Journal of Power Sources 96 (2001) 113±120

Methods for state-of-charge determination and their applications

Sabine Piller*, Marion Perrin, Andreas Jossen
Center for Solar Energy and Hydrogen Research (ZSW), Division 3: Electrochemical Energy Storage and Energy Conversion,
Helmholtzstr. 8, 89081 Ulm, Germany
Received 8 January 2001; accepted 10 January 2001

Abstract

State-of-charge (SOC) determination becomes an increasingly important issue in all the applications that include a battery. Former
operation strategies made use of voltage limits only to protect the battery against deep discharge and overcharge. Currently, battery
operation is changing to what could rather be called battery management than simply protection. For this improved battery control, the
battery SOC is a key factor.
Much research work has been done in recent years to improve SOC determination. Operational conditions differ for batteries in, for
example, photovoltaic applications, (hybrid)-electric vehicles or telecommunications. Hence, a given method for SOC calculation will be
more suitable for a certain application than for another. The authors introduce commonly used methods for SOC determination and
establish a relationship between the advantages of the different methods and the most common applications. As a main illustration, the
analysis of Kalman ®lter technique for lead-acid battery SOC determination are presented and some results for other calculation methods as
well. # 2001 Elsevier Science B.V. All rights reserved.

Keywords: Lead-acid batteries; State-of-charge

1. Introduction Given the importance of knowing the SOC of a battery,

another question arises: how to measure it? And before
Most of us use a mobile phone, a laptop computer, measuring the SOC, we need to know what it actually is.
perhaps a cordless shaver, maybe even an electric car for In a recent paper [1] the different de®nitions for the capacity
the luckiest of us. All these devices are now part of our daily and SOC were summarised. According to this paper, we will
life and bring a big contribution to our mobility and our make reference to the SOC as: the ratio between the
freedom (even if this point is more questionable). difference of the rated capacity and the net amount of charge
But besides the comfort that is gained by all these devices discharged from a battery since the last full SOC on the one
that use the electrical energy delivered by an accumulator, a hand, and the rated capacity on the other hand. Due to this
new source of stress has appeared: will I be able to shave this de®nition, the full SOC is reached when (according to DIN
morning, will the battery last until I have written the last 43539), the battery current is not changing during 2 h at a
page of my paper? In most systems that use a battery, an constant charge voltage and constant temperature.
important point is the knowledge of the state-of-charge This de®nition leaves behind the problem of battery
(SOC) of the battery or more simply: how long do I have ageing. In fact, the capacity that can be delivered by a
until my device stops working? battery may change in the course of its life due to problems
In addition to the immediate displaying of the SOC to the like the loss of charge acceptance of the active material on
user, the knowledge of the remaining battery capacity is of either of the electrodes, changes in the physical properties
importance for its management. Namely, many systems are of the electrolyte or corrosion of the current conductors. In
sensitive to deep discharge or overcharge because these this paper, we will also not deal with the state-of-health
states of extremely high or too low SOC can lead to (SOH) of batteries, which is another topic of actual
irreversible damage in the battery. research.
The paper summarises different techniques that exist for
*
Corresponding author. Tel.: ‡49-731-9530-201;
determining the SOC of a battery. Additionally, the most
fax: ‡49-731-9530-666. suitable ®eld of application for each method is presented.
E-mail address: [email protected] (S. Piller). The main focus is put on the lead-acid system.

0378-7753/01/$ ± see front matter # 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 8 - 7 7 5 3 ( 0 1 ) 0 0 5 6 0 - 2
114 S. Piller et al. / Journal of Power Sources 96 (2001) 113±120

2. Methods for determining the state-of-charge A current-loss calculation approach for photovoltaic (PV)
applications was developed by Jossen [2]. The Butler±
The determination of the SOC of a battery may be a Volmer equation is used to calculate the major losses during
problem of more or less complexity depending on the battery charging, i.e. the gassing current. Since in PV applications
type and on the application in which the battery is used. the currents are small, the Butler±Volmer equation could be
Since this paper will focus on lead-acid accumulators, the modi®ed and normalised to become ®nally that shown in
SOC determination methods are described explicitly for this Eq. (2)
type of battery system, but apply partially also for the other  
Ubatt UN T TN
electrochemical systems. Iloss ˆ I0 exp K2 ;
K1 T TN
 
2.1. Discharge test UN K2
I0 ˆ I0 exp (2)
K1 TN
The most reliable test for the determination of the SOC of where Iloss is the current consumed by the loss reactions; K0,
a battery Ð i.e. its remaining capacity Ð is a discharge test K1, K2 are the constants; Ubatt and UN are the battery and
under controlled conditions. But such a test, which usually rated battery voltage; T and TN are the battery temperature
includes a consecutive recharge, is too time consuming to be and temperature under standard conditions.
considered for most applications. As a second drawback As mentioned above, the errors for this method can be
during testing the system function is interrupted. kept low if points for re-calibration can be identi®ed. Since
in PV applications the time for recharge is limited by the
2.2. Ampere hour counting (including loss calculation) length of daylight, full charge is seldom achieved. Concern-
ing this problem, in the same publication [2] a so-called
This is the most common technique for calculating the remaining charge current technique for re-calibration is
SOC. Since the charge and discharge are directly related to presented which allows re-calibrations if the SOC is above
the supplied or withdrawn current, the idea of balancing the 90%. For more details see [3].
battery current is evident. If a starting point (SOC0) is given, The same loss calculation was used in a SOC algorithm
the value of the current integral (Eq. (1)) is a direct indicator published on [4]. This algorithm was developed only for PV
for the SOC. applications, i.e. currents in the range of I10, and measure-
Z t ment intervals of about 1 min or more. It is not designed for
1
SOC ˆ SOC0 ‡ Ibatt Iloss † dt (1) online use because `future' data is necessary to calculate the
CN t0
SOC. The reference data for the linear model [5] and the
where CN is the rated capacity, Ibatt the battery current, and Kalman ®lter [6] approach (both described below) was
Iloss is the current consumed by the loss reactions. generated with this programme.
Two main complications arise with this method: ®rstly, Another procedure based on Ah counting is reported in
incorrect current measurement could add up to a large error [7]. This publication focuses on discharges during the
and accurate current measurement is expensive. Secondly, operation of electric vehicles. Three empirical gauge func-
not all current supplied to the battery is consumed by tions are established concerning the problems of tempera-
charging and the corresponding losses have to be taken into ture in¯uence, maximum deviation from the nominal
account. The ®rst point can be overcome by investing money capacity and SOC dependence on discharge rate.
in measuring equipment, while for the second one, many Ah counting is the most common method applied to most
different approaches have been developed. As an example, systems and applications. Indeed, the method is easy and
two loss calculations will be presented below. The errors can reliable as long as the current measurement is accurate and
be kept low if points for re-calibration are reached, e.g. the enough re-calibration points are available. For example, for
SOC is set to one if a full charge is detected or open circuit EV application, the method can take bene®t of the regular
voltage measurement (see below) is used to correct the SOC full recharges under controlled conditions. In that ®eld of
value. application, the technique ®nds a limit in the case of high
The most simple loss estimation method is to apply a temperature effects and/or high current variations but it can
constant charge factor to the battery at each recharge, i.e. a be applied to all the battery systems used for the EV
constant loss is assumed and this loss is additionally returned application. (i.e. lead-acid, Ni/Cd, Ni/MH, Zebra and
to the battery. Such a method is only suitable for systems that lithium systems).
are not too sensitive to overcharge. In Ni/MH batteries, for In the same way, because of its simplicity, Ah counting is
example, a value of 1.3 is often used to assure full charge of used in most consumer applications.
the battery. Ni/MH batteries are not as sensitive to over-
charge as are lead-acid or especially Li-based batteries. With 2.3. Measurement of the electrolytes physical properties
lead-acid batteries, a charge factor of 1.05±1.2 is used,
depending on the battery type. Additionally, this method In a lead-acid battery, the electrolyte takes part in
implies that the charging operation is controlled. the reactions during charge and discharge. The linear
 S. Piller et al. / Journal of Power Sources 96 (2001) 113±120 115

relationship between the change of acid density and the SOC circuit voltage for a fully charged and a ¯ooded lead-acid
can be used to determine the latter. This method is feasible cell is about 100 mV. Especially in VRLA batteries long
only with vented lead-acid batteries, while methods for density times (several hours) are needed to reach a steady state and
measurement within VRLA batteries are not yet available. this can cause problems. A second point to mention is the
Possible applications are, therefore, in stationary batteries question of what is meant by rest period. Often a minimal
with liquid electrolyte. The density is measured directly or current ¯ow is required for monitoring devices: clocks, etc.
indirectly by ion-concentration, conductivity, refractive In such a case, the open circuit voltage is never reached.
index, viscosity, ultrasonics, etc. A detailed description of Finally, like with all methods which use directly or indirectly
the different methods is given, for example, in [8]. the monitoring of acid concentration, acid strati®cation can
Problems related to this technique are the occurrence of generate inaccurate results.
acid strati®cation, water loss and the long term stability of
the sensors. The ®rst two can be avoided by electrolyte 2.5. Heuristic interpretation of measurement curves
circulation and automatic water re®ll systems. Additionally,
the measurement of the physical properties of the acid The following gives a brief, non-comprehensive, selection
cannot be done in the pores of the electrodes where the of techniques that use the electrical discharge/charge char-
acid is actually either consumed or produced. This means acteristics to calculate the SOC. Some methods were devel-
that during high current processes, slow electrolyte diffusion oped to get data on the SOC by interpreting only parts of the
is a source of errors for this technique. discharge curves. The chosen parts are understood to be
signi®cant for the battery SOC.
2.4. Open circuit voltage
2.5.1. Coup de fouet
Similar to the acid density measurements described in In [10] the so-called coup de fouet region is used to
Section 2.3, the open circuit voltage relates in a linear calculate not exactly the SOC but the capacity that can
manner to the SOC. Fig. 1 shows this linear dependence be delivered after a full recharge for a discharge at a given
as obtained from four 12 V, 52 Ah (C/5) VRLA batteries of current and temperature. The coup de fouet is the short
the Optima type when tested at the University of Ulm. voltage drop at the beginning of discharge following a full
In applications where relatively long rest periods are charge of a lead-acid battery. Two parameters are estab-
common, this method is promising. Since the rest periods lished, which are shown to have a linear relationship to the
will only occur from time to time, the open circuit voltage deliverable capacity: ®rstly the trough voltage, i.e. the
measurement is usually combined with other techniques [9] minimum voltage during the occurrence of coup de fouet
to ensure a continuous indication of SOC. In such a combi- and secondly the plateau voltage, i.e. the maximum voltage
nation, the open circuit voltage measurement can be utilised reached during voltage recovery at the end of the coup de
to adjust the other technique(s). The difference in the open fouet. For one cell type it has been shown that the linear

Fig. 1. Dependence of the state-of-charge on the open circuit voltage from four VRLA batteries of the same type.
116 S. Piller et al. / Journal of Power Sources 96 (2001) 113±120

equation determined by regression stays within an error limit 2.5.3. Artificial neural network
of 12.5% for high discharge-current variations and temp- The utilisation of arti®cial neural networks for SOC
erature changes. Since the coup de fouet region occurs only determination is presented in [5,11,12]. Since arti®cial
after a full charge, this method can be used if a full charge is neural networks establish a relationship between input/out-
frequently reached during operations. For static discharge- put data of any kind, this method can be utilised for all
currents this method is advantageous, because it gives an battery systems and for all applications, providing that
estimation of the actual available capacity, which depends training data for the net is available. In [5], an arti®cial
on the discharge-current and temperature. The authors neural network is presented which is trained before use,
suggest this technique for batteries in telecommunication whereas in [11,12] adaptable arti®cial neural networks are
installations. used. If an adaptable arti®cial neural network is applied,
other methods for SOC determination are used to provide
2.5.2. Linear model training data at selected states of operation. Errors depend
In [5] a linear relationship (Eq. (3)) was established strongly on the training data and the training method. Since
between the variation of SOC, the intermediate electrical the training methods usually minimise functions that lead to
measurements on the battery and the previous SOC value. an evenly distributed error, the error is usually no function of
the SOC. However, in Fig. 3 (taken from [5]) error extremes
DQ i† ˆ b0 ‡ b1 U i† ‡ b2 I i† ‡ b3 Q i 1†; occur at high SOCs, because the net was trained with data
Q i† ˆ Q i 1† ‡ DQ i† (3) obtained from a battery with slightly larger capacity and
better electrolyte circulation. But the SOC curves of a
where Q(i) is the state-of-charge, DQ(i) the SOC-difference, battery of the same type as the one corresponding to the
U the voltage and I is the current measurements. training data show smaller and evenly distributed errors.
The factors b0 ; . . . ; b3 are determined from reference data
by least-mean-square calculations. The model was devel- 2.6. Impedance spectroscopy
oped for PV applications, i.e. for low currents and slow SOC
changes, and is characterised by high robustness in relation Much research work has been carried out on impedance
to measurement errors and wrong initial conditions, as spectroscopy. This method is a common measurement tech-
shown in Fig. 2. nique to investigate electrochemical processes and has been
It is important to note that the b-factors do not describe studied for all battery systems not only for SOC but also for
physical parameters. SOH determination. A review of impedance measurements
The linear model can be applied to various battery types for determination of the SOC for lead-acid and nickel/
and to batteries at different stages in their lives. However, cadmium batteries is given in [13]. Fig. 4 shows a Nyquist
best results are achieved if reference data from the same diagram of the complex impedance of a lead-acid battery
battery type is used to calculate the b-factors. (OPzS 150) during discharge.

Fig. 2. SOC curves of the linear model, showing the reference and the error between the two curves.
 S. Piller et al. / Journal of Power Sources 96 (2001) 113±120 117

Fig. 3. SOC curves of back-propagation artificial neural network, showing the reference and the error between the two curves.

A combination of impedance spectroscopy with fuzzy cycle number was the input and for the available reference
logic methodology is presented in [14]. Two battery systems points of 0, 25 and 100% an error below 10% was achieved.
were investigated, a lithium/sulphur dioxide and a nickel/ Impedance curves are strongly in¯uenced by temperature
metal hydride system. A fuzzy model was used to establish a effects. Therefore, the best utilisation of this method is with
relationship between battery model parameters derived from batteries in temperature controlled environment, e.g. large
impedance spectroscopy measurements and the SOC. For stationary battery installations.
the lithium/sulphur dioxide cells the imaginary component Surprisingly, in spite of the quantity of papers written on
of the impedance at three different frequencies was used as the subject, impedance spectroscopy is seldom implemented
input for the SOC calculation and an accuracy of 5% was for practical SOC determination and it stays still a subject
achieved, for the nickel/metal hydride cells the C2 capaci- of debate. In a review [13], Huet concluded that the
tance (also derived from impedance measurement) and the temperature in¯uence is so high that practical application

Fig. 4. Impedance measurements from a lead-acid battery (OPzS 150) during discharge (from [8]).
118 S. Piller et al. / Journal of Power Sources 96 (2001) 113±120

of impedance spectroscopy can only be made in ranges of

high frequencies. This technique seems to be more suitable
for lead-acid systems than for nickel/cadmium and is more
promising for the determination of SOH rather than for the
precise measurement of SOC.
Another review on this subject is given in [17]. Besides
conventional systems (i.e. lead-acid, nickel/cadmium,
nickel/metal hydride), initial measurements on lithium-ion Fig. 5. Battery model for SOC determination in PV applications.
batteries are presented. For this system, impedance spectro-
scopy seems to be a possible method for SOC determination.
In fact, conductance measurements at a given frequency divided by the current change during the same (short) time
are performed on batteries for quality control and provide interval. The value of the resistance depends heavily upon
information about state-of-health. They can also be used for the chosen time interval. For a time interval smaller than
SOC evaluation provided they are taken within a convenient 10 ms, only Ohmic effects are measured. If the interval is
frequencies range (e.g. in Fig. 4, the SOC varies only at low extended, other effects such as transfer reactions or acid
frequencies). diffusion are involved and the resistance becomes complex.
In this case it would be better to use the previously described
2.7. Internal resistance impedance spectroscopy instead of the voltage current ratio.
For lead-acid batteries the change in the internal (Ohmic)
Related to the impedance spectroscopy is the calculation resistance between full SOC and SOC ˆ 0 is only some mO
of the internal resistance of the battery, i.e. the voltage drop per cell. This kind of measurement is more useful for

Fig. 6. Battery model for dynamic applications such as HEV and EV.

Fig. 7. Current and voltage curves measured on four 12 V lead-acid battery blocks used for Kalman filter SOC determination (simulated hybrid vehicle
cycle).
 S. Piller et al. / Journal of Power Sources 96 (2001) 113±120 119

Fig. 8. SOC curve calculated by the Kalman filter technique compared with Ah balance with loss calculation.

providing statements on the state-of-health of the batteries. dynamic system and one of the inner states is the SOC. The
It can also provide some information about the SOC pro- estimation is based on a model of the dynamic system. At
vided the battery is not fully charged [18]. the ZSW, Kalman ®lters with two models are specially
investigated. The ®rst is a model based on [15], which is
2.8. Kalman filters seen as appropriate for PV applications and the second one
is a more dynamical model, being a simpli®cation of the
A Kalman ®lter is an algorithm to estimate the inner battery model in [16]. The models are shown in Figs. 5
states of any dynamic system. In our case, the battery is the and 6.

Fig. 9. Summery of the different techniques for determination of state-of-charge, presented with their field of application, advantages and drawbacks.
120 S. Piller et al. / Journal of Power Sources 96 (2001) 113±120

The results for the SOC determination from the Kalman  The artificial neural network that can be implemented for
®lter using the ®rst model (Fig. 5) where presented in [5]. any battery and battery system provided enough training
It was shown that the errors could be kept below 10% for data can be supplied.
data of a 12 V, 125 Ah ¯ooded lead-acid battery, which
was cycled on a typical PV regime over a period of 2.5
years. References
To apply the Kalman ®lter technique for data of a higher
dynamic, e.g. for a hybrid electric vehicle (HEV), the [1] D.U. Sauer, G. Bopp, A. Jossen, et al., State-of-charge Ð what do we
really speak about? INTELEC, 1999.
second battery model (Fig. 6) has to be used. Since ®eld [2] A. Jossen, BetriebsfuÈhrung photovoltaischer Anlagen mit Energie-
data of HEVs are not available, the results presented are speichern, Dissertation, UniversitaÈt Stuttgart, 1994.
obtained using measurements performed on four 12 V, [3] A. Jossen, H. DoÈring, J. Garche, Ladezustandsbestimmung von
52 Ah (Optima Yellow Top) batteries by means of synthetic Bleibatterien an zwei ausgewaÈhlten Beispielen, 10, Symposium
hybrid cycles, generated with the simulation tool FAHR- Photovoltaischer Solarenergie, Staffelstein, 1995.
[4] ZSW (Hrsg.) & ISET & ISE, Referenzdaten zum Test von Verfahren
SIM [19]. Fig. 7 shows the current and voltage evolution zur Ladezustandsbestimmung von Bleibatterien, CD-ROM, 1999.
over time. The results of the SOC calculation with this data [5] Ch. Ehret, S. Piller, W. Schroer, A. Jossen, State-of-charge
and the dynamic model is shown in Fig. 8. Because no other determination for lead-acid batteries in PV-applications, in: Proceed-
reference exists, the Kalman SOC is compared to the Ah ings of the 16th European Photovoltaic Solar Energy Conference,
balance with the above described loss calculation based on Glasgow, 2000.
[6] S. Piller, A. Jossen, Kalman-filter zur Ladezustandsbestimmung von
[2]. In total, only about 10 Ah were taken from the batteries. Bleibatterien in PV-Anlagen, 14, Symposium Photovoltaischer
Both SOC curves are very similar, the maximum difference Solarenergie, Staffelstein, 1999.
is 2% at the lowest SOC. Other cycles showed comparable [7] J. Alzieu, H. Smimite, D. Glaize, Improvement of intelligent battery
results. controller: state-of-charge indicator and associated functions, J.
These are ®rst results with dynamic data and SOC deter- Power Sources 67 (1997) 157±161.
[8] H.-P. HoÈnes, Elektrochemische Energiespeicher in photovoltaischen
mination using the Kalman ®lter technique. At ZSW further Anlagen, Dissertation, UniversitaÈt Stuttgart, 1994.
work will be carried out to con®rm the suitability of this [9] M. Rothert, B. Willer, C. Schmitz, et al., Verschiedene AnsaÈtze zur
method for high dynamic applications. Ladezustandsbestimmung, Forschungsverbund Sonnenenergie Work-
shop Elektrochemische Speichersysteme fuÈr regenerative Energie-
versorgungsanlagen, May 1999.
[10] Ph.E. Pascoe, A.H. Anbuky, Estimation of VRLA Battery Capacity
3. Conclusion Using the Analysis of Coup de Fouet Region, INTELEC, 1999.
[11] D. Kray, D.U. Sauer, Neuro-Methoden zur Prognose und Fehler-
The paper has given a short overview of the existing erkennung durch Analyse vor Daten aus Monitoring-Systemen,
techniques for the evaluation of battery SOC. Emphasis was Internationales Sonnenforum, KoÈln, 1998.
given to the lead-acid system even though many techniques [12] M. Stoll, Ein SchaÈtzverfahren uÈber den inneren Zustand geschlossener
Bleiakkumulatoren, VDI Fortschritt-Berichte, Reihe 21, Nr. 167, 1994.
are also suitable for other systems. Fig. 9 summarises this [13] F. Huet, A review of impedance measurements for determination of
overview. the state-of-charge or state of health of secondary batteries, J. Power
The most used technique at this time for all systems is Ah Sources 70 (1998) 5949.
counting because it is the most direct and transparent [14] A.J. Salkind, C. Fennie, P. Singh, et al., Determination of state-of-
method and quite easily implemented. It also gives satisfy- charge and state-of-health of batteries by fuzzy logic methodology, J.
Power Sources 80 (1999) 293±300.
ingly accurate results for short-time applications, especially [15] W. Burkner, Verfahren zur Ladezustandsbestimmung von Blei-
if used in the range of low to medium SOC. The determina- Batterien in Photovoltaikanlagen, IfE Schriftenreihe, Heft 27,
tion of SOC by the means of impedance spectroscopy, Lehrstuhl fuÈr Energiewirtschaft und Kraftwerkstechnik, Technische
including ac and dc inductance measurements, is still a UniversitaÈt, MuÈnchen, 1994.
subject of debate because of its temperature sensitivity [16] R. Gretsch, Ein Beitrag zur Gestaltung der elektrischer Anlage in
Kraftfahrzeugen Habilitationsschrift, Friedrich-Alexander-Universi-
and the dif®culty of its online implementation. New promis- taÈt Erlangen, NuÈrnberg, 1978.
ing methods are: [17] S. Rodrigues, N. Munichandraiah, A.K. Shukla, A review of state of
charge indication of batteries by means of ac impedance measure-
 The linear model that is impressive for its simplicity and ment, J. Power Sources 87 (2000) 12±20.
[18] A. Jossen, H. DoÈring, J. Garche, Ladezustandsanzeiger fuÈr elek-
because it delivers satisfying results for photovoltaic
trochemische Akkumulatoren, UECT, 1995.
applications. [19] H. Kabza, C. Koerner, P. Seger, A. Wagener, FAHRSIM: A Versatile
 The Kalman filter that gives perspectives for high Simulator for Drivetrain Modeling and Analysis, EVS 17, Montreal,
dynamic usage (HEV, EV). 2000.