You are on page 1of 15

See

discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/227788717

Dynamic battery model for photovoltaic


applications
ARTICLE in PROGRESS IN PHOTOVOLTAICS RESEARCH AND APPLICATIONS MAY 2003
Impact Factor: 7.58 DOI: 10.1002/pip.480

CITATIONS

READS

52

177

2 AUTHORS, INCLUDING:
Santiago Silvestre
Polytechnic University of Catalonia
65 PUBLICATIONS 615 CITATIONS
SEE PROFILE

Available from: Santiago Silvestre


Retrieved on: 16 November 2015

PROGRESS IN PHOTOVOLTAICS: RESEARCH AND APPLICATIONS


Prog. Photovolt: Res. Appl. 2003; 11:193206 (DOI: 10.1002/pip.480)

Applications

Dynamic Battery Model for


Photovoltaic Applications
D. Guasch and S. Silvestre*,y
Semiconductor Devices Group of the Electronic Engineering Department, Universitat Polite`cnica de Catalunya,
Campus Nord UPC, C/Jordi Girona 13, Mo`dul C4, 08034 Barcelona, Spain

The main purpose of this paper is to offer a useful battery model for simulation of
stand-alone photovoltaic applications. At present, many models for battery behaviour
simulation are available. Owing to the complex response of this element, these models are mainly concerned with stationary working point conditions. This paper presents an enhancement of a generic battery model, achieving a dynamic battery model
for photovoltaic applications. It includes the use of automatic parameter extraction
techniques and the solving of numerical calculation problems. It also introduces
some new concepts, such as the maximum available capacity and the level of energy.
Finally, a battery state of health estimation, has been added to complete the battery
behaviour description, including non-ideal effects, such as battery capacity reduction
and self-discharging current. Copyright # 2003 John Wiley & Sons, Ltd.

INTRODUCTION

his paper seets to offer a model for batteries working in photovoltaic environments. Despite the fact that
batteries are widely used, the behaviour of their electrochemical reactions hide an unexpected complexity. The problem of simulating leadacid batteries by means of equivalent electrical circuits has been
described in the literature,14 and different models can be found to have different degrees of complexity and simulation quality. Some models offer a good compromise between complexity and precision, but the problem of modelling batteries for the analysis of photovoltaic systems using leadacid batteries has not been satisfactorily solved
yet. Maybe because most of the models have been developed for constant current battery charge and discharge,
but not taking into account the dynamical behaviour of leadacid batteries in photovoltaic applications.
The paper proposes an electrical battery model and a new technique for characterizing a leadacid battery
operating within a photovoltaic system. The core is based on the well-known model of Copetti et al.3 This
model considers the battery behaviour as a sequence of steady states, disregarding transient effects and assuming constant currents and temperatures in all tests. Therefore, numerical discontinuities appear in the transitions
between steady stages in dynamical applications. Moreover, all numerical values of the model are fitted to a
specific battery, and must be adjusted for every new battery.
After an exhaustive study of the Copetti model, the deficiency mentioned led us to develop many solutions for
dynamical simulations. On the one hand, numerical discontinuities are avoided by considering some constants
in the Copetti model as new variables in this model, and by adopting linear approaches between steady states.
On the other hand, for fitting the numerical values of model, the LevenbergMarquardt algorithm5,6 is applied,
obtaining higher results. Furthermore, some parameters have been redefined (SOC, state of charge) and new
ones have been added (LOE, level of energy, and SOH, state of health).
* Correspondence to: S. Silvestre, Electronic Engineering Department, Universitat Polite`cnica de Catalunya, Campus Nord UPC, C/Jordi
Girona 13, Mo`dul C4, 08034 Barcelona, Spain.
y
E-mail: santi@eel.upc.es
Published online 30 January 2003
Copyright # 2003 John Wiley & Sons, Ltd.

Received 28 February 2001


Revised 2 November 2001

194

D. GUASCH AND S. SILVESTRE

Results have been compared with experimental data. Some of the experiments have been repeated, and are
included in this paper. Thus, an enhanced battery model is discussed for dynamical photovoltaic solar energy
applications.

BATTERY MODEL
A basic electrical approach to a battery is shown in Figure 1 and Equation (1). It consists of a voltage source Vg
and a resistor Rg. This first generic model includes the main variables of the system: the battery state of charge
SOC, the current flowing across the battery I, and the working temperature T.

Vbat Vg IRg

Vg f SOC
Rg f I; SOC; T

The voltage source Vg represents the voltage at open circuit between the battery terminals. This voltage is due
to energy stored into the battery through the electrochemical reactions. Obviously, this term depends directly on
the energy stored. On the other hand, Rg represents the resistance that the battery offers to the energy flow. This
value includes the effects of working point (I, SOC, T) and the health of the battery. A damaged battery shows a
high value of resistance irrespective of its working point. Also, the resistance is inversely proportional to the
state of charge. Furthermore, at the same time as the battery is discharging, the resistance value is increasing.
Note that all physical effects are transformed to electrical space and are modelled in these terms.
The state of charge must be considered as an indicator of the electrical charge stored by the battery. The value
range available is 0 < SOC < 1. Equations ( 24) describe the behaviour of this indicator. Terms involved are:
C(t) the battery capacity, c the charging efficiency and I(t) the current flowing through it. Also Cnominal is the
rated battery capacity ( at n hours), Ct coef, Acap and Bcap are model constants, T is the temperature variation
from the reference value of 25 C, Inominal the discharge current corresponding to Cnominal rated capacity, n is the
time in hours, and , are the temperature coefficients.
Note that SOC must be understood as the relation between energy accepted and the capacity available at all
times. The inner integral term models the energy accepted over the battery working life. Also, the outer integral
term models the battery capacity due to the working point environment at any given time. Both terms are functions of time and are evolving continuously. When SOC is unity the battery cannot accept more energy from the
system, because the energy stored fills all the battery capacity. And when SOC is zero the battery has no energy.
SOCti 

Ct

1
Cti

ti

c t It @t

1



Cnominal Ct coef
2

Bcap 1 C Tt C Tt
jItj
1 Acap
Inominal

Inominal

Cnominal
n

Figure 1. Basic equivalent circuit for the battery


Copyright # 2003 John Wiley & Sons, Ltd.

Prog. Photovolt: Res. Appl. 2003; 11:193206

DYNAMIC BATTERY MODEL FOR PV APPLICATIONS

195

For a more intuitive use of the SOC concept in photovoltaic applications, a new indicator has been introduced:
LOE, the level of energy. This indicator shows the amount of energy available in the battery under normal working conditions. Equations (5, 6) show how LOE is defined, where T1 and T2 define the known temperature range
of battery operation. Note that LOE depends only on the constitutive parameters of the device and the accumulated charge over time, not on the working environment of the battery. The LOE calculus may be carried out by
considering the maximum available battery capacity Cn, obtained by taking into account the range of possible
current and temperature values. So, Cn is evaluated from Equation (3), with battery current equal to zero and
maximum temperature value (if a standard temperature range [40 , 40 C] is considered for a first theoretical
approach, then T 40 C) as is reflected in Equation (6). Thus, LOE is not limited to the higher limit of unity,
but LOE values near or greater than unity are undesirable in order to avoid damaging the battery.
LOE represents directly the energy stored in the battery (Cn taken as constant). Therefore SOC and LOE are
complementary. Furthermore, a battery can have SOC 1 and LOE 6 1; this indicates that the battery is saturated, but only at a percentage of nominal capacity.
LOEti 

1
Cn

ti

c t It @t

1


Cn maxC

I0

TT1 ;T2 

These equations show that this model is using the Coulomb counting method for determining SOC and LOE.
Once the initial value of SOC is determined, the battery evolution is evaluated as a function of the working zone
(saturation, overcharge, charge, discharge, overdischarge or exhaustion) described below. These areas are an
enhancement of features of the Copetti model.3 Figure 2 illustrates the battery voltage evolution due to current

Figure 2. Working areas of the battery


Copyright # 2003 John Wiley & Sons, Ltd.

Prog. Photovolt: Res. Appl. 2003; 11:193206

196

D. GUASCH AND S. SILVESTRE

Table I. Battery working zone conditions


Battery voltage Vbat

Working zone

Vsc

Saturation zone
Overcharge zone
Charge zone
Changing from charge to discharge or
from discharge to charge
Discharge zone
Overdischarge
Exhaustion

Vc
Vcdc
Vdc

Zone conditions

Ibat > 0
Ibat 0
Ibat < 0

Vbat Vec
Vec 5 Vbat 5 Vg
Vbat < Vg
Vc 5 Vbat 5 Vdc

c 0
0 < c < 1

Vbat > 09VN


09VN 5 Vbat 5 07VN
Vbat < 07VN

c  1

flowing through it. Thus, different possible working zones for a 2 V element can be observed. For the first 16 h
the current is flowing inside the battery, and it evolves into charging, overcharging and saturation zones. From
16 to 27 h the current is flowing outside the battery and it evolves into discharging, overdischarging and exhaustion zones.
Table I shows the notation used to identify the battery voltage in each of the possible working zones. Key
parameters included are battery current Ibat, nominal battery voltage VN, gassing voltage Vg, maximum battery
voltage Vec. The charge efficiency c has been included in order to clarify its behaviour. All these terms and
parametric equations governing the described working areas are described in the following paragraphs.
Discharge zone
In this area the battery is delivering energy to the system. Together with the charging zone, these are the best
areas for the battery to work in to avoid damage. The battery voltage evolution is determined by Equation (7).
Note that in the equation, current is always taken as an absolute value. Also, SOC is used instead of LOE as it is
more realistic and it describes better the behaviour of the battery, its voltage does not represent the exact amount
of energy, but the amount of energy is appropriate to the working point. For example, at high discharge, current
rates can extract less energy from battery than at lower current rates.
!
jItj
P1dc
P3dc
Vdc t Vbodc  Kbodc 1  SOCt 

P5dc 1  rdc Tt


C10 1 jItjP2dc SOCtP4dc

Overdischarge zone
The battery works in this area if excessive energy has been extracted from it. Voltage is decreasing quickly,
owing to the nonlinear effects of electrochemical reactions, as indicated in Equation (7). This is a dangerous
area for the battery and must be avoided by disconnecting the battery from the system. Usually, battery controllers carry out this disconnection until new energy can fill the battery again.
Exhaustion zone
This is the most dangerous zone and can damage seriously the battery. Despite the fact that Equation (7) continues to be valid, SOC must not be zero in order to avoid a singularity. Also a voltage remainder can appear, but
no energy can be extracted from the battery.
Charge zone
The battery is placed in this area when there is excessive energy on the system. The energy is absorbed and
causes the voltage to increase, as Equation (8) describes. As described above, together with the discharging
area, these are the most secure zones for battery function, and system controllers must ensure the battery does
not leave them. A charge efficiency factor, c t, must be considered in order to reflect that only a fraction of the
Copyright # 2003 John Wiley & Sons, Ltd.

Prog. Photovolt: Res. Appl. 2003; 11:193206

DYNAMIC BATTERY MODEL FOR PV APPLICATIONS

197

theoretical energy is really stored. Equation (9) describes the relationship between the efficiency factor, the
current rate and SOC.
!
It
P1c
P3c
Vc t Vboc Kboc SOCt

P5c 1  rc Tt


8
C10 1 ItP2c 1  SOCtP4c
2

6
c t 1  exp4

acmt
7
5SOCt  1
It
bcmt
I10

Overcharge zone
If too much energy is stored in the battery it can start to saturate, decreasing the charge acceptance due to the
gassing effect in the electrolyte. Equations (1014) describe this effect. Note that this does not imply that the
battery has a lot of energy, only that the available capacity is almost full. The model assumes that a battery enters
this zone when Vbat 5 Vg, where Vg is the gassing voltage. This is a dangerous zone, owing to the gassing effects
which liberate extra hydrogen to the atmosphere with the risk of active mass losses, including explosions.
This effect appears when SOC(t) C(t) 95% of the maximum charge. Some battery models consider this
value of battery capacity as a constant included in the equations governing the overcharge zone.3 This standard
approximation to determine the start of battery overcharge can lead battery models to critical situations, in the
worst cases a battery voltage discontinuity can appear in the change from charge to overcharge zones, and
numerical algorithms will not solve it.
In order to solve this discontinuity between charge and overcharge zones, the consideration that the battery
must achieve at least a constant of 95% of charge3 to enter the overcharge area must be avoided. Determination
of the beginning of battery overcharge is based on different criteria3,7 by means of the initial evaluation of gassing voltage Vg.
A better solution for the evaluation of the battery voltage at the overcharge zone, eliminating the possibility
of discontinuity effects when the battery changes from charge to overcharge zones, is shown in Equation (12),
taking into account the battery SOC at gassing starting, defined as SOCVg. The state of charge of the battery
corresponding to the beginning of electrolyte gassing, as shown in Equation (14). Figure 3 shows the evolution
of SOCVg as a function of the battery temperature and current.
Thus Equation (12) describes the battery voltage evolution for Vbat when Vbat 5 Vg. Where , defined in
Equation (13), is a time factor for the overcharge phenomenon.




It 
Vg t Agas Bgas ln 1
1  gas Tt
C10

10



It
1  fc Tt
Vec t Afonsc Bfonsc ln 1
C10

11



LOEtCn  SOCVg tCt
Vsc t Vg t Vec t  Vg t 1  exp
Itt

12

t

Asc
 Csc
It
1 Bsc
C10

SOCVg SOCjVcVg
Copyright # 2003 John Wiley & Sons, Ltd.

13

14
Prog. Photovolt: Res. Appl. 2003; 11:193206

198

D. GUASCH AND S. SILVESTRE

Figure 3. Variation of SOC corresponding to gassing effect as functions of current and working temperature

Saturation
The battery cannot accept energy indefinitely. Finally it does not accept any more energy. At this point its voltage is maximized. Note that both Equations (11) and (12) converge to the same voltage. Also this area markedly affects battery health, and typically the system controller disconnects the solar panel to force battery
discharge through the loads.

Changing from charge to discharge or from discharge to charge


When the battery current in dynamical applications forces the transition from the charging to the discharging
zone or vice versa, voltage equations of both zones can introduce a numerical singularity.
Vc jI0 battery charging; Equation 8 6 Vdc jI0 battery discharging Equation
Then, if a strong reliability on dynamic working is required to the model, as can be the case for photovoltaic
applications, this effect may cause great problems in simulation numerical algorithms. This effect appears in
most battery models14 mentioned in the literature. The discontinuity at this point has a simple solution: to
allow linear battery voltage evolution at the change of operation mode. In order to avoid the possibility of a
battery voltage discontinuity appearing, a threshold current value jI j is defined, to identify the boundary
between charge and discharge battery operation areas, where battery voltage is governed by the respective
equations. When jIj < jI j, a new equation must be formulated to solve the possible discontinuity in battery
voltage. Considering a sufficiently small jI j, Equation (15) can be formulated to evaluate the battery voltage
Copyright # 2003 John Wiley & Sons, Ltd.

Prog. Photovolt: Res. Appl. 2003; 11:193206

DYNAMIC BATTERY MODEL FOR PV APPLICATIONS

199

Table II. Description of experiments


Experiment

Number of days tested


Date of test start
Number of samples

5
30/3/01
3614

5
22/5/01
3583

4
7/5/01
2965

4
20/6/01
2885

evolution in this area, where jIj < jI j, when the battery is changing from charge to discharge or from discharge
to charge.
Vcdc

VcjI  VdcjI
VcjI VdcjI
I
2I
2

15

Where Vc and Vdc are the battery voltage equations for charge and discharge zones, respectively.

PARAMETER EXTRACTION
An inherent problem of modelling is the accuracy of the model parameters with regard to the reality of the
device under characterization. The use of nominal values for a battery family, or of individually adjusted values
from the manufacturer, can introduce an error that may be important, depending of working conditions and the
lifetime.
In the model presented above, no values have been proposed for any of the constants involved in the model
equations. These constants can be considered as model parameters. A method for automatic model parameter
adjustment is discussed below. Note that this algorithm is valid for static tests or for free-running stand-alone
PV systems.
In order to ensure that parameter deviations are appropriate, nominal values, when these nominal values are
known, or in order to estimate their values, when these values are not fixed anywhere, an automatic parameter
extraction method is proposed. To implement the parameter extraction function, the LevenbergMarquardt
algorithm5,6 has been chosen because of to its tested efficiency. Figure 4 shows these differences for simulation
results.
As can be seen in Figure 4, the battery voltage evolution has been measured over a days operation in a standalone PV system where loads are connected at night. The measured battery voltage is compared with simulation
results of the battery model using two sets of parameter values:
*
*

nominal values from the literature3,8,9 (see Table III),


parameter values obtained from the parameter extraction function.

As Figure 4 shows, a better approach is obtained by the model simulation after automatic extraction of the
model parameter values.
Parameters considered under readjustment may be all or only a part of the constants included in the battery
model equations described. The parameter extraction procedure developed follows these steps:
1.
2.
3.
4.
5.
6.

To
To
To
To
To
To

obtain real data from PV system monitoring;


assign initial values for: LOE, SOC and C10 parameters;
assign initial values to the rest of the parameters: constants in model equations;
recalculate initial LOE, SOC and C10, using the LevenbergMarquardt algorithm;
recalculate the rest of the parameter values using the LevenbergMarquardt algorithm;
apply a statistical process to parameter values to obtain final values.

In order to validate the simulation results obtained, several experiments were carried out at the Semiconductor Devices Group of the Electronic Engineering Department at the Universitat Polite`cnica de Catalunya. In
each of these experiments, samples of voltage and current in all nodes of the system under test were monitored.
Copyright # 2003 John Wiley & Sons, Ltd.

Prog. Photovolt: Res. Appl. 2003; 11:193206

200

D. GUASCH AND S. SILVESTRE

Figure 4. Comparison between measured voltage, simulation results from nominal values3 and simulation results using
parameters adjusted by automatic extraction function

The monitored photovoltaic system was composed of: 8 solar panels Atersa A-85 (36 cells in series, Isc 44 A,
Voc 205 V, 2 panels in series, 4 panels in parallel); a leadacid battery Tudor 7TSE70 (12 elements of 2 V in
series, C10 550 A h); a battery controller Atersa Leo 1 12/24; an Atersa Taurus 1024 (1 KVA) inverter and
different combinations of resistive loads. Four different experiments were carried out during the spring and
summer of 2001 in Barcelona city (see Table III). The sampling rate used was one sample every 2 min. The
irradiance and battery current profiles were monitored for each of these experiments. The corresponding irradiance and battery current profiles are shown in Figures 5 and 6, respectively.
Starting from an initial analysis of the model, 11 parameters of the proposed battery model were included in
the parameter extraction algorithm: C10, SOC0 (initial SOC), LOE0 (initial LOE) and the constants Vbodc, Kbodc,
P1dc, P2dc, Vboc, Kboc, P1c, P2c. Note that there are three groups of parameters. The first group (C10, Vbodc, Kbodc,
Vboc and Kboc) affects the battery constitutive conditions, such as water level; the second (P1dc, P2dc, P1c and
P2c) is related to resistive losses; and the third (SOC0 and LOE0) shows the initial conditions of the experiment.
Table III. Nominal values for model parameters3 applied to leadacid battery Tudor 7TSE70
Cnominal 550 A h
n 10 h
Ctcoef 167
C 0005 C1
 C 0 C2
Acap 067
Bcap 09
Vbodc 2085 V
Kbodc 012 V

P1dc 4 VA h
P2dc 13
P3dc 027 V h
P4dc 15
P5dc 002 V h
acmt 2073
bcmt 055
Vboc 2 V
Kboc 016 V

Copyright # 2003 John Wiley & Sons, Ltd.

P1c 6 VA h
P2c 086
P3c 048 V h
P4c 12
P5c 0036 V h
Agas 224 V
Bgas 1970 V h
gas 0002 C1
fc 0002 C1

rdc 0007 C1


rc 0025 C1
Afonsc 245 V
Bfonsc 2011 V h
A sc 173 h
B sc 852 h
C sc 167

Prog. Photovolt: Res. Appl. 2003; 11:193206

DYNAMIC BATTERY MODEL FOR PV APPLICATIONS

201

Figure 5. Irradiance profiles corresponding to the four experiments carried out

Figure 6. Current profile, measured at the battery terminals, due to the energy flowing from solar panels, load consumption
and battery controller management

Copyright # 2003 John Wiley & Sons, Ltd.

Prog. Photovolt: Res. Appl. 2003; 11:193206

202

D. GUASCH AND S. SILVESTRE

Table IV. Calculated parameter values by the LevenbergMarquardt algorithm


Experiment

C10(A h)
SOC0
LOE0
Vbodc (V)
Kbodc (V)
P1dc (VA h)
P2dc
Vboc (V)
Kboc (V)
P1c [VA h]
P2c

5506983
09279
07684
2147
0284
4083
6634
1980
0149
5923
0024

5499392
08741
07238
2160
0281
3938
6141
1948
0214
9797
0292

5499671
09068
07711
2188
0329
9561
7726
2048
0160
8304
0854

5509557
09617
08846
2129
0235
4475
10656
1952
0167
6469
0018

Results obtained from the automatic parameter extraction function, after applying the LevenbergMarquardt
algorithm to the proposed battery model with initial parameter values listed in Table III, are shown in Table IV.
Note that the procedure described above was applied step by step to all the experiments.
Introducing parameter values shown in Table IV to the battery model, Matlab simulations of the PV system
described above,10 have been carried out for each of the four experiments. A second set of Matlab simulations of
the same system has been done, considering battery parameter values described in Table III as nominal parameters.3 Finally, all simulation results obtained have been compared for each experiment with measured data.
Figure 7 shows the measured battery voltage evolution and simulation results obtained from the battery
model, using both nominal parameters and parameter values obtained from the automatic parameter extraction
algorithm.
The mean error values between measured battery voltage evolution and simulation results have been evaluated from Equation (16):
Mean error

n
1X
jVmb  Vsb j
n i1
Vmb

16

where n is the total number of samples, Vmb is the measured battery voltage and Vsb the battery voltage obtained
as result of the simulations.
The mean error was reduced by the parameter extraction procedure as Table V shows, where mean error is
shown as a percentage with to respect nominal 24 V. Note that results obtained from a healthy battery, such as
the Tudor 7TSE70 battery used, are even better than the first results. Thus, it is demonstrated that the proposed
model really achieves improvements in simulation accuracy as well as in calculus reliability.
Furthermore, in Figure 8 an approach to SOC and LOE is plotted for each experiment. Note that LOE directly
indicates the amount of energy in the system and SOC shows only the ability to accept or deliver energy of the
battery.

BATTERY STATE OF HEALTH


Electrochemical effects, such as corrosion or water losses, lead to a variation of internal constitutive parameters.
So a battery state of health indicator (SOH) must be considered in the model. Two main effects have been taken
into account: reduction of the battery capacity and the presence of a self-discharge current.
A realistic approximation to the battery behaviour, over long time periods, needs to consider two additional
factors: the working temperature and the working zone. Then, both are to be included into the model through
two new effects: effective battery capacity reduction and a self-discharge battery current. So, the state of health
becomes the reference indicator for model implementation. SOH can be evaluated as a result of the influence of
Copyright # 2003 John Wiley & Sons, Ltd.

Prog. Photovolt: Res. Appl. 2003; 11:193206

DYNAMIC BATTERY MODEL FOR PV APPLICATIONS

203

Figure 7. Battery voltage measurements of four experiments and simulation results obtained

both health factors, as shown in Equation (17), where a battery in a perfect state corresponds to SOH 1 and a
fully damaged battery corresponds to SOH 0.
SOHti 1 

ti

T wz @t

17

1

where T s1 and wz s1 are the temperature health factor and the working zone health factor, respectively,
defined below.
Table V. Mean error between simulation results and real measurements
Experiment
Mean error (% nominal parameter values)
Mean error (% extracted parameter values)
Copyright # 2003 John Wiley & Sons, Ltd.

192
067

182
066

205
116

186
087

Prog. Photovolt: Res. Appl. 2003; 11:193206

204

D. GUASCH AND S. SILVESTRE

Figure 8. Simulation results for SOC and LOE corresponding to the simulation of the four experiments; note that despite
being similar, they show different behaviour

Temperature is a very important factor in battery life, a typical approximation is that for every 10 C temperature increment the useful life of the battery decreases by half. Taking this into account, a temperature health
factor is proposed, as Equation (18) shows. Where the reference temperature Tref is 10 C and T ( C1s1) and
T (s1) are temperature coefficients. The offset factor  T is included to take into account additional additives to
the battery (antifreezes, etc.):
T T jT  Tref j T

18

Usually, the battery works in the charging and discharging zones, but inevitably can pass to dangerous zones
in overcharges or deep overdischarges. These situations can, depending on duration and depth, strongly damage
the device. So a working zone health factor wz is proposed, based on Table I, from empirical experience and
battery manufacturers information about battery life tests, as is described in Table VI.
Effective battery capacity reduction
The influence of SOH on the capacity calculus is proposed, using a linear approximation. The criteria taken into
account consider that capacity can decrease to 25% of its nominal value when the battery is fully damaged, as
expressed in Equation (19), where a battery capacity reduction coefficient, C10 , is defined.
Copyright # 2003 John Wiley & Sons, Ltd.

Prog. Photovolt: Res. Appl. 2003; 11:193206

DYNAMIC BATTERY MODEL FOR PV APPLICATIONS

205

Table VI. Values for wz as a function of working zone


Working zone

wz(s1)

Saturation and exhaustion zones


Overcharge and overdischarge zones
Charge and discharge zones

55  106
55  107
27  107

Figure 9. Battery equivalent circuit including self-discharge current effects

C10 075 SOH 025

19

To include this effective battery capacity reduction in the model, the coefficient, C10 , must be included in
Equation (3), as shown in Equation (20).
Ct


Cnominal Ctcoef C10 
2

Bcap 1 C Tt C Tt
jItj
1 Acap
Inominal

20

Self-discharging current
Depending on the accumulated charge and the battery state of health, a self-discharging current must be considered. This self-discharging battery current can be evaluated as shown in Equation (22), where q is the selfdischarge current coefficient defined in Equation (21). The proposed approximation considers that the battery
loses 011% of its charge per day, depending on the battery SOH.
q 001  0009 SOH

21

Qt
t
24 h

22

Iadc t q

The effect of the self-discharging current must be taken into account by the battery model, as shown in
Figure 9.

CONCLUSIONS
The proposed battery model makes it possible to reproduce the battery behaviour in real dynamic working conditions, for battery current profiles with fast variations. This makes this battery model especially useful for the
simulation of stand-alone photovoltaic applications. This study has demonstrated that the automatic parameter
Copyright # 2003 John Wiley & Sons, Ltd.

Prog. Photovolt: Res. Appl. 2003; 11:193206

206

D. GUASCH AND S. SILVESTRE

extraction procedure introduced is an important tool in order to reproduce the real characteristics of batteries.
Furthermore, a key point for system diagnosis and energy management algorithms is available, applying the
mentioned parameter extraction philosophy, not only for photovoltaic system applications, but for any battery
applications, such as in the automotive or aerospace industries.
The addition of battery health effects to the model offers the possibility of evaluating the battery response as a
function of its use over long time periods. In consequence, the battery model allows one to reproduce the battery
voltage and current evolution in real working conditions, with a mean voltage error around 1%.

REFERENCES
1. Salameh ZM, Cassacca MA, Lynch WA. A mathematical model for leadacid batteries. IEEE Transactions on Energy
Conversion 1992; 7(1): 9397.
2. Cassacca MA, Salameh ZM. Determination of leadacid battery capacity via mathematical modeling techniques. IEEE
Transactions on Energy Conversion 1992; 7(3): 442446.
3. Copetti JB, Chenlo F, Lorenzo E. A general battery model for PV system simulation. Progress in Photovoltaics:
Research and Applications 1993; 1: 283292.
4. Ceraolo M. New dynamical models for leadacid batteries. IEEE Transactions on Energy Conversion 2000; 15(4):
11841190.
5. Levenberg K. A method for the solution of certain problems in least squares. Quarterly of Applied Mathematics 1944; 5:
Quarterly Journal 164168.
6. Marquardt D. An algorithm for least squares estimation of nonlinear parameters. SIAM Journal on Applied Mathematics
1963; 11: 431441.
7. Michel E, Bayetti V, Monvert B, Lagarde P, McCarthy S, Vela N, Alonso M, Lenain P, Marchand M. New battery charge
control for PV systems. Proceedings of the 16th European Photovoltaic Solar Energy Conference, Glasgow, May 2000;
24672469.
8. Copetti JB, Chenlo F, Lorenzo E. Comparison between charge and discharge models and real data for PV applications.
Proceedings of the 11th European PV Solar Energy Conference, Montreux, 1216 October 1992; 11311134.
9. Copetti JB, Chenlo F. Internal resistance characterization of leadacid batteries for PV rates. Proceedings of the 11th
European PV Solar Energy Conference, Montreux, 1216 October 1992; 11161119.
10. Silvestre S, Guasch D, Goethe U, Castaner L. Improved PV battery modelling using Matlab. Proceedings of the 17th
European Photovoltaic Solar Energy Conference and Exhibition, Munich, October 2001; pp. 507509.

Copyright # 2003 John Wiley & Sons, Ltd.

Prog. Photovolt: Res. Appl. 2003; 11:193206

You might also like