TSINGHUA SCIENCE A N D TECHNOLOGY

ISSN 1007-0214 12/23 pp204-208

V o l u m e 10, N u m b e r 2, April 2005

Artificial Neural Network Maximum Power Point Tracker

for Solar Electric Vehicle

Theodore Amissah O C R A N , C A O Junyi

CAO Binggang ( f ^ B ' J ) , SUN Xinghua ( # ^ )

R e s e a r c h a n d D e v e l o p m e n t C e n t e r for E l e c t r i c V e h i c l e , X i ' a n J i a o t o n g U n i v e r s i t y , X i ' a n 7 1 0 0 4 9 , C h i n a

Abstract: This paper proposes an artificial neural network maximum power point tracker (MPPT) for solar
electric vehicles. The MPPT is based on a highly efficient boost converter with insulated gate bipolar transis-
tor (IGBT) power switch. The reference voltage for MPPT is obtained by artificial neural network (ANN) with
gradient descent momentum algorithm. The tracking algorithm changes the duty-cycle of the converter so
that the PV-module voltage equals the voltage corresponding to the MPPT at any given insolation, tempera-
ture, and load conditions. For fast response, the system is implemented using digital signal processor (DSP).
The overall system stability is improved by including a proportional-integral-derivative (PID) controller, which
is also used to match the reference and battery voltage levels. The controller, based on the information sup-
plied by the ANN, generates the boost converter duty-cycle. The energy obtained is used to charge the lith-
ium ion battery stack for the solar vehicle. The experimental and simulation results show that the proposed
scheme is highly efficient.

Key words: artificial neural network; maximum power point tracker (MPPT); photovoltaic module; digital
signal processor; solar electric vehicle

(dc-dc or dc-ac converter) for load interface. Therefore,

Introduction the overall system cost could be reduced drastically by
using highly efficient power conditioners, such as the
Photovoltaic (PV) generation is gaining increased im-
maximum power point tracker (MPPT), to extract and
portance as a renewable source of energy. The undesir-
maintain the peak power from the PV module even
able rapid changes of solar power, which usually occur
when the above-mentioned unfavorable conditions
in a running vehicle, arise as a result of shade from
occur.
buildings, large trees, and clouds in the sky. Conven-
Various methods of maximum power tracking have
tional PV systems have difficulties in responding to
been considered in PV power applications^ " ^. Among 1 12

rapid variations due to shade. The main drawbacks of

the hill climbing methods , the perturb and observe
[1_5]

PV systems are that the initial installation cost is con-

(P&O) method tracks maximum power point (MPP) by
siderably high and the energy conversion efficiency
repeatedly increasing or decreasing the output voltage
(from 12% to 29%) is relatively low. Furthermore, in
at MPP of the PV module. This method requires calcu-
most cases, PV systems require a power conditioner
lation of dP/dFto determine the MPP ' ' . Though the [1 2 4]

Received: 2004-03-24 method is relatively simple to implement, it cannot

* * To whom correspondence should be addressed. track the MPP when the irradiance changes rapidly.
E-mail: caojy@mailst.xjtu.edu.cn;
Also, the method oscillates around the MPP instead of
Tel: 86-29-87570252
 Theodore Amissah OCRAN et al: Artificial Neural Network Maximum Power Point Tracker 205

directly tracking it. The incremental conductance

technique (ICT) is the most accurate ' [6 7]
among the
other methods. This method gives good performance
under rapidly changing conditions. However, the com-
plex calculation of dl/dV and the complicated
algorithm require use of a digital signal processor
(DSP), which will usually increase the total system
Fig. 1 Equivalent circuit of a solar cell
co§Ehe M P P tracking method using the short circuit
current of the P V module utilizes the fact that the op-
erating current at M P P of the P V module is linearly exp
AkT
proportional to the short circuit current of the P V mod-
u l e ^ . Under rapidly changing atmospheric conditions, (1)
this method has a fast response speed of tracking the where 7 a n d
o V 0 are the output current and output
MPP, but the control circuit is complicated. The M P P voltage of the solar array; / is the generated current

tracking method using open circuit voltage of the solar under a given insolation; 7 sat is the reverse saturation
current; q is the charge of an electron; k is the
panel [11]
utilizes the fact that the operating voltage at
Boltzmann constant; A is the ideality factor for a P-
M P P is almost linearly proportional to open circuit
N junction; Τ is the array temperature; R and R
voltage at M P P of the P V module (using 7 6 % of open
s sh

are the intrinsic series and shunt resistances of the solar

circuit voltage as the M P P voltage). This method is
array.
very simple and cost-effective, but the reference volt-
The saturation current of the solar array varies with
age does not change between samplings. MPPTs using
temperature according to the following equation:
the fuzzy l o g i c ' [13 14]
and the artificial neural network
(ANN) [ 1 5 ]
have been reported. These studies show that qE G

exp (2)
such modern control algorithms are capable of improv- Bk
ing the tracking performance as compared to
I = 17 + ^ _ ( r - 2 5 ) l — (3)
the conventional methods.
In this paper, we propose an A N N M P P T for solar
L so
8 /v ; J 1 0 0

where Tr is the reference temperature; Ior is the

electric vehicle P V systems. The tracking algorithm saturation current at T ; r E G0 is the band-gap en-
changes the duty ratio of the converter so that the P V ergy of the semiconductor used in the solar array; Β is
module voltage equals the voltage corresponding to the also an ideality factor; 7 sc is the short circuit current
M P P at that atmospheric condition. This adjustment is at 25°C; Kj is the short-circuit current temperature
carried out by using the back-propagation A N N . The coefficient and λ is the insolation in m W / c m . 2

reference voltage to the M P P is obtained by an offline
trained A N N . The controller generates the boost
converter duty-cycle.
1 PV Module Characteristics
The solar array characteristics significantly influence
the design of the converter and the control system, so
the P V characteristics will be briefly reviewed here.
The solar array is a nonlinear device and can be repre-
sented as a current source model, as shown in Fig. 1.
The traditional I-V characteristics of a solar array,
when neglecting the internal shunt resistance, are given
by the following equation:
Equations (l)-(3) are used in the development of
computer simulations for the solar array. The Matlab
programming language is used. Figure 2 show the
simulated current-voltage and power-voltage curves
for the solar array at different insolations and different
temperatures. These curves show that the output char-
acteristics of the solar array are nonlinear and greatly
affected by the solar radiation, temperature, and load
condition. Each curve has a maximum power point
(TW), which is the optimal operating point for the ef-
ficient use of the solar array.
2 Artificial Neural Network
A N N technology has been successfully applied to
 206 Tsinghua Science and Technology, April 2 0 0 5 , 10(2): 2 0 4 - 2 0 8

5-1000 W/m' j when neuron j is a hidden unit; e. (n) is the er-

ror signal at the output of neuron j for iteration η ;
and the set C includes all the neurons in the outer
layer of the network. The correction Aw (η) to the ß

synaptic weight w (η) is given by ß

Aw (η) = aAw (η -1) + ηδ {n)y (η)

ß ß ] t (5)
where a is the momentum constant; η is the learn-
ing-rate parameter of the back-propagation algorithm;
Sj(n) is the local gradient. The error signal at the out-
put is defined as
e (n) = d (n)-y (n)
J J J (6)
where d. {n) is the desired response or wanted tar-
gets and y. (n) is the output signal. Adjustment of the
weights for these layers is given by
w (n +1) = w (n) + a*w (n -1) + ηδ [n)y (n)
ß ß ß ] i

(7)
From Eqs. (4) and (6), the mean squared error per-
formance index can be rewritten as

s(n) = \(V (n)-V (n)f.

K( A

The network training is performed repeatedly until the

performance index £ = (V f-V ) rQ A
2
falls below a
specified value, ideally to zero. In other words ε —> 0
implies (V -V )
re{ A
2
^ 0 , and then the connecting
weights of the network are adjusted in such a way that
the array voltage V A is identically equal to the maxi-
mum power point voltage V . At this stage the refer- mp

ence voltage F ref becomes equal to the maximum

power point voltage V . mp

0 20 40 60 80
UN The configuration of the proposed three-layer feed-
(c) Power-voltage curves
forward neural network function approximator is
Fig. 2 C u r r e n t - v o l t a g e a n d p o w e r - v o l t a g e c u r v e s for shown in Fig. 3. The neural network is used to obtain
t h e s o l a r a r r a y at d i f f e r e n t i n s o l a t i o n s a n d different
the voltage of the maximum power V (n) of the so- mp
t e m p e r a t u r e s (S is t h e s o l a r r a d i a t i o n )
lar panel. The network has three layers: an input, a
solve very complex problems. Recently, its application hidden, and an output layer. The numbers of nodes are
in various fields is increasing rapidly ' . [16 17]
two, four, and one in the input, the hidden, and output
The instantaneous sum of error squares or error en- layers, respectively. The reference-cell open circuit
ergy at iteration η is given by voltage V {n) and the time parameter Τ (n) are
oc

supplied to the input layer of the neural network. These

<η) = \γή{η) (4) signals are directly passed to the nodes in the next hid-
Z
J t C den layer. The node in the output layer provides the
where neuron j lies in a layer to the right of neuron identified maximum power point voltage V (n) . The
i, and neuron k lies in a layer to the right of neuron nodes in the hidden layer get signals from the input
 Theodore Amissah OCRAN et al: Artificial Neural Network Maximum Power Point Tracker 207

layer and send their output to the node in the output output layer mapping. During the training, the con-
layer. The sigmoid activation function is utilized in the necting weights are modified recursively until the best
layers of the network. The training program calculates fit is achieved for the input-output patterns in the train-
the connecting weights ^{1,1} with the bias b{\} ing data. The training of the net was accomplished
for the input to hidden layer mapping, the connecting off-line using Matlab.
weights W {2,1} with bias b{2} for the hidden layer to
L

Input Hidden Output

* \ ^,{1,1} W {2,1}
L

^oc(»)
Γ
T{n)
) J >(±>
b{2}

2 4 1

Fig. 3 Feed-forward neural network function approximator

7=0.1; momentum factor a =0.9; number of training

3 Experimental/Simulation Results iterations=10 000; error goal=0.000 001. The conver-
gence error for the training process is shown in Fig. 4.
A PV array used for the collection of the experimental
data is JDG-M-45 (Germany) type modules. The mod-
ule has a maximum power output of 45 W and a 20-V
open-circuit voltage at an irradiation of 1000 W/m 2

and a 25°C temperature. The PV module specifica-

tions provided by the manufacturer in Table 1.

Table 1 E l e c t r i c a l s p e c i f i c a t i o n s for t h e P V module

(JDG-M-45)

M a x i m u m power, P m (W) 45 0 2000 4000 6000 8000 10 000

Iterations
S h o r t circuit current, 7 SC (A) 2.93

O p e n circuit v o l t a g e , V (V) 20 Fig. 4 Convergence error for t h e neural network

oc

training process
V o l t a g e at m a x p o w e r p o i n t , V mp (V) 17.1
C u r r e n t at m a x p o w e r p o i n t , 7 (A) 2.64
m p
The parameters used for the training of the ANN and
Size ( m m X m m X m m ) 971X441X28
the output values given by the ANN after the training
M a s s (kg) 5
can be found in Table 2. Various sets of reference cell
The feed-forward back-propagation ANN, as shown open circuit voltage V and a time parameter Τ (as
oc

in Fig. 3 was trained with values obtained from ex- shown in Table 2) are supplied as the input to the ANN.
perimental data of the reference cell. Gradient descent In order to validate the learning capability of the ANN,
algorithm was used in training as it improves the per- other sets of V different from the one in Table 2 were
oc

formance of the ANN, reducing the total error by also supplied to the ANN, which gave out values of
changing the weights along its gradient. The training V as expected. The software Matlab was used in the
mp

parameters are as follows: learning rate parameter training of the ANN.

Table 2 T h e results of the trained data given by the A N N

H o u r of the day

9:00 9:30 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00

Time parameter -1 -0.9 -0.8 -0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0

V oc measured 0.555 0.519 0.516 0.529 0.517 0.514 0.516 0.510 0.511 0.509 0.508

V mp measured 0.422 0.394 0.392 0.402 0.393 0.391 0.392 0.388 0.388 0.387 0.386
Vmp given by A N N 0.421 0.379 0.388 0.416 0.395 0.394 0.392 0.388 0.388 0.386 0.385
 208 Tsinghua Science and Technology, April 2 0 0 5 , 10(2): 2 0 4 - 2 0 8

The weights to the hidden layer 1 from input 1 are as LA, U S A , 2000: 115-1156.

follows: [6] T s e Κ Κ , H o Μ Τ, C h u n g Η S, H u i S Y . A n o v e l m a x i m u m

^{1,1} = "2.1706 -5.3865 -3.2326 -5.1766" p o w e r p o i n t t r a c k e r for P V p a n e l s u s i n g s w i t c h i n g fre-

1
[109.8309 -27.9435 97.7250 -8.5640_ q u e n c y m o d u l a t i o n . IEEE Trans. Power Electron., 2002,
17(6): 980-989.
The weights to the output layer 2 are ^{2,1} =
[7] Brambilla A, Gambarra M , Garutti A, Ronchi F. N e w ap-
[-0.2039 0.0065 0.0561 -0.3029]. The bias to
proach to photovoltaic arrays m a x i m u m power point track-
layer 1 is b{\} = [-60.0887 12.996 -54.4753 ing. In: P r o c . I E E E P o w e r E l e c t r o n . S p e c . Conf., C h a r l e s -
-1.5969]. The bias to layer 2 is b{2} = [-0.0397]. ton, SC, U S A , 1999: 632-637.

[8] M u t o h N , M a t u o T, O k a d a K, S a k a i M . P r e d i c t i o n - d a t a -

4 Conclusions b a s e d m a x i m u m p o w e r p o i n t t r a c k i n g m e t h o d for p h o t o -
v o l t a i c p o w e r g e n e r a t i o n s y s t e m s . In: P r o c . I E E E P o w e r
An artificial neural network MPPT for charging the E l e c t r o n . S p e c . Conf., C a i r n s , A u s t r i l i a , 2 0 0 2 : 1 4 8 9 - 1 4 9 4 .
battery stack of a solar (hybrid) vehicle has been pro- [9] N o g u c h i T, T o g a s h i S, N a k a m o t o R. S h o r t - c u r r e n t p u l s e -
posed in this paper. An off-line ANN, trained using a b a s e d m a x i m u m p o w e r p o i n t t r a c k i n g m e t h o d for m u l t i p l e
back-propagation with gradient descent momentum al- p h o t o v o l t a i c a n d c o n v e r t e r m o d u l e s y s t e m . IEEE Trans.
gorithm, is utilized for online estimation of reference Ind. Electron., 2002, 49(1): 217-223.
voltage for the feed-forward loop. Experimental data is [10] L e e D Y , N o h Η J, H y u n D S, C h o y I. A n improved
used for the offline training of the ANN, and software M P P T c o n v e r t e r u s i n g c u r r e n t c o m p e n s a t i o n m e t h o d for
Matlab is used in the training of the net. The precision small scaled PV-applications. In: P r o c . IEEE Applied
of the estimation has been verified by the graph of the P o w e r E l e c t r o n . Conf., D a l a s , T X , U S A , 2 0 0 2 : 5 4 0 - 5 4 5 .
convergence error. The proposed method has several [11] E n s l i n J Η R, W o l f Μ S, S n y m a n D B , S w e i g e r s W . I n t e -
advantages over the conventional methods, particularly grated photovoltaic m a x i m u m power point tracking con-
in that there is no need for voltage and current sensors, verter. IEEE Trans. Ind. Electron., 1997, 44(6): 769-773.
and in that it avoids a complex calculation of power. [12] C h e n Y , S m e d l e y K, V a c h e r F , B r o u w e r J. A n e w m a x i -
The experimental and simulation results show that the mum power point tracking controller for photovoltaic
proposed scheme is highly efficient. p o w e r g e n e r a t i o n . In: P r o c . I E E E A p p l i e d P o w e r E l e c t r o n .
Conf., M i a m i B e a c h , F L , U S A , 2 0 0 3 : 5 6 - 6 2 .
References
[13] N a f e h A , F a h m y F H , E l - Z a h a b Ε Μ A . E v a l u a t i o n of a

[1] H u a C, L i n J, S h e n C. I m p l e m e n t a t i o n of a D S P - c o n t r o l l e d p r o p e r c o n t r o l l e r p e r f o r m a n c e for m a x i m u m p o w e r p o i n t

p h o t o v o l t a i c s y s t e m w i t h p e a k p o w e r t r a c k i n g . IEEE Trans. t r a c k i n g o f a s t a n d - a l o n e P V s y s t e m . International Journal

Ind. Electron., 1998, 45(1): 99-107. of Numerical Modelling, 2002, 15(4): 385-398.

[2] K o u t r o u l i s E , K a l a i t z a k i s K, V o u l g a r i s Ν C. D e v e l o p m e n t [14] Veerachary M, Senjyu Τ S, U e z a t o K. Feedforward

of a m i c r o c o n t r o l l e r - b a s e d p h o t o v o l t a i c m a x i m u m p o w e r m a x i m u m p o w e r p o i n t t r a c k i n g of P V s y s t e m s u s i n g fuzzy

p o i n t t r a c k i n g c o n t r o l s y s t e m . IEEE Trans. Power Elec- c o n t r o l l e r . IEEE Trans. Aerosp. and Electron. Syst., 2002,

tron., 2 0 0 1 , 1 6 ( 1 ) : 4 6 - 5 4 . 38(3): 969-981.

[3] K u o Y C, L i a n g Τ J, C h e n J F . N o v e l m a x i m u m p o w e r [15] H i y a m a T, K o u z u m a S, O r t m e y e r T. E v a l u a t i o n o f n e u r a l

p o i n t t r a c k i n g c o n t r o l l e r for p h o t o v o l t a i c e n e r g y c o n v e r - network based real time m a x i m u m p o w e r tracking control-

sion s y s t e m . IEEE Trans. Ind. Electron., 2001, 48(3): 594- ler for P V s y s t e m . IEEE Trans, on Energy Com., 1995,

601. 10(3): 543-548.

[4] S u l l i v a n C R, P o w e r s Μ J. A h i g h - e f f i c i e n c y maximum [16] B o s e Β Κ. Artificial n e u r a l n e t w o r k a p p l i c a t i o n s in p o w e r

p o w e r p o i n t t r a c k e r for p h o t o v o l t a i c a r r a y s in a solar- e l e c t r o n i c s . In: P r o c . I E E E Ind. E l e c t r o n . Conf., D e n v e r ,

p o w e r e d r a c e v e h i c l e . In: P r o c . I E E E P o w e r E l e c t r o n . S p e c . CO, USA, 2 0 0 1 : 1631-1638.

Conf., Seattle, W A , U S A , 1 9 9 3 : 5 7 4 - 5 8 0 . [17] H a y k i n S. N e u r a l N e t w o r k s — A C o m p r e h e n s i v e F o u n d a -

[5] S i m o e s Μ G, F r a n c e s c h e t t i Ν Ν . A rise-microcontroller tion, 2 n d

Edition. N e w York: Prentice Hall Inc., 1999.

b a s e d p h o t o v o l t a i c s y s t e m for i l l u m i n a t i o n a p p l i c a t i o n s . In:
P r o c . I E E E A p p l i e d P o w e r E l e c t r o n . Conf., N e w O r l e a n s ,