752 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 25, NO.

2, MARCH 2017

Adaptive State of Charge Estimation of Lithium-Ion Batteries

With Parameter and Thermal Uncertainties
Hicham Chaoui, Senior Member, IEEE, and Hamid Gualous, Member, IEEE

Abstract— In this brief, an adaptive state of charge (SOC) drawbacks, it remains the simplest approach for real-time
estimation strategy is proposed for lithium-ion batteries. The industrial applications [4]. Another rational way to determine
proposed methodology makes use of adaptive control theory to the SOC is to use the open-circuit voltage (OCV) since the
track online parameter variation. The convergence and stability
of the proposed estimator are guaranteed by Lyapunov’s direct battery’s voltage is directly correlated with the battery’s charge
method as opposed to many existing procedures. Since temper- status [7], [8]. However, this correlation holds only when the
ature variations introduce a drift in the battery’s parameters, battery reaches an equilibrium state (i.e., no current flows
a compensation methodology is also proposed to cope with this through the battery for several minutes or hours). A com-
effect. Therefore, robustness to both parameter and temperature bination between the aforementioned two methods yields a
variations is obtained, which yields precise SOC estimation.
The proposed estimation scheme is validated through a set of hybrid estimation technique. Thus, the Coulomb counting
experiments under different currents and temperatures. The technique is then used and whenever the battery reaches
experimental results reveal high performance in determining the an equilibrium state, a reset of the accumulated errors is
SOC with high accuracy. performed by updating the SOC with the OCV technique. Yet,
Index Terms— Lithium-ion batteries, Lyapunov stability, some applications require a continuous operation and do not
parameter estimation, state of charge (SOC). allow batteries to reach an equilibrium state, which raises the
urgency of considering other alternatives of SOC estimation.
I. I NTRODUCTION Various advanced estimation strategies are proposed at the
cost of a higher design complexity [9]–[13]. A sliding mode

L ITHIUM-ION batteries offer a higher power density and

energy efficiency as opposed to other types of batteries
such as lead acid, nickel metal hydride (NiMH), and nickel
observer is proposed in [14] to compensate for modeling
uncertainties. On the other hand, the SOC is derived from the
charge/discharge experimental data under different constant
cadmium [1], [2]. They have received an increasing inter- currents and temperatures for a NiMH battery [15]. In [16],
est from the scientific community because of their several measured current/voltage profiles are used with an optimiza-
other advantages such as rapid charge capability, low steady- tion procedure to estimate online the battery’s parameters.
state float current, wide temperature operation range, small As such, the model captures the battery’s parameter variation.
size, light weight, low self-discharge rate, long life cycle, Another SOC estimation technique is proposed in [17] using
no memory effects, and the absence of hydrogen outgassing, a reduced-order observer. However, it requires the knowledge
which make them good candidates for many applications such of battery’s parameters, which results in accuracy reduction as
as electric vehicles, mobile phones, and laptops [3]. SOC is batteries age. This drawback has been overcome in [18], where
considered as the battery’s energy gauge, which is crucial an adaptive SOC estimation strategy is proposed for lead-
information since a bad SOC estimation may ultimately result acid batteries. Particle filter (PF) is a sequential Monte Carlo
in damaging the battery and reducing its lifespan. method that uses weighted random samples (particles) to
Conventional SOC estimation procedures are used for their estimate the probability distribution function of any nonlinear
simplicity. For instance, a straightforward way to estimate a system. Several PF-based battery SOC estimation methods are
battery’s SOC is the ampere-hour balancing technique, also suggested in the literature for lithium-ion batteries [19]–[21].
called the Coulomb counting method [4], [5]. In this technique, In [19], the battery is considered as a nonlinear dynamic
SOC is determined by integrating through time the battery’s system with the SOC of the battery as the only state variable.
entering and departing currents. However, the accumulation Classical estimation approaches such as the Kalman filter
of the startup and current sensor errors leads to a drift and show limitations in handling nonlinear and non-Gaussian error
poor precision [6]. Although this technique has some serious distribution problems. In addition, uncertainties in the battery
Manuscript received February 29, 2016; revised May 13, 2016; accepted model parameters must be taken into account to describe
May 20, 2016. Date of publication June 14, 2016; date of current ver- the battery degradation. In [20], a model-based method is
sion February 8, 2017. Manuscript received in final form May 20, 2016. presented combining a sequential Monte Carlo filter with
Recommended by Associate Editor S. Varigonda.
H. Chaoui is with the Intelligent Robotic and Energy Systems adaptive control to determine the cell SOC and its electric
Research Group, Department of Electrical and Computer Engineering, impedance. The applicability of this dual estimator is verified
Tennessee Technological University, Cookeville, TN 38505 USA (e-mail: using measurement data acquired from a commercial LiFePO4
[email protected]).
H. Gualous is with the Laboratoire Universitaire des Sciences Appliquées cell. Due to a better handling of the hysteresis, the results show
de Cherbourg, University of Caen Normandy, Caen 50130, France the benefits of the proposed method against the estimation with
(e-mail: [email protected]). an extended Kalman filter. In [21], another state estimation
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. technique is presented for lithium-iron phosphate batteries
Digital Object Identifier 10.1109/TCST.2016.2572362 where a PF overcomes the problem of the variance and
1063-6536 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: National Chung Hsing Univ.. Downloaded on June 19,2024 at 06:53:39 UTC from IEEE Xplore. Restrictions apply.
CHAOUI AND GUALOUS: ADAPTIVE SOC ESTIMATION OF LITHIUM-ION BATTERIES 753

the mean of a Gaussian probability density function using

Monte Carlo sampling. Several other model-based estimators
are proposed in the literature. A reduced-order thermal model
is proposed in [22] to estimate temperature distribution inside
a cylindrical battery. Then, a combination of a Kalman filter
and an extended Kalman filter is carried out for the system’s
identification and the estimation of the battery core tempera-
ture. In [23], reduced-order Luenberger and learning observers
are also proposed for lithium-ion batteries. In [24], an electro-
Fig. 1. Equivalent electric circuit of a lithium battery.
chemical model-based SOC and capacity estimation method is
presented for a composite electrode lithium-ion battery. This
work shows that electrochemical-based models can maintain generally provides an accurate representation of the system’s
an accurate battery model under varying degradation aging dynamics [7], [30].
modes. The voltage–current characteristic dynamic mathematical
Besides, soft-computing tools such as neural networks and model can be described by the following equations [31], [32]:
fuzzy logic have been credited in various applications as 1 1
powerful tools for systems subjected to uncertainties [25], [26]. V̇ p = V p − Ib (1)
RC C
Several neural network models have been applied for the Vb = Voc + V p + Rb Ib (2)
SOC and SOH estimation problems, which have led to a
satisfactory performance [27], [28]. However, neural networks where Voc is the OCV, Vb and Ib are, respectively, the voltage
remain incapable of incorporating any human-like expertise and the current at battery terminals, Rb is the internal resis-
already acquired about the dynamics of the system in hand, tance, R and C are the equivalent resistance and capacitance,
which is considered one of the main weaknesses of such respectively, and V p is the voltage across the RC network.
methodologies. This weakness has been overcome in [29]
with a fuzzy neural network. However, these tools achieve B. Problem Statement
outstanding performance at the expense of a heavy computa- This brief aims to estimate the OCV Voc since it is directly
tion. Furthermore, they are based on heuristic and tuning may correlated with the battery’s SOC. This objective is to be
not be trivial. In addition, many soft-computing-based esti- achieved under temperature variations. In this work, the para-
mation algorithms lack stability proofs in several estimation meters R, C, and Rb are assumed to be a priori unknown
applications. and V p is not measurable. The system’s measurable states are
The contribution of this brief is to propose an adaptive SOC the battery’s voltage Vb and current Ib . Temperature is also
estimation methodology for lithium-ion batteries. Therefore, assumed to be measurable.
the proposed algorithm tracks the battery’s parameters as they Remark 2: The current Ib is taken as positive in charge
vary over time ensuring robustness to parametric uncertainties. mode and negative otherwise.
Unlike many other conventional and advanced estimation Assumption 1: The battery’s voltage Vb and current Ib
strategies, the convergence and stability are guaranteed by along with their derivatives V̇b and I˙b are continuous and
Lyapunov’s direct method. Moreover, a compensation tech- bounded.
nique is proposed to cope with temperature variations, which is Assumption 2: The estimation algorithm sampling fre-
usually ignored in many estimators available in the literature. quency is high enough such that V̇oc ≈ Ṙb ≈ Ṙ ≈ Ċ ≈ 0.
This is one of the first attempts, if any, to achieve Since the battery’s parameters vary with the SOC, online
SOC estimation with guaranteed stability under both paramet- parameter estimation is then required to track the battery’s
ric and temperature uncertainties. parameter variation. Thus, robustness to parametric uncertain-
ties is achieved, which yields high SOC accuracy. In addition,
II. L ITHIUM BATTERIES temperature variations introduce a drift in the estimates and
lead to a bad SOC estimation. Therefore, a temperature com-
A. Modeling pensation technique is needed to guarantee the high estimation
The electric circuit model of a lithium battery is shown precision. The proposed estimation scheme is outlined in
in Fig. 1. The RC network characteristics are modeled with Fig. 2. As such, an adaptive parameter identification strategy
an equivalent resistance and capacitance, while the internal estimates online the battery’s OCV using the voltage Vb and
battery’s resistance is represented by Rb . On the other hand, current Ib . Then, the battery’s OCV is fed to a temperature
OCV–SOC characteristic is represented in this model by a compensation algorithm to correct the battery’s voltage before
current-controlled current source, a battery storage capac- determining the SOC using a lookup table. Next, the proposed
ity Cc and a self-discharge resistance Rd . A voltage-controlled online parameter estimation approach is described in detail
voltage source is also used to bridge SOC to OCV [30]. with its stability proof.
Remark 1: It is noteworthy that several models presented in
the literature result in a drastic increase in the system’s com- III. O NLINE PARAMETER I DENTIFICATION
plexity. As it is shown experimentally in [31] and references Here, an adaptive parameter identification technique is pro-
therein, the circuit model in Fig. 1 with a single RC network posed based on adaptive control theory to estimate online

Authorized licensed use limited to: National Chung Hsing Univ.. Downloaded on June 19,2024 at 06:53:39 UTC from IEEE Xplore. Restrictions apply.
754 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 25, NO. 2, MARCH 2017

Fig. 2. Block diagram of the proposed estimation scheme.

the battery’s parameters. Henceforth, the proposed strategy’s This can be written in a state-space form as
stability is guaranteed by Lyapunov’s direct method.
Substituting V p from (2) into (1) and using Assumption 2 Ẋ = A X + B U (7)

1 Rb 1 1 where X ∈ R2 = [ eb , eb ]T is the state vector and U ∈ R =
V̇b − Vb − Rb I˙b + Ib + Ib + Voc = 0. ˜ is the state-space input. A ∈ R2×2 , which is a stable
β̂ T 
RC RC C RC
Multiplying by RC yields matrix, and B ∈ R2 are given by
 
0 1
Vb = RC V̇b − Rb RC I˙b + (R + Rb )Ib + Voc =  T  (3) A=
−K i −(K p + β̂)
where  ∈ R4 is a vector of known functions (regressor) and  
0
 ∈ R4 = [RC, − Rb RC, R + Rb , Voc ] is a vector of B = .
1
parameters. Therefore, the battery’s voltage estimation law is
defined as Henceforth, the estimator’s gains K p and K i may be chosen
to place the closed-loop poles at their desired locations using
V̂b = R̂ Ĉ V̇r − R̂b R̂ Ĉ I˙b + ( R̂ + R̂b )Ib + V̂oc (4) a pole placement technique or by solving the algebraic Riccati
where •ˆ is the estimate of • and equation.
 Theorem 1: Consider a nonlinear system in the form
V̇r = V̇ˆb − K p eb − K i eb (5) of (1) and (2) with the estimation law (4). The estimation error
asymptotic stability and convergence to zero are guaranteed
where K p and K i are the proportional and integral gains, with the following adaptation law:
respectively, and eb = Vb − V̂b is the battery’s voltage ˙ˆ = − β̂ B T P X
estimation error to be driven to zero by the adaptive estimator.  (8)
Substituting V̇r from (5) into (4) leads to
 where  = [γ1 , γ2 , . . . , γ4 ] and γi is a positive constant gain.
ˆ
V̂b = R̂ Ĉ V̇b − R̂ Ĉ K p eb − R̂ Ĉ K i eb
P is a symmetric positive definite matrix chosen to satisfy the
following Lyapunov equation:
− R̂b R̂ Ĉ I˙b + ( R̂ + R̂b )Ib + V̂oc .
A T P + P A = −Q (9)
Adding and subtracting R̂ Ĉ V̇b
 with Q being a positive definite matrix.
V̂b = − R̂ Ĉ ėb − R̂ Ĉ K p eb − R̂ Ĉ K i eb Proof: Choose the following Lyapunov candidate:
˜ T  −1 .
V = XT P X +  ˜
+ R̂ Ĉ V̇b − R̂b R̂ Ĉ I˙b + ( R̂ + R̂b )Ib + V̂oc . (6)
Taking the derivative of V yields
Subtracting (3) from (6) leads to
 ˙ˆ
˜ T  −1 .
V̇ = Ẋ T P X + X T P Ẋ + 2 (10)
R̂ Ĉ ėb + ( R̂ Ĉ K p + 1)eb + R̂ Ĉ K i eb
Since the parameters  are considered to be constant or slowly
= RC V̇b − R̂ Ĉ V̇b − Rb RC I˙b ˙ˆ Substituting Ẋ from (7)
˙˜ = .
time varying (Assumption 2), 
+ R̂b R̂ Ĉ I˙b + (R + Rb )Ib − ( R̂ + R̂b )Ib + Voc − V̂oc . yields
Therefore, using the linear regression yields ˙ˆ
˜ T  −1 .
V̇ = [ AX + BU ]T P X + X T P[ AX + BU ] + 2

˜
ėb + (K p + β̂)eb + K i eb = β̂ T  ˜ implies that
Therefore, setting U = β̂ T 

˜ = − ˆ and β = (1/RC). In this way, the estimation ˜ T  β̂ B T P X + 2

V̇ = X T [ A T P + P A]X + 2 ˙ˆ
˜ T  −1 .
where 
law (4) leads to the following closed-loop dynamics:
 Setting A T P + P A = −Q as in (9) leads to
ėb + (K p + β̂)eb + K i eb = 0. ˙ˆ
˜ T [ T β̂ B T P X +  −1 ].
V̇ = −X T Q X + 2

Authorized licensed use limited to: National Chung Hsing Univ.. Downloaded on June 19,2024 at 06:53:39 UTC from IEEE Xplore. Restrictions apply.
CHAOUI AND GUALOUS: ADAPTIVE SOC ESTIMATION OF LITHIUM-ION BATTERIES 755

Setting the adaptation law as defined in (8) implies that

V̇ = −X T Q X < 0.
Since Q is positive definite, then V̇ < 0 ∀X = 0 so
that X = 0 is a globally asymptotically stable equilibrium
point. A positive Lyapunov function V , which is decreasing
(V̇ < 0), must converge to a finite limit. Therefore, the
system is asymptotically
 stable in the sense of Lyapunov.
Hence, X and so, e, e, and  ˜ are also bounded and converge
to finite values. Since  is bounded, it implies from (7) that Ẋ
is also bounded. Therefore, V̈ is also bounded.
Lemma 1 (Barbalat): If the differentiable function V (t) has
a finite limit as t → ∞, and if V̇ (t) is uniformly continuous,
then V̇ (t) → 0 as t → ∞.
From Lemma 1, V has a finite limit as t → ∞ and V̇
is uniformly continuous. Therefore, limt →∞ V̇ = 0, and
hence, lim t →∞ X = 0. It shows that limt →∞ eb = 0 Fig. 3. Experimental setup.
and limt →∞ ėb = 0. Therefore, lim t →∞ V̂b = Vb .
IV. SOC E STIMATION /T EMPERATURE C OMPENSATION
SOC estimation is crucial to determine the battery’s energy
level. In general, SOC estimation is achieved by Coulomb
counting in operation and the OCV method is used only
when the battery is in equilibrium, which limits its use in
real-time industrial applications. Unlike these techniques, the
SOC estimation is achieved in this brief by online battery’s
OCV estimation. It is noteworthy that accurate estimation of
the parameter 4 ultimately leads to the estimation of the
battery’s OCV. It is also important to note that accurate SOC Fig. 4. Illustration of the experimental setup.
estimation is guaranteed since the adaptive parameter estimator
tracks the battery’s parameters as they vary over time. shown in Fig. 3 is designed to charge and discharge the battery.
Unlike other methods, where the effect of temperature is At each discharge cycle, the battery’s voltage is measured
ignored or embedded in the system’s model [10], [12], [33], at different operating temperatures using a USB-based data
this brief proposes a forecompensation and a postcom- acquisition board (NI USB-6008) as an interface with the
pensation technique, which yields complexity reduction. LiFePO4 battery (Fig. 4). Therefore, the validation process
Temperature variation introduces an offset in both the consists in placing the battery inside a climatic chamber and
Vbat –SOC and Voc –SOC curves. Therefore, temperature com- characterizing it at different currents (6, 10, and 20 A) and
pensation is carried out as temperatures (10 °C, 25 °C, and 40 °C). As it is illustrated
25
Vbat °C
= Vbat + η1 T in [34], the estimation of the battery’s parameters can be
25 °C
Voc = Voc + η2 T achieved regardless of the discharge rate. The characterization
protocol consists of charging the battery according to the man-
where T = Tactual − 25 °C, Vbat 25 °C and V 25 °C are the
oc ufacturer’s guidelines, i.e., constant current constant voltage.
battery’s terminal and OCV, respectively, at 25 °C (77 °F), and As such, the battery is charged to 3.65 V using a constant
η1 and η2 are compensation coefficients. It is noteworthy that current and then a constant voltage of 3.65 V is applied until
the compensation strategy can be used either as a forecompen- the battery’s current reaches 0.01 A, which indicates a fully
sation or a postcompensation to cope with the temperature drift charged battery. Next, the battery is fully discharged using con-
effect. As it is illustrated in Fig. 2, the temperature compen- tinuous discharging until its voltage reaches 2.8 V. For every
sation is applied to the battery’s voltage Vb when Sb is closed discharge current (6, 10, and 20 A), the battery’s voltage is
and Sc is open and to the battery’s OCV estimate V̂oc when measured. These steps are repeated for different temperatures
Sb is open and Sc is closed. The compensation coefficients (10 °C, 25 °C, and 40 °C). The battery’s SOC is estimated by
η1 and η2 are extracted from experimental data measured at current integration to show its relationship with the battery’s
different temperatures. voltage characterization measured values. While this procedure
is known to drift due to current sensor gain/offset errors if no
V. E XPERIMENTAL R ESULTS
reset is performed after a long period of time, the duration
A. Setup of experiments is not sufficient to introduce a serious drift.
In this brief, a 20-Ah, 3.2 V lithium-iron phosphate In addition, the current sensor used in experiments is calibrated
battery (LiFePO4 ) is used to validate the proposed estima- and its offset is eliminated in the software before integration
tion scheme. For that, the hardware-in-the-loop testbench begins. In order to validate the effectiveness of the proposed

Authorized licensed use limited to: National Chung Hsing Univ.. Downloaded on June 19,2024 at 06:53:39 UTC from IEEE Xplore. Restrictions apply.
756 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 25, NO. 2, MARCH 2017

Fig. 5. Experimental results for discharge currents of (a)–(c) 6, (d)–(f) 10, and (g)–(i) 20 A.

approach, the necessary time to discharge the battery by 10%

is calculated. Then, the battery is discharged each time by a
10% step for each discharge current and is left at idle for
about 1 h to reach equilibrium before its OCV is measured.
These measurements are used as a benchmark to demonstrate
the performance of the proposed strategy.
The SOC algorithm sampling time is set to 0.1 s, which
yields reduced computation burden. The parameters’ estimate
and adaptation rate vectors are initialized to (10−3 , 10−4 ,
10−3 , 3.5) and (10−7 , 10−7 , 10−7 , 10−4 ), respectively. The
estimator’s gains are set to k p = 10−2 , ki = 10−3 , η1 =
3.75 × 10−3 , and η2 = 2.9 × 10−3 .

B. Results Fig. 6. Experimental results: voltage estimation error e.

As indicated, the battery is subjected to various tempera-
tures, i.e., 10 °C, 25 °C, and 40 °C. For each operating tem- voltage in this case is depicted in Fig. 5 for 10 °C, 25 °C, and
perature, the battery is completely discharged at different rates 40 °C. It is important to note that temperature introduces an
(6, 10, and 20 A) until its voltage reaches 2.8 V. The battery’s offset on the battery’s voltage [Fig. 5(a), (d), and (g)]. As it is

Authorized licensed use limited to: National Chung Hsing Univ.. Downloaded on June 19,2024 at 06:53:39 UTC from IEEE Xplore. Restrictions apply.
CHAOUI AND GUALOUS: ADAPTIVE SOC ESTIMATION OF LITHIUM-ION BATTERIES 757

Fig. 7. Experimental results for OCV estimate with discharge currents of (a)–(d) 6, (b)–(e) 10, and (c)–(f) 20 A.

observed, temperature variations have an effect on the battery’s decreases to a negligible magnitude. The advantage behind the
capacity as it is well known that batteries exhibit lower use of the adaptive estimation approach is clearly shown in this
capacity at cold temperatures. The battery’s SOC obtained experiment. Faster convergence can be achieved by increasing
through current integration is also shown and as it can be adaptation parameter γi in (8) at the expense of more noise
observed, the time when the battery’s voltage gets to the cutoff in the estimates. Therefore, there is a tradeoff between fast
of 2.8 V, which is an indicator of a 0% SOC, corresponds to convergence at startup and noise rejection. Since the SOC
the predicted time of the SOC estimation procedure. Then, the is a slowly time-varying process, fast convergence is less
correlation between the battery’s voltage and its SOC, called important than the steady-state accuracy, which yields reduced
the Vbat –SOC curve, is established for various temperatures, noise in the estimates. Fig. 7 illustrates the performance of
as shown in Fig. 5(b), (e), and (h). The performance of the the parameter estimation under different temperatures. As the
proposed temperature compensation method when applied to battery discharges, the OCV decreases [Fig. 7(a)–(c)]. As it
the battery’s voltage Vb (i.e., Sb is closed and Sc is open) is expected, temperature variation introduces again a drift in
is depicted in Fig. 5(c), (f), and (i). As it is illustrated, the the OCV–SOC curve. Using the proposed temperature com-
forecompensation technique is able to cope with the drift pensation method, the corrected OCV–SOC correlations are
introduced by temperature variations. shown in Fig. 7(d)–(f). More importantly, a good mapping is
Next, three sets of experiments are carried out to study shown with offline OCV measurements taken as a benchmark
the performance of the proposed estimation scheme. For each to demonstrate the high performance of the proposed strategy.
experiment, a fully charged LiFePO4 battery is subjected again Remark 3: State of Health (SOH) estimation is also an
to different temperatures (10 °C, 25 °C, and 40 °C) and is important aspect to determine the battery’s end of life.
completely discharged at various rates (6, 10, and 20 A) until In general, offline SOH estimation is accomplished using
its voltage reaches the preset cutoff, i.e., 2.8 V. The system’s AC signal injection. However, the interruption of the system
response is studied taking into account the battery’s voltage operation is required to perform a capacity check. Moreover,
estimation error e and the OCV estimate 4 ≈ Voc . Since additional and costly measurement instrumentation is neces-
the tracking error behavior is similar for all temperatures, it sary [35]. Another offline procedure is to monitor the time to
is then shown in Fig. 6 only for one case (10 A at 25 °C) completely discharge a fully charged battery with a constant
and for the first 25 s to highlight convergence at startup. As it current, which usually takes several minutes or hours. Unlike
is illustrated, the battery’s estimation tracking error gradually these techniques, online SOH estimation can be achieved with

Authorized licensed use limited to: National Chung Hsing Univ.. Downloaded on June 19,2024 at 06:53:39 UTC from IEEE Xplore. Restrictions apply.
758 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 25, NO. 2, MARCH 2017

resistance measurement while battery is in operation, which [10] Y. Wang, H. Fang, Z. Sahinoglu, T. Wada, and S. Hara, “Adaptive
removes restrictions compared with other procedures. Studies estimation of the state of charge for lithium-ion batteries: Nonlinear
geometric observer approach,” IEEE Trans. Control Syst. Technol.,
have reported a battery’s resistance increase as an indication vol. 23, no. 3, pp. 948–962, May 2015.
of SOH decline. However, it is noteworthy that low battery’s [11] Z. Chen, Y. Fu, and C. C. Mi, “State of charge estimation of lithium-ion
resistance values combined with low sensor precision and batteries in electric drive vehicles using extended Kalman filtering,”
IEEE Trans. Veh. Technol., vol. 62, no. 3, pp. 1020–1030,
measurement noise make accurate estimation a difficult task Mar. 2013.
to undertake and many estimation procedures fail to predict [12] M. Corno, N. Bhatt, S. M. Savaresi, and M. Verhaegen, “Electrochemical
it with high accuracy [36]. Therefore, future work envisions model-based state of charge estimation for Li-ion cells,” IEEE Trans.
Control Syst. Technol., vol. 23, no. 1, pp. 117–127, Jan. 2015.
online estimation of the battery’s SOH through accurate iden-
[13] H. He, R. Xiong, X. Zhang, F. Sun, and J. Fan, “State-of-charge
tification of the battery’s equivalent series resistance. estimation of the lithium-ion battery using an adaptive extended Kalman
filter based on an improved Thevenin model,” IEEE Trans. Veh. Technol.,
vol. 60, no. 4, pp. 1461–1469, May 2011.
VI. C ONCLUSION [14] I. S. Kim, “Nonlinear state of charge estimator for hybrid electric vehicle
In this brief, an adaptive parameter identification approach battery,” IEEE Trans. Power Electron., vol. 23, no. 4, pp. 2027–2034,
Jul. 2008.
is developed for lithium-ion batteries. The proposed strategy [15] A. Szumanowski and Y. Chang, “Battery management system based
makes use of the capabilities of adaptive control theory to on battery nonlinear dynamics modeling,” IEEE Trans. Veh. Technol.,
achieve online parameter estimation. Therefore, robustness to vol. 57, no. 3, pp. 1425–1432, May 2008.
[16] P. J. van Bree, A. Veltman, W. H. A. Hendrix, and
parameter variation is achieved and unlike many estimation P. P. J. van den Bosch, “Prediction of battery behavior subject to
techniques, the convergence and stability are guaranteed by high-rate partial state of charge,” IEEE Trans. Veh. Technol., vol. 58,
Lyapunov’s direct method. Furthermore, it is easier to be no. 2, pp. 588–595, Feb. 2009.
implemented as opposed to other estimation methodologies [17] H. Chaoui and P. Sicard, “Accurate state of charge (SOC) estimation
for batteries using a reduced-order observer,” in Proc. IEEE Int. Conf.
with a similar performance, such as soft-computing-based Ind. Technol. Southeastern Symp. Syst. Theory (ICIT-SSST), Mar. 2011,
estimators. The proposed estimation technique shows high pp. 39–43.
tracking and estimation accuracy based on only the mea- [18] H. Chaoui, P. Sicard, and H. J. N. Ndjana, “Adaptive state of
charge (SOC) estimation for batteries with parametric uncertainties,”
surements of the battery’s voltage and current. However, in Proc. IEEE/ASME Adv. Intell. Mechatronics Int. Conf., Jul. 2010,
temperature effects tend to introduce an offset leading to a pp. 703–707.
bad estimation. Therefore, a simple temperature compensation [19] M. Gao, Y. Liu, and Z. He, “Battery state of charge online estimation
based on particle filter,” in Proc. 4th Int. Congr. Image Signal Process.,
methodology is proposed to cope with these effects. The Shanghai, China, Oct. 2011, pp. 2233–2236.
experimental results highlight the performance of the proposed [20] J. Li, J. K. Barillas, C. Guenther, and M. A. Danzer, “Sequential
estimator in determining the SOC with high accuracy. The Monte Carlo filter for state estimation of LiFePO4 batteries based
on an online updated model,” J. Power Sour., vol. 247, pp. 156–162,
effectiveness of the proposed approach is validated using Feb. 2014.
a practical temperature range, i.e., from 10 °C to 40 °C. [21] S. Schwunk, N. Armbruster, S. Straub, J. Kehl, and M. Vetter, “Particle
To extend this work, the future plan might envision to cover filter for state of charge and state of health estimation for lithium–
temperatures under 0 °C with a discharge rate higher than 1C. iron phosphate batteries,” J. Power Sour., vol. 239, pp. 705–710,
Oct. 2013.
[22] Y. Kim, S. Mohan, J. B. Siegel, A. G. Stefanopoulou, and Y. Ding,
R EFERENCES “The estimation of temperature distribution in cylindrical battery cells
under unknown cooling conditions,” IEEE Trans. Control Syst. Technol.,
[1] M.-Y. Kim, C.-H. Kim, J.-H. Kim, and G.-W. Moon, “A chain structure vol. 22, no. 6, pp. 2277–2286, Nov. 2014.
of switched capacitor for improved cell balancing speed of lithium-ion [23] W. Chen, W.-T. Chen, M. Saif, M.-F. Li, and H. Wu, “Simultaneous fault
batteries,” IEEE Trans. Ind. Electron., vol. 61, no. 8, pp. 3989–3999, isolation and estimation of lithium-ion batteries via synthesized design of
Aug. 2014. Luenberger and learning observers,” IEEE Trans. Control Syst. Technol.,
[2] A. Kuperman, U. Levy, J. Goren, A. Zafransky, and A. Savernin, vol. 22, no. 1, pp. 290–298, Jan. 2014.
“Battery charger for electric vehicle traction battery switch station,” [24] A. Bartlett, J. Marcicki, S. Onori, G. Rizzoni, X. G. Yang, and T. Miller,
IEEE Trans. Ind. Electron., vol. 60, no. 12, pp. 5391–5399, Dec. 2013. “Electrochemical model-based state of charge and capacity estimation
[3] A. H. Ranjbar, A. Banaei, A. Khoobroo, and B. Fahimi, “Online for a composite electrode lithium-ion battery,” IEEE Trans. Control Syst.
estimation of state of charge in Li-ion batteries using impulse response Technol., vol. 24, no. 2, pp. 384–399, Mar. 2016.
concept,” IEEE Trans. Smart Grid, vol. 3, no. 1, pp. 360–367, Mar. 2012. [25] H. Chaoui and P. Sicard, “Adaptive fuzzy logic control of permanent
[4] T. Hansen and C.-J. Wang, “Support vector based battery state of charge magnet synchronous machines with nonlinear friction,” IEEE Trans. Ind.
estimator,” J. Power Sour., vol. 141, no. 2, pp. 351–358, Mar. 2005. Electron., vol. 59, no. 2, pp. 1123–1133, Feb. 2012.
[5] S. Duryea, S. Islam, and W. Lawrance, “A battery management system [26] H. Chaoui, P. Sicard, J. Lee, and A. Ng, “Neural network modeling of
for stand-alone photovoltaic energy systems,” IEEE Ind. Appl. Mag., cold-gas thrusters for a spacecraft formation flying test-bed,” in Proc.
vol. 7, no. 3, pp. 67–72, Jun. 2001. IEEE Conf. Ind. Electron. Soc., Oct. 2012, pp. 2619–2621.
[6] L. Liu, L. Y. Wang, Z. Chen, C. Wang, F. Lin, and H. Wang, “Integrated [27] M. Charkhgard and M. Farrokhi, “State-of-charge estimation for lithium-
system identification and state-of-charge estimation of battery systems,” ion batteries using neural networks and EKF,” IEEE Trans. Ind.
IEEE Trans. Energy Convers., vol. 28, no. 1, pp. 12–23, Mar. 2013. Electron., vol. 57, no. 12, pp. 4178–4187, Dec. 2010.
[7] M. Gholizadeh and F. R. Salmasi, “Estimation of state of charge, [28] H.-T. Lin, T.-J. Liang, and S.-M. Chen, “Estimation of battery state of
unknown nonlinearities, and state of health of a lithium-ion battery based health using probabilistic neural network,” IEEE Trans. Ind. Informat.,
on a comprehensive unobservable model,” IEEE Trans. Ind. Electron., vol. 9, no. 2, pp. 679–685, May 2013.
vol. 61, no. 3, pp. 1335–1344, Mar. 2014. [29] F.-J. Lin, M.-S. Huang, P.-Y. Yeh, H.-C. Tsai, and C.-H. Kuan,
[8] M. Shahriari and M. Farrokhi, “Online state-of-health estimation of “DSP-based probabilistic fuzzy neural network control for Li-ion battery
VRLA batteries using state of charge,” IEEE Trans. Ind. Electron., charger,” IEEE Trans. Power Electron., vol. 27, no. 8, pp. 3782–3794,
vol. 60, no. 1, pp. 191–202, Jan. 2013. Aug. 2012.
[9] A. A. Hussein, “Capacity fade estimation in electric vehicle Li-ion [30] M. Chen and G. A. Rincon-Mora, “Accurate electrical battery model
batteries using artificial neural networks,” IEEE Trans. Ind. Appl., capable of predicting runtime and I-V performance,” IEEE Trans. Energy
vol. 51, no. 3, pp. 2321–2330, May/Jun. 2015. Convers., vol. 21, no. 2, pp. 504–511, Jun. 2006.

Authorized licensed use limited to: National Chung Hsing Univ.. Downloaded on June 19,2024 at 06:53:39 UTC from IEEE Xplore. Restrictions apply.
CHAOUI AND GUALOUS: ADAPTIVE SOC ESTIMATION OF LITHIUM-ION BATTERIES 759

[31] J. Kim, J. Shin, C. Chun, and B. H. Cho, “Stable configuration of a [34] N. Omar et al., “Lithium iron phosphate based battery—Assessment
Li-ion series battery pack based on a screening process for improved of the aging parameters and development of cycle life model,” Appl.
voltage/SOC balancing,” IEEE Trans. Power Electron., vol. 27, no. 1, Energy, vol. 113, pp. 1575–1585, Jan. 2014.
pp. 411–424, Jan. 2012. [35] W. Huang and J. A. A. Qahouq, “An online battery impedance
[32] M. Petzl and M. A. Danzer, “Advancements in OCV measurement and measurement method using DC–DC power converter control,”
analysis for lithium-ion batteries,” IEEE Trans. Energy Convers., vol. 28, IEEE Trans. Ind. Electron., vol. 61, no. 11, pp. 5987–5995,
no. 3, pp. 675–681, Sep. 2013. Nov. 2014.
[33] H. Rahimi-Eichi, F. Baronti, and M.-Y. Chow, “Online adaptive para- [36] H. Chaoui, A. El Mejdoubi, A. Oukaour, and H. Gualous, “Online
meter identification and state-of-charge coestimation for lithium-polymer system identification for lifetime diagnostic of supercapacitors with
battery cells,” IEEE Trans. Ind. Electron., vol. 61, no. 4, pp. 2053–2061, guaranteed stability,” IEEE Trans. Control Syst. Technol., vol. PP, no. 99,
Apr. 2014. pp. 1–9, 2016, doi: 10.1109/TCST.2016.2520911.

Authorized licensed use limited to: National Chung Hsing Univ.. Downloaded on June 19,2024 at 06:53:39 UTC from IEEE Xplore. Restrictions apply.