EL-401: B.

TECH PROJECT-I
MID-SEMESTER
EVALUATION REPORT

Smart PV Fault Detection and Power

Optimisation

Project Mentor
Dr. Anup Kumar Mandpura
Project Members
Khushbu Pahwa (Roll No. 2K16/EL/038)
Mankaran Singh (Roll No. 2K16/EL/041)
Mayank Sharma (Roll No. 2K16/EL/042)
INTRODUCTION:

The rapid growth of the solar industry over the past several years has expanded the significance
of photovoltaic (PV) systems. Fault analysis in solar photovoltaic (PV) arrays is a fundamental
task to increase reliability, efficiency, and safety of PV systems and, if not performed, may not
only reduce power generation and accelerate system ageing but also threaten the availability of
the whole system. Due to the current-limiting nature and nonlinear output characteristics of PV
arrays, faults in PV arrays may not be easily detected. The efficient use of photovoltaic energy
requires the optimization of photovoltaic systems. In fact, the diagnosis of the PV plants is
required in order to maintain optimum performance. The performance of photovoltaic systems
is affected by several circumstances such as shading, dirt and degradation of cells. To overcome
the drop of the PV plant yield, multiple methods favour the detection and localization of any
losses and faults occurring. In order to minimize energy losses in installed photovoltaic system,
the observation and detection of PV system plays a very crucial role. The output power of a
PV system relies mainly on irradiance and surface temperature of PV cells. If some faults exist
in a PV system which result in lower power output, users may ascribe this decrement and think
of it due to presence of low solar irradiance or other changing condition. Even if the users know
that the PV system has a fault condition, they may not able to identify which fault it has, nor
have a method to determine what fault is present. Besides decreasing the generation of output
power, faults in a photovoltaic system also threaten the availability and reliability of the whole
system. Therefore, in an effort to increase the output power to maximum, identifying faults in
a photovoltaic system is extremely important. Various factors such as effect of temperature,
shading, insulation etc. may affect the performance of the PV array.

LITERATURE REVIEW:

In recent years, artificial intelligence algorithms have attracted the attention of many research
scholars. These algorithms mainly include artificial neural networks (ANNs) and other machine
learning methods to detect PV array faults.

Chineetal. and Mekkietal. used an artificial neural network method to detect short circuit
faults and partial shading faults in a PV array.

Dhimish et al. introduced a method to distinguish short circuit faults and partial shading faults
of a PV array based on fuzzy logic. They also proposed a statistical method to detect and locate
accurately different faults types including a PV module fault, the PV strings faults, and a faulty
maximum power point tracker (MPPT) unit.

Harrou et al. presented a statistical monitoring approach to discriminate short-circuit, open-

circuit and shading faults of a PV array. However, these algorithms require a longer
computation time and larger memory size for the process of fault detection.

The mathematical model analysis method compares actually measuring output values with
analytically computing output values to detect the fault status of a PV array. There are some
mathematical modelling approaches for fault detection.

Kang et al. proposed a novel method to diagnose output power lowering in a PV array based
on using the Kalman-filter algorithm. But the effectiveness of these mathematical model-based
fault detection methods depends heavily on the accuracy of the models.
Silvestre et al. and Yahyaoui and Segatto presented a new procedure for automatic fault
detection in grid-connected PV systems based on the evaluation of current and voltage
indicators where the computational complexity is reduced through minimizing the number of
monitoring sensors. However, their method cannot distinguish between degradation faults and
partial shading faults of the PV modules in a PV array.

Sunil Rao et al. presented a Cyber Physical Systems approach to fault detection in
Photovoltaic PV arrays by developing Smart Monitoring Devices (SMDs) with sensors and
actuators that measure current, voltage and temperature.

OBJECTIVE:

In this project, we attempt to do a comprehensive analysis of the most common faults in a PV

array system. We seek to classify the faults by employing various Machine Learning
techniques. We simulate, classify and discuss the electrical faults first. To study the utility and
the purpose of the system in the field of classification and fault detection of PV systems, various
faulty states were established. Faults in the direct current (DC) side of a PV system, such as
open-circuit, short-circuit, degradation and shading faults, are often difficult to avoid and can
result in system energy loss, PV module lifespan reduction, or even serious safety concerns.
Hence, the development of a fault detection method for the PV array faults is particularly
significant for improving the energy conversion efficiency of the PV system, increasing the
service life of the PV modules, and reducing maintenance cost. Six fault types were considered
in the following task. The PV module is subject to different module fault, faults of protection
diodes and different shading configurations.

MODELLING OF A PV CELL (MODULE):

Numerous models of solar cells that efficiently replicate the theoretical PV equations have been
reported in literature. The single diode model of a solar cell is the most common one. This
model contains a photo-generated current source connected in parallel with a diode. To this
model, a serial resistance is added on account of recombination of carriers in the PN junction
region of the diode, material resistance and contact resistance, and a parallel resistance is added
on account of the leakage current at the edge. The single diode model of a solar cell is shown
in Figure 1.

Figure 1. Single diode model of a solar cell

SIMULATION:

In our study, we modelled a 6x7 PV array using the PV block available in the Simscape
Electrical Library in Simulink. The PV array block implements an array of photovoltaic (PV)
modules. The array is built of strings of modules connected in parallel, each string consisting
of modules connected in series. The PV array block is a five-parameter model using a current
source IL (light-generated current), diode (with parameters I0 and nI), series resistance Rs, and
shunt resistance Rsh

Figure 2. Simscape PV Array Block

The diode I-V characteristics for a single module are defined by the equations:

where

Id Diode Current (A)

Vd Diode Voltage(V)
Io Diode Saturation Current (A)

nI Diode ideality factor, a number

close to 1.0

k Boltzman constant=1.3806 x 10-23

J.K-1

q Charge on an electron = 1.6022 x

10-19 C
T Cell Temperature (K)
Ncell Number of cells connected in series
in a module
SIMULATION PARAMETERS:

Figure 3. PV Block Simulation Parameters

i. Parallel strings: Number of strings of series-connected modules that are connected in parallel. The
default value is 40.

ii. Series-connected modules per string: Number of PV modules connected in series in each string. The
default value is 10.

iii. Maximum Power (W): Power obtained at maximum power point (Vmp, Imp). Pmax is computed as
Pmax = Vmp × Imp. The default value is 213.15 W.

iv. Cells per module (Ncell): Number of cells per module. The default value is 60.

v. Open circuit voltage Voc (V): Voltage obtained when array terminals are left open. The default value
is 36.3V.

vi. Short-circuit current Isc (A): Current obtained when array terminals are short circuited. The default
value is 7.84 A.

vii. Voltage at maximum power point Vmp (V): Voltage at maximum power point. The default value is
29 V.

viii. Current at maximum power point Imp (A): Current at maximum power point. The default value is
7.35 A.

ix. Temperature coefficient of Voc (%/deg.C): Defines variation of Voc as a function of temperature. The
open-circuit voltage at temperature T is obtained as VocT = Voc (1 + beta_Voc(T–25)), where Voc is the
open-circuit voltage at 25 degrees C, VocT is the open-circuit voltage at temperature T (in degrees C),
beta_Voc is the temperature coefficient (in %/degrees C), and T is the temperature in degrees C. The
default value is -0.36099 %/deg.C.

x. Temperature coefficient of Isc (%/deg.C): Defines variation of Isc as a function of temperature. The
short-circuit current at temperature T is obtained as IscT = Isc (1 + alpha_Isc(T–25)), where Isc is the
short-circuit current at 25 degrees C, IscT is the short-circuit current at temperature T (in degrees C),
alpha_Isc is the temperature coefficient (in %/degrees C) and T is the temperature in degrees C. The
default value is 0.102 %/deg.C.
 xi. Irradiances (W/m2): This parameter is available only if Display I-V and P-V characteristics of is set
to one module @ 25 deg.C & specified irradiances or array @ 25 deg. C & specified irradiances. Enter
a vector of irradiances in W/m2. The default value is [ 1000 500 100 ].

xii. T_cell (deg. C): This parameter is available only if Display I-V and P-V characteristics of is set to
array @ 1000 W/m2 & specified temperatures. Enter a vector of temperatures in degrees C. The default
value is [45 25] deg. C.

Figure 4. I-V and P-V characteristics of the PV array module under STC

We simulate each fault one by one on Simulink, and generate a dataset with 200 points of each
fault

1) Partial Shading Fault: The shading patterns can be very complicated due to non-
uniform insolation. Under partial shading conditions, the short-circuit current for two
cases remains identical, while the open-circuit voltage slightly decreases with the
increase in the number of shaded modules. I-V curves of all shaded groups have
multiple steps, while P-V curves of shaded groups are characterized by multiple peaks,
whose number is equal to the number of solar insolation levels received by string,
respectively. When the PV array is partially shaded, the MPP current of the PV array
declines obviously, but the short-circuit current and the open-circuit voltage of the PV
array are basically invariant.

Figure 5. I-V and P-V characteristics of the PV array module with partial shading
 2) Bypass Diode Fault: This fault occurs because of short circuit of the bypass diode in
case of incorrect connection. If one full module is shorted by bypass diode, the
maximum power and Voc of the PV array drops significantly and short-circuit current
remains the same as other normal strings.

Figure
6. I-V and P-V characteristics of the PV array module with bypass diode fault

3) Bridge Fault: This fault is described as low- resistance connection between two points
of different potential in string of module or cabling. Bridging faults usually involve
reduced array voltage (Voc) but have much small reduction in array current (Isc). The
fault with larger voltage difference between two fault points will lead to larger reduction
in Voc and Impp and Vmpp.

Figure 7. I-V and P-V characteristics of the PV array module with bridge fault

4) Temperature fault: The output power of a PV system decreases as the module

temperature increases. An increasing cell temperature at constant irradiance causes a
reduction in VOC at a rate approximately of 2–2.5 mV/C. The short-circuit current, on
the other hand, remains nearly constant.

Figure 8. I-V and P-V characteristics of PV array

module with Temperature fault
 5) Complete Shading fault: The short-circuit current varies nonlinearly with irradiance.
It is observed that with the increase in the solar irradiance, the cell-voltage and cell-
power increases.

Figure 9. I-V and P-V characteristics of the PV array module with

complete shading

6) Short Circuit Fault: A short-circuit fault is mainly due to bad wiring in a PV string or
between PV strings. Additionally, aging, vibration and abrasion of PV modules are also
the significant sources of short-circuit faults. When a short circuit fault appears, the
open-circuit voltage of the PV array reduces rapidly

Figure 10. I-V and P-V characteristics of the PV array

module with short circuit fault

FAULT CLASSIFICATION: Up until mid-semester evaluation, we have used one machine

learning technique (Decision Tree Learning) to classify our faults.
Steps undertaken -
1) Data Pre-processing: After the dataset has been imported, we use MinMax
Normalization, which is a normalization strategy which linearly transforms x to y = (x-
min)/(max-min), where min and max are the minimum and maximum values in X, where
X is the set of observed values of x. This means, the minimum value in X is mapped to
0 and the maximum value in X is mapped to 1. This means that the entire range of
values of X from min to max are mapped to the range 0 to 1. In our project, for example,
Pmax and Temperature are in very different ranges. When we do further analysis, Pmax
 will intrinsically influence the result more due to its larger value. But this doesn’t
necessarily mean it is more important as a predictor. That is why we normalize the data.

2) Algorithm Implementation: Decision Trees are a type of Supervised Machine

Learning (that is you explain what the input is and what the corresponding output is in
the training data) where the data is continuously split according to a certain parameter.
The tree can be explained by two entities, namely decision nodes and leaves. The leaves
are the decisions or the final outcomes. And the decision nodes are where the data is
split. We have used the CART (Classification and Regression Trees) Algorithm that
splits data according to a criterion known as ‘Gini Index’.
Gini Index measures the degree or probability of a particular variable being wrongly
classified when it is randomly chosen. If all the elements belong to a single class, then
it can be called pure. The degree of Gini index varies between 0 and 1, where 0 denotes
that all elements belong to a certain class or if there exists only one class, and 1 denotes
that the elements are randomly distributed across various classes. A Gini Index of 0.5
denotes equally distributed elements into some classes
𝑛

𝐺𝑖𝑛𝑖 = 1 − ∑(𝑝𝑖 ) ^2
𝑖=1

where pi is the probability of an object being classified to a particular class.

3) Validation and Performance Evaluation: Cross-validation is a resampling procedure
used to evaluate machine learning models on a limited data sample. It is primarily used
in applied machine learning to estimate the skill of a machine learning model on unseen
data. That is, to use a limited sample in order to estimate how the model is expected to
perform in general when used to make predictions on data not used during the training
of the model.
The k-Fold Cross Validation has a single parameter called k that refers to the number
of groups that a given data sample is to be split into. When a specific value for k is
chosen, it may be used in place of k in the reference to the model, such as k=10
becoming 10-fold cross-validation.
The general procedure is as follows:
1. Shuffle the dataset randomly.
2. Split the dataset into k groups
3. For each unique group:
a) Take the group as a hold out or test data set
b) Take the remaining groups as a training data set
c) Fit a model on the training set and evaluate it on the test set
d) Retain the evaluation score and discard the model

4. Summarize the skill of the model using the sample of model evaluation scores
The evaluation scores used in this project can be explained by an understanding of the
confusion matrix. As the name suggests, it is a matrix which describes the complete
performance of the model.
● True Positives: The cases in which we predicted
YES and the actual output was also YES.
● True Negatives: The cases in which we predicted
NO and the actual output was NO.
● False Positives: The cases in which we predicted
YES and the actual output was NO.
● False Negatives: The cases in which we predicted
NO and the actual output was YES.
Based on TP, FP, FN and TN, the following evaluation scores are used:
𝑻𝑷+𝑻𝑵
a) Accuracy:
𝑻𝑷+𝑭𝑵+𝑻𝑵+𝑭𝑷

𝑻𝑷
b) Precision:
𝑻𝑷+𝑭𝑷

𝑻𝑷
c) Recall:
𝑻𝑷+𝑭𝑵

𝟐∗𝑷𝑹𝑬𝑪𝑰𝑺𝑰𝑶𝑵∗𝑹𝑬𝑪𝑨𝑳𝑳
d) F1-Score:
𝑷𝑹𝑬𝑪𝑰𝑺𝑰𝑶𝑵+𝑹𝑬𝑪𝑨𝑳𝑳

The results of the 5-fold cross validation are summarised in the table below:

K-Fold Accuracy Precision Recall F1-Score

Validation

1 95.83% 0.96 0.96 0.96

2 95.00% 0.95 0.95 0.95

3 94.58% 0.95 0.95 0.95

4 96.66% 0.97 0.97 0.97

5 97.50% 0.97 0.97 0.97

Averages 95.91% 0.96 0.96 0.96

FUTURE SCOPE:
Moving ahead, we would like to classify the same dataset using other machine learning
techniques such as Random Forest, K-Nearest Neighbour and Artificial Neural Networks. We
would then perform a comparative analysis of the results and choose the algorithm that best
classifies the PV faults. We are also planning to expand the dataset by including another PV
fault in the dataset, the Line-to-Line Fault, in which modules of the same string are shorted
together. We are also looking towards the feasibility of validating theoretical data with practical
data that is actually collected from a solar panel installation.