Design of an intelligent MPPT based on ANN using
an experimental photovoltaic system data
Sadeq D. Al-Majidi 1,2 Maysam F. Abbod 1 Hamed S. Al-Raweshidy 1
[Link]-Majidi@[Link] [Link]@[Link] [Link]-Raweshidy@[Link]
1
Department of Electronic and Computer Engineering, College of Engineering, Design and Physical Sciences, Brunel University
London, Uxbridge UB8 3PH, United Kingdom
2 Department of Electrical Engineering, College of Engineering, University of Misan, Amarah 62001, Iraq
Abstract- Maximum power point tracking (MPPT) methods are 16% of global production of electricity by 2050s [2]. Unlike,
an essential part in photovoltaic (PV) system design for increasing the major problem is a low efficiency when installing this
the output power of a PV array. Whilst several methods have resource, because the generated power from a PV system
been introduced, the artificial neural network (ANN) is one of the depends upon the irradiance level and temperature operation,
most powerful method for MPPTs due to its less oscillation and which can result in losses of energy of up to 25% [3]. The
fast response. However, accurate training data is a big challenge
to design an optimized ANN-MPPT technique. In this paper, an
most effective method to improve the efficiency of a PV
ANN-MPPT technique based on a large experimental training system is to employ a maximum power point tracking MPPT
data is proposed to avoid the system from having a high training with a PV array, as shown in Fig 1, thereby achieving
error. Those data are collected during one year from maximum power generation under varying weather conditions.
experimental tests of a grid-connected PV system installed at Basically, the MPPT technique is a controller, which generates
Brunel University, London, United Kingdom. The irradiance and an appropriate duty ratio (D) to a DC-DC convertor for the
temperature of weather conditions are selected as the input, and output (current and voltage) and/or input (irradiance and
the maximum power from the PV system as the output of the temperature) of the PV array to achieve continuous maximum
ANN model. To assess the performance, the Perturb and Observe power production. In general, cost, efficiency, lost energy, and
(P&O) and the proposed ANN-MPPT methods are simulated
using a MATLAB/Simulink model for a PV system. The results
type of implementation are the major key issues when aiming
show that the proposed ANN method accurately tracks the to propose the MPPT method for PV systems [4]. Taking these
optimal maximum power point and avoids the phenomenon of into account, many types of MPPT methods have been
drift problem, whilst achieving a higher output power when proposed, which can be divided into two types: classical
compared with P&O-MPPT method. techniques, such as Perturbation and Observation (P&O) [5],
and Incremental Conductance (IC) [6]; and Artificial
Keywords— Artificial Neural Network (ANN), Maximum power Intelligent (AI) techniques, for instance, Fuzzy Logic
point tracking (MPPT), photovoltaic (PV) and Perturb and Controller (FLC) [7], Artificial Neural Network (ANN) [8]
Observe (P&O). Adaptive Neural-fuzzy Interference System (ANFIS) [9].
The P&O-MPPT is a widely technique used for PV-MPPT
I. INTRODUCTION due to its simple implementation and low cost [10]. However,
this faces several problems, such as lower response time, high
Nowadays, the global demand for energy is growing fluctuation, and drift problem phenomenon [11]. Hence,
dramatically due to population growth. Moreover, global several modifications have been proposed to address those
warming phenomenon has been intensified because of the CO2 issues, such as in Refs. [2, 12-14], but they are considered as
emissions from traditional fossil fuels. To address those issues, non-optimal solutions. Consequently, the AI techniques based
many scholars have turned to renewable energies to face the on MPPT have been developed to solve these limitations.
issue of lack of energy in coming years and to minimize the Furthermore, these methods do not need accurate parameters
side effects of burning traditional fossil fuels. The major and complex mathematics when managing the system. The
renewable energy resources are a photovoltaic (PV) systems, FLC-MPPT is one of the most powerful intelligent controllers
wind turbines, hydropower, and geothermal power. However, for a PV system due to its high response and lesser oscillation,
the PV system is one of the most engaging renewable energy when compared with traditional MPPT methods. However, the
resources, owing to its provision of sustainable, clean and safe major disadvantages are the phenomenon of drift problem
energy [1]. Regarding to the International Energy Agency associated with a change in the irradiance level and
(IEA), global energy generation from PV resources will reach temperature operation [7]. This is because it heavily depends
978-1-5386-2910-9/18/$31.00 ©2019 IEEE
on the good knowledge of PV systems, resulting in inaccurate II. PV SYSTEM
membership functions. To address this issue, many
modifications have been proposed, for example, an adaptive The studied PV system comprises a 925 W, 5 modules
and optimized membership function of the traditional FLC- (Sharp NU-S5E3E 185) of PV array installed at the Brunel
MPPT, as evidence in [7, 15-17]. However, the implementation University London, United Kingdom, as shown in Fig. 2. A
is become over complex. The ANN technique is another dynamic simulation model of the PV system is designed under
powerful intelligent method for a nonlinear system, such as a MATLAB/ Simulink to analyse and discus the performances
PV module. The major benefits of using ANN with PV-MPPT of the proposed ANN-MPPT method.
systems are that there is no requirement for knowledge on the
internal system parameters, less computational time and it
provides a compact solution for multivariable problems [18].
Hence, the ANN- MPPT method has a faster response time and
less oscillation around the maximum power point. However,
getting accurate training dataset is a big challenge to design an
optimized ANN-MPPT technique [19]. Therefore, several
types of ANN-MPPT methods have been proposed using
various types of training data, such as in Refs. [20-22].
Fig. 2. The studied PV array installed at Brunel University London
A PV cell is a crucial part in a PV system that converts the
light into electrical energy. In ideal PV cell system, parallel
and series resistances are not included, but in practical case,
they are introduced due to a leakage current and ohmic
resistances as shown in Fig. 3. The main contributor of the
Fig. 1 a scheme diagram of PV system based MPPT.
shunt resistance Rsh is that a p–n junction of Photovoltaic diode
is non-optimal, while the series resistance Rs is the bulk
In this paper, ANN technique is used to determine the resistance of semiconductor material and interconnections.
maximum power point of a PV array based on large real Kirchhoff’s law, as given in equation (1) can find the output
training dataset. Those for the ANN model are collected from current from the solar PV cell:
experimental tests of a PV array installed at Brunel University,
London, United Kingdom. The irradiance level and
temperature operation are selected as the input, and the
reference power (Pref.) of the PV array at the maximum power
point as the output. In the same environmental conditions, where IL is the light generated current is given as equation (2):
actual PV power (Pact.) is measured using a sensed the voltage
and current of a PV Simulink operation. These two power
sources are compared, and the error (e) is given to a
proportional integral (PI) controller to generate the duty ratio
of a DC-DC boost convertor, and this duty ratio is converted to where, G is the solar irradiance level, T is the ambiance
the signal (S) by a Pulse-width modulation (PWM) to adjust temperature operation, ISC is the PV short circuit current, ka is
the operating the maximum power point of the PV array, as the temperature coefficient, TSTC is the ambiance temperature
shown in Fig. 1. The results demonstrate that the proposed for the PV cell under standard test conditions (STC), and Id is
ANN method has the lower tracking time, less oscillation the diode current which is given as equation (3):
around the maximum power point, and the higher output
power. Moreover, it is the most accurate for tracking the
maximum power point and avoiding the phenomenon of drift
problem. The rest of this paper is organized as follows.
Section II covers the PV system, while Section III and section
IV introduce the MPPT using the P&O and ANN algorithms. where I0 is the reverse saturation current of the PV diode, and
In section V, the results are provided and discussed. Finally, Vd is the Voltage across the diode, k is the Boltzmann's
section VI presents the conclusion. constant (1.38×10-23 J/K), q is the electric charge (1.69×10-19
C), and n is the diode factor. The general equation that
describes the P-V characteristic curve of the PV cell is given in
equation (4):
where IPV is the PV output current, and VPV is the PV output
voltage. PV cells are connected in series and parallel to obtain
desired voltage and current respectively for the PV panel, and
then this PV panels are connected in series and/or parallel to (a)
give different configurations of photovoltaic array.
(b)
Fig. 3. Equivalent circuit of a PV cell.
Fig. 4 P-V curve of a PV array under; (a) varying values of irradiance at
constant temperature of 25 C°, (b) varying values of temperature at constant
As shown in Fig. 4, there are unique points on the P-V curve irradiance of 1000 W/m2,
of the PV array, which are known as the maximum power point
(MPP) and the location of those points depends on irradiance
and temperature of weather conditions: the maximum power III. P&O-MPPT ALGORITHM
increases as irradiance level increases, conversely a PV
generator better for low operating temperature than raised one. The Perturb and Observe (P&O) method is widely used for
On top of that, the power generator of the PV array depends PV-MPPT technique due to its simple implementation and low
upon the impedance of the load. When the PV array is cost. The principle work of this method, as shown in Fig. (5),
connected to the load, it drops to a new operating point. To is that it calculating the PV power by using the sensed values
address those issues, power conversion system with MPPT of the voltage and current, and then, it compares with the
technique is employed between PV array and the load or previous PV power and voltage, with the direction of the
invertor, as shown in Fig.1. In general, there are several types algorithm being adjusted accordingly and a reference voltage
of DC-DC converters, the boost DC-DC converter is widely of the boost converter being adjusted as equation (6):
used for PV systems because of its easily adapted MPPT
controller and high efficiency. The DC-DC boost converter is
used to regulate and provide the output voltage that its level is
more than the input voltage. The voltage gain of DC-DC boost
converter is given as equation (5): where Vref.k+1 and Vref.k, are the next and previous
perturbation of the reference voltage respectively, and ΔV is
the incremental increase in the reference voltage. If the
tendency of change in the PV power and PV voltage increase
regarding to an increase in the reference voltage, the control
system moves in the same direction; otherwise the operating
point moves in the opposite direction. The processing work of
where Vo is the output voltage, Vi is the input voltage, and D the algorithm is continued until it reaches to the optimized
is the duty ratio of MPPT controller, which is converted to a MPP point and then it oscillates around it. In general, there are
signal by the PWM to adjust the operating the maximum three main problem facing its operation: a low response time,
power point of the PV array. high oscillation around the MPP point and a phenomenon of
drift problem associated with the input irradiance changing
rapidly. In other hand, this problem is happened when the weather station and PV system are controlled and monitored
input solar irradiance increases only [7]. To address these through a supervisory control and data acquisition (SCADA)
issues, several proposed P&O-MPPT methods have been cum data logger connected to a PC computer terminal and
developed such as a variable step size and modified P&O linked to the internet though Brunel university’s local area
algorithm. However, they are considered non-optimized network (LAN). The 48,000 datasets were collected daily at
solutions to address all of these issues. Consequently, artificial each 5 minutes’ intervals for one year to get the accurate
intelligence techniques based on MPPT method have been training data. This data has been divided into sub-data; 70%,
developed to overcome the limitations of the traditional P&O- 15% and 15% to train, test and validate the ANN network
MPPT method. respectively. To get the optimized activation function in
hidden layer, five neurons is used which achieve less tolerance
of Mean Square Error (MSE) about 0.0075, as shown in Fig.
7. In general, the MPPT technique based on ANN is designed
to solve the limitations of a classical methods. Additionally,
the tracking power of ANN-MPPT has a faster response and
less oscillation under varying weather conditions. However,
accurate training data is a big challenge when designing an
optimal ANN-MPPT method.
Fig. 5. Flowchart of a P&O algorithm
Fig. 6. A block diagram of the ANN mode
IV. ANN-MPPT METHOD
ANN based on MPPT controller has been widely used for
PV systems due to its ability to track the MPP under different
weather conditions. The developed architecture of ANN
model consists of three layers; input, hidden and output as
shown in fig.6. In this work, the input layer consists the
irradiance and temperature of weather conditions, while the
output layer is the reference power of the installed PV array at
the MPP. In the same environmental conditions, PV power of
Simulink operation is measured using a sensed voltage and the
current. These two powers are compared, the error (e) is given
to PI controller to generate the duty ratio of a DC-DC boost
convertor, and this duty ratio is converted to the signal by
PWM to adjust the operating MPP of the PV array. The ranges
of the irradiance and temperature are collected according to
the latitude and longitude of Uxbridge, London, United
Kingdom, which are 51.531 and -0.474, respectively. In the
grid-connected Brunel laboratory, the PV system and weather
station are connected to a Sunny Boy data logger. Both the Fig. 7. Mean squared error versus epochs for the proposed ANN model
Table 1. A comparison of the properties of the ANN and P&O-MPPTs V. SIMULATION RESULTS
MPPT Tracking Oscillation Drift Output
time (s) problem power (W) To evaluate the performance, the P&O and ANN-MPPT
ANN-MPPT 0.06 low avoidance 924.2 methods are simulated using a MATLAB/Simulink model for a
PV system. This PV system consists of a PV array, DC–DC
P&O-MPPT 0.12 High suffering 922.5
boost converter, MPPT micro-controller and a load. The
installed PV array consists of five PV modules connected in
series. The main parameters of the PV module under standard
test conditions are 185W maximum power, 36.21 V optimized
voltage, 5.11 A optimized current, 44.9 V open circuit voltage
and 5.75 A short circuit current. As shown in Fig. 8, the input
irradiance level of the simulation scenario was faster
decreased rapidly from 1000 to 200 W/m2 at 1 to 2 s, and then
increased rapidly from 200 to 1000 W/m2 at 3 to 4 s, and the
temperature operation was kept at 25 °C. As shown in in the
zoom in of Fig. 9(a), the tracking time of the ANN-MPPT
method is the faster when compared to P&O-MPPT method,
being about 0.07 s, 0.12 s, respectively. In addition, the power
Fig.8. the input of solar irradiance level.
tracker of the ANN-MPPT method turned out to be accurate in
finding the right direction, whilst that of the P&O-MPPT
method was lost it when the input irradiance level increased
rapidly.
As a result, the later takes a longer time than MPPT based
on ANN to solve the drift problem, as shown in Fig. 9(a).
Moreover, it has the lesser smooth oscillation around the
optimal MPP for steady-state conditions, thus resulting in less
computation time, as shown in the zoom in of Figs. 9(b).
Alternatively, the fluctuation problem is the higher in the
P&O-MPPT method because of the continuous perturbation of
(a)
the P&O tracker for reaching the optimal MPP. Consequently,
the lost power in ANN-MPPT is lesser than for P&O-MPPT
technique. Hence, the output power of ANN-MPPT and P&O-
MPPT methods, after they have reach the optimal MPP, are
about 924.2 W and 922.5 W, respectively, as shown in the
zoom in of Fig. 9(a). To assess ANN-MPPT method further,
table 1 compares its properties with P&O-MPPT method. As
can be seen, ANN-MPPT has the faster tracking time, the
lesser oscillation around the MPP and the higher output power
generation. Moreover, it is the most accurate in tracking the
optimized MPP and avoiding the drift problem.
(b)
VI. CONCLUSION
An intelligent maximum power point tracking method based
on an artificial neural network using an experimental data of a
PV system has been proposed. The large training dataset is
collected during one year from the experimental testing of the
PV system installed at Brunel University, London, United
Kingdom. The irradiance and temperature of weather
conditions are selected as the inputs, whilst the maximum
(c) power from the PV system as the output of ANN model.
Under the same environmental conditions, actual PV power is
Fig. 9. PV array output for a conventional P&O-MPPT VS the proposed measured using a sensed voltage and the current of a PV
ANN-MPPT under fast change in solar irradiance; (a) power, (b) voltage, and
(c) duty ratio. Simulink operation. These two power outputs are compared,
and the error is given to a proportional integral controller to
generate the signal of a DC-DC convertor by the PWM MPPT for high tracking efficiency of partially shaded PV module,”
Sol. Energy, vol. 163, no. November 2017, pp. 570–580, 2018.
generator, to adjust the operating MPP of the PV array. To [14] M. Killi and S. Samanta, “Modified Perturb and Observe MPPT
sum up, the mythology of collected data and the tuning of Algorithm for Drift Avoidance in Photovoltaic Systems,” IEEE
proposed ANN model are explained. The P&O and the Transactions on Industrial Electronics, vol. 62, no. 9, pp. 5549–
proposed ANN-MPPT methods are simulated using a 5559, 2015.
[15] T. H. Kwan and X. Wu, “Maximum power point tracking using a
MATLAB/Simulink. The results demonstrate that the proposed variable antecedent fuzzy logic controller,” Sol. Energy, vol. 137,
ANN method exhibits higher output power, and no divergence pp. 189–200, 2016.
from the optimal maximum power point during varying [16] O. Guenounou, B. Dahhou, and F. Chabour, “Adaptive fuzzy
weather conditions. controller based MPPT for photovoltaic systems,” Energy Convers.
Manag., vol. 78, pp. 843–850, 2014.
[17] Y. Soufi, M. Bechouat, and S. Kahla, “ScienceDirect Fuzzy-PSO
ACKNOWLEDGMENT controller design for maximum power point tracking in photovoltaic
system,” Int. J. Hydrogen Energy, vol. 42, no. 13, pp. 8680–8688,
2016.
The corresponding author is grateful to University of Misan [18] C. Ben Salah and M. Ouali, “Comparison of fuzzy logic and neural
and the Iraqi Ministry of Higher Education and Scientific network in maximum power point tracker for PV systems,” Electr.
Research for financially supporting the current research. Power Syst. Res., vol. 81, no. 1, pp. 43–50, 2011.
[19] T. Hiyama, S. M. Ieee, and K. Kitabayashi, “Neural Network Based
Estimation of Maximum Power Generation from PV Module Using
Environmental Information" IEEE Transactions on Energy
REFERENCES Conversion,” vol. 12, no. 3, 1997.
[20] F. Almonacid, E. F. Fernández, and P. Rodrigo, “Estimating the
maximum power of a High Concentrator Photovoltaic ( HCPV )
[1] V. Verma, A. Kane, and B. Singh, “Complementary performance module using an Arti fi cial Neural Network,” Energy, vol. 53, pp.
enhancement of PV energy system through thermoelectric 165–172, 2013.
generation,” Renew. Sustain. Energy Rev., vol. 58, pp. 1017–1026, [21] A. Chaouachi, R. M. Kamel, and K. Nagasaka, “A novel multi-
2016. model neuro-fuzzy-based MPPT for three-phase grid-connected
[2] S. D. Al-Majidi, M. F. Abbod, and H. S. Al-Raweshidy. “A photovoltaic system,” Sol. Energy, vol. 84, no. 12, pp. 2219–2229,
Modified P&O-MPPT based on Pythagorean Theorem and CV- 2010.
MPPT for PV Systems” 2018 53rd International Universities Power [22] A. Gupta, P. Kumar, R. K. Pachauri, and Y. K. Chauhan,
Engineering Conference (UPEC), 2018. “Performance Analysis of Neural Network and Fuzzy Logic Based
[3] E. Román, R. Alonso, P. Ibañez, S. Elorduizapatarietxe, and D. MPPT Techniques for Solar PV Systems,” 2014 6th IEEE Power
Goitia, “Intelligent PV Module for Grid-Connected PV Systems,” India Int. Conf., pp. 1–6, 2014.
IEEE Transactions on Industrial Electronics, vol. 53, no. 4, pp.
1066–1073, 2006.
[4] Z. Salam, J. Ahmed, and B. S. Merugu, “The application of soft
computing methods for MPPT of PV system: A technological and
status review,” Appl. Energy, vol. 107, pp. 135–148, 2013.
[5] N. S. D. Souza, L. A. C. Lopes, and X. Liu, “Comparative study of
variable size perturbation and observation maximum power point
trackers for PV systems,” Electric Power System Research, vol. 80,
pp. 296–305, 2010.
[6] M. A. Elgendy, B. Zahawi, S. Member, and D. J. Atkinson,
“Assessment of the Incremental Conductance Maximum Power
Point Tracking Algorithm,” IEEE Transactions on Sustainable
Energy, vol. 4, no. 1, pp. 108–117, 2013.
[7] S. D. Al-Majidi, M. F. Abbod, and H. S. Al-Raweshidy, “A novel
maximum power point tracking technique based on fuzzy logic for
photovoltaic systems,” International Journal of Hydrogen Energy,
2018.
[8] M. Seyedmahmoudian et al., “State of the art artificial intelligence-
based MPPT techniques for mitigating partial shading effects on PV
systems – A review,” Renew. Sustain. Energy Rev., vol. 64, pp.
435–455, 2016.
[9] M. A. Enany, M. A. Farahat, and A. Nasr, “Modeling and
evaluation of main maximum power point tracking algorithms for
photovoltaics systems,” Renew. Sustain. Energy Rev., vol. 58, pp.
1578–1586, 2016.
[10] F. Paz and M. Ordonez, “Zero oscillation and irradiance slope
tracking for photovoltaic MPPT,” IEEE Trans. Ind. Electron., vol.
61, no. 11, pp. 6138–6147, 2014.
[11] V. Kamala Devi, K. Premkumar, A. Bisharathu Beevi, and S.
Ramaiyer, “A modified Perturb & Observe MPPT technique to
tackle steady state and rapidly varying atmospheric conditions,” Sol.
Energy, vol. 157, no. August, pp. 419–426, 2017.
[12] . K. Kollimalla, S. Member, M. K. Mishra, and S. Member, “A
Novel Adaptive P & O MPPT Algorithm Considering Sudden
Changes in the Irradiance,” IEEE Transactions on Energy
Convertions, vol. 29, no. 3, pp. 602–610, 2014.
[13] R. Alik and A. Jusoh, “An enhanced P&O checking algorithm