International Conference on Smart Manufacturing Application

April. 9-11, 2008 in KINTEX, Gyeonggi-do, Korea

Maximum Power Point Tracking Control of PV System for DC Motors
Drive with Neural Network
Jae-Sub Kol and Byung-Jin Jung2 and Ki-Tae Park3 and Chung-Hoon Choi4 and
Dong-Hwa Chung5
Department of Electrical Control Engineering, Sunchon University, Suncheon, Korea
(Tel: +82-61-750-3540; E-mail: kojs250 1 ggmail.com)
2Department of Electrical Control Engineering, Sunchon University, Suncheon, Korea
(Tel: +82-61-750-3540; E-mail: jokzebeeggmail.com)
3 DaeyangTech corp., Kwang yang, Korea
(Tel: +82-61-793-5925; E-mail: bosomggmail.com)
4 DaeyangTech corp., Kwang yang, Korea
(Tel: +82-61-793-5925; E-mail: kokos22gnaver.com)
5Department of Electrical Control Engineering, Sunchon University, Suncheon, Korea
(Tel: +82-61-750-3540; E-mail: hwa777gsunchon.com)
Abstract: This paper presents an application of a Neural Network(NN) for Maximum Power Point Tracking (MPPT)
of PV supplied DC motor. A variation of solar irradiation is most important factor in the MPPT of PV system. That is
nonlinear, aperiodic and complicated. NN was widely used due to easily solving a complex math problem.
The paper consists of solar radiation source, DC-DC converter, DC motor and load(cf, pump). NN algorithm apply to
DC-DC converter through an Adaptive control of Neural Network, calculates Converter-Chopping ratio using an
Adaptive control of NN. The results of an Adaptive control of NN compared with the results of Converter-Chopping
ratio which are calculated mathematical modeling and evaluate the proposed algorithm. The experimental data show
that an adequacy of the algorithm was established through the compared data.

Keywords: PV system, Neural Network, MPPT, DC-DC converter

1. INTRODUCTION in many parts. The power of the neural network in

modeling complex mappings and in system
The interest of renewable energy in environmental identification has been demonstrated (Narendra and
issues such as solar energy is increasing. Photovoltaic Parthasarathy, 1990)[5]. Neural network based real time
(PV) generation is gaining increased importance as maximum power point tracking controller for PV grid
renewable source which have the merits like non-using connected systems has been reported (Takashi and
of fuel cost, no noise without moving parts and little Imakubo, 1995)[6]. The studies stress that the solar cell
maintenance. Especially, PV systems are rapidly array operating point is moved to its maximum power
expanding and have increasing roles in electric power point by using a voltage control type inverter, which is
technologies, Especially, PV systems are rapidly identified by the neural network. Various methods of
expanding and have increasing roles in electric power maximum power tracking have been reported in
technologies, literature. Perturbation and Observation method (P&O)
Optimum operation of the solar cell array and a dc moves the operating point towards the maximum power
shunt motor is achieved by means of a switching point by periodically increasing or decreasing the array
procedure of the solar cell array modules, direct current voltage and it is also proper for slowly varying solar
transformers and controlling the motor fluxes[1]. isolations. The incremental conductance method offers
Appelbaum (1986) has analyzed the performance of dc good performance with rapidly changing solar radiation.
motors supplied from PV sources[2]. These studies But it requires the high-resolution sensors for
show that the DC shunt motor supplied by solar cells measurement and the phase of instrumentation imposing
has an inferior performance and a separately excited DC an economic restriction.
motor driving a centrifugal pump is the suitable drive. This paper brings out the identification of optimal
Performance analysis (dynamic and steady state) of point of the PV supplied separately excited DC motor
DC motors proposed from PV supplies through fed from intermediate power converter driving
intermediate power converter( Veerachary, 2000)[3]. centrifugal pump or volumetric pump. The
Saied (1988) formulated the rules to construct the motor converter-chopping ratio is suitably selected through
v-i characteristics about daily maximum total using neural network to get maximum power from solar
mechanical energy output (TME) and to determine the cell array or TME output from the combined system.
optimal parameters of proper motors to match the solar This on-line training approach avoids the move of
generator[4]. Neural networks are widely recognized as optimal operating point through trial and error
a technology suggesting an alternative way to solve
procedure and does not require high-resolution sensors.
complex problems and have been successfully applied

514
Also it proved propriety of algorism with outcome Where Kins is percentage of insolation.
analysis.
R
2. MATHEMATICAL MODEL OF THE
SYSTEM
The constructed system consists of Solar Cell Array,
DC-DC converter, DC motor coupled onto either
centrifugal pump or volumetric pump load as shown in
Fig. 1. Fig. 2 Equivalent circuit of the PV generator
The explained functions about MP or TME operations
and the mathematical modeling for each combined -------- 00% ------ -0- - - -80 60%

elements are contained in the next section. ------ 40% - ---- 20%

PV Source

0.0 0.0

0 0.2 0.4 0.6 0.8 1.0

Current(p.u)
Fig. 1 Block diagram of the system Fig. 3 The v-i/P-i characteristics of PV generator at
different solar insolations
2.1 PV generator model
2.2 Power converter model
The PV generator consists of the combination of
many PV cells connected in series and parallel fashion The DC-DC converter is Buck-Boost converter that
to provide desired value of output voltage and current. can be variation duty or chopping ratio. This converter
PV generator exhibits a non-linear solar radiation is generated a chopped DC voltage output and
depending on v-i characteristic. The v-i characteristic controlled mead voltage and current of motor. Also, the
(Appelbaum, 1986) with Np cells in series and converter continually matches the output characteristic
Ns cells in parallel is of the PV generator to the input characteristic of the
motor so that maximum power is extracted form the
solar cell array or the gross mechanical energy per day
Vg =IgRs Nj<+( s
I nI +Ns;h;
( I ) ofthe system is maximum.
Assuming the DC-DC converter is ideal, voltage and
current of converter about duty ratio 5 is concerned to
solar cell voltage( Vg ), current( Ig ) and as following
where A = q I AKT; -electric charge; -Completion
factor; K-Boltzman constant; T-Absolute temperature. V = (3)
The equivalent circuit of PV generator is shown in Fig.
2. The output v-i, P-i characteristics of the solar cell
array (insolation of 1000 W/m2= 10000 insolation) with
solar insolation as a parameter is shown Fig. 3. Where
Ig, Vg, Pg respectively are the output current, voltage = ( ) (5)
and power of the solar array. When the solar insolation
is changing, the moving of the maximum power point of
the solar array is shown in Fig. 3. The PV generator (I= - (6)
considered in these studies consists of 18 parallel paths
and each path contains 324 cells in series. Substituting
the cell constants in (1) then the simplification of the Where, 5 is duty ration, Y is chopping ration, Tp
result of v-i characteristic is is switching period of the converter, for 5 < 0.5 buck
operation and 5 > 0.5 for boost operation
Vg = -0.9Ig + 23.697 Int1.0 + 123.456(13.45Kns -Ig)1 (2)
2.3 DC motor model

515
 then the motor armature voltage equation is
When the dc separately excited motor is received
from a PV source by means of intermediate, the motor Eb Ra
voltage and torque equations are following
= +I (15)
°mp °mp
Vav=Eb +IavRa (7) Rearranging the above equation,
Te = C,Iav (8)
Eb = CeCO (9) Vm62p-Eb6mp-ImRa =0 (16)
2.4 Pump loads model The duty rate is equal to the following.
Pump loads is used pump of volumetric or centrifugal
types. The pump load will generate speed depended on
torques. The speed-torque characteristics of centrifugal ZIr2
(V )(2 ) |1 2 (17)
( TL1 ) and volumetric pump ( TL2) loads including
friction torque are given by
Motor armature current is obtained from (18), (19) so
as to give solar cell array maximum power.
TL1 = Al + Bl+CI+C1c8 Nm (1OA)
TL2 =A1+Blw Nm (lOB) Pm = Eav + I 2Ra (18)
RaIav+ EbIav -Pm =
3. MAXIMUM POWER OPERATION OF
SOLAR CELL ARRAY
I =v + + (19)
To maximize utilization of solar cell array, the power
converter is used in between solar cell array and motor.
The duty ration of the converter is changed to match
motor load to solar cell array. If we assume that power
is idea (the motor is employed from all array power) at
maximum power point, the power adsorbed by motor
equals the power delivered by solar cell array. In other
words,

P. = vaIaV =
V.I, (11)

Where Vf7, Iav the motor armature voltage and

are
current. Vm, I,m respectively solar cell array
are
voltage, current at maximum power point of solar cell
array. The motor armature voltage and currents are Current(p.u)
expressed in terms of solar cell array voltage, current at Fig. 5 MP and TME operation points trajectory of solar
maximum power point as. cell array at different solar insolations

vav = 3m, Vm (12) Where Eb is denoted by (9). The duty rate of

Im
(13) converter depends on the motor back emf changing
iav =c according to the motor load. When connecting the dc
°mp
(14) motor to load(centrifugal or volumetric pump-load), the
Vav ==Eb +IavRa
back emf at a given solar cell array power( Pm) is
obtained by solving (8), (10-A) and (19) in case of
PV Source I I
centrifugal pump load and (8), (10-B) and (19) for
volumetric pump load. The back emf is calculated by
Converter V, Pm, and the duty rate of converter is obtain from (17).
For training the Neural Network, calculated values is
Fi I
used as reference patterns.
Fig. 4 Equivalunt circuit of the system
4. TOTAL MECHANICALLY ENERGY
In Fig.4 transforming the motor equivalent circuit to
solar cell array side by substituting (12), (13) in (14) OUTPUT FROM SOLAR CELL ARRAY

516
 To make the solar cell array and the motor to operate
at maximum power points in all solar insolations is W w,

impossible for a given value of flux coefficient of the

machine. This is due to the motor v-i characteristic
dependent on motor flux coefficient and copper losses
in the machines. In case of the system make a
point( Pm, VI,4, ) of gross mechanical energy output Thbreshold

per day operate for a given daily insolation curve. The -

a~1
trajectory of operating point is shown in Fig.5. These Fig. 6 Schematic diagram of a neural network
operating points are VM > V, Im< In For a
considered machine, the optimal parameters Input vectors x= (xl ,x2, x,.. .,,)
is applied to the
Ce = 0.6626 make connected system operated at input layer. Net input in the hidden layer "j" is
maximum gross mechanical energy output[4]. The
voltage, current( Vm, Im ) corresponding to MP netI =
n
..XwI +t' (25)
i=l
i=l
operation at a given solar insolation is determined. The
voltage and current corresponding to TME operation
with Vm, Im are calculated by the following Where w is weight connected from ith input. ojh
equations. about j = 1,2,...Nh is the threshold value for the
hidden layer neurons. The output of neurons in hidden
layer is
Vm (2Ra
* VM= + 89.8Im 127 )
(Ra + Im0
2
) (20) h7 ( It h)
(26)
I*= 89.8I0127)
(Ra + ImO127 )
i

(21)
P =V I* And the input of neurons in the output layer is
(22)
Nh
I* = Eb + jEb + neto E
j=l
WkY' + Llk (27)
(23)
Where 0k, is the threshold value for neuron in the
where the back emf(E*) is calculated using (23), output layer. The output of neuron in the output layer is
(10-A), (8) in case of centrifugal pump load and (23),
(10-B), (8) in case of volumetric pump load at different Nh
solar insolations. When knowing Eb, the duty rate of Yk=f Y Wki Yj + k (28)
j=l
buck-boost converter is obtained from the following
equations for TME operation. The error function is as following.
/
E; 2 I2
5.E NEU IT
2V* LK2V;)
vM )j (24) Ep=Ep=-y(tp
2Z(-j_op )2) (28)
m
5. NEURAL NETWORKS E= EP (30)
P=1
In this paper, optimization chopping ratio for MP or
TME operation is determined by neural network. Neural Training courses of neural network is as following.
network has capability that can be training nonlinear Step 1. Construct the network and initialize the
mapping, advantage of high speed response and robust. synaptic weights with random values.
Neural network is high speed calculation time due to its Step 2. Apply an input vector the network and
high speed response time. Therefore, neural network is calculate the corresponding output values(25, 28).
very useful chopping ratio calculation of power Step 3. Compare the actual outputs with the desired
converter. Fig. 6 shows schematic diagram of neural outputs and determine a measure of the error.
network that has two input layers, four hidden layers, Step 4. Determine the amount by which each weight
one output layer. is to be changed and make corrections to each weight.
Step 5. Repeat step 2 to step 4 with all the training
vectors until the error for the vectors in the training set
is reduced to an acceptable value.

517
 The neural network training block diagram for MP or
TME operation shows in Fig. 7. Using mathematical
model the reference pattern for MP or TME operation is
calculated. Neural network training repeat until error is
less than the specified value by comparison between
reference pattern and neural network output. The used
learning rate and momentum factor is separately 0.55, -,a,

0.80.
Reference
Patterns for
MP/E & GA/E
operations

Solar InsolationE() c
Fig. 9 Comparison of chopping ratios for MP operation

/tontroller fMYp, (or) Y,M

Fig. 7. Block diagram for neural network training.

6. PERFORMANCE RESULT OF SYSTEM

A 120V, 9.2A, 1500 rpm PV supplied dc separately R---

excited motor driving (i) centrifugal pump load (ii)

volumetric pump load is considered in this paper. In the .,

above two cases of the calculated values are denoted in

Fig 8. In section 4, the mentioned following procedures
are training to an arbitrary synaptic. The training data Fig. IO Comparison of chopping ratios for TME
for neural network is obtained from Fig.8. The chopping operation
ratios corresponding to MP & TME operations are 1 00

obtained this trained adaptive neural network controller 80

for different solar insolations. As result of MP & TME 60

operations are respectively given to the implement of 40

comparison in Fig.9, Fig. 10. It is able to see from these 20

results that the predicted value of chopping ratio is
almost equal to the calculated values. The percentage 34 5 6 7 8 9 0 2 3 4 5 6 7 8 9

error in the calculation of chopping ratios from neural

-20

network controller is denoted in Fig. 11 and Fig. 12.

-40

According to these results, percent errors are shown that -60

errors under NN expectation are smaller than 2%, 700 -80

with centrifugal pump and volumetric pump load. It is -1 00

|-errin the prediction forvoueti pump Ioad --X--%errin the prediction for centrifuglpmpla

observed from research that medium results gained from Fig. I11 Error in the tracking of chopping ratio from NN
NN even learning to the tenth solar radiations is in controller for MP operation
ranges.

Fig. 12 Error in the tracking of chopping ratio from NN

Fig. 8 Computed values of chopping ratios controller for TME operation

518
 7. CONCLUSION
This paper is proposed PV optimization point
tracking of power converter that feed power to DC
motor witch drive pump loads. Neural network is
output optimization duty or chopping ratio for
MP/TME operation through training witch error
compared with mathematical model and neural
network output make loss than specified value.
This paper is computed chopping ratio for volumetric,
centrifugal pump loads and than compare to tracking
value by neural network. Compared value shows
excellent performance due to error rage is 2-7% and
prove validity of this paper.

This research was financially supported by the Ministry

of Commerce, Industry and Energy(MOCIE) and Korea
Industrial Technology Foundation(KOTEF) through the
Human Resource Training Project for Regional
Innovation

REFERENCES
[1] Zinger Z. and Braunstein A, "Optimum operation
of a combined system of a solar cell array and a dc
motor," IEEE Trans, on PAS 100, 1193-1197,
1981
[2] Appelbaum J, "Starting and steady-state
characteristics of dc motors powered by solar cell
generators,"IEEE Trans, on Energy conversion 1,
17-24, 1986
[3] Veerachary M, "Steady state and dynamic
performance of PV supplied dc motors fed from
intermediate power converters," Solar Energy
Materials and Solar Cells 61, 365-381, 2000
[4] Saied M.M, "Matching of dc motors to
photovoltaic generators for maximum daily gross
mechanical energy,"IEEE Trans, on Energy
Conversion 3, 1988
[5] Narendra K. S. and Parthasarathy K,
"Identification and control of dynamical systems
using neural network", IEEE Trans, on Neural
Networks 1(1), 4-27, 1990
[6] Takashi H. and Imakubo T, "Identification of
optimal operating point of PV modules using
neural network for real time maximum power
tracking control,"IEEE Trans, on Energy
Conversion 10(2), 360-367, 1995

519