Download as pdf or txt
Download as pdf or txt
You are on page 1of 19

Solar Energy 201 (2020) 279–297

Contents lists available at ScienceDirect

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

A fault locating method for PV arrays based on improved voltage sensor T


placement

Tingting Peia,c, , Jiangfeng Zhangb, Li Lic, Xiaohong Haoa
a
College of Electrical and Information Engineering, Key Laboratory of Gansu Advanced Control for Industrial Processes, National Demonstration Center for Experimental
Electrical and Control Engineering Education, Lanzhou University of Technology, Lanzhou 730050, China
b
Department of Automotive Engineering, Clemson University, SC 29607, USA
c
School of Electrical and Data Engineering, University of Technology Sydney, NSW 2007, Australia

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

Keywords: Solar photovoltaic (PV) power generation has been widely used because of its environment-friendly advantages.
Photovoltaic arrays However, operational faults in PV arrays have always been one of the critical factors affecting the PV system’s
Voltage sensors power-generation efficiency and life cycle. This paper proposes a fault locating strategy, which does not need the
Fault locating current sensor in every PV string and can also minimize the number of voltage sensors between strings, to
Open circuit and short circuit faults
accurately locate the faulty PV modules in a PV array. A variety of faults including open circuit, short circuit,
Degradation and shading faults
degradation and partial shading faults are considered, and a universal method is proposed in this paper to locate
the faulty modules under these faults. In this method, the voltage sensors are deployed through differentiating
the parity of the number of strings for the PV array, while the fault locating rules are formulated under each of
the afore-mentioned faults, respectively. Compared with existing fault locating methods, the proposed locating
technique is shown to be effective in the application to PV arrays with any size and capacity, and it can accu-
rately locate each faulty module for those faults, especially for degradation and partial shading faults. Thus, it is
helpful in PV array dynamic reconfiguration and maintenance cost reduction.

1. Introduction economically eliminate the faults, a prerequisite is to locate the faulty


PV modules. Therefore, based on the above considerations, it is parti-
Due to the continuous increase in energy demand and the rapid cularly important to develop a fault locating strategy with higher ac-
depletion of fossil fuels, solar photovoltaic (PV) power generation is curacy, lower computational complexity and minimum additional
recognized as one of the best solutions to extract energy from the nat- equipment (Sevilla-Camacho et al., 2019).
ural environment. Energy harvested from PVs will occupy 70% of the Many fault detections and locating techniques have been developed
total electrical energy yield by the year 2050 (Dincer, 2011). The in- to identify the faults in the DC side of a PV system. Existing fuses and
stalled PV power capacities will be more than 4 TW by 2025 and 21.9 insulation monitoring devices are applied to identify the ground faults
TW by 2050, and solar PV energy will continue to grow faster than the of PV arrays (Flicker and Johnson, 2013, Alam et al., 2015). Over-
overall energy demand in the coming years (Jäger-Waldau, 2019). PV current protecting devices can isolate faults if the PV system carries a
power generation can be considered as the most essential and pre- large fault current (Zhao et al., 2012). However, the accuracy of these
requisite source of electricity owing to the subsidies offered by gov- methods is not satisfactory and the results are easily affected by the
ernments and technological advancements that have reduced the cost leakage current. The thermal infrared detection method can distinguish
per watt appreciably (Pandey et al., 2016). However, grid-connected PV and locate the modules with abnormal surface temperature caused by
(GCPV) plants often work under complicated and changeable climate faults. In (Wang and Zheng, 2010), a method based on infrared image
conditions which may cause potential PV panel faults. In particular, analysis is presented that can automatically identify the working status
faults in the direct current (DC) side of PV systems are unavoidable, of PV arrays by combining a fuzzy reasoning technique. By utilizing the
which can seriously affect system efficiency and productivity as well as infrared images of the PV modules, faults such as cracks, broken grids
the entire plant’s security and reliability. In order to timely and and black spots of the modules can be differentiated and located (Nian


Corresponding author at: College of Electrical and Information Engineering, Key Laboratory of Gansu Advanced Control for Industrial Processes, National
Demonstration Center for Experimental Electrical and Control Engineering Education, Lanzhou University of Technology, Lanzhou 730050, China.
E-mail address: peitt52@163.com (T. Pei).

https://doi.org/10.1016/j.solener.2020.03.019
Received 4 January 2020; Received in revised form 4 March 2020; Accepted 5 March 2020
Available online 10 March 2020
0038-092X/ © 2020 International Solar Energy Society. Published by Elsevier Ltd. All rights reserved.
T. Pei, et al. Solar Energy 201 (2020) 279–297

et al., 2010). The high costs are the main limitations to apply the in-
frared images analysis method to fault detection. The time-domain re-
flectometry method detects degradation faults and locates the positions
of the corresponding faulty modules (Takashima et al., 2009), and the
earth capacitance measurement method can locate the disconnection
position in faulty strings (Takashima et al., 2008). However, both
strategies can only be operated in offline modes.
In practice, online fault detecting methods with the lowest im-
plementation cost are urgently desired. For this purpose, the power loss
analysis approach is reported in (Chouder and Silvestre, 2009) to esti-
mate the mismatch effect on the entire PV array. The method about the
current-voltage (I − V ) characteristics analysis is described in (Bressan
et al., 2016), which compares the I − V curve under fault-free and
partial shading conditions to discriminate abnormal problems caused
by shading faults. A threshold-based strategy involving the evaluation
of current and voltage indicators is proposed to perform online fault
detection for grid-connected PV systems in (Yahyaoui and Segatto,
2017). A mathematical model-based method is presented by the com-
parison of predicted and measured values to distinguish the faults of PV
arrays in (Platon et al., 2015). Besides, some intelligent algorithms like
fuzzy logic and artificial neural networks are utilized to detect and
classify one or more faulty modules, and partial shading faults (Dhimish Fig. 1. Configuration of a PV array.
et al., 2017, Dhimish et al., 2018). Furthermore, a two-stage support
vector machine is used to recognize line-to-line faults of PV arrays 2. Configuration and fault characteristics of PV arrays
under low-illumination conditions (Yi and Etemadi, 2017). In (Harrou
et al., 2018), statistical monitoring approaches are applied to identify The configuration of an m × n PV array is presented in Fig. 1 where
the short-circuit, open-circuit and shading faults. However, the above the series diode within each string is to prevent reverse current flowing
mentioned online methods can only detect that the whole PV array has in the strings, m is the number of modules in a string and n is the
faults, but cannot accurately locate the faulty strings or modules. number of strings. In practice, a module is formed by PV cells connected
In order to solve the fault locating challenges of PV arrays, the in series and then in parallel with a bypass diode (Wu et al., 2013). A
terminal current-voltage curves of PV arrays are divided into high- string is composed of the modules (see PV1,1~P Vn, m in Fig. 1) connected
voltage and low-voltage sections to locate the faulty modules (Hu et al., in series; then, all the strings are connected in parallel to form the PV
2015). On this basis, a cost-effectiveness analysis of the proposed fault array.
locating technique is carried out (Madeti and Singh, 2017). However, Common faults occurring at the DC side of PV systems have been
the voltage detecting rules are only suitable for the configuration widely studied (Mansouri et al., 2018). Fig. 2 shows the short circuit,
structure of 3× 3 PV arrays. A voltage-based protection scheme that can open circuit, degradation and partial shading faults in a 3× 2 PV array.
recognize and locate short-circuit, open-circuit and shading faults in Open circuit faults may be caused by a broken wire between the PV
utility-scale PV arrays is presented in (Saleh et al., 2017). In (Chen and cells, as denoted by ‘FO ’. Short circuit faults are mainly due to bad
Wang, 2018), a fault locating algorithm is developed based on a gen- wiring in strings, as represented by ‘FS ’. Additionally, abrasion, vibra-
eralized local likelihood ratio test, which can capture arc and short- tion and aging of modules are also critical sources to bring about short-
circuit faults at each string. Due to technical levels and environmental circuit faults (Lu et al., 2019). Degradation faults are often caused by
factors, the energy conversion efficiency of PV arrays is still not sa- some tiny cracks on PV cells, the delamination phenomenon resulting
tisfactory. In this situation, faults occurring in any modules of PV arrays from bond failures between different layers of a module and frequent
will lead to less output power from the PV system. Under practical temperature changes of modules with internal series resistance in-
conditions, faults in the DC side such as open circuit, short circuit, creasing denoted by ‘FD’ (Huang et al., 2019). Partial shading faults will
degradation or partial shading faults are most possible (Garoudja et al., lead to uneven illumination on PV arrays represented by ‘FP ’ (Chen
2017). However, under these four common faults, the locating method et al., 2019).
which can point out each faulty module has not been reported. In order The output characteristics of the PV array under common fault
to accurately and timely locate any faulty modules under the mentioned conditions are shown in Fig. 3. The output voltage of the PV array at the
faults, this paper presents a fault locating method based on an improved maximum power point (MPP) under a short circuit or degradation fault
voltage sensor placement method. Different locating criterions of the is far less than its fault-free voltage at the MPP, which is because the
corresponding faults are designed according to an odd or even number voltage of the short-circuited module is zero, while the voltage of the
of strings in the PV array. The proposed method can acquire the degraded module is lower than its fault-free voltage. For an open circuit
terminal characteristic of each module so that it can accurately locate fault, the voltage of the module in the faulty string is equal to its open
any faulty modules, and is applicable to the PV arrays with any scales circuit voltage, but the voltages of the healthy strings are equal to those
and capacities. Simulation tests are performed to validate the effec- under the fault-free condition. For a partial shading fault, the voltage of
tiveness of the proposed method under various fault scenarios. the faulty module is between zero and its open circuit voltage. Hence,
This paper is organized as follows. Section 2 introduces the PV array the voltages of the modules at the MPP are selected as fault char-
configuration and fault characteristics. In Section 3, an improved vol- acteristic quantities to implement PV array fault locating methods in
tage sensor placement method and the mathematical mechanism of this paper.
fault locating are proposed. Section 4 completes the design of the fault
location strategy for PV arrays. Simulation results and discussions are
provided in Section 5. Finally, Section 6 presents the main conclusions 3. Improved voltage sensor placement and mathematical
along with suggestions for future works. mechanism of fault locating

In order to identify fault locations in a PV array, the voltages of a

280
T. Pei, et al. Solar Energy 201 (2020) 279–297

Fig. 2. PV array under different fault conditions.

large number of PV modules need to be collected. In this paper, an


improved voltage sensor deployment is presented to utilize fewer vol-
tage sensors, and it is helpful to reduce both wiring complexity and
maintenance cost.

3.1. Improved voltage sensor placement method

The improved sensor placement method is presented according to


the following guidelines:

(a) The connection wire between the adjacent modules in a string is


regarded as a node;
(b) All the nodes must be connected by voltage sensors;
(c) In a PV string, a node is connected to as few voltage sensors as
possible.
(d) A voltage sensor must be connected to different isoelectric points
from two strings.

The sensor placement method is different if the parity of the number


of strings is different. For an m × n PV array, when n is even, it can be
divided into n 2 blocks of m × 2 sub-arrays connected in parallel. The
sensor placement is illustrated in Fig. 4. Each sub-array needs m − 1 Fig. 4. Voltage sensor placement method for even n.
voltage sensors. Hence, the whole PV array needs n (m − 1) 2 voltage
sensors, which is proven the minimum number of voltage sensors
needed (Hu et al., 2015).
When n is odd, the PV array can be divided as one m × 3 sub-array

Fig. 3. The output characteristics of the PV array under common fault conditions. (a) Power-voltage curves; (b) Current-voltage curves.

281
T. Pei, et al. Solar Energy 201 (2020) 279–297

Fig. 5. Voltage sensor placement method for odd n.

connected in parallel with (n − 3) 2 blocks of m × 2 sub-arrays, where ⎧VPV _pq = VPV − (m − p) Voc , a fault occurs between (1, q) and (p , q)
it is assumed n ⩾ 3. Fig. 5 shows the sensor placement for this case. The ⎨
⎩VPV _pq = pVoc , a fault occurs between (p + 1, q) and (m , q)
(n − 3) 2 blocks of m × 2 sub-arrays need to use (n − 3)(m − 1) 2
voltage sensors, which follows the same reasons as the case that n is (2)
even. For the m × 3 sub-array, the sensors V1,2j − 1 are placed between where Voc is the open circuit voltage of a PV module.
strings 1 and 2, for j = 1, 2, ...,m − 2 ; the sensors V1,2j are placed be- Under degradation faults, the terminal voltage of a degraded
tween strings 2 and 3, for j = 1, 2, ...,m − 2 ; the sensor V1,2(m − 1) − 1 is module is higher than zero and lower than Vmax . When the q-th string
placed between strings 1 and 3; thus the m × 3 sub-array needs has a degradation fault, VPV _pq is given as,
2(m − 1) − 1 sensors. Hence, (n + 1)(m − 1) 2 − 1 voltage sensors are p−1 p
applied to the PV array. It is worth noting that at least one node is ⎧ m − 1 VPV < VPV _pq < V ,
m PV
a fault occurs between (1, q) and (p , q)
covered by two different sensors when n is odd. ⎨ p VPV < VPV _pq <
p
V , a fault occurs between (p + 1, q) and (m , q)
⎩m m − 1 PV

(3)
3.2. Mathematical mechanism of the proposed fault locating method Under partial shading faults, the terminal voltage of a shaded
module is between zero and Voc . If a partial shading fault happens in the
The corresponding faulty modules can be located by acquiring the q-th string, VPV _pq is described by,
voltage change patterns of the layout sensors under open circuit, short
p−1
circuit, degradation and shading faults. In order to locate each faulty ⎧ m − 1 VPV < VPV _pq < pVoc, a fault occurs between (1, q) and (p , q)
module, it is assumed that there is at most one type of faults occurring ⎪ p
V < pVoc and VPV _pq < m − 1 VPV ,
in any PV module at any considered time instant. To facilitate the ⎨ PV _pq
⎪ a fault occurs between (p + 1, q) and (m, q)
discussion for the above m × n PV array, we introduce the location ⎩ (4)
coordinate ( p , q ) for the module sitting in the p-th row of the q-th
Given the voltage value of VPV _kl similar to VPV _pq , the reading of the
string.
voltage sensor connecting module ( p , q ) and module (k , l ), denoted by
According to the characteristics of short circuit faults, the terminal
Vkl, pq , can be calculated as follows,
voltage of a short-circuited module is equal to zero. When the q-th
string has a short circuit fault, the voltage from the top of module ( p , q ) Vkl, pq = VPV _kl − VPV _pq (5)
to the bottom of module (1, q ), denoted by VPV _pq , can be calculated as
follows,
4. Proposed fault locating strategy of PV arrays
p−1
⎧VPV _pq = V ,
m − 1 PV
a fault occurs between (1, q) and (p , q)
The fault locating method of a PV array is developed according to
⎨VPV _pq = p
V ,
m − 1 PV
a fault occurs between (p + 1, q) and (m , q) (1) the types of faults as detailed below, where it is assumed that only one

type of faults happens in the PV array and the total PV system performs
where VPV = mVmax is the total voltage of the PV array, and Vmax is the maximum power point tracking (MPPT) strategy at the considered time
module voltage at the MPP. instant.
According to the characteristics of open circuit faults, the terminal
voltage of an open-circuited module is equal to its open circuit voltage. 4.1. Open circuit faults
If an open circuit fault occurs in the q-th string, the voltage VPV _pq can
be defined by, For an m × n PV array, assume that both the q-th string and the l-th

282
T. Pei, et al. Solar Energy 201 (2020) 279–297

string belong to a sub-array, i.e., l = q + 1 or l = q + 2 , while the q-th the m × 3 sub-array 1 as follows, and Table 2 shows the voltage rules
string is open-circuited and the l-th string is fault-free. Then the value of for this case.
Vkl, pq is calculated by, In case that string 1 has an open circuit fault (represented as “100”
in Table 2), the sensor values of V1,2i − 1 are (i + 1) VPV m − iVoc for
Vkl, pq = VPV _kl − VPV _pq i = 1, 2, ...,m − 2 , and V1,2(m − 1) − 1 is VPV m − (m − 1) Voc , which presents
k
⎧ m VPV − (VPV − (m − p) Voc ), similar rules as the case “10” in Table 1; then the values of V1,2i are
⎪ equals to VPV m for i = 1, 2, ...,m − 2. Similar open circuit locating rules
⎪ a fault occurs between (1, q) and (p, q) can be observed for those cases “010” and “001” in Table 2.
= k
⎨ VPV − pVoc, a fault occsurs between (p + 1, q) When both string 1 and string 2 have open circuit faults, which is
⎪m
⎪ and (m , q) denoted as “110” in Table 2, the readings of V1,2i , i = 1, 2, ...,m − 2 , and
⎩ (6) V1,2(m − 1) − 1 is similar to the case “10” in Table 1, and the readings of
If both the l-th string and the q-th string are open-circuited, Vkl, pq is V1,2i − 1, i = 1, 2, ...,m − 2 , are equal to Voc . For those cases “101” and
calculated by, “011”, similar rules can be observed. All strings in sub-array 1 have
simultaneously open circuit faults, denoted as “111”, the readings from
Vkl, pq = VPV _kl − VPV _pq V11 to V1,2(m − 1) − 2 are Voc and V1,2(m − 1) −1 is − (m − 2) Voc .
⎧ kVoc − pVoc , faults occur in (1, q) to (p , q) and (1, l)
⎪ 4.2. Short circuit faults
⎪ to (k , l)

⎪ kVoc − pVoc , faults occur in (p + 1, q) to (m , q) If a short circuit fault happens in the q-th string while the l-th string
⎪ and (k + 1, l) to (m , l) is fault-free, the value of Vkl, pq can be calculated as follows,
=
⎨ kVoc − (VPV − (m − p) Voc ),
⎪ Vkl, pq = VPV _kl − VPV _pq
⎪ faults occur in (1, q) to (p , q) and (k + 1, l) to (m , l) k p−1
⎪ ⎧ m VPV − m − 1 VPV , a fault occurs between (1, q) and (p , q)
⎪ (VPV − (m − k ) Voc ) − pVoc , ⎪
⎪ = k VPV − p VPV , a fault occurs between (p + 1, q)
⎨m
⎩ faults occur in (p + 1, q) to (m , q) and (1, l) to (k , l)
m−1
(7)
⎪ and (m , q)
⎩ (8)
Due to space constraints, the following discusses only open circuit
faults occurring at the top of strings. The other types of open circuit The short circuit fault locating rules will be discussed based on (8)
faults are similar, and thus is omitted. The open circuit locating as follows.
methods based on (6) and (7) can be given as follows.
(1) n is even
(1) n is even
As mentioned in Section 4.1, we only need to consider the fault
The PV array can be divided into n 2 blocks of m × 2 sub-arrays and locating method for sub-array 1.
each sub-array possesses similar sensor deployment in Fig. 4. Therefore, Table 3 shows the short circuit fault locating rules of the modules of
we only need to consider the open circuit fault locating method for sub- string 1. When a short circuit fault happens in the module PV1,1, the
array 1. Table 1 shows the voltage rules for open circuit faults. reading of V1,1 is equal to 2VPV m , V1, m −1 is (1 m − (m − 2) (m − 1)) VPV ;
If string 1 has an open circuit fault and string 2 is healthy, this case and the other corresponding sensor readings exhibit an arithmetic
is denoted as “10” in Table 1. The voltage readings from V1,1 to V1, m −2 progression with a common difference (1 m − 1 (m − 1)) VPV . If a short
are an arithmetic progression with VPV m − Voc as its common differ- circuit fault happens in the module PV1, i , i = 2, ...,m − 2 , the voltage
ence, and the voltage value of V1, m −1 is equal to VPV m − (m − 1) Voc . If difference between V1, i − 1 and V1, i is equal to VPV m , and the other sensor
string 1 is healthy and string 2 has an open circuit fault, this case is readings present an arithmetic progression with a common difference
denoted as “01”. The readings from V1,1 to V1, m −2 are also an arithmetic (1 m − 1 (m − 1)) VPV , while V1, m −1 is still (1 m − (m − 2) (m − 1)) VPV .
progression with Voc − VPV m as its common difference, and the voltage It is worth emphasizing that if a short circuit fault happens in the
value of V1, m −1 is Voc − (m − 1) VPV m . If both string 1 and string 2 have module PV1, m − 1 or PV1, m , the reading of V1, m −1 is equal to
open circuit faults, denoted as “11”, the readings from V1,1 to V1, m −2 are (1 m − (m − 2) (m − 1)) VPV or (1 m − 1) VPV , respectively, and other
all Voc , and V1, m −1 is − (m − 2) Voc . The case “00” represents the scenario corresponding sensor readings are an arithmetic progression with a
that both strings are healthy. common difference (1 m − 1 (m − 1)) VPV . Table 4 shows the short
circuit fault locating rules of the modules of string 2. Similar rules on
(2) n is odd arithmetic progression can be observed.

Assuming n ⩾ 3, the PV array in Fig. 5 can be divided as one block (2) n is odd
of m × 3 sub-array, and (n − 3) 2 blocks of m × 2 sub-arrays whose
sensor placement methods are similar to the sub-arrays of a PV array Similar to Section 4.1, we only need to consider the short circuit
with an even n. Hence, the fault locating strategy is only formulated for locating strategy for the m × 3 sub-array 1. Tables 5–7 show the

Table 1
Voltage rules for open circuit faults in sub-array 1 (n is even).
String1 ~ String2 V1,1 V1,2 V1,3 … V1, m −2 V1, m −1

00 1
V
1
V
1
V … 1
V −
m−2
VPV
m PV m PV m PV m PV m
10 2
VPV − Voc
3
VPV − 2Voc
4
VPV − 3Voc … m−1
VPV − (m − 2) Voc
1
V − (m − 1) Voc
m m m m m PV
01 2Voc −
1
V 3Voc −
2
V 4Voc −
3
V … (m − 1) Voc −
m−2
VPV Voc −
m−1
VPV
m PV m PV m PV m m
11 Voc Voc Voc … Voc − (m − 2) Voc

283
T. Pei, et al. Solar Energy 201 (2020) 279–297

Table 2
Voltage rules for open circuit faults in sub-array 1 (n is odd).
String1 ~ String3 V1,1 V1,2 … V1,2(m − 1) − 3 V1,2(m−1) − 2 V1,2(m − 1) −1

000 1
V
1
V … 1
V
1
V −
m−2
VPV
m PV m PV m PV m PV m
100 2
VPV − Voc
1
V … m−1
VPV − (m − 2) Voc
1
V
1
VPV − (m − 1) Voc
m m PV m m PV m
010 2Voc −
1
V
2
VPV − Voc … (m − 1) Voc −
m−2
VPV
m−1
VPV − (m − 2) Voc −
m −2
VPV
m PV m m m m
001 1
V 2Voc −
1
V … 1
V (m − 1) Voc −
(m − 2)
VPV Voc −
m−1
VPV
m PV m PV m PV m m
110 Voc 2
V − Voc … Voc m−1
VPV − (m − 2) Voc
1
VPV − (m − 1) Voc
m PV m m
101 2
V − Voc 2Voc −
1
V … m−1
VPV − (m − 2) Voc (m − 1) Voc −
m−2
VPV − (m − 2) Voc
m PV m PV m m
011 2Voc −
1
V Voc … (m − 1) Voc −
m−2
VPV Voc Voc −
m−1
VPV
m PV m m
111 Voc Voc … Voc Voc − (m − 2) Voc

corresponding sensor readings for the cases that the short circuit fault degradation fault, see the details in Table 8.
happens in strings 1, 2 and 3, respectively. Again, the arithmetic pro- As illustrated in Table 9, when a degradation fault occurs in the
gressions within the readings is observed. module PV2,1 of string 2, the voltage readings from V1,1 to V1, m −1 are
bigger than the corresponding sensor values under short circuit faults,
but smaller than the corresponding fault-free values. When any module
4.3. Degradation faults
PV2, j has a degradation fault, 2 ⩽ j ⩽ m , the values of V1, m − 1 and those
from V1,1 to V1, j − 2 are higher than the corresponding fault-free readings
The majority of degradation faults may cause an increase in the
and lower than those readings under short circuit faults. The readings
series equivalent resistances of the faulty modules. However, severe
from V1, j − 1 to V1, m −2 maintain similar rules as the corresponding sensors
degradation faults can result in short circuits within the faulty modules.
under PV2,1 degradation fault.
In this paper, we consider only the case that degradation faults cause
the increase of the series resistance in the faulty modules. If the q-th
(2) n is odd
string has a degradation fault and the l-th PV string is fault-free, Vkl, pq is
obtained by,
Similarly, it suffices to consider the degradation fault locating
strategy for the m × 3 sub-array 1.
⎧Vkl, pq = VPV _kl − VPV _pq
⎪V k The corresponding range of the degradation faults in strings 1, 2 and
⎪ PV _kl = m VPV 3 of sub-array 1 are listed in Tables 10–12, respectively. In case that a
⎨ p − 1 VPV < VPV _pq < p VPV , a fault occurs between (1, q) and (p , q) degradation fault occurs in string 1, the readings of V1,2i are all equal to
⎪ mp − 1 m
p
⎪ m VPV < VPV _pq < m − 1 VPV , a fault occurs between (p + 1, q) and (m , q) VPV m , for i = 1, 2, ...,m − 2 , and V1,2i − 1, i = 1, 2, ...,m − 1, are similar to
⎩ the case in Table 8. In case that string 2 has a degradation fault, the
(9) readings of V1,2i − 1 are similar to the case in Table 9 for i = 1, 2, ...,m − 2,
The degradation faults locating rules based on (9) can be listed as and V1,2i are similar to the case in Table 8 for i = 1, 2, ...,m − 2 , while
follows. the reading of V1,2(m − 1) − 1 is − (m − 2) VPV m . In case that a degradation
fault occurs in string 3, V1,2i − 1, i = 1, 2, ...,m − 2, are equal to VPV m ,
and V1,2i , i = 1, 2, ...,m − 2, and V1,2(m − 1) − 1 are similar to those cases in
(1) n is even
Table 9.
Similarly, we only need to consider the degradation fault locating
method for sub-array 1. 4.4. Partial shading faults
If a degradation fault occurs in the module PV1,1 of string 1, the
voltage values from V1,1 to V1, m −1 are bigger than the corresponding In general, when a module is exposed to partial shading conditions,
sensor readings under fault-free conditions and smaller than those the module will receive non-uniform irradiance. Under extreme climate
readings under short circuit faults. If any module PV1, j has a degradation conditions, the received irradiance of the faulty module in a string may
fault, 2 ⩽ j ⩽ m , the values from V1,1 to V1, j − 1 are higher than the cor- be zero, showing an open circuit. This will further lead to the dis-
responding sensor readings under short circuit faults, and also lower connection of the overall string since it cannot reach the PV array’s
than those readings under fault-free condition. The values from V1, j to required voltage even though the other healthy modules in this string
V1, m −1 still keep the same patterns as the corresponding sensors of PV1,1 can work properly. In this paper, we focus only on the situation that the

Table 3
Voltage rules for short circuit faults in string 1 (n is even).
PV1,1 ~PV1, m V1,1 V1,2 V1,3 … V1, m −3 V1, m −2 V1, m −1

PV1,1 2 …
V
m PV ( 3
m

1
m−1 )VPV ( 4
m

2
m−1 )V
PV ( m−2
m

m−4
m−1 )V
PV ( m−1
m

m−3
m−1 )V
PV ( 1
m

m−2
m−1 )V
PV

PV1,2 …
( 2
m

1
m−1 )VPV ( 3
m

1
m−1 )VPV ( 4
m

2
m−1 )V
PV ( m−2
m

m−4
m−1 )V
PV ( m−1
m

m−3
m−1 )V
PV ( 1
m

m−2
m−1 )V
PV

… … … … … … … …
PV1, m − 2 …
( 2
m

1
m−1 )VPV ( 3
m

2
m−1 )VPV ( 4
m

3
m−1 )V
PV ( m−2
m

m−3
m−1 )V
PV ( m−1
m

m−3
m−1 )V
PV ( 1
m

m−2
m−1 )V
PV

PV1, m − 1 …
( 2
m

1
m−1 )VPV ( 3
m

2
m−1 )VPV ( 4
m

3
m−1 )V
PV ( m−2
m

m−3
m−1 )V
PV ( m−1
m

m−2
m−1 )V
PV ( 1
m

m−2
m−1 )V
PV

PV1, m …
( 2
m

1
m−1 )VPV ( 3
m

2
m−1 )VPV ( 4
m

3
m−1 )V
PV ( m−2
m

m−3
m−1 )V
PV ( m−1
m

m−2
m−1 )V
PV ( 1
m )
− 1 VPV

284
T. Pei, et al. Solar Energy 201 (2020) 279–297

Table 4
Voltage rules for short circuit faults in string 2 (n is even).
PV2,1~PV2, m V1,1 V1,2 V1,3 … V1, m −3 V1, m −2 V1, m −1

PV2,1 … m−1
( 1
m−1

1
m )V
PV ( 2
m−1

2
m )V PV ( 3
m−1

3
m )V
PV ( m−3
m−1

m−3
m )VPV ( m−2
m−1

m−2
m )V PV −
m
VPV
PV2,2 …
( 1
m−1

1
m )V
PV ( 2
m−1

2
m )V PV ( 3
m−1

3
m )V
PV ( m−3
m−1

m−3
m )VPV ( m−2
m−1

m−2
m )V PV ( 1
m−1

m−1
m )V
PV

PV2,3 …
( 2
m−1

1
m )V
PV ( 2
m−1

2
m )V PV ( 3
m−1

3
m )V
PV ( m−3
m−1

m−3
m )VPV ( m−2
m−1

m−2
m )V PV ( 1
m−1

m−1
m )V
PV

… … … … … … … …
PV2, m − 1 …
( 2
m−1

1
m )V
PV ( 3
m−1

2
m )V PV ( 4
m−1

3
m )V
PV ( m−2
m−1

m−3
m )VPV ( m−2
m−1

m−2
m )V PV ( 1
m−1

m−1
m )V
PV

PV2, m …
( 2
m−1

1
m )V
PV ( 3
m−1

2
m )V PV ( 4
m−1

3
m )V
PV ( m−2
m−1

m−3
m )VPV ( m−1
m−1

m−2
m )V PV ( 1
m−1

m−1
m )V
PV

Table 5
Voltage rules for short circuit faults in string 1 (n is odd).
PV1,1 ~PV1, m V1,1 V1,2 V1,3 V1,4 … V1,2(m − 1) −3 V1,2(m − 1) −2 V1,2(m−1) − 1

PV1,1 2 1 1 … 1
V
m PV
V
m PV ( 3
m

1
m−1 )V PV V
m PV ( m−1
m

m−3
m−1 )V
PV V
m PV ( 1
m

m−2
m−1 )V
PV

PV1,2 1 1 … 1
( 2
m

1
m−1 )V PV V
m PV ( 3
m

1
m−1 )V PV V
m PV ( m−1
m

m−3
m−1 )V
PV V
m PV ( 1
m

m−2
m−1 )V
PV

… … … … … … … …
PV1, m − 2 1 1 … 1
( 2
m

1
m−1 )V PV V
m PV ( 3
m

2
m−1 )V PV V
m PV ( m−1
m

m−3
m−1 )V
PV V
m PV ( 1
m

m−2
m−1 )V
PV

PV1, m − 1 1 1 … 1
( 2
m

1
m−1 )V PV V
m PV ( 3
m

2
m−1 )V PV V
m PV ( m−1
m

m−2
m−1 )V
PV V
m PV ( 1
m

m−2
m−1 )V
PV

PV1, m 1 1 … 1
( 2
m

1
m−1 )V PV V
m PV ( 3
m

2
m−1 )V PV V
m PV ( m−1
m

m−2
m−1 )V
PV V
m PV ( 1
m )
− 1 VPV

Table 6
Voltage rules for short circuit faults in string 2 (n is odd).
PV2,1~PV2, m V1,1 V1,2 … V1,2(m − 1) −3 V1,2(m − 1) −2 V1,2(m−1) − 1

PV2,1 2 … m−2
( 1
m−1

1
m )V PV V
m PV ( m−2
m−1

m−2
m )VPV ( m−1
m

m−3
m−1 )V PV −
m
VPV
PV2,2 … m−2
( 1
m−1

1
m )V PV ( 2
m

1
m−1 )V PV ( m−2
m−1

m−2
m )VPV ( m−1
m

m−3
m−1 )V PV −
m
VPV
PV2,3 m−2
( 2
m−1

1
m )V PV ( 2
m

1
m−1 )V PV ( m−2
m−1

m−2
m )VPV ( m−1
m

m−3
m−1 )V PV −
m
VPV
… … … … … … …
PV2, m − 2 … m−2
( 2
m−1

1
m )VPV ( 2
m

1
m−1 ) VPV ( m−2
m−1

m−2
m )VPV ( m−1
m

m−3
m−1 )VPV −
m
VPV
PV2, m − 1 … m−2
( 2
m−1

1
m )V PV ( 2
m

1
m−1 )V PV ( m−2
m−1

m−2
m )VPV ( m−1
m

m−2
m−1 )V PV −
m
VPV
PV2, m … m−2
( 2
m−1

1
m )V PV ( 2
m

1
m−1 )V PV ( m−1
m−1

m−2
m )VPV ( m−1
m

m−2
m−1 )V PV −
m
VPV

Table 7
Voltage rules for short circuit faults in string 3 (n is odd).
PV3,1~PV3, m V1,1 V1,2 V1,3 V1,4 … V1,2(m − 1) −3 V1,2(m − 1) −2 V1,2(m−1) − 1

PV3,1 1 1 … 1 m−1
V
m PV ( 1
m−1

1
m )V PV V
m PV ( 2
m−1

2
m )VPV V
m PV ( m−2
m−1

m−2
m )VPV −
m
VPV
PV3,2 1 1 … 1
V
m PV ( 1
m−1

1
m )V PV V
m PV ( 2
m−1

2
m )VPV V
m PV ( m−2
m−1

m−2
m )VPV ( 1
m−1

m−1
m )V
PV

PV3,3 1 1 … 1
V
m PV ( 2
m−1

1
m )V PV V
m PV ( 2
m−1

2
m )VPV V
m PV ( m−2
m−1

m−2
m )VPV ( 1
m−1

m−1
m )V
PV

… … … … … … … … …
PV3, m − 1 1 1 … 1
V
m PV ( 2
m−1

1
m )V PV V
m PV ( 3
m−1

2
m )VPV V
m PV ( m−2
m−1

m−2
m )VPV ( 1
m−1

m−1
m )V
PV

PV3, m 1 1 … 1
V
m PV ( 2
m−1

1
m )V PV V
m PV ( 3
m−1

2
m )VPV V
m PV ( m−1
m−1

m−2
m )VPV ( 1
m−1

m−1
m )V
PV

285
T. Pei, et al. Solar Energy 201 (2020) 279–297

Table 8
Voltage rules for degradation faults in string 1 (n is even).
PV1,1 ~PV1, m V1,1 V1,2 … V1, m −2 V1, m −1

PV1,1 1
V <
1
V < … 1
V < −
m −2
VPV <
m PV m PV m PV m
2
V1,1 < VPV
m
V1,2 < ( 3
m

1
m−1 )V
PV V1, m − 2 < ( m−1
m

m−3
m−1 )V
PV V1, m − 1 < ( 1
m

m−2
m−1 )V
PV

PV1,2 1
V >
1
V < … 1
V < −
m −2
VPV <
m PV m PV m PV m

V1,1 > ( 2
m

1
m−1 )V PV V1,2 < ( 3
m

1
m−1 )V
PV V1, m − 2 < ( m−1
m

m−3
m−1 )V
PV V1, m − 1 < ( 1
m

m−2
m−1 )V
PV

… … … … … …
PV1, m − 1 1
V >
1
V > … 1
V > −
m −2
VPV <
m PV m PV m PV m

V1,1 > ( 2
m

1
m−1 )V PV V1,2 > ( 3
m

2
m−1 )V
PV V1, m − 2 > ( m−1
m

m−2
m−1 )V
PV V1, m − 1 < ( 1
m

m−2
m−1 )V
PV

PV1, m 1
V >
1
V > … 1
V > −
m −2
VPV >
m PV m PV m PV m

V1,1 > ( 2
m

1
m−1 )V PV V1,2 > ( 3
m

2
m−1 )V
PV V1, m − 2 > ( m−1
m

m−2
m−1 )V
PV V1, m − 1 >
1
m (−1 )V PV

Table 9
Voltage rules for degradation faults in string 2 (n is even).
PV2,1~PV2, m V1,1 V1,2 … V1, m −2 V1, m −1

PV2,1 … m−1
( 1
m−1

1
m) VPV < ( 2
m−1

2
m) VPV < ( m−2
m−1

m−2
m ) VPV < −
m
VPV <
1 1 1 m−2
V1,1 < VPV V1,2 < VPV V1, m − 2 < VPV V1, m − 1 < − VPV
m m m m
PV2,2 …
( 1
m−1

1
m) VPV < ( 2
m−1

2
m) VPV < ( m−2
m−1

m−2
m ) VPV < ( 1
m−1

m−1
m )
VPV >
1 1 1 m−2
V1,1 < V V1,2 < V V1, m − 2 < VPV V1, m − 1 > − VPV
m PV m PV m m
… … … … …
PV2, m − 1 …
( 2
m−1

1
m) VPV > ( 3
m−1

2
m) VPV > ( m−2
m−1

m−2
m ) VPV < ( 1
m−1

m−1
m )
VPV >
1 1 1 m−2
V1,1 > V V1,2 > V V1, m − 2 < VPV V1, m − 1 > − VPV
m PV m PV m m
PV2, m …
( 2
m−1

1
m) VPV > ( 3
m−1

2
m) VPV > ( m−1
m−1

m−2
m ) VPV > ( 1
m−1

m−1
m )
VPV >
1 1 1 m−2
V1,1 > VPV V1,2 > VPV V1, m − 2 > VPV V1, m − 1 > − VPV
m m m m

Table 10
Voltage rules for degradation faults in string 1 (n is odd).
PV1,1 ~PV1, m V1,1 V1,2 … V1,2(m − 1) −3 V1,2(m − 1) −2 V1,2(m−1) − 1

PV1,1 1
V <
1
V … 1
V <
1
V −
m −2
VPV <
m PV m PV m PV m PV m
2
V1,1 < VPV
m
V1,2(m − 1) − 3 <( m−1
m

m−3
m−1 )V
PV V1,2(m − 1) − 1 <
m(
1

m−2
m−1 )V
PV

PV1,2 1
V >
1
V … 1
V <
1
V −
m −2
VPV <
m PV m PV m PV m PV m

V1,1 > ( 2
m

1
m−1 )V PV V1,2(m − 1) − 3 <( m−1
m

m−3
m−1 )V
PV V1,2(m − 1) − 1 <
m(
1

m−2
m−1 )V
PV

… … … … … … …
PV1, m − 1 1
V >
1
V … 1
V >
1
V −
m −2
VPV <
m PV m PV m PV m PV m

V1,1 > ( 2
m

1
m−1 )V PV V1,2(m − 1) − 3 >( m−1
m

m−2
m−1 )V
PV V1,2(m − 1) − 1 <
m(
1

m−2
m−1 )V
PV

PV1, m 1
V >
1
V … 1
V >
1
V −
m −2
VPV >
m PV m PV m PV m PV m

V1,1 > ( 2
m

1
m−1 )V PV V1,2(m − 1) − 3 >( m−1
m

m−2
m−1 )V
PV V1,2(m − 1) − 1 >
m
1
(− )
1 VPV

received irradiance of the shaded module is greater than zero but less Based on (10), the partial shading fault locating rules will be en-
than 1000W m2 . If a partial shading fault happens in the q-th string, and umerated as follows.
the l-th string is fault-free, Vkl, pq can be calculated by,
(1) n is even
⎧Vkl, pq = VPV _kl − VPV _pq
⎪V k
= V It is sufficient to consider only the partial shading fault locating
⎪ PV _kl m PV
⎪ p−1 method for sub-array 1. Partial shading faults occurring in strings 1 or 2
VPV < VPV _pq < pVoc , a fault occurs between (1, q) and (p , q)
⎨m−1 of sub-array 1 have similar rules to the degradation faults in Tables 8
⎪ VPV _pq < pVoc and VPV _pq < p VPV , and 9; see the details in Tables 13 and 14.
⎪ m−1
⎪ a fault occurs between (p + 1, q) and (m, q)
⎩ (2) n is odd
(10)

286
T. Pei, et al. Solar Energy 201 (2020) 279–297

Table 11
Voltage rules for degradation faults in string 2 (n is odd).
PV2,1~PV2, m V1,1 V1,2 … V1,2(m − 1) −3 V1,2(m − 1) −2 V1,2(m − 1) −1

PV2,1 1 … 1 m −2
( 1
m−1

1
m) VPV < V <
m PV ( m−2
m−1

m−2
m )V PV < V <
m PV

m
VPV
2
V1,1 <
1
V
m PV
V1,2 < VPV
m V1,2(m − 1) − 3 <
1
V
m PV
V1,2(m − 1) −2 < (m−1
m

m−3
m−1 )V PV

PV2,2 1 … 1 m −2
( 1
m−1

1
m) VPV < V >
m PV ( m−2
m−1

m−2
m ) VPV < V <
m PV

m
VPV

V1,1 <
1
V
m PV
V1,2 > ( 2
m

1
m−1 )VPV V1,2(m − 1) − 3 <
1
V
m PV
V1,2(m − 1) −2 < (m−1
m

m−3
m−1 )V PV

… … … … … … …
PV2, m − 1 1 … 1 m −2
( 2
m−1

1
m) VPV > V >
m PV ( m−2
m−1

m−2
m )V PV < V >
m PV

m
VPV

V1,1 >
1
m
VPV V1,2 > ( 2
m

1
m−1 )VPV V1,2(m − 1) − 3 <
1
V
m PV
V1,2(m − 1) −2 > (m−1
m

m−2
m−1 )V PV

PV2, m 1 … 1 m −2
( 2
m−1

1
m) VPV > V >
m PV ( m−1
m−1

m−2
m )V PV > V >
m PV

m
VPV

V1,1 >
1
m
VPV V1,2 > ( 2
m

1
m−1 )VPV V1,2(m − 1) − 3 >
1
V
m PV
V1,2(m − 1) −2 > (m−1
m

m−2
m−1 )V PV

Table 12
Voltage rules for degradation faults in string 3 (n is odd).
PV3,1~PV3, m V1,1 V1,2 … V1,2(m − 1) −3 V1,2(m − 1) −2 V1,2(m−1) − 1

PV3,1 1 … 1 m−1
V
m PV ( 1
m−1

1
m) VPV < V
m PV ( m−2
m−1

m−2
m )VPV < −
m
VPV <
1 1 m−2
V1,2 < VPV V1,2(m − 1) − 2 < V V1,2(m − 1) − 1 < − VPV
m m PV m
PV3,2 1 … 1
V
m PV ( 1
m−1

1
m) VPV < V
m PV ( m−2
m−1

m−2
m )VPV < ( 1
m−1

m−1
m )VPV >
1 1 m−2
V1,2 < V V1,2(m − 1) − 2 < V V1,2(m − 1) − 1 > − VPV
m PV m PV m
… … … … … … …
PV3, m − 1 1 … 1
V
m PV ( 2
m−1

1
m) VPV > V
m PV ( m−2
m−1

m−2
m )VPV < ( 1
m−1

m−1
m )VPV >
1 1 m−2
V1,2 > VPV V1,2(m − 1) − 2 < V V1,2(m − 1) − 1 > − VPV
m m PV m
PV3, m 1 … 1
V
m PV ( 2
m−1

1
m) VPV > V
m PV ( m−1
m−1

m−2
m )VPV > ( 1
m−1

m−1
m )VPV >
1 1 m−2
V1,2 > VPV V1,2(m − 1) − 2 > V V1,2(m − 1) − 1 > − VPV
m m PV m

Table13
Voltage rules for partial shading faults in string 1 (n is even).
PV1,1 ~PV1, m V1,1 V1,2 … V1, m −2 V1, m −1

PV1,1 2
V − Voc < V1,1 <
3
V − 2Voc < V1,2 < … m−1
VPV − (m − 2) Voc < V1, m − 2 <
1
V − (m − 1) Voc < V1, m − 1 <
m PV m PV m m PV
2
V
m PV
3
m( −
1
m−1 )V PV ( m−1
m

m−3
m−1 )V PV ( 1
m

m−2
m−1 )VPV

PV1,2 2
V1,1 > VPV − Voc
3
V − 2Voc < V1,2 < … m−1
VPV − (m − 2) Voc < V1, m − 2 <
1
V − (m − 1) Voc < V1, m − 1 <
m m PV m m PV

V1,1 >
2
m ( −
1
m−1
VPV ) 3
m( −
1
m−1 ) VPV ( m−1
m

m−3
m−1 ) VPV ( 1
m

m−2
m−1 )VPV

… … … … … …
PV1, m − 1 V1,1 >
2
V − Voc V1,2 >
3
V − 2Voc … V1, m − 2 >
m−1
VPV − (m − 2) Voc
1
V − (m − 1) Voc < V1, m − 1 <
m PV m PV m m PV

V1,1 > ( 2
m

1
m−1 )
VPV V1,2 > ( 3
m

2
m−1
VPV ) V1, m − 2 >
m(
m−1

m−2
m−1
VPV ) ( 1
m

m−2
m−1 )VPV

PV1, m 2
V1,1 > VPV − Voc
3
V1,2 > VPV − 2Voc … V1, m − 2 >
m−1
VPV − (m − 2) Voc
1
V1, m − 1 > VPV − (m − 1) Voc
m m m m

V1,1 >
2
m ( −
1
m−1
VPV ) V1,2 >
3
m (−
2
m−1
VPV ) V1, m − 2 >
m−1
m (−
m−2
m−1
VPV ) V1, m − 1 >
1
m
−1 ( )V PV

Now consider the partial shading fault locating strategy for the during ten consecutive minutes, the PV array may have faults (Matam
m × 3 sub-array 1 which can be discussed similarly to the degradation and Barry, 2018). Secondly, use fault detection methods to detect po-
faults in Tables 10–12, and the corresponding results are reported in tential faults (Silvestre et al., 2014), and utilize the deployed voltage
Tables 15–17. sensors to measure the relative voltage between strings (Vm ). Then, the
proposed fault locating strategy is employed to calculate the theoretical
4.5. Application of the proposed strategy in fault locating value of the relative voltage (Vkl, pq ) under different fault situations. If
the voltage deviation Vε between Vm and Vkl, pq is smaller than 5% of Vkl, pq
The implementation of the fault location strategy for a PV array is (Hu et al., 2015), the corresponding location of the faulty modules will
summarized in the flowchart in Fig. 6. Firstly, the MPPT strategy is be identified. Finally, output the locations of the faulty modules.
performed (Pei et al., 2018), and the difference between the theoretical
MPP power (Pt ) and the actual MPP power (Pm ) is checked every minute.
If the power deviation (Pε = Pm − Pt ) is always greater than 10–15% of Pt

287
T. Pei, et al. Solar Energy 201 (2020) 279–297

Table 14
Voltage rules for partial shading faults in string 2 (n is even).
PV2,1~PV2m V1,1 V1,2 … V1, m −2 V1, m −1

PV2,1 … m−1
( 1
m−1

1
m )VPV < V1,1 < ( 2
m−1

2
m )V
PV < V1,2 < ( m−2
m−1

m−2
m )V
PV < V1, m − 2 < −
m
VPV < V1, m − 1 <
1 2 m−2 m−1
2Voc − V 3Voc − V (m − 1) Voc − VPV Voc − VPV
m PV m PV m m
PV2,2 … m−1
( 1
m−1

1
m )VPV < V1,1 < ( 2
m−1

2
m )V
PV < V1,2 < ( m−2
m−1

m−2
m )V
PV < V1, m − 2 < V1, m − 1 < Voc −
m
VPV

2Voc −
1
V
m PV
3Voc −
2
V
m PV
(m − 1) Voc −
m−2
m
VPV V1, m − 1 < ( 1
m−1

m−1
m
VPV )
… … … … …
PV2, m − 1 1 2 … m−1
V1,1 < 2Voc − V
m PV
V1,2 < 3Voc − V
m PV ( m−2
m−1

m−2
m )V
PV < V1, m − 2 < V1, m − 1 < Voc −
m
VPV

V1,1 < ( 2
m−1

m
1
VPV) V1,2 < ( 3
m−1

m
2
)VPV (m − 1) Voc −
m−2
m
VPV V1, m − 1 < ( 1
m−1

m−1
m
VPV)
PV2, m 1
V1,1 < 2Voc − VPV
2
V1,2 < 3Voc − VPV … V1, m − 2 < (m − 1) Voc −
m−2
VPV V1, m − 1 < Voc −
m−1
VPV
m m m m

V1,1 <
2
m−1 (

m
1
VPV ) V1,2 < (
3
m−1

m
2
VPV ) V1, m − 2 < (
m−1
m−1

m−2
m
VPV ) V1, m − 1 <
1
m−1( −
m−1
m
VPV )

Table 15
Voltage rules for partial shading faults in string 1 (n is odd).
PV1,1 ~PV1, m V1,1 V1,2 … V1,2(m − 1) −3 V1,2(m − 1) −2 V1,2(m−1) − 1

PV1,1 2
V − Voc < V1,1 <
1
V … m−1
VPV − (m − 2) Voc < V1,2(m − 1) −3 <
1
V 1
V − (m − 1) Voc < V1,2(m−1) − 1 <
m PV m PV m m PV m PV
2
V
m PV ( m−1
m

m−3
m−1 ) VPV ( 1
m

m−2
m−1 )V PV

PV1,2 2
V1,1 > VPV − Voc
1
V … m−1
VPV − (m − 2) Voc < V1,2(m − 1) −3 <
1
V 1
V − (m − 1) Voc < V1,2(m−1) − 1 <
m m PV m m PV m PV

V1,1 >
2
m (−
1
m−1
VPV ) ( m−1
m

m−3
m−1 ) VPV ( 1
m

m−2
m−1 )V PV

… … … … … … …
PV1, m − 1 V1,1 >
2
V − Voc
1
V … V1,2(m − 1) −3 >
m−1
VPV − (m − 2) Voc
1
V 1
V − (m − 1) Voc < V1,2(m−1) − 1 <
m PV m PV m m PV m PV

V1,1 > ( 2
m

1
m−1
VPV ) V1,2(m − 1) −3 >
m−1
m ( −
m−2
m−1
VPV ) ( 1
m

m−2
m−1 )V PV

PV1, m 2
V1,1 > VPV − Voc
1
V … V1,2(m − 1) −3 >
m−1
VPV − (m − 2) Voc
1
V V1,2(m−1) − 1 >
1
V − (m − 1) Voc >
m m PV m m PV m PV

V1,1 >
2
m (−
1
m−1
VPV ) V1,2(m − 1) −3 >
m−1
m (−
m−2
m−1
VPV ) V1,2(m−1) − 1 > ( 1
m )
− 1 VPV

Table 16
Voltage rules for partial shading faults in string 2 (n is odd).
PV2,1~PV2, m V1,1 V1,2 … V1,2(m − 1) −3 V1,2(m − 1) −2 V1,2(m − 1) - 1

PV2,1 2 … m−1 V1,2(m − 1) −1 <


( 1
m−1

1
m )V PV < V
m PV
− Voc < ( m−2
m−1

m−2
m )V PV < m
VPV − (m − 2) Voc <
m −2
2 − VPV
V1,1 < 2Voc −
1
V
m PV
V1,2 < V
m PV V1,2(m − 1) −3 < (m − 1) Voc −
m−2
m
VPV V1,2(m − 1) −2 < ( m−1
m

m−3
m−1 )V PV m

PV2,2 2 … m−1 V1,2(m − 1) −1 <


( 1
m−1

1
m )V PV < V1,2 > V − Voc
m PV ( m−2
m−1

m−2
m )V PV < m
VPV − (m − 2) Voc <
m −2
− VPV
V1,1 < 2Voc −
1
V
m PV
V1,2 > ( 2
m

1
m−1 )
VPV V1,2(m − 1) −3 < (m − 1) Voc −
m−2
m
VPV V1,2(m − 1) −2 < ( m−1
m

m−3
m−1 )V PV m

… … … … … … …
PV2, m − 1 1 2 … m−1 V1,2(m − 1) −1 <
V1,1 < 2Voc − V
m PV
V1,2 > V − Voc >
m PV ( m−2
m−1

m−2
m ) VPV < V1,2(m − 1) −2 >
m
VPV − (m − 2) Voc
m −2
− VPV
V1,1 < ( 2
m−1

m
1
VPV ) V1,2 > ( 2
m

1
m−1 )
VPV V1,2(m − 1) −3 < (m − 1) Voc −
m−2
m
VPV V1,2(m − 1) −2 >
m−1
m (−
m−2
m−1
VPV ) m

PV2, m V1,1 < 2Voc −


1
V V1,2 >
2
V − Voc … V1,2(m − 1) −3 < (m − 1) Voc −
m−2
VPV V1,2(m − 1) −2 >
m−1
VPV − (m − 2) Voc V1,2(m − 1) −1 <
m PV m PV m m m −2
− VPV
V1,1 < ( 2
m−1

m
1
)VPV V1,2 > ( 2
m

1
m−1 )
VPV V1,2(m − 1) −3 < ( m−1
m−1

m−2
m )V PV V1,2(m − 1) −2 >
m−1
m (−
m−2
m−1
VPV ) m

5. Results and discussion sensors from V1,1 to V1,5 are compared with the corresponding fault lo-
cating rules for an odd n (n = 3) when faults occur in sub-array 1, and
To inspect the performance of the proposed method for PV array’s the readings from V2,1 to V2,3 are compared with the corresponding lo-
fault locating, a GCPV plant of 1.2 kWp at an experimental base of cating rules for an even n (n = 2 ) when faults occur in sub-array 2.
China electric power research institute is simulated in the Matlab/
Simulink platform. The PV system includes a 5 × 5 PV array with the 5.1. Open circuit faults
proposed voltage sensor placement, and its configuration is shown in
Fig. 7, which can be divided as one 5× 3 sub-array 1 and one 5× 2 sub- In fault scenario 1 with open circuit faults, the PV array receives
array 2. The critical parameters of the modules are reported in Table 18 irradiance of 1000W m2 under temperature 25°C . When string 4 of sub-
(Patel and Agarwal, 2008). For simulation purposes, 4 typical fault array 2 has an open circuit fault (denoted by ‘FO1’ in Fig. 7(a)), the
scenarios such as open circuit, short circuit, degradation, and partial outputs of V2,1, V2,2 and V2,3 are 13.3V , 9.5V and − 45.7V , respectively,
shading faults are studied. Meanwhile, the readings of the voltage and Vref 21 is the fault-free reference voltages for V2,1 and V2,2 , and Vref 22 is

288
T. Pei, et al. Solar Energy 201 (2020) 279–297

Table 17
Voltage rules for partial shading faults in string 3 (n is odd).
PV3,1~PV3, m V1,1 V1,2 … V1,2(m − 1) −3 V1,2(m − 1) −2 V1,2(m−1) − 1

PV3,1 1 … 1 m−1
V
m PV ( 1
m−1

1
m )VPV < V1,2 < V
m PV ( m−2
m−1

m−2
m )V
PV < V1,2(m − 1) −2 < −
m
VPV < V1,2(m−1) − 1 <
1 m−2 m−1
2Voc − V (m − 1) Voc − VPV Voc − VPV
m PV m m
PV3,2 1 … 1 m−1
V
m PV ( 1
m−1

1
m )VPV < V1,2 < V
m PV ( m−2
m−1

m−2
m )V
PV < V1,2(m − 1) −2 < V1,2(m−1) − 1 < Voc −
m
VPV

2Voc −
1
V
m PV
(m − 1) Voc −
m−2
m
VPV V1,2(m−1) − 1 < ( 1
m−1

m−1
m
VPV)
… … … … … … …
PV3, m − 1 1 1 … 1 m−1
V
m PV
V1,2 < 2Voc − V
m PV
V
m PV ( m−2
m−1

m−2
m )V
PV < V1,2(m − 1) −2 < V1,2(m−1) − 1 < Voc −
m
VPV

V1,2 < ( 2
m−1

m
1
)VPV (m − 1) Voc −
m−2
m
VPV V1,2(m−1) − 1 < ( 1
m−1

m−1
m )
VPV
PV3, m 1
V V1,2 <
1
2Voc − VPV < … 1
V V1,2(m − 1) −2 < (m − 1) Voc −
m−2
VPV V1,2(m−1) − 1 < Voc −
m−1
VPV <
m PV m m PV m m

V1,2 < ( 2
m−1

1
m
VPV ) V1,2(m − 1) −2 < (
m−1
m−1

m−2
m
VPV ) V1,2(m−1) − 1 < ( 1
m−1

m−1
m )
VPV

Fig. 6. Flowchart of the proposed fault location strategy.

the reference for V2,3 , as illustrated in Fig. 8(a). The values of V2,1, V2,2 showing that string 4 and string 5 have open circuit faults.
and V2,3 satisfy the open circuit locating rules of the case “10” in Fig. 9 shows the voltage sensor readings under an open circuit fault
Table 1, and thus it can be concluded that string 4 has an open circuit in sub-array 1. The notation Vref 11 is the reference of V1,1 and V1,3 , Vref 12 is
fault. When open circuit faults happen simultaneously in string 4 and the reference of V1,5 , and Vref 13 is the reference of V1,2 and V1,4 under fault-
string 5 denoted by ‘FO2 ’, the outputs of V2,1, V2,2 and V2,3 are equal to free conditions. When string 2 has an open circuit represented by ‘FO3 ’,
21V , 21V and − 41.9V , respectively, as shown in Fig. 8(b). These values the readings of V1,1, V1,3 and V1,5 are 24.8V , 28.6V and − 34.3V , respec-
are close to the corresponding locating rules of the case “11” in Table 1, tively, and the readings of V1,2 and V1,4 are 13.4V and9.5V , respectively.

289
T. Pei, et al. Solar Energy 201 (2020) 279–297

Fig. 7. Configuration of the PV array with the proposed voltage sensor placement. (a) Open circuit and short circuit faults; (b) Degradation and shading faults.

Table 18 the outputs of V2,1, V2,2 and V2,3 are 5.4V , 10.8V , and − 27.1V , respec-
PV module specifications. tively, as illustrated in Fig. 11(b). The actual sensor values satisfy the
Parameters Value short circuit locating rules of the second module in Table 4. Hence, the
effectiveness of the proposed short circuit locating strategy is demon-
Maximum power 59.9 W strated when the number of strings for a sub-array is even.
Voltage at open circuit 21 V Fig. 12 shows the voltage values for a short circuit fault occurring in
Current at short circuit 3.74 A
Voltage at the MPP 17.1 V
string 2 represented by ‘FS3 ’. The values of Vref 11, Vref 12 and Vref 13 are still
Current at the MPP 3.5 A the reference voltages with fault-free conditions. The readings of V1,1,
Cell per module 36 V1,3 and V1,5 are 27.1V , 31.9V , and − 31.9V , respectively, and the read-
ings of V1,2 and V1,4 are 10.8V and 5.4V , respectively. The sensor values
agree with the short circuit locating rules of the m-th module in Table 6.
These readings follow the open circuit locating rules of the case “010” Fig. 13 shows the voltage values for a short circuit fault happening in
in Table 2. Fig. 10 shows the voltage sensor readings in the case of string 3 represented by ‘FS4 ’. The readings of V1,1 and V1,3 are 16.3V , and
multi-string open circuit faults in sub-array 1. When open circuit faults the readings of V1,2 , V1,4 and V1,5 are 5.4V , 10.8V , and − 48.7V , respec-
happen simultaneously in strings 1 and 2 represented by ‘FO4 ’, the tively. These values match the short circuit locating rules of the first
readings of V1,1 and V1,3 are identically equal to 21V , and the readings of module in Table 7. Hence, the proposed strategy is verified when the
V1,2 , V1,4 and V1,5 are 13.1V , 9.1V , and− 45.9V , respectively. These values number of strings for a sub-array is odd.
agree with the proposed locating rules of the case “110” in Table 2.
5.3. Degradation faults
5.2. Short circuit faults
In fault scenario 3 with degradation faults, Fig. 14 shows the voltage
In fault scenario 2 with short circuit faults, Fig. 11 shows the sensor sensor readings in the case of degradation faults in sub-array 2. Vref 21
values for short circuit faults in sub-array 2. Similarly, notations Vref 21 and Vref 22 are still the reference voltages, and V21S , V22S and V23S are the
and Vref 22 are the fault-free reference voltages. If a short circuit fault corresponding voltages in the case of short circuit faults in PV4,2 or PV5,3 .
occurs in string 4 (denoted by ‘FS1’ in Fig. 7(a)), the outputs of V2,1, V2,2 As presented in Fig. 14(a), when a degradation fault happens in string 4
and V2,3 are 10.8V , 5.4V and− 27.1V , respectively, as illustrated in (denoted by ‘FD1’ in Fig. 7(b)), the outputs of V2,1, V2,2 and V2,3 are 13.6V ,
Fig. 11(a), which match the short circuit locating rules of the (m-1)-th 22.7V and − 30.2V , respectively. The value of V2,1 is higher than V21S and
module in Table 3. If string 5 has a short circuit fault denoted by ‘FS2 ’, lower than Vref 21. The values of V2,2 and V2,3 are bigger than Vref 21 and

Fig. 8. Open circuit faults in sub-array 2. (a) Mono-string open circuit; (b) Multi-string open circuit.

290
T. Pei, et al. Solar Energy 201 (2020) 279–297

Fig. 9. A mono-string open circuit fault in sub-array 1. (a) The values of V1,1, V1,3 and V1,5 ; (b) The values of V1,2 and V1,4 .

Fig. 10. Multi-string open circuit faults in sub-array 1. (a) The values of V1,1 and V1,3 ; (b) The values of V1,2 , V1,4 and V1,5 .

Fig. 11. A short circuit fault in sub-array 2. (a) String 4 short circuit; (b) String 5 short circuit.

Vref 22 , respectively, and smaller than V22S and V23S , respectively. This degradation fault happens in string 2 represented by ‘FD3 ’. Vref 11, Vref 12
verifies the locating rules of the second module degradation in Table 8. and Vref 13 are the fault-free reference voltages. V11S to V15S are the cor-
In Fig. 14(b), when string 5 has a degradation fault denoted by ‘FD2 ’, the responding voltages in the case of short circuit faults in PV2,4 . The va-
outputs of V2,1, V2,2 and V2,3 are 22.7V , 13.6V and − 30.2V , respectively. lues of V1,1, V1,3 and V1,5 are 22.7V , 25.7V and − 34V , respectively, and
The value of V2,2 is higher than V22S and lower than Vref 21. The values of the values of V1,2 and V1,4 are 13.6V and 10.6V , respectively. The values
V2,1 and V2,3 are bigger than Vref 21 and Vref 22 , respectively, and smaller follow the locating rules of the m-th module degradation of string 2 in
than V21S and V23S , respectively. Again, this is consistent with the lo- Table 11. Fig. 16 shows the readings when string 1 has a degradation
cating rules of the (m − 1)-th module degradation in Table 9. There- fault represented by ‘FD4 ’. V11S to V15S are the corresponding voltages in
fore, the proposed degradation fault locating method is verified when the case of short circuit faults in PV1,1. The values of V1,1, V1,3 and V1,5 are
the string number of a sub-array is even. 25.7V , 22.7V and − 30.2V , respectively, and the values of V1,2 and V1,4 are
Fig. 15 shows the voltage readings in sub-array 1 when a 16.9V . These values satisfy the locating rules of the first module

291
T. Pei, et al. Solar Energy 201 (2020) 279–297

Fig. 12. String 2 short circuit in sub-array 1. (a) The values of V1,1, V1,3 and V1,5 ; (b) The values of V1,2 and V1,4 .

Fig. 13. String 3 short circuit in sub-array 1. (a) The values of V1,1 and V1,3 ; (b) The values of V1,2 , V1,4 and V1,5 .

Fig. 14. A degradation fault in sub-array 2. (a) String 4 degradation fault; (b) String 5 degradation fault.

degradation of string 1 in Table 10, which verifies the proposed strategy V22O and V23O are the corresponding voltages in the case of open circuit
when the string number of a sub-array is odd. faults in string 4. The values of V2,1, V2,2 and V2,3 are 14.7V , 11.7V and
− 32.5V , respectively. This verifies the locating rules of the shaded
5.4. Partial shading faults (m − 1)-th module in Table 13. If a shading fault occurs in PV5,2 de-
noted by ‘FP2 ’, the values of V2,1, V2,2 and V2,3 are 11.7V , 14.7V and
In fault scenario 4 with partial shading faults, the received irra- − 32.5V , respectively, as shown in Fig. 17(b). According to the shading
diation of the healthy modules are still 1000W m2 and the shaded locating rules in Table 14, it is proven that the second module has a
modules are 100W m2 under temperature 25°C . Fig. 17(a) shows the shading fault.
voltage readings when PV4,3 in sub-array 2 has a shading fault (denoted Fig. 18 shows the voltage readings when PV2,4 in sub-array 1 has a
by ‘FP1’ in Fig. 7(b)). The values of V21S , V22S and V23S are the corre- shading fault represented by ‘FP3 ’. V21S , V22S and V23S are the corre-
sponding voltages in the case of short circuit faults in PV4,3 , and V21O , sponding voltages in the case of short circuit faults in PV2,4 , and V21O ,

292
T. Pei, et al. Solar Energy 201 (2020) 279–297

Fig. 15. String 2 degradation fault in sub-array 1. (a) The values of V1,1, V1,3 and V1,5 ; (b) The values of V1,2 and V1,4 .

Fig. 16. String 1 degradation fault in sub-array 1. (a) The values of V1,1, V1,3 and V1,5 ; (b) The values of V1,2 and V1,4 .

Fig. 17. A partial shading fault in sub-array 2. (a) String 4 partial shading fault; (b) String 5 partial shading fault.

V22O and V23O are the corresponding voltages in the case of open circuit Fig. 19 shows the voltage readings when a shading fault occurs in PV3,1
in string 2. The values of V1,1 and V1,3 are 23.8V and 26.9V , respectively, of sub-array 1 represented by ‘FP4 ’. V21S , V22S and V23S are the corre-
and the values of V1,2 , V1,4 and V1,5 are 14.7V , 11.7V and − 35.5V , re- sponding voltages in the case of short circuit faults in PV3,1, and V21O ,
spectively. Compared with the shading locating rules presented in V22O and V23O are the corresponding voltages in the case of open circuit
Table 16, the voltage values satisfy the locating rules of m-th module. in string 3. The values of V1,1 and V1,3 are 17.8V , and the values of V1,2 ,

293
T. Pei, et al. Solar Energy 201 (2020) 279–297

Fig. 18. String 2 partial shading fault in sub-array 1. (a) The values of V1,1 and V1,3 ; (b) The values of V1,2 , V1,4 and V1,5 .

Fig. 19. String 3 partial shading fault in sub-array 1. (a) The values of V1,1 and V1,3 ; (b) The values of V1,2 , V1,4 and V1,5 .

V1,4 and V1,5 are 11.7V , 14.7V and − 44.7V , respectively. Compared with
the shading locating rules illustrated in Table 17, the sensor values
agree with the locating rules of the first module shading. Hence, the
locating strategy of PV modules is verified under partial shading faults.

5.5. Comparison study

In order to verify the superiority of the proposed fault locating


method, two different cases are arranged to perform the comparison
studies respectively.
Case 1: Compare the proposed method with that in (Hu et al., 2015).
A 3 × 3PV array is built with three voltage sensors deployed, and ‘FO ’,
‘FS ’, ‘FD ’ and ‘FP ’ denote the designed fault scenarios, as illustrated in
Fig. 20. Under the fault scenarios, the comparison results are sum-
marized in Table 19. The cross notation ‘X’ indicates the locating failure
of the relevant method. As shown, the proposed locating strategy pos-
sesses higher accuracy, especially for degradation and partial shading
faults.
Case 2: Compare the proposed method with that in (Madeti and
Singh, 2017). Table 20 summarizes the comparison results of the two
Fig. 20. Configuration of the 3 × 3 PV array under the mentioned faults. methods based on the fault scenarios of sub-array 2 (see in Fig. 7) under
open circuit, short circuit, degradation and partial shading faults. Si-
milar effectiveness on fault locating for the proposed method can be
observed.

294
T. Pei, et al. Solar Energy 201 (2020) 279–297

Table 19
Comparison of fault locating methods for case 1.
Faults Techniques Sensor readings Fault location

FO Proposed method V1 = 17V ; V2 = 24.8V ; V3 = −13.1V String 3


Ref. (Hu et al., 2015) V1 = 17V ; V2 = 24.8V ; V3 = −13.1V String3

FS Proposed method V1 = 8.3V ; V2 = 33V ; V3 = −16.5V PV2,1


Ref. (Hu et al., 2015) V1 = 8.3V ; V2 = 33V ; V3 = −16.5V PV2,1

FD Proposed method V1 = 23.4V ; V2 = 13.2V ; V3 = −16.8V PV2,3


Ref. (Hu et al., 2015) V1 = 33.5V ; V2 = 8.4V ; V3 = −16.8V ×

FP Proposed method V1 = 24.7V ; V2 = 16.6V ; V3 = −12.6V PV1,1


Ref. (Hu et al., 2015) V1 = 12.8V ; V2 = 16.6V ; V3 = −24.2V ×

Table 20
Comparison of fault locating methods for case 2.
Faults Techniques Sensor readings Fault location

FO1 Proposed method V2,1 = 13.3V ; V2,2 = 9.5V ; V2,3 = −45.7V String 1
Ref. (Madeti and Singh, 2017) V2,1 = 34.2V ; V2,2 = 51.3V ; V2,3 = 17.1V String1

FO2 Proposed method V2,1 = 21V ; V2,2 = 21V ; V2,3 = −41.9V String 1 and String 2
Ref. (Madeti and Singh, 2017) V2,1 = 0V ; V2,2 = 0V ; V2,3 = 0V String 1 and String 2

FS1 Proposed method V2,1 = 10.8V ; V2,2 = 5.4V ; V2,3 = −27.1V PV4,3
Ref. (Madeti and Singh, 2017) V2,1 = 11.4V ; V2,2 = 5.7V ; V2,3 = −28.5V PV4,3

FS2 Proposed method V2,1 = 5.4V ; V2,2 = 10.8V ; V2,3 = −27.1V PV5,2
Ref. (Madeti and Singh, 2017) V2,1 = 5.7V ; V2,2 = 11.4V ; V2,3 = −28.5V PV5,2

FD1 Proposed method V2,1 = 13.6V ; V2,2 = 22.7V ; V2,3 = −30.2V PV4,2
Ref. (Madeti and Singh, 2017) V2,1 = 11.4V ; V2,2 = 28.5V ; V2,3 = −28.5V ×

FD2 Proposed method V2,1 = 22.7V ; V2,2 = 13.6V ; V2,3 = −30.2V PV5,3
Ref. (Madeti and Singh, 2017) V2,1 = 28.5V ; V2,2 = 11.4V ; V2,3 = −28.5V ×

FP1 Proposed method V2,1 = 14.7V ; V2,2 = 11.7V ; V2,3 = −32.5V PV4,3
Ref. (Madeti and Singh, 2017) V2,1 = 11.4V ; V2,2 = 5.7V ; V2,3 = −28.5V ×

FP2 Proposed method V2,1 = 11.7V ; V2,2 = 14.7V ; V2,3 = −32.5V PV5,2
Ref. (Madeti and Singh, 2017) V2,1 = 5.7V ; V2,2 = 11.4V ; V2,3 = −28.5V ×

Fig. 21. String 1 open circuit in the first sub-array. (a) The values of V1,1, V1,2 , V1,3 ; (b) The value of V1,99 .

5.6. Case study for a large-scale PV array needs 99 voltage sensors. Since each sub-array possess similar voltage
sensor deployment, the fault locating rules are also similar for these
For practical GCPV plants, the capacity of the PV arrays is generally sub-arrays.
at the megawatt level. To verify the applicability of the proposed As mentioned in Sections 4.3 and 4.4, severe degradation faults can
method for a PV array with large capacity and size, consider a 100 × 100 result in short circuits, and the full shading of the modules in a string
PV array whose parameters are listed in Table 18. According to the may be treated as open circuit. Therefore, the proposed locating
proposed sensor placement method, the PV array can be divided into 50 method will be verified only under open circuit and short circuit faults.
blocks of 100 × 2 sub-arrays connected in parallel and each sub-array Assume that the fault occurs in sub-array 1 of the 100 × 100 PV array

295
T. Pei, et al. Solar Energy 201 (2020) 279–297

Fig. 22. String 1 short circuit in the first sub-array. (a) The values of V1,1, V1,2 , V1,3 , V1,96 , V1,97 , and V1,98 ; (b) The value of V1,99 .

and the placed sensors are V1,1 ~ V1,99. Fig. 21 shows the corresponding influence the work reported in this paper.
voltage values in the case of an open circuit fault in string 1. The no-
tations Vref 1 and Vref 2 are the fault-free reference voltages, and the Acknowledgments
readings of V1,1, V1,2 , V1,3 , …, V1,96 , V1,97 , V1,98 and V1,99 are 13.36V , 9.42V ,
5.48V , …, − 360.91V , − 364.85V , − 368.79V , and − 2082.73V , respec- We are very grateful for the helpful comments from the anonymous
tively. The sensor readings satisfy the open circuit locating rules for reviewers. This work was supported by the China Scholarship Council
string 1 of the case “10” in Table 1. Fig. 22 shows the corresponding [grant number 201908620089] and the National Natural Science
values for a short circuit fault occurring in string 1. The outputs of V1,1, Foundation of China [grant number 61540033]. The first author would
V1,2 , V1,3 , …, V1,96 , V1,97 , V1,98 and V1,99 are 16.25V , 16.08V , 15.92V , …, 0.5V , like to thank the School of Electrical and Data Engineering, University
0.34V , 0.16V , and − 1625.18V , respectively, which matches the short of Technology Sydney, for providing the support to carry out the re-
circuit locating rules of the m-th module in Table 3. Hence, the effec- search work.
tiveness of the proposed strategy is demonstrated for large size PV ar-
rays. References

6. Conclusions Alam, M.K., Khan, F., Johnson, J., Flicker, J., 2015. A comprehensive review of cata-
strophic faults in PV arrays: types, detection, and mitigation techniques. IEEE J.
Photovolt. 5 (3), 982–997.
In this paper, an accurate fault locating method is developed to Bressan, M., Basri, Y.E., Galeano, A.G., Alonso, C., 2016. A shadow fault detection method
locate any faulty modules of PV arrays under open circuit, short circuit, based on the standard error analysis of I-V curves. Renew. Energy 99, 1181–1190.
degradation, or partial shading faults, and it can also reduce the Chen, L., Wang, X., 2018. Adaptive fault localization in photovoltaic systems. IEEE Trans.
Smart Grid 9 (6), 6752–6763.
maintenance cost and wiring complexity for PV systems. From the ac- Chen, Z., Chen, Y., Wu, L., Cheng, S., Lin, P., 2019. Deep residual network based fault
quired readings of the deployed voltage sensors and the formulated detection and diagnosis of photovoltaic arrays using current-voltage curves and
locating rules, the corresponding faulty module can be accurately ambient conditions. Energy Convers. Manage. 198, 111793–111812.
Chouder, A., Silvestre, S., 2009. Analysis model of mismatch power losses in PV systems.
identified. The fault locating technique has been demonstrated on a J. Sol. Energy Eng. 131 (2), 24504–24508.
GCPV plant simulation system of 1.2kWp at an experimental base of Dincer, F., 2011. The analysis on photovoltaic electricity generation status, potential and
China electric power research institute. The results show the ability of policies of the leading countries in solar energy. Renew. Sust. Energy Rev. 15 (1),
713–720.
the proposed strategy to identify the locations of faulty modules under Dhimish, M., Holmes, V., Mehrdadi, B., Dales, M., Mather, P., 2017. Photovoltaic fault
the corresponding fault scenarios. Besides, the applicability of the detectionalgorithm based on theoretical curves modelling and fuzzy classification
proposed method on PV arrays of large scale and high capacity is ver- system. Energy 140, 276–290.
Dhimish, M., Holmes, V., Mehrdadi, B., Dales, M., 2018. Comparing Mamdani Sugeno
ified by a 100 × 100 PV array under open circuit and short circuit faults. fuzzy logic and RBF ANN network for PV fault detection. Renew. Energy 117,
Compared with existing methods, the proposed approach can not 257–274.
only locate the faulty modules under all the mentioned fault conditions Flicker, J., Johnson, J., 2013. Analysis of fuses for blind spot ground fault detection in
photovoltaic power systems. Sandia National Laboratories Report.
for a PV array of any size and capacity, but also have a higher fault
Garoudja, E., Harrou, F., Sun, Y., Kara, K., Chouder, A., Silvestre, S., 2017. Statistical fault
locating accuracy, especially for degradation and partial shading faults. detection in photovoltaic systems. Sol. Energy 150, 485–499.
The acquired modules’ state information can assist in implementing Harrou, F., Sun, Y., Taghezouit, B., Saidi, A., Hamlati, M., 2018. Reliable fault detection
MPPT and dynamical reconfiguration of PV arrays so that the identified and diagnosis of photovoltaic systems based on statistical monitoring approaches.
Renew. Energy 116, 22–37.
faulty modules can be repaired or cleared on time. In order to further Hu, Y., Zhang, J., Cao, W., Wu, J., Tian, G., Finney, S., 2015. Online two-section PV array
reduce the investment costs and wiring complexities in large-scale PV fault diagnosis with optimized voltage sensor locations. IEEE Trans. Ind. Electron. 62
arrays, locating fault areas instead of specific modules will be in- (11), 7237–7246.
Huang, J., Wai, R., Gao, W., 2019. Newly-designed fault diagnostic method for solar
vestigated in our future work. Meanwhile, we will devote to identifying photovoltaic generation system based on IV-curve measurement. IEEE Access 7,
the shading faults caused by dust or soiling and so on. 70919–70932.
Jäger-Waldau, A., 2019. Snapshot of photovoltaics—February. Energies 12 (5), 769–776.
Lu, X., Lin, P., Cheng, S., Lin, Y., Chen, Z., Wu, L., 2019. Fault diagnosis for photovoltaic
Declaration of Competing Interest array based on convolutional neural network and electrical time series graph. Energy
Convers. Manage. 196, 950–965.
The authors declare that they have no known competing financial Madeti, S., Singh, S., 2017. Online fault detection and the economic analysis of grid-
connected photovoltaic systems. Energy 134, 121–135.
interests or personal relationships that could have appeared to

296
T. Pei, et al. Solar Energy 201 (2020) 279–297

Mansouri, M., Hajji, M., Trabelsi, M., Harkat, M., Ai-khazraji, A., Livera, A., 2018. An for PV arrays based on frequency analysis. IEEE Access 7, 72050–72061.
effective statistical fault detection technique for grid connected photovoltaic systems Silvestre, S., Da Silva, M., Chouder, A., Guasch, D., Karatepe, E., 2014. New procedure for
based on an improved generalized likelihood ratio test. Energy 159, 842–856. fault detection in grid connected PV systems based on the evaluation of current and
Matam, M., Barry, V., 2018. Improved performance of Dynamic Photovoltaic Array under voltage indicators. Energy Convers. Manage. 86, 241–249.
repeating shade conditions. Energy Convers. Manage. 168, 639–650. Takashima, T., Yamaguchi, J., Ishida, M., 2008. Disconnection detection using earth
Nian, B., Fu, Z., Wang, L., Cao, X., 2010. Automatic detection of defects in solar modules: capacitance measurement in photovoltaic module string. Prog. Photovolt. Res. Appl.
image processing in detecting. In: International Conference on Wireless 16 (8), 669–677.
Communications Networking & Mobile Computing. Takashima, T., Yamaguchi, J., Otani, K., Oozeki, T., Kato, K., Ishida, M., 2009.
Pandey, A., Tyagi, V., Jeyraj, A., Selvaraj, L., Rahim, N., Tyagi, S., 2016. Recent advances Experimental studies of fault location in PV module strings. Sol. Energy Mat. Sol. C.
in solar photovoltaic systems for emerging trends and advanced applications. Renew. 93 (6–7), 1079–1082.
Sustain. Energy Rev. 53, 859–884. Wang, P., Zheng, S., 2010. Fault analysis of photovoltaic array based on infrared image.
Patel, H., Agarwal, V., 2008. MATLAB-based modeling to study the effects of partial Acta Energiae Solaris Sinica 31 (2), 197–202.
shading on PV array characteristics. IEEE Trans. Energy Convers. 23 (1), 302–310. Wu, C., Zhou, D., Li, Z., 2013. Hot spot detection and fuzzy optimization control method
Pei, T., Hao, X., Qu, G., 2018. A novel global maximum power point tracking strategy of PV module. Proc. CSEE 33 (36), 50–61.
based on modified flower pollination algorithm for photovoltaic systems under non- Yahyaoui, I., Segatto, M.E.V., 2017. A practical technique for on-line monitoring of a
uniform irradiation and temperature conditions. Energies 11 (10), 2708–2723. photovoltaic plant connected to a single-phase grid. Energy Convers. Manage. 132,
Platon, R., Martel, J., Woodruff, N., Chau, T., 2015. Online fault detection in PV systems. 198–206.
IEEE Trans. Sustain. Energy 6 (4), 1200–1207. Yi, Z., Etemadi, A., 2017. Line-to-line fault detection for photovoltaic arrays based on
Saleh, K., Hooshyar, A., El-Saadany, E., Zeineldin, H., 2017. Voltage-based protection multiresolution signal decomposition and two-stage support vector machine. IEEE
scheme for faults within utility-scale photovoltaic arrays. IEEE Trans. Smart Grid 9 Trans. Ind. Electron. 64 (11), 8546–8556.
(5), 4367–4382. Zhao, Y., De Palma, J.F., Mosesian, J., Lyons, R., Lehman, B., 2012. Line–line fault ana-
Sevilla-Camacho, P.Y., Zuñiga-Reyes, M.A., Robles-Ocampo, J.B., Castillo-Palomera, R., lysis and protection challenges in solar photovoltaic arrays. IEEE T. Ind. Electron. 60
Muñiz, J., Rodriguez-Resendiz, J., 2019. A novel fault detection and location method (9), 3784–3795.

297

You might also like