a review.
Renew. Sustain. Energy Rev. 67, 1180–1207.
Madeti, S.R., Singh, S.N., 2017. A comprehensive study on different types of faults and
detection techniques for solar photovoltaic system. Sol. Energy 158, 161–185.
Mercure, J.-F., Salas, P., 2012. An assessment of global energy resource economic po-
tential. Energy 46, 322–336.
Oliva, D., Cuevas, E., Pajares, G., 2014. Parameter identification of solar cells using ar-
tificial bee colony optimization. Energy 72, 93–102.
Ortiz-Rivera, E.I., Peng, F.Z., 2005. Analytical model for a photovoltaic module using the
electrical characteristics provided by the manufacturer data sheet. Power Electron.
Specialists Conf. 2087–2091.
Petrone, G., Spagnuolo, G., Vitelli, M., 2007. Analytical model of mismatched photo-
voltaic fields by means of Lambert W-function. Sol. Energy Mater. Sol. Cells 91 (18),
1652–1657.
Polo, F.A., del Rosario, J.J., García, G.C., 2010. Supervisory control and automatic failure
detection in grid-connected photovoltaic systems. In: International Conference on
Industrial, Engineering and Other Applications of Applied Intelligent Systems, pp.
458–467.
PVPS, Iea, 2015. Performance and reliability of photovoltaic systems. Report IEA PVPS
Task 13.
Salem, F., Awadallah, M.A., 2014. Parameters estimation of photovoltaic modules:
comparison of ann and anfis. Int. J. Ind. Electron Drives 1 (2), 121–129.
S.R. Madeti, S.N. Singh
Salmi, T., Bouzguenda, M., Gastli, A., Masmoudi, A., 2012. Matlab/simulink based
modeling of photovoltaic cell. Int. J. Renew. Energy Res. (IJRER) 2 (2), 213–218.
Saravanan, C., Panneerselvam, M., 2013. A comprehensive analysis for extracting single
diode pv model parameters by hybrid ga-pso algorithm. Int. J. Comput. Appl. 78 (8).
Schmidt, C., Bartl, S., Speil, M., Straer, J., Ernst, G., Spitzlsperger, G., 2004. Fault de-
tection and classification (FDC) for a via-etching-process,” In: Proc. 5th Eur. AEC/
APC Conf., Dresden, Germany, 2004.
Solórzano, J., Egido, M.A., 2013. Automatic fault diagnosis in PV systems with distributed
MPPT. Energy Convers. Manage. 1 (76), 925–934.
Solórzano, J., Egido, M.A., 2013. Automatic fault diagnosis in PV systems with distributed
MPPT. Energy Convers. Manage. 76, 925–934.
Villalva, M.G., Gazoli, J.R., Ruppert, Filho E., 2009. Comprehensive approach to mod-
eling and simulation of photovoltaic arrays. IEEE Trans. Power Electron. 24 (5),
1198–1208.
Walker, G., 2001. Evaluating MPPT converter topologies using a MATLAB PV model. J.
Electr. Electron. Eng., Australia 21 (1), 49.
Xiao, W., Dunford, W.G., Capel, A., 2004. A novel modeling method for photovoltaic cells.
151
Solar Energy 173 (2018) 139–151
Power Electron. Specialists Conf., IEEE 3, 1950–1956.
Zagrouba, M., Sellami, A., Bouaicha, M., Ksouri, M., 2010. Identification of pv solar cells
and modules parameters using the genetic algorithms: application to maximum
power extraction. Sol. Energy 84 (5), 860–866.
Zdravkovic, M., Vasic, A., Dolicanin, C., Stankovic, K., Osmokrovic, P., 2009.
Temperature effects on photovoltaic components characteristics. Ser. A: Appl. Math.
Inform. Mech. 1 (1), 29–36.
Zhao, Y., De Palma, J.F., Mosesian, J., Lyons, R., Lehman, B., 2013. Line–line fault ana-
lysis and protection challenges in solar photovoltaic arrays. IEEE Trans. Ind. Electron.
60 (9), 3784–3795.
Zhao, Y., de Palma, J.F., Mosesian, J., Lyons, R., Lehman, B., 2013. Line–line fault ana-
lysis and protection challenges in solar photovoltaic arrays. IEEE Trans. Ind. Electron.
60 (9), 3784–3795.
Zhao, Y., Ball, R., Mosesian, J., de Palma, J.F., Lehman, B., 2015. Graph-based semi-
supervised learning for fault detection and classification in solar photovoltaic arrays.
IEEE Trans. Power Electron. 30 (5), 2848–2858.