You are on page 1of 10

PROGRESS IN PHOTOVOLTAICS: RESEARCH AND APPLICATIONS, VOL I, 283-292 (1993)

Applications A General Battery Model for


P V System Simulation
J. B. Copetti', E. Lorenzo' and F. Chenlo'

' C I E M A T , lnstituro de Energias Renouables, Avda. Complurense 22,Madrid, Spain: ' E T S I Telecomunicacion.
Instiruto de Energia Solar, U P M ,Ciudad Uniuersitaria, Madrid 28040, Spain

A battery model is proposed as a tool to simulate and optimize photovoltaic


( P V)/storuge systems. The normalized form of the equations with respect to the
batterji capacity allows u s to generalize its use for any type and size of lead-acid batteries.
The validity of this model to represent the battery voltage evolution during charge,
o vercharge arid discharge processes and to predict the performance of solar systems
under djfltrent operational conditions is analysed. Moreover, the battery eficiency
losses are presented as a function of the upper regulation thresholds of the charge
controllers und the size of the array and storage systems in a domestic application in
the climate of Madrid.

INTRODUCTION

L et us consider a stand-alone photovoltaic (PV) system as shown in Figure 1, composed of a PV


array, a battery and a charge regulator. The merits of such a system should be judged in terms
of the reliability of the electricity supply to the load and also in terms of the long-term efficiency
of both the PV array and the battery.
For our present purposes it is convenient to establish coulombic definitions for these parameters. In
this way, the reliability is quantified by the concept of loss of load probability (LLP), defined as the
ratio between the ampere-hours deficit and the ampere-hours demand, both with respect to the load,
over a long period of time. The battery efficiency, qB, is defined as the ratio of the ampere-hours removed
from the battery to the ampere-hours restored to the battery, for the same final and initial conditions
(state of charge and temperature). It is worth noting that 1 - qB represents the fraction of the
ampere-hours restored to the battery that is devoted to overcharge. Regarding the PV array, rather than
an efficiency defined classically as the ratio of output to input, we define a PV array utilization factor,
Fa,, as the ratio of the ampere-hours given by the PV array in real operation conditions to the
ampere-hours that the PV array would give at maximum power point conditions.
It is clear that a battery model for PV design purposes must be able to predict reliably the LLP, qB
and Fa, values when considering different sizes of the PV array and battery under different operation
conditions.
Several authors have proposed models for representing the battery operation during the charge and
discharge processes. The validity of such models is usually analysed in terms of their ability to represent
the battery voltage evolution during constant current and constant temperature charge and discharge.
Along these lines, we previously analysed' the models proposed by Shepherd,' F a ~ i n e l l iMenga
,~ et a1.4
and Mayer and Biscaglia.' Their agreement with experimental data obtained from tests with lead-acid
batteries at currents typical in PV applications and at different temperatures were verified. The correct
parameter values used in each model were fitted and the results obtained allow us to conclude that these
models can adequately reproduce the behaviour of batteries during discharge (the root mean square

l062-7995/93/040283-10 $10.00 Received 9 March 1993


0 1993 by John Wiley & Sons, Ltd Revised 20 May 1993
284 J . B. COPETTI ET AL.

charge controller

I
--------- I

photovoltalc
generator
V load

T
Figure 1. Schematic description of a stand-alone PV system

error ( R M S E see Appendix for definition) remains at less than 40mV per cell in all the models), but
significant limitations occur when they come to modelling charge, including overcharge (RMSE of the
order of 100 mV per cell for all the models). As a result, we proposed a modified model, based on the
Shepherd model, where equations for internal resistance, capacity, correction for temperature effects in
different parameters and an expression for overcharging are included. With this model it is possible to
reach a good agreement with actual data (the RMSE remains at less than 30 mV per cell for both charge
and discharge).
All these models have to be characterised for each individual battery. In other words, they use some
parameters whose values have to be fitted for each particular battery design and capacity. This requires
a detailed testing procedure that is often beyond the standard data sheets of the manufacturer and is
hence likely to be both expensive and time consuming, i.e. a major inconvenience for general PV
simulation purposes. Hence, despite the lower accuracy, it is desirable to have a normalized model with
a battery capacity that would keep the values of the parameters constant and therefore be valid for any
size of battery. An attempt to do just this was presented by Macomber.6 We also analysed this model
and found that the agreement with experimental data is poor (an RMSE of about 150 and 600 mV per
cell for discharge and charge, respectively).
In this paper we present a new normalized model with a battery capacity that shows reasonable
agreement with actual data. Furthermore, this model leads to adequate prediction for LLP, qe and Fa,
values. The study focuses on the lead-acid batteries, because at present these are the most commonly
used batteries in PV application owing to their relatively low cost and wide availability.

THE NEW PROPOSED MODEL


We carried out a set of experiments with several lead-acid batteries (Fulmen EF2050, Clo = 50 A.h;
Varta Vb624, C , , = 100 A-h; and ATSA Tudor, C , , = 180 A h) indicated for PV solar applications in
order to verify the voltage variations and the behaviour of internal resistance during charge and discharge
processes for different currents and tempexatures. Currents of I(C5)-I(Cl ,), where I(C,) is the charge
or discharge current expressed as a function of n hours of capacity, e.g. I(Cloo) = Cloo/lOO, and
temperatures of 5-45C were considered.
The complete description of the tests, procedures and their corresponding results may be found in
previous articles.
The experimental study allows us to find relations between voltage, current, state of charge and
BATTERY MODEL FOR PV SYSTEM SIMULATJON 285

temperature. The equations represent the set of experimental data concerning both processes with a
good degree of accuracy, but the values of coefficients are different for each battery tested.
In order to generalize the equations with respect to battery capacity, we verified the variation of the
resistance parameter values with the battery capacity and it was found that this variation conformed to
the equation RACA = RBCB(where R is the internal resistance, C is the capacity and subscripts A and
B represent different battery capacities). We consider that this assumption can be applied to different
types of lead-acid battery. The self-discharge was assumed to be negligible because in a PV system the
battery is in continuous operation and this decreases its effects.
The equations were rewritten as functions of 10 h of rated capacity (Clo), which is standard on the
manufacturers data sheet. Therefore, for this case, the capacities C, and C, correspond to C,, for
different batteries (sizes and/or types). In the following section we present the equations applied to a
single-cell battery (2 V).

= C2.085 - 0.12(1 - SOC)] -


c,,
~
~

1 + 11.3
+ soc1.5
~

The first term represents the voltage variation with the state of charge (SOC) (electrolyte concentration)
and the second the variation due to internal resistance variation.
Specific tests were carried out that made it possible to obtain a relationship between the internal
resistance and temperature, state of charge and current during charge and discharge. Owing to the current
rates utilized, the battery behaviour was considered as a sequence of steady states, disregarding the
transient effects. In Equation (l), the internal resistance is represented by the sum of series resistances
that correspond to different phenomena.
The temperature variation is AT = T - cef,
where the reference temperature (cef)
is 25C.
The SOC indicates how much electric charge is stored in the cell at a given time, defined as

sot=(,-;)
where the ratio between the charge delivered at the time of interest (Q = It) and the battery capacity
(C) represents the depth of discharge (DOD) or the fraction of discharge, i.e. DOD = 1 - SOC.
The efficiency of a battery during discharge is assumed to be 100%; however, the total amount of
useful charge available during discharge is limited by the current rate and temperature given by the
capacity equation. The capacity equation, normalized with respect to discharge current corresponding
to C1, rated capacity ( I l o ) is
C 1.67
_ -- + 0.005AT)
C,, 1 + 0.67(1/Z,0)0.9(1 (3)

When the discharge current tends to zero, the maximum capacity that can be removed is about 67%
over the C , , capacity at 25C.

(-/lfrrl/;/l</ I o/flrc/c cqlrtrliorl

For the charging process up to overcharging, the parameter values also differ from those used in
discharging
0.48
V, = [2 + 0.16SOC] + ~
(4)

SOC can be calculated easily at any point during the discharge period; nevertheless, during (re)charge
it is much more difficult, Generally, the efficient region is where SOC is below 0.7 and V , is less than
286 J. B. COPETTI ET AL.

2.3 V per cell. The efficiency drops to zero at full charge. A function that represents the charge efficiency
(v,) variation with state of charge and current rate is8
- -
qc = 1 - exp
a
~1/11,, b + - '1
where a and b are recharge constants that depend on the specific battery construction. We found a = 20.73
and b = 0.55 by fitting this equation to actual data for the three measured batteries, whose principal
construction characteristics are both their tubular positive plates and their low-antimony alloys.

O~YW/I~/IY
r r l+i r,r r l i o r i

The tests included charging where gassing occurred and the results demonstrated that the final charge
voltage ( V,,) increases with the current intensity and with decreasing temperature. The same argument
can be used to write a function for the gassing voltage (5) with different parameter values. These
dependencies can be described by the following equations
V,, = C2.45 + 2.011 ln(1 + l/Clo)](l - 0.002AT) (6)
V, = C2.24 + 1.97 ln(1 + l/Clo)](l - 0.002AT) (7)
The overcharge phenomenon (gassing evolution) can be represented by an exponential function, such as

where Ahrestored represents the ampere-hours stored in the battery with regard to the battery capacity C
for the charge current I during this hour. We assumed that 95% of the capacity was already restored at
the start of the overcharge.
The time constant of the phenomenon t is inversely proportional to charge current intensity and, as
an approximation, can be written

Therefore, the voltage evolution during the charge process is represented by Equation (4), up to the start
of gassing ( V , < V,) and by Equation (8) for overcharging ( V , > V,) until a constant final voltage ( 5,)
is reached.

7 1 i ~ ~ ~r / i A t r /o/'
i ~/ ~/ I ~( , t rriotlrd

We verify how this model represents the battery behaviour during the processes for the complete set of
data (three batteries, five current rates and three temperatures), using the values of mean bias error
(MBE; see Appendix for definition) and RMSE.
Table I shows the agreement between the normalized model and the set of actual data.

Table I. Comparison between values measured and calculated by the normalized model
~ ~________

Charge (V cell- I) Discharge (V cell - I )

5C 25C 45C 5C 25C 45C

MBE -0.013 -0.0053 -0.019 0.0096 -0.0034 0.014


RMSE 0.058 0.041 0.050 0.073 0.018 0.024
BATTERY MODEL FOR PV SYSTEM SIMULATION 287

It can be observed that the average value of RMSE for the three temperatures remains less than 50
and 40 mV per cell for charge and discharge, respectively. We believe that these values are reasonable
for a model that avoids characterization of each individual type of battery.
Furthermore, we explore the usefulness of this model for PV design purposes. For this, we developed
a PV system simulation program, working on an hourly basis, that allows us to compute monthly mean
values of LLP, qe and Fa, once the load and climatological conditions have been defined. We implemented
it for a typical domestic load profile' and for the Typical Meteorological Year (TMY) of Madrid," we
analysed the behaviour of a specific PV system, including just the above-mentioned tested batteries, and
we compare the different results obtained when incorporating different battery models into the simulation
program. The results corresponding to the model already mentioned, characterized for each type of
battery, are considered to be the results for the reference model, because they are able to provide an
accurate representation of the battery current and voltage progression.
Tables 11, I11 and IV summarize the results for a 12 V, PV system composed of a 282 Wp PV array
(six modules in parallel, each composed of 30 series-associated solar cells), a Tudor-STTH 180 A . h (Clo)
battery and a common series-interrupting charge controller, for which regulation thresholds are 15.2 V
and 12 V for overcharge and 10.8 and 12.3 V for overdischarge. The daily load consumption is equal to
67.5 A.h. Besides the values of LLP, qB and Fa,, we also calculated the fractions of time, over the full
length of the day, when the switches remain off, designated as Foc and Fod for overcharge and
overdischarge, respectively.

Table 11. Simulation results using the battery reference model

Month LLP VB (%) Fa, (%I Foc (%) (%)

1 0.243 87.55 94.52 3.63 23.39


2 0.191 83.75 94.81 3.72 18.05
3 0 78.25 85.10 11.96 0
4 0 74.4 82.31 14.86 0
5 0 83.0 80.08 16.53 0
6 0 81.0 74.32 17.78 0
7 0 72.34 75.05 13.30 0
8 0 74.16 76.18 10.48 0
9 0 79.6 76.02 15.55 0
10 0 83.18 88.10 10.88 0
11 0.106 99.0 99.45 0 9.58
12 0.145 100 100.0 0 14.25

Table 111. Simulation results using Macomber's model of the battery

Month LLP VB (%) Fa, (%) Foc (%I (%)

1 0.203 80.37 99.35 0 18.01


2 0.129 74.93 99.10 0 12.50
3 0 39.67 98.50 0 0
4 0 31.90 98.03 0 0
5 0 33.80 98.31 0 0
6 0 26.13 96.80 0 0
7 0 24.63 95.42 0 0
8 0 25.25 94.86 0 0
9 0 21.32 95.90 0 0
10 0 41.92 98.15 0 0
11 0.050 93.40 99.85 0 4.58
12 0.151 100 100.0 0 14.27
288 J. B. COPETTI ET AL.

Table IV. Simulation results using the new normalized model proposed

Month LLP VB (%) Fa, (7% Foc (%) Fod (%)

1 0.269 92.16 91.74 8.06 25.0


2 0.188 93.34 89.64 13.09 18.30
3 0 92.59 77.61 19.89 0
4 0 90.83 73.72 24.44 0
5 0 91.0 76.74 24.19 0
6 0 83.50 72.79 21.53 0
I 0 73.53 74.64 16.67 0
8 0 63.0 84.86 4.17 0
9 0 69.56 83.00 10.14 0
10 0 92.12 83.11 16.40 0
11 0.146 100 99.37 2.64 14.03
12 0.154 100 100.0 0 14.52

As the battery efficiency qB cannot be calculated from the previous definition, we used another concept
that considers the charge current (I,) and the charge efficiency qc (Equation (5))

where the second term represents the fraction of the charge current lost in gassing.
Three battery models are compared: the reference model, Macomber's normalized model and the
normalized model proposed here.
The LLP prediction depends on the reliability desirable for a given load profile. For the case presented,
where the dail load consumption is constant for all of the period analysed, the results obtained for the
LLP are of the same order for the three models and this would be in agreement with the widely extended
practice of using simple battery models for sizing stand-alone PV systems."-'
However, for the months with higher radiation (spring and summer), considering the same daily load
consumption, the qB, FOcand Fa, values are principally affected by the charge and overcharge model. In
this sense, it can be observed to be a general good performance of the normalized model, which indicates
the realistic situation of the smaller battery efficiency owing to a major number of hours in overcharge,
in contrast to Macomber's model, which is not able to represent the charge process.

MODEL USEFULNESS FOR P V SYSTEM DESIGNERS


The success and acceptance of PV systems by the consumer is typically judged by the reliability and
performance of these systems, which are mainly affected by their size and by battery life, which depends
on the operation conditions. The proper operation of the battery is mainly imposed by the specific
method and algorithm of the charge controllers, which should prevent the overcharge or overdischarge
operation.
The PV system sizing problems generally deal with estimation of LLP values. This is mainly related
to the random nature of the solar radiation. Because of the natural variability associated with the climate,
the differences observed in LLP estimations when using different battery models (see Tables I1 and 111)
cannot be important and simple energy transfer models would suffice.
In contrast, the battery behaviour, and consequently the battery life, is strongly related to the
environment of operation. There is, at present, a lack of information in the open literature concerning
this, which means that the specification of charge controllers is left entirely to the intuition of the PV
designer. Some experimental work is under way14 to cast some light on this problem. We believe that
BATTERY MODEL FOR P V SYSTEM SIMULATION 289

a simulation tool such as the battery model proposed here can be of enormous help to generalize lessons
and conclusions based on such experiences.
As an example, let us analyse the impact of the upper threshold voltage for overcharge limitation on
battery maintenance requirements, again for stand-alone PV systems devoted to domestic applications
in the climate of Madrid.
For this it is useful to adopt a PV array and battery definitions related to the load: on a daily basis,
the PV array capacity, C,, is defined as the ratio between the mean PV array energy production and
the mean load energy demand. The storage capacity, C,, is defined as the maximum energy that can be
taken out from the battery divided by the mean load energy demand. Thus

c, = VAG.
~ , c,=-C
L L
where A is the PV array area, q is the PV array efficiency, G is the mean daily irradiation on the PV
array, corresponding to the worst month, L is the mean daily energy consumption and C is the rated
battery capacity.
The use of such definitions gives a generalized form to the results. All the I-V equations governing
the electrical behaviour of the PV system can be arranged as functions of such definitions.
Figure 2 shows the yearly mean value of the battery efficiency loss ( 1 - qB) versus the overcharge
upper threshold for two different cases: C, = 1, C, = 2 and C, = 1, C, = 5. The increase of the voltage
threshold value guarantees that the battery reaches the full charge state, but, on the other hand, it means
an increase of the gassing losses and a decrease of the battery efficiency. For the same array capacity
(C,) but more storage capacity (C,), there is an increase of both the overcharge and battery efficiency
loss because it is necessary to take a long time to reach the specific voltage threshold.
Figure 3(a) shows the monthly mean values of 1 - qa, corresponding to C, = 1 and C, = 2. As can
be expected, July (the month that registers the major level of radiation) presents the major losses in
efficiency; in this period, the battery is in overcharge continuously. Figure 3(b) shows the deficit of energy
concerning the same period. The battery is always discharged to the lower thresholds of the overcharge
and it is not able to supply the energy for the load.

25

20

n
8 15
W

c-
I
3 0

5
r
0 I-- I3 2 I3 6 14 14 4
Regulation thresholds for overcharge (V)
14 8 1s 7.

mcs = 2 u c s = 5
Figure 2. Yearly mean value of 1 - qe and overcharge upper thresholds for different system configurations
290 J . B. COPETTI ET AL.

15 -

n
#
W
-
Po
I I
4 I L M
5 -
I A

oo
:v)
13.2 13.6 14

Regulation thresholds fw overchargc


I 15.2

m f e b =may n j u l m o c t

40

' 112
I
13.6
I
14
I
14.4
Regulation thresholds for overcharge (V)
14.8
I
1s.2

Figure 3. Monthly mean values of 1 - qB (a) and deficit (b) versus the overcharge upper thresholds

The 1 - qB figures can be correlated with water loss using the basic principle derived from Faraday's
law: 1 A . h of overvoltage = 0.336 ml of water loss. For the previous cases, C, = 1, C, = 2 and C, = 1,
C, = 5; in Figure 4 we present the yearly water consumption as a function of the regulation overcharge
threshold.
BATTERY M O D E L FOR P V SYSTEM SIMULATION 291

h
t
2 1
W
/

c /
0

/
E
/

,/ /
/
Y I

/ ,/ /
/
,/-
I _/- /
I
1 _----*
O * r1 6 - 14 I4 4 14 8 15 2
Regulation thresholds for overcharge (V)
+ cs=5 + cs=2
Figure 4. Yearly water consumption for different overcharge upper thresholds

A set of data obtained from tests carried out on lead-acid batteries at different currents and temperatures
allows us to write a model to represent the battery behaviour during the charge, overcharge and discharge
processes, which makes it possible to reach good agreement with actual data. However, the difficulty of
using this model, owing to the large number of parameters to be identified for each type of battery, leads
us to rewrite the equations as a function of battery capacity. Thus, the values of parameters can be fixed
and the model is valid for any size of lead-acid battery.
We verify the validity of this normalized battery model usinga PV system simulation program to
estimate the reliability of the electricity supply to the load ( U P ) and the efficiency of the battery (qa)
and the PV array (Fa).The results were compared with the previous model, which required specific
parameters for each type of battery, and to Macombers model. Despite having less accuracy compared
to the specific model, it presents good performance with regard to the battery operation. This is in
contrast to Macombers model, which presents significant limitations in the charging process.
With this new tool it is possible to realize mainly those studies concerning battery behaviour, such as
we present in our example, where the variation of the upper threshold regulation of the charge controller
is connected to the battery efficiency losses for the overcharging operation. In this way, other results
could be evaluated for different system configurations and parameter variations when considering these
effects in the life of the battery.

Acknowledgements
J. B. Copetti is grateful to CNPQ (Conselho Nacional de Desenvolvimento Cientifico e Tecnologico),
Brazil for sponsoring this research.

A PPEIYDIX
The models agreement with experimental data is evaluated using statistical indicators: mean bias error
( M B E ) and root mean square error ( R M S E ) , defined as
292 J. B. COPETTI ET A L .

l N
MBE=- C Ci-Mi
N i=,

where Ci and Mi are calculated and measured values, respectively, a n d N is the number of values.

REFERENCES

1. J. B. Copetti, F. Chenlo and E. Lorenzo, Comparison between charge and discharge battery models and real
data for PV applications, 11th European Photouoltaic Solar Energy Conference, Harwood Academic Publishers,
Montreux, Switzerland, 12-16 October 1992, pp. 1131-1134.
2. C. M. Shepherd, Design of primary and secondary cells, J . Electrochem. SOC., 112, 657-664 (1965).
3. W. A. Facinelli, Modeling and simulation of lead-acid batteries for PV systems, 18th Intersociety Energy
Conversion Engineering Conference, Orlando, Florida, August 1983, pp. 1582- 1588.
4. P. Menga, R. Buccianti, R. Giglioli and L. Thione, Model of the lead-acid battery, 7th EVS - Electric Vehicles
Symposium, June 1984, Versailles, France, pp. 26-29.
5. D. Mayer and S. Biscaglia, Use of the modeling of lead-acid battery operation for the development of a state
of charge meter, 10th European Photouoltaic Solar Energy Conference, Kluwer Academic Publishers, Lisbon,
Portugal, 8-12 April 1991, pp. 1209-1213.
6. H. L. Macomber, Engineering Design Handbook for Stand-alone PV Systems, Report no. M108, Monegon Ltd.,
Gaithersburg, Maryland, 1981.
7. J. B. Copetti and F. Chenlo, Internal resistance characterization of lead-acid batteries for PV rates, 11th
European Photouoltaic Solar Energy Conference, Harwood Academic Publishers, Montreux, Switzerland, 12-16
October 1992, pp. 1116-1119.
8. R. Kaushik and I. G. Mawston, Coulombic efficiency of lead-acid batteries particularly in remote-area
power-supply (RAPS) systems, J. of Power Sources, 35, 377-383 (1991).
9. A. Krenzinger and M. Monteiro, Energy consumption patterns of the rural PV market in Spain, 7th European
Photovoltaic Solar Energy Conference, D. Reidel Publishing Company, Seville, Spain, 27-3 1 October 1986, pp.
382-386.
10. M. H. Macagnan, E. Lorenzo and C. Jimenez, Solar radiation in Madrid, Int. J. Solar Energy, in press.
1 1. R. N. Chapman, Development of sizing nomograms for stand-alone photovoltaic-storage systems, Solar
Energy, 43(2), 71-76 (1989).
12. M. Egido and E. Lorenzo, The sizing of stand-alone PV systems: a review and a proposed new method, Sol.
Energy Mater. and Sol. Cells, 26, 51-69 (1992).
13. M. Hill and S . McCarthy, PV Battery Handbook, Hyperion Energy Systems Ltd., Ireland, 1991.
14. J. Dunlop, W. Bower and S. Harrington, Performance of battery charge controllers: first year test report, 22nd
I E E E Photouoltaic Specialists Conference, IEEE Publishers, Las Vegas, Nevada, 7-1 1 October 1991, pp. 640-645.

You might also like