2020 International Conference for Emerging Technology (INCET)

Belgaum, India. Jun 5-7, 2020

Comparison of Maximum Power Tracking using

Artificial Intelligence based optimization controller
in Photovoltaic Systems
Ayush Verma1, Shivam Yadav2, Ankita Arora3, Kartik Singh4
1,2,3,4
Dept of Electrical Engineering, Delhi Technological University, Delhi, India
1
[email protected], [email protected], [email protected], [email protected]

Abstract—This paper analyzes performance of Artificial under the varying weather conditions and solar irradiations
Intelligence based optimization controller for the comparative and thus delivering maximum power, as required for the
study of maximum power point tracking (MPPT) in PV experimental and study purpose.
Systems. Artificial Neural Network (ANN) and Particle Swarm
Optimization (PSO) control methods are the two such II. SYSTEM DESIGN
techniques used, and are simulated in MATLAB-Simulink
Basic layout and configuration of the photovoltaic
using Trina Solar TSM-250PD05.08. The simulation results
suitably depict the performance of these methods on the basis
system, which is being used to transfer the maximum power
of some parameters like their rise time, settling time, time to the load are depicted in Fig. 1. First component is the PV
taken to reach maximum power point and their efficiency. It is module, which collects and transmits the solar radiations to
found that maximum power point is tracked in PV systems the DC-DC converter, whose main feature is to convert a DC
with greater efficiency using PSO as compared to ANN. voltage from one level to another. Here, output power varies
due to variations in the temperature and irradiance
Keywords—Maximum Power Point Tracking (MPPT); intensities. Hence, in order to extract maximum power from
Photovoltaic System; Artificial Neural Network (ANN); Particle the PV module, different and effective maximum power
Swarm Optimization (PSO); point tracking (MPPT) algorithms have to be used. These
algorithms extract the maximum power at any point to
I. INTRODUCTION overcome the challenges imposed due to the variations in the
Renewable energy is a kind of clean energy source which physically collected radiations. Finally, the output is passed
has the capability of meeting our increased energy demands. to the PWM controller and the load segment.
These energy sources have attacked much attention as fossil
fuels are depleting at a very high rate. Among them, solar
energy is present in ample amount and does not have any
harmful effects on the environment. Solar energy conversion
system requires negligible maintenance and does not produce
any noise.
A photovoltaic (PV) system basically converts sunlight
into electricity whose generated DC power is primarily Fig. 1. PV System’s General Configuration
dependent on the factors like solar irradiance and
temperature. Maximum power point trackers (MPPT)are III. MATHEMATICAL SOLAR ARRAY MODELLING
seldom deployed in a way to enhance the system's efficiency
A solar cell is considered as the basic building block of
and performance. This is achieved by continually
the PV Arrays. Several solar cells are organized and
transferring the maximum power, irrespective of varying
connected together in various arrangements like the series
weather conditions, and to overcome the fluctuations in the
and parallel combinations to generate electrical energy.
generated power as the photovoltaic system varies
throughout the day.
There has been a plethora of maximum power point
tracking (MPPT) algorithms and methods. Some of these are:
Perturbation and Observation/ Hill Climbing, Constant
Voltage Method Fuzzy logic, Neural Network and
Incremental Conductance. Each of the algorithm comes with
its characteristics, constraints, challenges, weaknesses.
This paper aims to present a comparative study of the Fig. 2. Equivalent Circuit of a Solar Cell
performance of MPPT methods using Artificial intelligence
approaches i.e. ANN (Artificial Neural Networks) and PSO (1)
(Particle Swarm Optimization) based controller. Both of the
Where, Isc represents the current generated by photon,
above methods are well suited for our study as they operate
diode current is represented by Id and the shunt resistor is
at a swift rate as compared to other conventional methods.
represented by Rp.
Moreover, they precisely track the maximum power point

978-1-7281-6221-8/20/$31.00 ©2020 IEEE 1

Authorized licensed use limited to: Universitas Brawijaya. Downloaded on January 11,2024 at 07:28:54 UTC from IEEE Xplore. Restrictions apply.
 (2)
V signifies the PV module's output voltage, I signifies the
PV module's output current, series resistance is represented
by Rs , and shunt resistance is represented by Rp. q is
electron charge (and is equal to numeric value
). Number of cells connected in series are
denoted by Ns, while number of cells connected in parallel
are denoted by Np. Io denotes the saturation current, T (in Fig. 4. ANN Diagram
Kelvin) denotes the cell temperature, K denotes the
The ANN network's input layer is fed with the PV
Boltzmann constant ( J/K).
module parameters like PV voltages and PV currents, along
The typical I-V and P-V characteristic curve of a with some environmental factor like irradiance, temperature,
photovoltaic array is generally non -linear in nature. These angle of incident radiation, day time, the position of the sun,
two I-V and P-V characteristic curves are depicted in Fig. 3. cloudy factors, or any of their combinations. The Feed-
Forward model is then trained using the Backpropagation
Levenberg-Marquardt algorithm. Various link weights are
adjusted during the training process for reducing the error
term during the ack propagation step. The trained model is
then used to track the maximum power as required.

Fig. 3. Typical I-V and P-V characteristics curve of a PV module

Fig. 5. Simulink Model of Artificial Neural Network
The varying temperature and irradiation influences both
these curves, hence a significant impact is laid on the Duty cycle is produced as the output of the ANN model
operating point as well. It has been observed and concluded and is thus used to track the maximum power. This process is
that there is a single operating point at which the maximum depicted in Fig. 5. Our model comprises of ten neurons in the
power is delivered to the load. And this point is regarded as hidden layer. Neurons in the hidden layer uses the tangent
the maximum power point (MPP). The voltage and current sigmoid function as their activation function. However,
corresponding to the MPP are known as MPP Voltage (Vm) linear activation function is used in the neurons of the output
and MPP Current (Im) correspondingly. The intersection layer.
point between the I-V curve and load line of PV array is
referred to as the single operating point.
MPP controller is a DC-DC converter. Maximum power
tracking in MPP is reached by changing the output voltage
by varying the duty cycle of the converter. The above
discussion supports our study and analysis with the well
proven and established theoretical concepts.
IV. MAXIMUM POWER POINT TRACKING
MATLAB-Simulink has been used to simulate the ANN
and MLP techniques, which have been utilized in our
comparative study. MATLAB is a powerful tool to simulate Fig. 6. Neural Network Structure
complex mathematical computations and graphically
visualize the algorithms as it progresses. The method used to analyze the performance of the
network and measure the errors component is the mean
A. ANN based on data collected from the Perturbation and square error (MSE). Its equation representation is as follows:
Observation (P&O) method
Artificial Neural Network (ANN) has the ability to learn
based on its experiences and mistakes. It has numerous (3)
application and practical uses, and well-established
theoretical references. A general ANN configuration consists Where, the target value to be achieved at sample k is
of three types of layers. The first layer is the input layer, denoted by and the actual output obtained at sample k is
which is being supplied with the initial parameters. These denoted by . The number of training samples used to
parameters and features are then processed to extract the carry out the training process are denoted by N.
high-level and low-level features and then passed to the
output layer, which presents them as the final output. This 200001data is collected from Perturbation Observation
configuration is depicted below in Fig. 4. (P&O) algorithm out of which 70% of the data is used to

2
Authorized licensed use limited to: Universitas Brawijaya. Downloaded on January 11,2024 at 07:28:54 UTC from IEEE Xplore. Restrictions apply.
train the Neural Network. Rest 15% is used for testing
purpose and 15% for carrying out the validation process. The
data samples used for the process of training are summarized
in Table I.

TABLE I. TRAINING DATA USED

PV PV DUTY
CURRENT VOLTAGE CYCLE
4.01907302 71.216894 0.180875
4.06050779 71.2725545 0.18075
3.95174891 71.4092727 0.180875
3.82814333 71.6059525 0.181
3.61957079 71.8494836 0.181125
3.39247008 72.1256269 0.18125
3.11421112 72.424077 0.181375
2.82528021 72.735061 0.1815
2.50667102 73.0512071 0.181625
2.18876795 73.3662453 0.18175

Based on the above experimental results, the Mean Fig. 8. Regression Plot
Square Error (MSE) value obtained is 1.091e-02 which
has also been depicted in Fig. 7. B. Particle Swarm Optimization (PSO) algorithm
Evolutionary computing and algorithm have been
developed in the recent few years and have proved to be very
efficient for the optimization problems. One such algorithm
is the Particle swarm optimization (PSO). This algorithm
was proposed by Eberhart and Kennedy as an alternative
method to Genetic Algorithm (GA) and inspired by the
social behavior of the bird swarms. This algorithm has an
ability to solve optimization problems efficiently in
multidimensional space that too with high convergence rate.
It begins with a random solution, finds the best solution
through iteration, and evaluates the of the quality solution
through fitness. The motivation behind the PSO is based on
the common swarm intelligence. The PSO algorithm is used
to guide the search towards the global optimum solution
point.
The equations of a standard PSO algorithm are given in
Eq. 3, Eq. 4 and Eq. 5 respectively:
Fig. 7. Error Histogram of ANN
(3)
There are various advantages of the proposed algorithm.
One of them is that it requires less amount of time for (4)
training the data obtained from P&O algorithm. Although
there is a disadvantage associated with it, i.e. it requires huge (5)
amount of memory for the computation purpose. Where,
The regression plot obtained using the data extracted by
Perturbation and Observation MPPT algorithm is as shown represents the inertia weight
in Fig. 8.
are the acceleration coefficients
The ideal fit has the regression coefficient R as 1. But in
practical scenarios, R=1 is never achieved. However, the fits are the two random numbers between (0,1)
that yield good enough results have R>0.95. Our training set
has the value of R as 0.97718, whereas that of Validation set represents particle's velocity in the current iteration
and the Testing set is 0.97733 and 0.97637 respectively. Our
overall regression comes as R = 0.97708. The straight line represents particle's velocity in the subsequent
passing through the origin is the perfect fit line, obtained in iteration
ideal scenarios. For practical purposes the outliers, shown by represents particle's position in the current iteration
the black circles should lie along the length of the perfect fit
line. = particle position in the subsequent iteration

represents the best fitness value (individually) of the

particle

3
Authorized licensed use limited to: Universitas Brawijaya. Downloaded on January 11,2024 at 07:28:54 UTC from IEEE Xplore. Restrictions apply.
 represents the fitness of all particles (global best value of ensures a good exploitation but with a very slow
fitness) convergence rate.
The values of the parameters used for the implementation V. MATLAB SETUP FOR TRACKING MAXIMUM
of the proposed PSO algorithm are shown in Table. 2. POWER
TABLE II. PARAMETERS USED FOR PSO SIMULATION MATLAB Simulink model has been used to track the
maximum power. It comprises of following components: a
Parameters Values Solar PV module, DC-DC Boost Converter, Pulse Width
Number of particles 3 Modulation (PWM)Controller and Maximum Power
Maximum Number of iterations 20 algorithm as already explained and depicted in Fig. 1. The
.7 Artificial Intelligence techniques that is ANN and PSO are
used to simulate the model. The parameters of the PV
.1
module are summarized in Table 3. The model consists of
.7 one parallel string and two series connected modules. The
.75 parameters selected for DC-DC Boost Converter are L =
2.9e-4 H, Cin=1.875e-5F ,Cout=1.875e-5 F and fs=10 Khz.
1.5
Simulation is accomplished at a max step size of 5e-06.

TABLE III. PV MODULE PARAMETERS

Parameters Values
Pmpp 249.86 W
Voc 37.6 V
Isc 8.55 A
Vmpp 31 V
Impp 8.06 A

VI. RESULT AND DISCUSSION

ANN and PSO algorithm are proved to be efficient and
effective to track the maximum power, even under the
varying weather conditions like solar irradiations. Further,
the behavior and performance of the entire system has been
enhanced by using the above algorithms, and the same can
be observed by studying the waveforms of these algorithms.
For the ease of study, the temperature is kept constant at 25
°C. The simulating model is visually presented in Fig. 10.
and Fig. 11.

Fig. 9. PSO method flowchart

The main feature of the PSO technique that attracts the

interest of the researchers is its simplicity and accuracy. The
inertia weight plays a vital role in the performance of the
PSO because of its role in controlling the search process and
ensuring a balance between the exploration and exploitation
phases of the required search process. Large value of
enables particles to explore the search area whereas small
Fig. 10. Irradiance Signal

4
Authorized licensed use limited to: Universitas Brawijaya. Downloaded on January 11,2024 at 07:28:54 UTC from IEEE Xplore. Restrictions apply.
Fig. 11. Simulink Model of (a) ANN and (b) PSO

Result obtained using ANN-based controller 0.0021 sec with an efficiency of 99.55%. Thus, this
technique shows better and improved efficiency as well as it
Fig. 12. illustrates the simulation result when the PV shows improvement in rise time.
module is controlled using the ANN algorithm, and its
maximum power point is tracked. This algorithm requires a
large amount of memory since the training of ANN has to be
done, and its rise time comes out 0.003 sec with an efficiency
of 98.69 %.

Fig. 13. PSO Result

A. Comparison between ANN and PSO algorithm

The results obtained from Artificial Intelligence-based
control algorithms, ANN and PSO have been compared. It
Fig. 12. ANN Result
has been found the maximum power tracked using PSO
algorithm is greater than the power tracked using ANN
Fig. 13. illustrates the simulation result when the PV algorithm keeping the experimental parameter fixed in both
module is controlled using PSO algorithm and its maximum the cases. Thus, the efficiency of the system is enhanced in
power point is tracked. The rise time of this comes out to be PSO controlled system.

5
Authorized licensed use limited to: Universitas Brawijaya. Downloaded on January 11,2024 at 07:28:54 UTC from IEEE Xplore. Restrictions apply.
 TABLE V. COMPARISON RESULTS FOR TIME TAKEN TO TRACK MPP

MPPT ALGORITHM TIME TAKEN TO TRACK MPP

(s)

ANN 0.0072

PSO 0.0053

VII. CONCLUSION
In this paper, we have successfully used and studied the
applications of Artificial Intelligence Techniques i.e.
Artificial Neural Network (ANN) and Particle Swarm
Optimization (PSO) control methods in the field of
maximum power tracking. Both the Artificial Intelligence
Fig. 14. Overlapping waveforms of PSO and ANN technique produces a satisfactory output. Many factors have
been used to evaluate and compare both the maximum power
TABLE IV. COMPARISON OF RESULT OBTAINED tracking (MPPT) algorithms . These parameters include:
settling time and rise time before reaching MPP, along with
MPPT RISE SETTLING MAXIMUM EFFICIENCY the time taken to track maximum power point (MPP),
ALGORITHM TIME TIME POWER (in %) computed efficiency.
(in sec) (in sec) TRACKED
(in W) The efficiency of ANN controller comes out to be 98.69
% and the power delivered is 493.2 Watt. Whereas, PSO has
ANN 0.0030 0.0058 493.2 98.69 an efficiency of 99.55 %& and power delivered is 497.5
PSO 0.0021 0.0036 497.5 99.55 Watt. Thus, PSO has been proven to be more efficient as
well as effective method to track the maximum power, and
the maximum power point of the PV System. Also, PSO
significantly improved the efficiency of the maximum power
tracking model with easier and faster computations.
REFERENCES
[1] Y. Hu, J. Liu and B. Liu, "A MPPT Control Method of PV System
Based on Fuzzy Logic and Particle Swarm Optimization," 2012
Second International Conference on Intelligent System Design and
Engineering Application, Sanya, Hainan, 2012, pp. 73-75.
[2] S. Gavali and A. Deshpande, "Particle swarm based optimization
algorithm for maximum power point tracking in photovoltaic (PV)
systems," 2017 2nd IEEE International Conference on Recent Trends
in Electronics, Information & Communication Technology
(RTEICT), Bangalore, 2017, pp. 1583-1586.
[3] Y. Liu, D. Xia and Z. He, "MPPT of a PV system based on the
(a) particle swarm optimization," 2011 4th International Conference on
Electric Utility Deregulation and Restructuring and Power
Technologies (DRPT), Weihai, Shandong, 2011, pp. 1094-1096.
[4] Arora and P. Gaur, "AI based MPPT methods for grid connected PV
systems under non linear changing solar irradiation," 2015
International Conference on Advances in Computer Engineering and
Applications, Ghaziabad, 2015, pp. 542-547.
[5] R. R. Bholane and P. S. Babu, "Grid connected PV system using FB-
PSO," 2018 International Conference on Smart Electric Drives and
Power System (ICSEDPS), Nagpur, 2018, pp. 93-98.
[6] M. Merchaoui, A. Sakly and M. F. Mimouni, "Improved fast particle
swarm optimization based PV MPPT," 2018 9th International
Renewable Energy Congress (IREC), Hammamet, 2018, pp. 1-7.
[7] S. Samal, P. K. Barik and S. K. Sahu, "Extraction of maximum power
from a solar PV system using fuzzy controller based MPPT
technique," 2018 Technologies for Smart-City Energy Security and
Power (ICSESP), Bhubaneswar, 2018, pp. 1-6.
[8] R. Sankar, S. Velladurai, R. Rajarajan and J. A. Thulasi, "II. PV
(b)
system description: Maximum power extraction in PV system using
Fig. 15. Comparison analysis of ANN and PSO on the basis of (a) Rise fuzzy logic and dual MPPT control," 2017 International Conference
time (b) Settling time on Energy, Communication, Data Analytics and Soft Computing
(ICECDS), Chennai, 2017, pp. 3764-3769.

6
Authorized licensed use limited to: Universitas Brawijaya. Downloaded on January 11,2024 at 07:28:54 UTC from IEEE Xplore. Restrictions apply.