Increase the photovoltaic conversion efficiency using

Neuro-fuzzy control applied to MPPT

Hamza Afghoul* Fateh Krim Djamel Chikouche

Department of Electronics University of setif Department of Electronics
University of Msila,Algeria Power electronics and industrial Technology faculty
Power electronics and industrial control laboratory (LEPCI) University ofMsila
control laboratory (LEPCI), Route de Bejai'a, 19000, Setif, Algeria
University of setif Algeria [email protected]
[email protected] [email protected]

Abstract- Photovoltaic (PV) electricity is seen as an important To obtain the maximum power from a photovoltaic array, a
source of renewable energy. The PV generator is an unstable maximum power point tracker (MPPT) is used.
power source depending on several parameters such temperature,
irradiation level etc... This paper deals with an advanced energy The perturbation and observation is one of the most
management of a PV plant based on an intelligent control method commonly used in MPPT methods for its simplicity and ease of
for the maximum power point tracking (MPPT).This latter is implementation. The P&O works well when the irradiance
based on a neuro-fuzzy controller of a DC-DC converter device. change slowly but it presents drawbacks such as slow response
The design of this controller is presented using simulation. Then speed, oscillation around the MPP in steady state, and even
a comparative study of the proposed method and Perturb and tracking in wrong way under rapidly changing atmospheric
Observe (PO) performance is achieved to demonstrate its quality conditions [4].
level in terms of energy optimization and response rapidity to
change in operation conditions.
Recently, fuzzy logic has been applied for tracking the
MPP of PV systems because it has the advantages of being
Keywords-component; Neuro-fuzzy logic (ANFIS); Perturb & robust, design simplicity, and minimal requirement for accurate
Observe (P&O); PV system; MPPT; DC-DC converter. mathematical model. It is found that fuzzy logic-based P&O
and hill climbing MPPT methods perform better due to
optimized perturbation. However, the fuzzy methods depend
I. INTRODUCTION
on careful selection of parameter, definition of membership
function, and the fuzzy rules table. Developing fuzzy method
The demand for electrical energy is increasing around also involves expert knowledge and experimentation in
the world; it has motivated the use of new renewable selecting parameters and membership functions. For this
sources of energy. Among the unconventional renewable reason, adaptive fuzzy logic control and parameter
energy sources that have been studied, PV energy is now optimization techniques such as genetic algorithm and particle
becoming a real economical renewable source of energy, since swam optimization have been introduced to overcome the
it offers many advantages such as incurring no fuel costs, problem in MPPT algorithms.
not being polluting, requiring little maintenance, and
emitting no noise [1,2]. A number of studies on MPPT have concentrated on the
application of artificial neural network (ANN). In most of these
PV systems have been developed to supply clean energy to ANN-based methods, large numbers of field data considering
fulfill the energy demand required by the modern society. atmospheric conditions are required to train the ANN.
However, the widespread use of PV systems poses several Moreover, the main problem of ANN-based method is it cannot
challenges such as increasing the efficiency of energy be implemented for PV arrays with different characteristics. In
conversion, ensuring the reliability of power electronic a related work, a voltage-based MPPT using ANN has been
converters, and meeting the requirements for grid connection developed in which an optimal instantaneous voltage factor
[2]. was determined from a trained ANN. The inputs of the ANN
Maximum power extraction can be obtained by realizing consist of temperature module and solar irradiation.
dynamic or static method. In the dynamic method the The paper presents a method based on Adaptive Neuro­
maximum power point (MPP) is achieved based on tracking fuzzy Inference System (ANFIS) to estimate the MPP using the
the sun movement. This approach may not be suitable for data collected from several experiments performed in different
energy conversion at a small to medium power range due to its environmental conditions. This paper utilizes statistical
high cost and energy consumption. In the static method, which analysis methods to analyze the data of the simulation that has
is suitable for small power range; the maximum power is been collected in different environmental conditions. The
tracked using power converter with high frequency to adjust statistical analysis will confirm the necessity of using both
continuously the operating point at the MPP [3]. (Voc) and short circuit current (Isc) as input data for ANFIS to

978-1-4673-6374-7/13/$31.00 ©2013 IEEE

estimate the maximum power point voltage (Vmax). The ANFIS There are three key operating points on the I-V
generates a duty cycle value for the pulse with modulation curve of a photovoltaic cell. They are the short circuit point,
(PWM) generator. The PWM is then applied to the switching maximum power point and the open circuit point. At the open -
of the boost converter connected to a PV array. circuit point on the I-V curve, V Voc and I O . After = =

substituting these values in the single exponential Eq.l the

II. GLOBAL SYSTEM equation can be obtained [6].

[exp ( ) 1]
A. PV cell model qVac » Vac
o 1ph - 10, - - (2)
Rp
=
A.K,T
The use of equivalent electric circuits makes it possible to
model characteristics of a PV cell. The methods used here are At the short circuit point on the I-V curve, I = Isc and V =

implemented in MATLAB programs for simulations. The same O. Similarly, using equation (1), we can obtain:
modeling technique is also applicable for modeling a PV
module. Shorting together the terminals of the cell, the photon 1 sc -
- 1ph - 1 o· [exp (---:;:v)
q(Rs.lsc»
- 1] - �
[scRs
(3)
generated current will follow out of the cell as a short-circuit
current (lsc). Thus, Iph Isc when there is no connection to At the maximum - power point of the I-V curve, we have
I Impp and V Vmpp . We can use these values to obtain the
=

= =
the PV cell (open-circuit), the photon generated current is
shunted internally by the intrinsic p-n junction diode. This following:
gives the open circuit voltage (Voc) [5]. The simplest model of
1mpp -
_
1ph - 1 o· [exp ( q(Vmpp +Rs.lmpp » ) - 1] -
vmpp +Rs.lmpp
a PV cell equivalent circuit consists of an ideal current source A,KI
�
in parallel with an ideal diode. The current source represents
(4)
the current generated by photons (often denoted as Iph or IL),
and its output is constant under constant temperature and All values of this PV panel are shown in the table below:
constant incident radiation of light. The PV panel is usually
represented by the single exponential model or the double TABLE I. PV MODEL PARAMETERS
exponential model. The single exponential model is shown in
Fig. I. Short circuit current at the 2.664 A
reference condition

+
Thennal voltage timing 5.472 V
completion factor

Series resistance 1.3240

Open circuit voltage of the 87.72 V

PV module at reference
Figure I. Single exponential model of a PV Cell.
condition
The mathematical model which relates the output current to Maximum power point Vmpref 70.731 V
the output voltage is given by the following expression: voltage at the reference

- 1ph - 1 o·
1 - [exp ( q(V+Rs.l» ) - 1] -
V+Rs.l
(1)
condition
Maximum power point 1mpref 2.448 A
Rp
--
A.K.T
current at the reference
Where: condition
Reference irradiance Eref 1000 W/m2
1ph : The photo generated current;
Reference temperature Tref 25°C
10 : The dark saturation current;
Cpv 4
Overall heat capacity per unit 5*10
1Rp : Current flowing in the shunt resistance;
area of the PV cell/module JI(OC*m2)
Rs : Cell series resistance;
Effective area of the PV A
Rp : The cell (shunt) resistance; cell/module
A : The diode quality factor;
Transmittance-absorption Kinpv 0.9
q : The electrical charge 1.6 * 1O-19C; product of PV cells
K : The Boltzmann's constant 1.38 * 10-23 J / K;
Overall heat loss coefficient Kloss
T : The ambient temperature, in Kelvin.

Eq.l is nonlinear. Furthermore, the parameters (lph ,Rs , Rp The following little figures at various module temperatures
simulated with the MATLAB model for our PV module are
and A) vary with temperature, irradiance and depend on
shown with the maximum power points identified on them.
manufacturing tolerance. Numerical methods and curve fitting
After that, more powerful tools are presented (P&O and Neuro­
can be used to estimate [6, 7].
fuzzy logic) to be used for finding the MPPs over most
probable range.
 180
180
140
(8)
120
�'00
J 80
80 Where C is the capacity, L is the inductance, RL is the
40 resistive load, DE[O 1lis the duty ratio, which is also the control
20
input. Vo is the output voltage and iL is the inductor current.
Note the equivalent serie resistance (ESR) of the inductor and
Vottage (V)
wiring resistance are neglected in the case, so iL is assumed to
Figure 2. Influence of the solar radiation for constant temperature.
be equal to the PV current (Ipv ).

III. MAxIMUM POWER POINT TRACING TECHNIQUES

A. Perturb and Observation (P&O)

The perturbation and observation method has been widely

used because its simple feedback structure and fewer measured
parameters which are required [8, 9]. It operates by periodically
perturbing (incrementing or decrementing) the array terminal
Voltage(V)
voltage and comparing the PV output power with that of the
Figure 3. Influence of the junction temperature for constant radiation. previous perturbation cycle. If the power is increasing the
perturbation will continue in the same direction in the next
B. Boost converter cycle, otherwise the perturbation direction will be reversed.
Consider a boost type converter connected to a PV module The flowchart of this method is represented by fig. 5 [10].
with a resistive load as illustrated in Fig. 4. The P&O method presents, in some cases, two drawbacks:
L 5=0
• By forcing the operating point to operate near the
+ MPP, oscillations around the MPP appear in steady
state. Such a drawback gives rise to the waste of some
5=1 c Vo
amount of available energy.
• It can confuse; it moves the operating point far from
the MPP instead of close to it under rapidly changing
atmospheric conditions [9, 11].
Figure 4. Boost converter.

The power switch is responsible to modulate the energy

transfer from the input source to the load by varying the duty
cycle D [3].
Applying Kirchhoffs rules around the loops, it can obtain
the ideal mathematical model of this circuit:
Measure of
VpvCid � V,.(k)
L1 -
-
and
{i" L
_

L When S = 0 (5) ip,,(k)

v..
01 �
-

�
-
-
C C*RL Set
I'(k)- V,.(k) >Ip.(k)

When S = 1 (6)

Eqs. 5 and Eq. 6 give Eq. 7:

X = (1 - D)X1 + DXz (7)

Where

D E [O 1].

Hence the dynamic equation of the system can be described Figure 5. Flow chart of the P&O method.
by:
B. MPPT using Neuro-Fuzzy Logic Control adaptive techniques, it guarantees continuity of the output
The Adaptive-Neuro Fuzzy Inference System is a hybrid surface, and it is well-suited to mathematical analysis.
system that combines the potential benefits of both the methods Fig. 7 shows the basic structure of the ANFIS algorithm for
ANN (Artificial Neural network) and FL (Fuzzy Logic) [12]. a Sugeno-type FIS. The various layers shown bellow:
This system has been employed in numerous modeling and
forecasting problems [13]. ANFIS starts its functionality with
the fuzzification of input parameters defming the membership
function parameters and design of fuzzy IF-THEN rules, by
effectively employing the learning capability of ANN for
automatic fuzzy rule generation and self adjustment of
membership functions [14].
In this work, the Sugeno method or Takagi-Sugeno-Kang
method of fuzzy inference has been used. The Sugeno method
Figure 7. Typical ANFIS structure.
was first introduced in 1985 [15]. It is similar to the Mamdani
method in many aspects. The first two parts of the fuzzy
• Layer 1 (Membership layer)
inference process, fuzzifying the inputs and applying the fuzzy
• Layer 2 (Fuzzy AND layer)
operator are exactly the same. The difference is that unlike the
• Layer 3 (Normalizing layer)
Mamdani method, in the Sugeno method the output MFs are
only constants or have linear relations to the inputs. With a • Layer 4 (Partial output layer)
constant output MF, this method is known as the zero-order • Layer 5 (Output layer)
Sugeno method, whereas with a linear relation, it is known as
the first-order Sugeno method [16]. Fig. 8 shows the fust order Sugeno model in MA TLAB­
Simulink with two inputs and one output.
A typical rule in a Sugeno fuzzy model has the following
form: FISEdltor glob -ox
File Edit \liew

If Input-l = x, and Input-2 = y, then, Output z = ax + by + C

For a zero-order Sugeno model, the output level z is a

constant (a b 0). The output level Zi of each rule is
IXX� nput1
,100
- feu)
= =
(sugeno)
weighted by the firing strength Wi of the rule. For example, for
an AND rule with Input-l x, and Input-2
= y, the firing
= IXXY npul2
...'"
strength is
I '""""
I
FlSNomo !jOO FISType:

Wi = AND(F1 (x),Fz(y)) ""'''''''' 1 ",ad a.ner1V�

!Ivl
"'''''''' Nomo � otJpul
Where,Fl (.) and F2 (.) are the MFs for Inputs 1 and 2. 1",- H �
.......
-
I"" vi T""
R.... 10_2090.4791
The final output of the system is the weighted average of ....- I"", vi
the outputs of all the rules, computed as De1uzmamn 1-
HI ..... 1 Ie""" rl
Renamed AS to "tJ«t
I I
A Sugeno rule operates as shown in Fig. 6. Figure 8. Matlab windows of the first order Sugeno model with five inputs
Input MF and one output.

IV. SIMULATION AND PERFORMANCE COMPARISON

All simulation results using model in Matlab/simulink and

SimPower System Blockset are presented.
Fig. 9 presents the model used in the simulation

--r----.i lp'
Figure 6. First order Sugeno-type inference system.

It is a great advantage of the Sugeno-type FIS, that it avoids

the use of a time consuming defuzzification, since it is a more INPUT SIGNALS PV

compact and computationally efficient representation than the

Mamdani system, the Sugeno system lends itself to the use of
adaptive techniques for constructing fuzzy models. These
adaptive techniques can be used to customize the MFs so that MPPT CONTROLLER

the fuzzy system accurately models the data. Some of the

Figure 9. System simulation model.
advantages of the Sugeno-type method are that it is
computationally efficient, it works well with linear techniques Fig 10 and Fig. 11 show the fuzzy membership function for
(e.g., PID control), it works well with optimization and inputs (V, I).
 rx:x:x�'·" Xl
To evaluate the reliability performance of the developed
Neuro-fuzzy network for different ranges of climatic conditions,
the change of solar irradiation is applied for many levels (Fig.
15).
_1 0 1 2 200

-
Figure 10. Fuzzy membership functions for inputs V.
150 - -P&O
-ANFIS
T\

\ �'-.....
{XJC)(J(J
�
Q) 100
�
Q.
V \ �
50 �
'iT
�
o 20

Figure I I. Fuzzy membership function for inputs I.

20 40 60 80 100 120 140 160 180 200 220
Time
Fig.12 shows the rule view interfaces used to access data
Figure 15. PV power output with solar irradiation for many levels.
cycle output values.

Rule VIewer glob -D In Fig. 15, Neuro-fuzzy method is faster at instants t=105,
File Edt. View Options t=130, t=175 and stable compared to the P&O yet there is a
i'1>Ul:1.01 ilplt2 65
� rulpoi z 0.198 sudden decline at the instant t=105 and t=165. We notice that
12
the system is losing a bit of its power and the duty cycle of Fig

�
34
,
, 15 is shown below:
7
II o.5
10119
,
II

1213
Ii
Ii
II
" 0.4 H=p&o jI wP!iI!-,l
�
14
1617
1S
Ii
Ii
Ii
II � o. 3
v
--ANFIS
IIIImmiiIi
IIiiiiiI iiII. \ ��
1920
fl ��
16

u
�

2122 �o.2
14
Zl

"

1FJiU:: 1[1.012;65.031
�
o
o. 1 ��JM ---J
\
�o 40 60 80 100 120
�I
140 160
� 180 200 220
Time

Figure 12. Matlab windows of the rule viewer with two inputs and one output. Figure 16. Duty cycle for variable irradiation.

In Fig. 13, the Neuro-fuzzy method is faster and more All comparisons can be summarized in Tab. II.
stable than P&O method.
TABLE II. COMPARATIVE STUDY BETWEEN P&O AND NEURO-FUZZY
� .-------.--.---� METHODS

P&O method Neuro-fuzzy method

150

Power power loss efficient

Response time slow fast

-----------,--------------,--------------,--------------,------------- Duty cycle not stable stable

Cost low High

��------�5�--�1�O--�15�--���-�
Implementation easy difficult
Figure 13. PV power output with constant solar irradiation.

The duty cycle corresponding to Fig 13 is shown in Fig.14 v. CONCLUSION

0.5 This paper presents a comparative study between P&O and
0.45
n n n .� n n � n n n. n n � n.

Neuro-Fuzzy methods, for PV installation. An optimized

0.4
0.35 I model is proposed which is independent of any specific
� 0.3 I manufacturer's model and hence gives a greater option for
i 0.25 I -ANFIS
component selection. The results show that the ANFIS model
g 0.2 I -P&o

provides an accurate power output, while the optimization

0.15 -J algorithm provides configuration with the lowest cost. It has
0.1 been shown that the system based on Neuro-fuzzy is able to
0.05
learn the PV properties gradually, changing solar radiation
10 15 20 25 conditions, and leads to a quick tracking of the MPP when the
Time

solar radiation is rapidly changing.

Figure 14. Duty cycle for a fixed irradiation (IOOOW1m2).
 VI. REFERENCES

[1] J. Salazar, F. Tadeo, C. Prada, L. Palacin, "Simulation and Control of a

PV System connected to a Low Voltage Network", Jornadas de
Automatica, September 08-10, Jaen, Spain, 2010.
[2] A.Subiyanto, H. Mohamed, H. Shareef, "Hopfield Neural Network
Optimized Fuzzy Logic Controller for MaximumPower Point Tracking
in a Photovoltaic System", International Joumal of Photoenergy, Article
10 798361, 2012, 13 pages.
[3] A. Chouder, F. Guijoan, S. Silvestre, "Simulation of fuzzy-based MPP
tracker and performance comparison with perturb & observe method",
Revue des Energies Renouvelables Vol. 11 N°4, 2008, pp. 577 - 586.
[4] B. Amrouche, M. Belhamel, A. Guessoum, "Artificial intelligence based
P&O MPPT method for photovoltaic systems", Revue des Energies
Renouvelables ICRESD-07 Tlemcen, 2007, pp II - 16.
[5] A. EI Shahat, "Maximum power point genetic identification function for
photovoltaic system" IJRRAS 3 (3), June 2010, USA.
[6] A. Keyhani, M N. Marwali, and M. Dai, "Integration of Green and
Renewable Energy in Electric Power Systems," Wiley, January 2010.
[7] Masters, Gilbert M. Renewable and Efficient Electric Power Systems
John Wiley & Sons Ltd, 2004.
[8] N. Femia, G. Petrone, G. Spagnolo and M. Vitelli, "Optimizing Duty­
Cycle Perturbation of P&O MPPT Technique", 35th Annual IEEE
Power Specialists Conference, Aachen, Germany, 2004.
[9] N. Femia, G. Petrone, G. Spagnolo and M. Vitelli, "Increasing the
Efficiency of P&O by Converter Dynamic Matching", 35th Annual
IEEE Power Specialists Conference, Aachen, Germany, 2004.
[10] B. Amrouche, M. Belhamel, A. Guessoum, "Artificial intelligence based
P&O MPPT method for photovoltaic systems", Revue des Energies
Renouvelables, ICRESD-07 Tlemcen, 2007, pp. 11 - 16.
[11] N. Femia, G. Petrone, G. Spagnolo and M. Vitelli, 'Optimization of
Perturb and Observe Maximum Power Point Tracking Method', IEEE
Transactions on Power Electronics, Vol. 20, N°4, July 2005, pp. 963 -
973.
[12] J. Moreno, "Hydraulic plant generation forecasting in Colombian power
market using ANFIS", Energy Economics 31, Elsevier, 2009, pp. 450-
455.
[13] M.A. Boyacioglu, D. Avci, "An Adaptive Network-Based Fuzzy
Inference System (ANFIS) for the prediction of stock market return: The
case of the Istanbul Stock Exchange, Expert Systems with
Applications", Elsevier, Vol. 37, 2010, pp. 7908-7912.
[14] V. Gaur, A. Soni, "Analytical Inference Model for Prediction and
Customization of Inter-Agent Dependency Requirements", ACM
SIGSOFT Software Engineering Notes, March 2012.
[15] M. Sugeno, Industrial applications of fuzzy control. Elsevier Science
Pub. Co., Amsterdam, 1985.
[16] L. Wang et aI., "Wind Power Systems, Green Energy and Technology",
Springer-Verlag Berlin Heidelberg 2010, pp. 255-295.