An Artificial Neural Network Based MPPT

Algorithm For Solar PV System

Jyothy Lakshmi P.N
Dr. M R Sindhu
Department of Electrical and Electronics Engineering Department of Electrical and Electronics Engineering
Amrita School of Engineering, Amrita Vishwavidyapeetham Amrita School of Engineering, Amrita Vishwavidyapeetham
Coimbatore, India Coimbatore, India
jyothyammu25@gmail.com sindhumadassery@gmail.com

Abstract—The need of renewable energy integration with power The objective of this paper is to develop an artificial
system is shooting up day by day. Solar PV generation has an network based MPPT algorithm and to compare with other
important role for battery charging, grid tied applications etc. In peak power point tracking algorithms.
order to intensify output power of a solar photovoltaic
arrangement, it is imperative to find the maximum possible II. EQUIVALENT CIRCUIT AND CHARACTERISTICS OF PV
energy harvest from photovoltaic panel. In this paper Maximum A solar cell consists of a current source parallel with
Power Point Tracking (MPPT) controller for solar photovoltaic a diode [3]. Since solar cell is not an ideal one, a shunt and a
system is developed by practicing artificial neural network series resistance are connected. Series connected resistance is
(ANN). Also the performance of an ANN based MPPT controller
to oppose the electron flow. The equivalent circuit diagram of a
is compared with Conventional MPPT methods. In particular hill
climbing method (perturb & observe), Incremental Conductance solar cell is shown in Fig. 1
method and fractional open circuit voltage method. Simulations
are done by using MATLAB/SIMULINK to analyze results.

Keywords— Artificial neural networks, DC-DC boost converters,

Maximum power point tracker algorithms, Solar PV panel.

I. INTRODUCTION
Due to environmental and economic benefits solar
photovoltaic (PV) system is considered as more efficient over
the other renewable energy sources [1]. Since photovoltaic Fig. 1. Equivalent circuit of a solar cell
electricity generation is clean, inexhaustible, pollution less and ౒
ሺ ౚሻ
requires less maintenance they are the most promising I = ୱୡ Ǧ୓ ሺ‡ ౡ౐ െ ͳሻሺͳሻ
renewable energy source. The main advantage of using solar
Rearrange equation (1),
PV system is that it does not consist of any moving parts and it
does not produce any noise. Solar photovoltaic array converts ౒శ౅౎౩
I =Ȃ ୓ (‡ሺ ౤ౡ౐
ሻ
െ ͳ)+ୱୡ (2)
solar energy obtained from sun to electrical energy. The
efficiency of solar PV system is low while comparing with Where, I is the solar cell current, V is the solar cell voltage, T
other renewable energy sources but it would be more beneficial is the temperature in Kelvin and n is the ideality factor. A
if it can track maximum power continuously for any change in typical I-V and P-V curve of a solar PV cell can be seen in Fig.
environmental conditions like solar emanation and 2.
temperature. That is why the concept of tracking maximum
possible power from solar PV arrangement is established. The three important parameters in P-V and I-V curve is
Therefore it is essential to track MPP online. Thevenin voltage (୭ୡ ), short circuit current (ୱୡ ) and
maximum power point (MPP) [4]. At both ୭ୡ andୱୡ points
Solar power mainly depends on temperature and solar the power generated is zero. The short circuit current ሺୱୡ ሻ is
emanation conditions [2]. For slight change in solar emanation defined as it is the current at V = 0 and Thevenin voltage is
and temperature values maximum possible energy extraction defined as it is the voltage at ୱୡ = 0. The maximum energy
from solar PV arrangement is different. MPP can be obtained harvest by solar cell is only at a unique operating point of the
from PV curve of PV arrangement. DC-DC boost converter P-V, I-V characteristic, where the power is maximum at that
helps to achieve MPP of PV array. The most popular MPPT point, known as the MPP and can be seen in Fig. 2.
algorithms are hill climbing method, Incremental Conductance
(IC) method, fractional open circuit voltage (FOC) method
fuzzy logic based MPPT, ANN based MPPT etc. [5].

978-1-5386-3695-4$31.00 2018
c IEEE 375
 III. REVIEW OF MAXIMUM POWER POINT TRACKING
ALGORITHMS
The different algorithms found to track the MPP of solar
PV system are:
1) Hill climbing method (P&O)
2) Incremental Conductance method (IC)
3) Fractional open circuit voltage method (FOC)
4) Neural networks
A. Perturb & observe method(P&O)
Fig. 2. I-V and P-V characteristics of solar PV system Since P&O is simple and easy to implement, it is the
commonly using technique [5]. The peak power point is
ୢ୔ ୢ୔
obtained when the condition = 0. The descent ( ) of the
ୢ୚ ୢ୚
power can be written as follows

ୢ୔ ୔ሺ୬ሻି୔ሺ୬ିଵሻ
(n) = (3)
ୢ୚ ୚ሺ୬ሻି୚ሺ୬ିଵሻ

Here, P(n) = V(n)I(n) (4)

The basic working of P&O method is if ᇞP is positive and
ᇞV is positive then keep on climbing towards MPP. If ᇞP is
positive and ᇞV is negative then climb towards MPP but in
the opposite direction. Similarly if ᇞP is negative and ᇞV is
positive then climb towards MPP from opposite direction. If
Fig. 3. V-I and V-P characteristics at constant temperature ᇞP is negative and ᇞV is negative, keep climb in the same
(25°C) and various irradiance conditions direction of MPP to reach maximum power point by
Two important factors to be considered are solar comparing consecutive samples of voltage and power.
temperature and irradiance. The effect of the emanation on the
characteristics of solar PV entity is shown in Fig. 3. The B. Incremental Conductance (IC) algorithm
output current of solar photovoltaic panel is directly The idea of IC algorithm is that measuring and comparing
proportional to the irradiance. Therefore, every increase in the values of incremental conductance (ᇞI/ᇞV) and
solar irradiance level lead to rise in the output current of solar instantaneous conductance (I/V) of solar PV module to find
PV array. Also ୱୡ of solar PV is directly comparable to output maximum power point of the system [6]. The basic working
current of solar PV array. Therefore short circuit current is ୢ୔
theory of IC method is that = 0 at MPP.
also directly comparable to emanation level. For this reason ୢ୚
V-I characteristics changes according to irradiance level Then,
(There is no effect in open circuit voltage).Change in
ୢ୔ ୢሺ୚୍ሻ ୢ୍ ୢ୚ ୢ୍
temperature will affect the change in voltage as shown in Fig. = =  + = +I (5)
ୢ୚ ୢ୚ ୢ୚ ୢ୚ ୢ୚
4.
At maximum power point, equation (5) will be permutated as,
୍ ୢ୍
െ = (6)
୚ ୢ୚
୍ ୢ୍
It is clear that the maximum power point is െ = . If dI<0
୚ ୢ୚
solar irradiance is less and voltage decreases. If dI>0 means
solar irradiance is high and voltage increases to reach
ୢ୍ ୍
maximum power point. Similarly if >  then voltage
ୢ୚ ୚
increases and vice versa. IC method will give better result than
P&O method.
C. Fractional open circuit voltage method (FOC)
The tracking of maximum power using FOC is by using
Fig. 4. V-I and V-P characteristics at constant irradiation (1 Thevenin voltage of solar PV. Thevenin voltage ୭ୡ of solar
kW/ଶ ) and various temperature conditions PV panel has a direct relationship to the maximum power
point voltage ୫୮୮ for any variation of temperature and
irradiance value.
୫୮୮ = k*୭ୡ (7)

376 4th International Conference on Electrical Energy Systems (ICEES)

The value of k depends on the solar PV panel type,
configuration and specification. k is the ratio of ୫୮୮ to the
୭ୡ . The advantages of using FOC method is that this method
is quite simple and robust, and provides almost approximate
MPP tracking.
D. Artificial neuralnetwork (ANN) based MPPT
Artificial neural network (ANN) is a kind of artificial
intelligence method. Artificial intelligence method has more
advantages than conventional methods [7]. The disadvantages
of conventional method are that they show slow responses for Fig. 6. Block diagram of solar MPPT
sudden change in solar temperature and irradiance conditions
and sometimes they may fail to track maximum power point A. PV Array
also. Fig. 5 gives the overview of ANN in MPPT. The input is
Solar photovoltaic array converts solar energy obtained
solar temperature and irradiance. Target of neural network is from sun to electrical energy. Input to PV panel is irradiance
duty ratio to DC-DC converter. For every change in solar and temperature. For every temperature and irradiance
temperature and irradiance value neural network will give a maximum power point varies. It is possible to track MPP by
particular value of duty ratio to get maximum power point. using different maximum power point algorithm developed by
researchers. In this system the specification of PV array used is
given in Table I.
TABLE I. Solar PV panel details
PARAMETERS VALUE
Maximum output power 70W
No. of Parallel strings 2
Fig. 5. Structure of an ANN in MPPT No. of Series connected modules 1
per string
The network is obtained through training by using the Cells per module 60
Levenberg-Marquardt algorithm. For different combination of Open circuit voltage (୭ୡ ) 22.1 V
solar irradiance and temperature value duty ratio is calculated Short circuit current (ୱୡ ) 3.99 A
and ANN is trained. The training of neural network means Voltage at maximum power 18.1 V
adjusting the weights of layers to get the target values. point (୑୔୔ )
Throughout the training procedure weights are adjusted to Current at maximum power 3.88 A
track the target values with minimum error. The performance point (୑୔୔ ))
function of ANN is mean squared error (MSE). If a is the
present output, t is the target, then MSE will be, B. Design of Boost converter
Boost power converter is employed to transfer the power
ଵ
F= σே ଶ
(10) available from solar PV to load. Since two panels are placed in
௜ୀଵሺ‫ݐ‬௜ െ ܽ௜ ሻ
୒ parallel and the peak power output is 140W as per data sheet.
So the maximum operating power of the converter is 140 W.
Steps for designing ANN is given,
Output voltage of the panel is 18.1 V at its MPP.
• Data collection Therefore the input current of the converter
• Selection of network structure ୔
‹୐ = and ο‹୐ = 10% of average value.
• Training the network ୚ୱ
• Testing the network
୚ୱ
In boost mode ୭ = (8)
IV. SYSTEM ANALYSIS AND SIMULATION ଵିୢ

The basic block diagram of the proposed system is given Where,୭ is the load voltage of 24 V. From the relation, the
in Fig. 6. The PV array converts solar energy to electrical
energy and is given to boost converter. ANN based MPPT is duty ratio is, d = 0.2467.
used in the proposed system. ANN control will change the
duty ratio to get maximum power point. ୚ୱ‫כ‬ୢ
Inductor value is given by ୫୧୬ = (9)
୤‫כ‬ο୧୐

Where, f is the switching frequency (20 kHz).

ୢ‫୍כ‬୭
Output capacitance for boost converter is C = (10)
୤‫כ‬ο୚୭
Where,ο‘is taken as 1%

4th International Conference on Electrical Energy Systems (ICEES) 377

 Designed value of boost converter is presented in Table II.

TABLE II. Component values

Inductance 300mH
Capacitance 1000ɊF
Resistor 4.5 Ohm
Output voltage 24V
Switching frequency 20kHz

C.ANN based MPPT

Neural network structure for MPPT is given in Fig. 7.

Fig. 8. ANN training by MATLAB nntool

Fig. 9 shows the generated ANN block and layers inside
the block.

Fig. 7. Neural network for MPPT

Input to neural network is solar temperature and irradiance.
Number of hidden layers is calculated by trial and error
method. Output is duty ratio to the boost converter. Training
points should be obtained to start work with any ANN
algorithm. It is trained by using incremental conductance
algorithm. The training points are obtained by varying the
irradiance and temperature inputs to the PV array and
collecting values of duty ratios to boost converter to track
maximum power from solar panel for different temperature
and irradiance conditions. Some training points are kept as test
points to test the neural network after training. Neural network
is trained by using “nntool” in MATLAB m-FILE. Fig. 8.
shows the neural network training with MATLAB toolbox. Fig. 9. ANN block and layers inside block
The network is obtained through training by using the Table III shows the test result after testing the trained ANN by
Levenberg-Marquardt algorithm. The performance function of using some test points.
ANN is mean squared error (MSE). To train neural network
2400 data are taken. i.e...Temperature range is from (20Ԩ to TABLE III. ANN test results
45Ԩ) and irradiance range is from (600 W/ଶ to 1000
W/ଶ ). The activation function of input layer and the output Temperature Irradiance ANN Actual ANN Actual
layer is “tansig” and purelin” respectively. The training data (Ԩ) (W/ଶ ) Duty Duty Output Output
ratio ratio power power
will be obtained by conducting the solar PV system simulation
by using IC method. (W) (W)
∗ Input : Irradiance and Temperature
∗ Output :Duty ratio
∗ Number of samples:2400 35 1000 0.312 0.315 127.07 128
∗ Number of hidden neurons:5 ͵Ͳ 900 0.262 0.26 117.08 118.1
∗ No of epochs:100 25 800 0.207 0.209 107.07 107
∗ Method : Feed forward method

378 4th International Conference on Electrical Energy Systems (ICEES)

 Fig. 10. ANN performance

In Fig. 10 ANN performance is given. From performance

of trained ANN it is clear that mean squared error (MSE) will Fig. 12. Simulation results from IC method
decrease when epochs increases. A well trained artificial
neural network will have very low mean squared error at the Fig. 12 shows the simulation result of incremental
end of training. The low mean squared error means that the conductance MPPT method. It is the most commonly using
desired output and neural network’s output are close to each MPPT method. IC method is more efficient than P&O. From
other. the simulation result as shown, the IC method takes more time
to settle, also steady state oscillations exist.
V. SIMULATION RESULTS AND COMPARISON
Fig. 11, Fig. 12, Fig. 13 and Fig. 14 show the simulation
results of P&O, IC, FOC, ANN methods respectively at 25Ԩ
temperature and 1000 W/ଶ irradiance. Waveforms show the
PV output voltage, boost converter output voltage and output
power as presented in figure.

Fig. 13. Simulation results from FOC method

Fractional open circuit voltage method requires less

settling time as shown in Fig. 13. While comparing with other
methods it has less output power. No steady state oscillations.
ANN based MPPT tracking is more reliable while
comparing with all other conventional control methods. As
shown in the Fig. 14 it requires less settling time, no steady
Fig. 11. Simulation results from P&O method state oscillations and more output power. Therefore, ANN
based MPPT algorithm gives better result while comparing
Fig. 11 shows the simulation outcome of MPPT tracking with other MPPT methods.
using hill climbing method at Light intensity of 1000 W/ଶ
and Cell temperature of 25Ԩ. PV output voltage, boost output
voltage and output power takes more time to settle. Steady
state oscillations also exist.

4th International Conference on Electrical Energy Systems (ICEES) 379

 VI. CONCLUSION
In this paper a maximum energy harvesting technique
utilizing artificial neural network is introduced and compared
with other 3 conventional MPPT methods to find which
algorithm gives better performance. Input to ANN is solar
irradiance and temperature and output is duty ratio for boost
converter. ANN is trained by using “nntool” in
MATLAB/SIMULINK model. Comparison results shows that
ANN based MPPT controller gives better performance like
less steady state error, fast response for sudden change in solar
temperature and irradiance etc… than the other maximum
power point tracking algorithms. From the result it is
concluded that while comparing with other conventional
MPPT techniques ANN based MPPT gives better performance
in terms of more output power, less steady state oscillations
Fig. 14. Simulation results from ANN method and requirement of less settling time.

TABLE IV. Comparison of different MPPT REFERENCES

Methods Output Output Settling time [1] R. Mahalakshmi, Aswin K.A, and Kumar A, “ Design of fuzzy logic
based maximum power point tracking controller for solar array for
voltage(V) power(W) (Seconds) cloudy weather conditions”, proceedings of Power and Engy System
P&O 24.1 130-132 Settled Conference, 2014
almost at [2] Vesselin Chobanov,The impact of PV orientation in smart grids,
0.06s but proceedings of IEEE International Power Electronics and Motion
Control Conference (PEMC) , IEEE conference publication. 2016, pp
varying 1055- 1059, 2016
IC 24.3 130-132.5 Settled [3] N. Femia, “Optimization of Perturb and observe Maximum PowerPoint
almost at tracking Method,” proceedings of IEEE Trans. Power Electron., Vol.
0.06s but 20, pp.963-973, July 2005.
varying [4] José Antonio Barros Vieira and Alexandre Manuel Mota, “Maximum
FOC 19.84 87.50 0.01s Power Point Tracker Applied in Batteries Charging with PV Panels”,
proceedings of IEEE International Symposium on Industrial
ANN 24.46 133 0.03s(steady) Electronics, pp. 202-207, 2008.
[5] N. Femia, G. Petrone, G. Spagnuolo and M. Vitelli, “Perturb and
Table IV shows the comparison result of different MPPT Observe MPPT Technique Robustness Improved”, proceedings of IEEE
algorithms. From that is clear that ANN gives best result. International Symposium on Industrial Electronics, Vol. 2, pp. 845-850,
2004.
Advantages of using ANN controllers:
[6] P.K Vineeth Kumar and C.A Asha,”An efficient solar power converter
• Fast response for maximum power point tracking. with high MPP tracking accuracy for rural electrification”, proceedings
• Steady state error will be less. of International Conference on Computation of Power, Energy,
Information and Communication, ICCPEIC 2014.
• Reprogramming is not necessary as the learning
mechanism is inherent. [7] Bendib, B.; Krim, F.; Belmili, H.; Almi, M.F.; Bolouma, S., "An
intelligent MPPT approach based on neural-network voltage estimator
Disadvantages of ANN controllers: and fuzzy controller, applied to a stand-alone PV system," Industrial
• Network needs to be intensely trained. Electronics (ISIE), proceedings of IEEE 23rd International Symposium
on , vol.no., pp.404,409, 1-4 June 2014.
• The neural network framework is sometimes
sophisticated in design.

380 4th International Conference on Electrical Energy Systems (ICEES)