Energies 2012, 5, 1-x manuscripts; doi:10.

3390/en50x000x 1
2
energies 3
ISSN 1996-1073 4
www.mdpi.com/journal/energies 5
Article 6
Estimation of State of Charge of Lithium-Ion Batteries used in 7
HEV using Robust Extended Kalman Filtering 8
Caiping Zhang
1,
*, Jiuchun Jiang
1
, Weige Zhang
1
, S.M. Sharkh
2
9
1
School of Electrical Engineering, Beijing Jiaotong University, Beijing, 100044, China 10
2
School of Engineering Sciences, University of Southampton, Highfield, Southampton, SO17 1BJ, 11
UK 12
E-Mails: jcjiang@bjtu.edu.cn; wgzhang@bjtu.edu.cn; S.M.Abu-Sharkh@soton.ac.uk 13
* Author to whom correspondence should be addressed; Tel.: +86-10-5168-3907; Fax: +86-10-5168- 14
3907, Email: cpzhang@bjtu.edu.cn 15
Received: / Accepted: / Published: 16
17
Abstract: Robust extended Kalman filter (EKF) is proposed as a method for estimation of 18
the state of charge (SOC) of lithium-ion batteries used in hybrid electric vehicles (HEV). An 19
equivalent circuit model of the battery, including its electromotive force (EMF) hysteresis 20
characteristic and polarization characteristics is used. The effect of the robust EKF gain 21
coefficient on SOC estimation is analyzed, and an optimized gain coefficient is determined 22
to restrain battery terminal voltage from fluctuating. Experimental and simulation results are 23
presented. SOC estimates using the standard EKF are compared with the proposed robust 24
EKF algorithm to demonstrate the accuracy and precision of the latter for SOC estimation. 25
Keywords: lithium-ion batteries; SOC estimation; robust estimation; EKF; HEV 26
27
1. Introduction 28
Lithium-Ion batteries have become a promising alternative power source in electric vehicles (EV) 29
and power assist units used in hybrid electric vehicles (HEV). Their advantages include high nominal 30
cell voltage, high energy density, long life and not having a memory effect. As one of the power 31
supplies on a vehicle, the performance of lithium-ion batteries will have direct impact on the driving 32
performance of the vehicle. It is necessary for the battery to be effectively managed to improve its 33
OPEN ACCESS
Energies 2012, 5


2
performance and extend its lifetime. The main components of a battery management system include 34
state of charge (SOC) estimation, cell balancing, thermal management, and safety control. The SOC, 35
which describes the percentage of the battery available capacity to its rated capacity, is a key parameter 36
in the battery management system. Accurate estimation of the SOC of the battery is also important for 37
accurate simulation and optimization, and real time energy management of HEV and EV. 38
Techniques for estimation of the SOC of a battery may be categorized as direct computational 39
methods or intelligent computational methods. Direct computational methods calculate the SOC of the 40
battery directly based on its relationship with measurable battery parameters, for example, using 41
ampere hour counting, open-circuit voltage, or internal impedance [1-3]. These methods are 42
extensively used in HEV and EV applications since they are easy to implement. However, they suffer 43
from relatively poor accuracy due to accumulative errors, especially when ampere hour counting is 44
used. 45
Intelligent computational methods include those using artificial neural networks and extended 46
Kalman filtering. The artificial neural networks approach has the advantage of adaptive learning, and 47
can cope with the batterys nonlinear characteristics during charging and discharging. It has been 48
investigated and used by many researchers [4, 5]. However, the algorithm requires a large amount of 49
data for training, and the accuracy of these models is affected significantly by the training data and 50
training method. 51
In the extended Kalman filtering (EKF) approach the battery is regarded as a dynamic system, and 52
the SOC is considered as an internal state of the dynamic system. Optimal state estimates can be 53
obtained by adjusting the filter gain [6]. Using EKF to estimate the SOC of batteries has been the 54
subject of extensive study in recent years due to its high accuracy and suitability for real time 55
implementation [7-9]. For an accurate estimate of the SOC, the EKF requires accurate system 56
modeling as well as knowledge of the statistical properties of the system noise. However, in practice, 57
the dynamic system model and the covariance matrices cannot be precisely determined especially in 58
the environment of HEV including a variety of interference noises, e.g. Schottky noise, thermal noise 59
and space electromagnetic noise, which may cause random modeling errors thus altering the statistical 60
properties of the system noise. Using regular EKF may result in accumulative errors in the states 61
estimates under the above conditions, and may even cause filter divergence. 62
In this paper we propose a robust EKF in which the state estimation is decoupled from the bias state 63
estimation. The state estimation error is gradually reduced to a minimum under a certain criterion even 64
if the system model contains indeterministic information. 65
This paper is organized as follows. The next section discusses the modeling of the lithium-ion 66
battery used in HEV as the foundation for SOC estimation. EMF hysteresis is included in the battery 67
model to improve SOC estimation. Section 3 introduces the robust state estimation EKF method. SOC 68
estimation based on the proposed battery model is performed. The effects of the bias vector on state 69
estimation and the acquisition of the bias constant are subsequently addressed. And then the 70
optimization of Kalman gain is proposed. The results of simulation and tests are discussed in section 4. 71
72
Energies 2012, 5


3
2. Modeling of the lithium-ion battery 73
74
Modeling of the battery aims to find the relationship between currents and voltages measured at the 75
terminals of the battery. Besides concentration and activation polarization effects lithium ion batteries 76
exhibit an equilibrium potential hysteresis phenomenon. The hysteresis characteristics of the lithium- 77
ion battery have a notable effect on SOC estimation accuracy, and it is therefore important that the 78
model should incorporate these characteristics. 79
2.1. Hysteresis characteristics 80
The cell equilibrium potential depends on its charge and discharge history, and in some batteries 81
exhibits hysteresis. There are many publications on the hysteresis characteristics for nickel-hydrogen 82
batteries; a detailed study is reported in [10, 11]. A preliminary study of lithium-ion battery hysteresis 83
characteristics is reported in [12-15]. 84
A series of experiments were conducted in order to study the open circuit voltage characteristics of 85
the battery and the influence of the current on its hysteresis during charging and discharging. The 86
lithium-ion battery studied in this paper is composed of 16 cells in series. Figure 1 (a) shows the open- 87
circuit cell voltage versus SOC under the same charge/discharge current of 30 A. Figure 1 (b) shows 88
the open circuit voltage versus SOC for different charge/discharge currents. The testing procedure in 89
Figure 1 (a) was as follows: at room temperature, the battery was fully charged, left it in the open- 90
circuit state for 2 hours, and then discharged by 10 % of the rated capacity at a current of 30A.The 91
battery is then kept in the open-circuit state and the open circuit voltage was observed. After 10 hours, 92
the measured battery voltage was regarded to be the equilibrium potential of the battery since its 93
growth rate was negligible. The process was subsequently repeated to get the equilibrium voltage 94
curve during discharge as shown in Figure 1 (a). Similarly the battery was charged, and the 95
equilibrium open circuit voltage of the battery was measured every 10% SOC to obtain the open circuit 96
voltage curve during charging. Using the same discharge test procedure described above, the 97
equilibrium potential was also measured when the discharge current was 100 A as shown in Figure 98
1(b). 99
Figure 1. Open-circuit cell voltages as a function of SOC at room temperature 100
101
From Figure 1 (a), it is evident that the equilibrium open circuit voltage of a cell for a given SOC 102
during charge and discharge are different. The OCV measured during charge is higher than that 103
0 0.2 0.4 0.6 0.8 1
3.3
3.4
3.5
3.6
3.7
3.8
3.9
4
4.1
4.2
SOC
(a)
O
C
V
(
V
)


OCV after discharge
OCV after charge
0 0.2 0.4 0.6 0.8 1
3.3
3.4
3.5
3.6
3.7
3.8
3.9
4
4.1
4.2
SOC
(b)
O
C
V
(
V
)


OCV after 30A discharge
OCV after 100A discharge
Energies 2012, 5


4
measured during discharge. It is evident in Figure 1(b) that the equilibrium potentials at the discharge 104
current of 30 A and 100 A are nearly the same and the biggest difference is only 3mV, which means 105
the equilibrium potential of the battery is essentially independent of the battery current, and the 106
difference of equilibrium potential between charge and discharge is an inherent characteristic of the 107
battery itself. Thus, for a specified open-circuit voltage there are two different SOC values, which 108
introduce uncertainty in SOC determination using the OCV method. Therefore, it is important to 109
analyze the hysteresis characteristics of the battery to reduce SOC estimation error. 110
Table 1. Comparison of the cell OCV during charging and discharging at SOC=0.5 111
Test conditions
OCV after
Charge/(V)
OCV after
Discharge/(V)
Difference
/(mV)
Average
OCV
/(V)
At room temperature, left at
open-circuit state for 10 h
3.979 3.955 24 3.967
112
Table 1 shows the equilibrium potential measured during charging process and discharging process 113
when the SOC is 0.5. We find that the OCV after charging is 24 mV higher than that measured after 114
discharging. The difference of 24 mV compared to the whole working voltage range of the battery is 115
small (the voltage range of single battery over the whole SOC working range is around 1200 mV). 116
However, the gradient of the measured OCV of the battery is relatively small in the middle region of 117
SOC and a small hysteresis of 24 mV can cause a significant error in the SOC estimate based on the 118
OCV curve. The practical capacity of the battery used in the experiment is 94 Ah, and the change in 119
capacity per 1 mV change of open-circuit voltage is 0.19Ah mV-1 in the vicinity of the 50% depth of 120
discharge (DOD). A 24 mV uncertainty in the OCV means a 4.56 Ah capacity estimate uncertainty, 121
with a significant corresponding SOC uncertainty of around 4.85%. It is therefore important that the 122
hysteresis characteristics are included in the battery model used to estimate the SOC. 123
2.2. Model formulation 124
The proposed equivalent circuit model including OCV hysteresis is shown in Figure 2. It comprises 125
three parts: (1) open-circuit battery voltage V
oc
, which is composed of an average equilibrium potential 126
V
e
and a hysteresis voltage V
h
, (2) internal resistance R
i
comprising the Ohmic resistance R
o
and the 127
polarization resistances, R
pa
and R
pc
. R
pa
represents effective resistance characterizing activation 128
polarization and R
pc
represents the effective resistance characterizing concentration polarization, (3) 129
effective capacitances C
pa
and C
pc
, which are used to describe the activation polarization and 130
concentration polarization, and used to characterize the transient response of the battery. In addition, 131
ideal diodes are added so that different resistance parameters are used during charging and discharging. 132
The resistances connected in series with the diodes have additional subscripts to indicate charging or 133
discharging. But in the equations these subscripts are omitted for conciseness. 134
By analyzing the hysteresis characteristics as the function of SOC, the electrical behavior of the 135
circuit can be expressed as follows: 136
Energies 2012, 5


5
,max
( )( ( ) )
pa
pa
pa pa pa
pc
pc
pc pc pc
h h h
t e h pa pc o
V
I
V
R C C
V
I
V
R C C
V sign I V sign I V
V V V V V IR
o

= +

= +

= +

= +
(1) 137
where I is the current through the battery, V
h,max
represents the maximum hysteresis voltage of the 138
battery as a function of SOC, which is defined as s , is hysteresis coefficient, and V
t
is the terminal 139
voltage of the battery. In addition, the average open-circuit voltage of the battery is considered as the 140
equilibrium potential V
e
in this paper. The identification of the model parameters was described in 141
detail in a previous paper by the authors [16]. 142
Figure 2. Proposed equivalent circuit model for the lithium-ion battery 143
V
e
V
h
(a)
R
SD
R
od
R
pad
C
pc
C
pa
V
oc
V
t
R
oc
R
pac
R
pcd
R
pcc
I
V
pc
V
pa
144
145
3. SOC estimation using robust extended Kalman filtering 146
3.1. Robust state estimation 147
The main idea of robust state estimation is that the modeling errors are regarded as constant bias 148
state vectors. The dynamic states of the system and bias states are estimated separately, and the 149
dynamic states are subsequently corrected using the bias states estimated values. To get the solution to 150
states and bias estimation of a dynamic system, Friedland proposed decoupling the bias estimation 151
from the state estimation to reduce computational complexity [17]. Hsieh and Chen generalized 152
Friedlands filter and proposed an optimal solution for a two-stage Kalman estimator by applying a 153
two-stage U-V transformation [18, 19]. To reduce the number of arithmetic operations in speed and 154
rotor flux estimation of induction machine, Hilairet proposed modified optimal two-stage Kalman 155
estimator derived from Hsieh and Chen algorithm [20]. Considering the nonlinear characteristics of the 156
lithium-ion battery model, we propose in this paper a robust state estimation including EKF, which is 157
suited for nonlinear systems. If modeling errors are neglected, the nonlinear system of interest can be 158
defined by: 159
Energies 2012, 5


6
1 1 1
,
( , , )
( , , )
~ (0, )
~ (0, )
k k k k
k k k k k
k x k
k k
x f x u w
y h x u v
w Q
v R

=

(2) 160
where x
k
is the vector of dynamic states, u
k
is the control input, w
k
represents process noise which is 161
assumed to be discrete-time Gaussian zero-mean white noise with covariance of Q
x,k
, v
k
represents 162
measurement noise which is assumed to be discrete-time Gaussian white noise with zero mean and a 163
covariance R
k
, F
k-1
and B
k-1
are system matrices. 164


( , , 0) ( )
( , , 0) ( )
[ ( , , 0) ]
k k
k k k k k k k
x x
k k k k k k k k
k k k k k k k k k
k k k k
h h
y h x u x x v
x v
h x u C x x M v
C x h x u C x M v
C x z v




c c
= + +
c c
= + +
= + +
= + +
(3) 165
where
k
z and
k
v are defined as follows: 166
( , , 0)
~ (0, )
k k k k k k
T
k k k k
z h x u C x
v N M R M

=

(4) 167
To allow for modeling errors constant matrices C A A A and F A are introduced such that the 168
linearized systems equations with modeling error is are given by: 169
1 1 1 1 1
( A) ( F)
k k k k k k
x A x F u w

= +A + +A + (5) 170
( C)
k k k k k
y C x z v = +A + + (6) 171
where: 172
T
1 n
A [a a ] A = ,
T
1 n
F [f f ] A = ,
T
1 n
C [c c ] A = . 173
where a
n
i
R e , f
q
i
R e , c
m
i
R e , 1, , i n = . 174
Define a constant bias vector
( )
b
n n q m
k
R
+ +
e such that
T T T T T T T
1 n 1 1
b [a a f f c c ]
k q m
= . The equations (5) 175
and (6) can be rearranged as follows: 176
1 1 1 1 1 1 1 k k k k k k k k
x A x F u B b w

= + + + (7) 177
k k k k k k k
y C x D b z v = + + + (8) 178
where
1 1
1
1 1
0 0
0 0
T T
k k
k
T T
k k
n n
x u
B
x u

(
(
=
(
(
(

,
0 0 0
0 0 0
T
m k
k
T
m k
m
x
D
x
(
(
=
(
(
(

. 179
The vector b
k
contains the constant bias states, which includes Gaussian white noise in practice, and 180
can be expressed by: 181
1 1 k k k
b b

= +, (9) 182
where
k
, is Gaussian zero-mean white noise, and 183
( )
T
i j b ij
E Q o = , ( ) ( ) ( ) 0
T T T
i j i j i j
E wv E w E v , , = = = . 184
The state estimation of the system with modeling error is converted to state estimation of the system 185
with constant bias through the transformation. The robust state estimation combined EKF based on 186
separate bias estimator are derived in detail in appendix A as described in [22]. Combining the 187
equations in appendix A, the estimates of x can be given by: 188
1 1 1 1 1 1

k k k k k k k
x A x F u B b
+ +

= + + (10) 189
Energies 2012, 5


7
, 1 , 1 1 1 , 1 1 , 1
T T
x k k x k k k b k k x k
P A P A B Q B Q
+

= + + (11) 190
1
, , ,
( )
T T
x k x k k k x k k k
K P C C P C R

= + (12) 191
, , ,
( )
x k x k k x k
P I K C P
+
= (13) 192
, , k x k k b k
K K V K = + (14) 193

( ( , , 0) )
k k k k k k k k
x x K y h x u D b
+
= + (15) 194
From equations (10) to (15), it is shown that the bias information
1 , 1 1
T
k b k k
B Q B

is added to estimate of 195
x . The vector b
k
appears as an input to the system and thus its associated noise Q
b
must be included in 196
the estimate of P. In equations (10) to (15), the matrices B
k
and D
k
contains the states
1

k
x
+

, which is 197
used instead of the states x
k
in this paper: 198
1 1
1 1
0 0
0 0
T T
k k
k
T T
k k
n n
x u
B
x u
+

+

(
(
=
(
(
(


1
1
0 0 0
0 0 0
T
m k
k
T
m k
m
x
D
x
+

(
(
=
(
(
(

. 199
From the equations, it is inferred that the dynamic states of the system and the bias states are 200
decoupled. They can be estimated based on the proposed robust estimation algorithm, which reduces 201
the dimensions of the dynamic state equations, and further decreases the computation compared to the 202
virtual noise compensation algorithm described in [23]. 203
3.2. SOC estimation based on the proposed battery model 204
SOC can be regarded as a state variable, which is added to the proposed battery model. Discretizing 205
the equivalent model of the battery, and the system can be expressed by [24]: 206
, , 1 1
, , 1 1
, , 1 ,max 1
1 1
,
exp( / ( )) (1 exp( / ( )))
exp( / ( )) (1 exp( / ( )))
exp( ) (1 exp( )) ( ) ( )
/
pa k pa k pa pa k pa pa pa
pc k pc k pc pc k pc pc pc
h k h k h k
k k i k N
t
V V t R C I R t R C
V V t R C I R t R C
V V t t sign I V s
s s I t C
V
o o
q




= A + A

= A + A

= A A

= A

, , ,
( )
k e k k o pa k pc k h k
V s I R V V V = +
(16) 207
where t A is the sampling interval;
k
s represents the SOC,
i
q stands for columbic efficiency, which 208
is a function of the battery current. Using curve fitting of the experiment data, the equilibrium potential 209
V
e
as the function of SOC was determined as follows: 210
4 3 2
( ) 5.2 0.86 12 15 59
e k k k k k
V s s s s s = + + (17) 211
To estimate the SOC using the Robust EKF, the mathematical model of the battery in (17) needs to 212
be rearranged as: 213
1 1 1 1
( , )
k k k k k
k k k
x A x F u
y h x u

= +

(18) 214
Where 215
T
pa pc h
x V V V s ( =

, u I = ,
t
y V = . 216
1
exp( / ( )) 0 0 0
0 exp( / ( )) 0 0
0 0 0
0 0 0 1
pa pa
pc pc
k
t R C
t R C
A
t o

A (
(
A
(
=
( A
(

exp(- )
, 217
T
1
(1 exp( / ( ))) (1 exp( / ( ))) 0 /
k pa pa pa pc pc pc i N
F R t R C R t R C t C q

( = A A A

. 218
Energies 2012, 5


8
The largest source of error is the SOC estimation error. There are two main possible sources of 219
modeling error. One is the change of the internal resistance caused by the effect of the working 220
environment such as temperature and aging of the battery. The other arises from inaccuracy of the 221
mathematical relationship between the equilibrium potential and SOC since the function is obtained by 222
curve fitting the experiment data; the measurement error and fitting error may lead to estimation error 223
of SOC. Referring to the system equations (5) and (6) including modeling error, and considering the 224
SOC error components of the battery system, the bias matrix of the battery system can be defined as: 225
| | 0 0 0 C A = 226
where is a bias constant. The bias vector will be therefore given by: 227
T
[0 0 0 ]
k
b = 228
The procedure of SOC estimation using the proposed robust EKF algorithm is shown in Figure 3. 229
Figure 3. Procedure of the Robust EKF algorithm in SOC estimation 230
1 , 1 1 , 1

, , ,
k b k k x k
b P x P
+ + + +

Caluculation
,
k k
B D
Prediction
, ,

, , ,
k bk k x k
b P x P

Caluculation
, ,
, , , ,
k k x k b k k
U S K K V
Caluculation
,

t k
V
Caluculation
, ,

, , ,
k b k k x k
b P x P
+ + + +

,
k k k k
x x b b
+ +
= =
231
3.3. Analysis of the effect of the bias vector on state estimation 232
As described above, the values of the bias vector will have significant effect on robustness of SOC 233
estimation using the Robust EKF algorithm. It is therefore necessary to investigate the influence of the 234
bias constant variation on the estimation of the states of the battery system. The battery discharge 235
current of 1C was used in the experiment. The results are shown in Figure 4. 236
The bias constant as described in Section 3.2 reflects measurement error and coefficient fitting 237
error of the battery system, which impact on the relationship between states (V
pa
, V
pc
, V
h
and SOC) 238
estimation of the battery dynamics and the parameter is illustrated in Figure 4. From Figure 4, it is 239
seen that the polarization voltage V
pa
and hysteresis voltage V
h
estimate are hardly affected by the 240
variation of the bias constant . But the change in the polarization voltage V
pc
and SOC estimate for a 241
bias constant value of 0.5 are 0.157 V and 0.01, respectively compared to their value when =0. It is 242
clear that the estimates of V
pc
and SOC are influenced significantly by the bias constant. The 243
Energies 2012, 5


9
polarization voltage V
pc
overestimated when >0 compared and underestimated when <0. In contrast, 244
the SOC estimate is underestimated when >0, and overestimated when <0. 245
Figure 4. Impact of bias constant on states estimation 246
247
We therefore conclude that the influence of the bias constant variation on the polarization voltage 248
V
pa
and hysteresis voltage V
h
estimate can be ignored. But the polarization voltage V
pc
and SOC 249
estimate errors tends to change linearly as the bias constant varying. The value of the bias constant 250
can be determined as discussed in the following section. 251
3.4. Determining the bias constant 252
Initialization of the covariance matrices, and determining with a covariance Q of the plant noise and 253
in the covariance R of the measurement noise, are important parts of the Robust EKF procedure since 254
it impacts significantly on its estimation performance. The dynamic states and bias state are separately 255
estimated in the Robust EKF algorithm. It is therefore necessary to choose the values of the filter 256
parameters for both the dynamic states and bias state parts. 257
A judicious choice of Q and R is attained from experimental studies under the simplifying 258
assumption that the noise processes {w
k
} and {v
k
} are uncorrelated, leading to diagonal Q and R. The 259
initial covariance matrices P
b
and Q
b
, and the initial weighted matrix were ultimately chosen to be 260
0.01
b
P = , 0.00001
b
Q = , (0.001, 0.001, 0.001, 0.001) V diag = . 261
And the initial covariance matrix
x
P , together with Q
x
and R were given by: 262
0.0001 0 0 0
0 0.0001 0 0
0 0 0.0001 0
0 0 0 0.0001
x
Q
(
(
(
=
(
(

,
10 0 0 0
0 10 0 0
0 0 0.1 0
0 0 0 0.04
x
P
(
(
(
=
(
(

, 5 R = . 263
Data obtained from tests of battery discharge at constant current was used in this paper to verify 264
SOC estimation using the proposed battery model with Robust EKF assuming the bias constant is 265
0 2 4 6 8 10
-0.05
0
0.05
0.1
0.15
Time (s)
P
o
l
a
r
i
z
a
t
i
o
n

v
o
l
t
a
g
e

V
p
a

(
V
)
0 2 4 6 8 10
-0.2
0
0.2
0.4
0.6
Time(s)
P
o
l
a
r
i
z
a
t
i
o
n

v
o
l
t
a
g
e

V
p
c

(
V
)
0 2 4 6 8 10
-0.012
-0.01
-0.008
-0.006
-0.004
-0.002
0
Time(s)
H
y
s
t
e
r
e
s
i
s

v
o
l
t
a
g
e

V
h

(
V
)
0 2 4 6 8 10
0.892
0.894
0.896
0.898
0.9
0.902
0.904
Time(s)
S
O
C
=0
=0.5
=-0.5
=-0.2
=-0.5
=0
=-0.2
=0.5
=0.2
=0.2
Energies 2012, 5


10
zero. The measured and estimated SOC of the battery for three rates of discharge of C/3, 1C and 1.5C, 266
respectively, are plotted in Figure 5. From Figure 5(a), (b) and (c), it is clear that the estimated SOC 267
are all less than the true value. The difference between the estimated and measured SOC increases with 268
the battery current. Based on the analysis of the impact of the bias vector on SOC estimation as 269
discussed in Section 3.3, the bias constant needs to be negative to reduce the steady error of SOC 270
estimate. A bias constant ranging from -0.8 to 0 was selected based on the practical results in Figure 271
5, and an optimal value of is ultimately chosen, based on Figure 6, to be -0.4. 272
Figure 5. The experimental and estimated SOC of the battery with Robust EKF at =0 (SOC0=0.9) 273
274
275
276
0 1000 2000 3000 4000 5000 6000 7000 8000
0
0.2
0.4
0.6
0.8
1
Time(s)
(a)
S
O
C


true
Robust EKF (=0)
0 500 1000 1500 2000 2500 3000
0
0.2
0.4
0.6
0.8
1
Time(s)
(b)
S
O
C


true
Robust EKF( =0)
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Time (s)
(c)
S
O
C


true
Robust EKF( =0)
Energies 2012, 5


11
Figure 6. Modified result of Robust EKF estimation with different bias value at the discharge 277
current of 1C (SOC
0
=0.9) 278
279
3.5. Optimization of the filter gain coefficient 280
The greatest advantage of EKF is to make the estimated state with initial error converge quickly to 281
the true value of the state. While the SOC value describing the state of charge of the battery is a 282
gradually changing state variable, the rate of change of the battery terminal voltage is relatively more 283
rapid especially when the current changes direction between charging and discharging, leading to large 284
fluctuations of the terminal voltage error of the battery model, which may further lead to oscillations of 285
the SOC estimate thus enlarging the error. To resolve this problem, an optimal filter gain coefficient r 286
is introduced to adjust the Kalman gain of the SOC in order to improve the stability of SOC estimation 287
while ensuring the required precision. Based on this idea, the states estimation is changed to: 288

( ( , , 0) )
k k k k k k k k k
x x K y h x u D b
+
= +I (19) 289
where
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0
k
r
(
(
(
I =
(
(

. 290
Figure7. Effect of gain coefficient values on SOC estimation with Robust EKF 291
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0 100 200 300 400 500
Time (s)
S
O
C
TRUE
r=0.3
r=0.1
r=0.05
r=0.01
292
0 500 1000 1500 2000 2500 3000
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Time (s)
S
O
C


true
=-0.05
=-0.2
=-0.4
=-0.8
Energies 2012, 5


12
The value of the gain coefficient r directly affects the accuracy and stability of SOC estimation. If 293
the value is too big, the fluctuation of SOC estimate might be enlarged, and the accuracy reduced. If it 294
is too small, filtering convergence speed to the true value of the state may be decreased, and the 295
accuracy of SOC estimation may also be affected. 296
The convergence speed and estimation accuracy were all taken into consideration during the 297
optimization of the gain coefficient. Based on comparison between experimental and simulation results 298
with different r values of ranging from 0.01 to 0.5 were investigated. 299
In Figure 7, the true SOC value during a hybrid pulse charge/discharge test is compared with the 300
estimated SOC obtained from the Robust EKF algorithm for different gain coefficients assuming the 301
same initial error. It is evident that the bigger the gain coefficient r, the faster the convergence speed to 302
the true value of SOC, however, the estimated SOC is easily influenced by the terminal voltage of the 303
battery. Taking r = 0.3 for example, during t = 60 s ~ 70 s, the changing from the discharging state to 304
charging state causes the terminal voltage of the battery to rapidly increase, which results in the SOC 305
being over estimated and vice versa. When r is small, SOC estimation result can reflect the true 306
tendency well, but the speed of convergence to the true value is slow. The trend of the estimated SOC 307
curve is the closest to the True plotted in Figure 7 when r = 0.01, however, the estimation error may 308
be unacceptably large if the initial error is relatively large. A gain coefficient of 0.1 was ultimately 309
selected to achieve a compromise between the speed of convergence and estimation accuracy. 310
4. Results discussion 311
The self-defined hybrid pulse power characterization (self-defined HPPC) shown in Figure 8 was 312
applied for validating SOC estimation using the Robust EKF. Figure 9(a) shows a comparison between 313
the SOC values estimated using both the Robust EKF and regular EKF. The estimation error is shown 314
in Figure 9 (b). From Figure 9 we find that using regular EKF, the SOC estimation changes with the 315
voltage fluctuations rapidly, and the maximum estimation error can be over 10 %, which does not 316
satisfy the requirement of electric vehicle. Using the proposed Robust EKF algorithm, the estimation 317
error is within 5% during the entire process of charging and discharging of the battery, thus meeting 318
the SOC accuracy requirement. 319
Figure 8. Profile of self-defined HPPC for one cycle 320
321
322
0 50 100 150 200 250 300 350 400 450 500
-400
-300
-200
-100
0
100
200
300
400
500
Time (s)
C
u
r
r
e
n
t

(
A
)
Charge
Discharge
Energies 2012, 5


13
Figure 9. Comparing results of SOC estimation with the proposed Robust EKF and EKF for self- 323
defined HPPC cycles 324
325
The DST driving cycles was also used to validate the SOC estimation performance using Robust 326
EKF algorithm. The measured SOC and the estimated SOC for different SOC initial values are 327
illustrated in Figure 10 (the true initial SOC value was 0.9). It is seen that the estimated SOC using the 328
Robust EKF converges faster to the measured SOC values, and the estimation accuracy is significantly 329
improved compared to SOC estimate using EKF only. When the initial SOC values are set to 0.8 and 330
0.7, the SOC estimation error with Robust EKF falls to within 6 % of the true value after 300 s and 600 331
s correction, respectively. Furthermore, the SOC estimation error using the Robust EKF is enlarged 332
when the SOC is in the range from 0.55 to 0.2. This is because the battery terminal voltage reduces and 333
the output current supplied by the battery increase (for given power output) with the passage of time, 334
which results in serious polarization of the battery, leading to large estimation error. The SOC 335
estimation error using Robust EKF is controlled within 5 % during the entire DST driving cycles. 336
Figure 10. Estimated and tested SOC for DST driving cycles (SOC
0
=0.9) 337
338
0 1000 2000 3000 4000 5000 6000 7000 8000
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Time (s)
(a)
S
O
C


True
EKF
Robust EKF
( =-0.2, r=0.1)
0 2000 4000 6000 8000
-15
-10
-5
0
5
10
15
Time (s)
(b)
E
s
t
i
m
a
t
i
o
n

e
r
r
o
r
(
%
)


EKF
Robust EKF
( =-0.2, r=0.1)
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Time (s)
S
O
C


True (SOCo=0.9)
Robust EKF(SOCo=0.9)
EKF(SOCo=0.9)
Robust EKF(SOCo=0.8)
Robust EKF(SOCo=0.7)
Energies 2012, 5


14
5. Conclusions 339
The paper presented an equivalent circuit model with two RC networks characterizing battery 340
activation and concentration polarization processes. The hysteresis voltage was also included in the 341
battery model to improve the accuracy of SOC determination. The paper proposed a Robust Extended 342
Kalman filter in which the steady error of the SOC estimation was investigated, and accounted for to 343
improve the estimation accuracy. Based on the knowledge of battery characteristics, a filter gain 344
coefficient was introduced to decrease the fluctuation of SOC estimation caused by terminal voltage 345
fluctuation, which occurs when a standard EKF is used without the gain coefficient. Simulation results 346
demonstrated the accuracy of SOC estimation with the proposed Robust EKF during both the self- 347
defined HPPC cycles and DST driving cycles. The error found to be less than 5% compared to nearly 348
10 % error achieved by the standard EKF. 349
Acknowledgments 350
The work was supported by the National High Technology Research and Development Program of 351
China (No. 2011AA05A108) and National Natural Science Foundation of China (No. 71041025). 352
References 353
1. Piller, S.; Perrin, M.; Jossen, A. Methods for state-of-charge determination and their applications. 354
J. Power Sources 2001, 96, 113-120. 355
2. Sato, S.; Kawamura, A. A New Estimation Method of State of Charge using Terminal Voltage and 356
Internal Resistance for Lead Acid Battery. In Proceedings of 2002 IEEE International Conference 357
on Power Conversion, Osaka, Japan, 2002. 358
3. Soon, Ng.K.; Moo, C.S.; Chen, Y.P.; Hsieh, Y.C. Enhanced coulomb counting method for 359
estimating stat-of-charge and state-of-health of lithium-ion batteries. Applied Energy 2009, 86, 360
1506-1511. 361
4. Shen, W.X. State of available capacity estimation for lead-acid batteries in electric vehicles using 362
neural network. Energy Convers. Manage. 2007, 48, 433-442. 363
5. Cai, C.; Du, D.; Liu, Z.; Ge, J. State-of-charge (SOC) estimation of high power Ni-MH 364
rechargeable battery with artificial neural network. In Proceedings of 9th IEEE International 365
Conference on Neural Information, Singapore, 2002. 366
6. Grewal, M.S.; Andrews, A.P. Kalman filtering, theory and practice. Prentice Hall: New Jersey, 367
USA; 1993. 368
7. Plett, G.L. Extended Kalman filtering for battery management systems of LiPB-based HEV 369
battery packs - Part 3: State and parameter estimation. J. Power Sources 2004, 134, 277-292. 370
8. Vasebi, A.; Bathaee, S.M.T.; Partovibakhsh, M. Predicting state of charge of lead-acid batteries 371
for hybrid electric vehicles by extended Kalman filter. Energy Convers. Manage. 2008, 49, 75-82. 372
9. Han, J.; Kim, D.; Sunwoo, M. State-of-charge estimation of lead-acid batteries using an adaptive 373
extended Kalman filtering. J. Power Sources 2009, 188, 606-612. 374
10. Ta, K.P.; Newman, J. Proton intercalation hysteresis in charging and discharging nickel hydroxide 375
electrodes. J. Electrochem. Soc. 1999, 146, 2769-2779. 376
Energies 2012, 5


15
11. Srinivasan, V.; Weidner, J.W.; Newman, J. Hysteresis during cycling of nickel hydroxide active 377
material. J. Electrochem.Soc. 2001, 148, A969-A980. 378
12. Murray, J.J.; Sleigh, A.K.; Mckinnon, W.R. Heats and hysteresis in calorimetry of Li/LixMnO2 379
cells. Electrochimica Acta. 1991, 36, 489-498. 380
13. Sleigh, A.K.; Murray, J.J.; Mckinnon, W.R. Memory effects due to phase conversion and 381
hysteresis in Li/LixMnO2 cells. Electrochimica Acta. 1991, 36, 1469-1474. 382
14. Zheng, T.; McKinnon, W.R.; Dahn, J.R. Hysteresis during lithium insertion in hydrogen- 383
containing carbons. J. Electrochem. Soc. 1996, 143, 2137-2145. 384
15. Inaba, M.; Fujikawa, M.; Abe, T.; Ogumi, Z. Calorimetric study on the hysteresis in the charge- 385
discharge profiles of mesocarbon microbeads heat-treated at low temperatures. J. Electrochem. 386
Soc. 2000, 147, 4008-4012. 387
16. Zhang, C.P.; Zhang, C.N.; Liu, J.Z.; Sharkh, S.M. Identification of dynamic model parameters for 388
lithium-ion batteries used in hybrid electric vehicles. High Technol. Lett. 2010, 16, 6-12. 389
17. Friedland, B. Treatment of bias in recursive filtering. IEEE Trans. on Automat. Contr. 1969, AC- 390
14, 359-367. 391
18. Hsieh, C.S.; Chen, F.C. Optimal solution of the two-stage Kalman estimator. IEEE Trans. On 392
Automat. Contr. 1999, 44, 194-199. 393
19. Chen, F.C.; Hsieh, C.S. Optimal multistage Kalman estimators. IEEE Trans. On Automat. Contr. 394
2000, 45, 2182-2188. 395
20. Hilairet, M.; Auger, F.; Berthelot, E. Speed and rotor flux estimation of induction machine using a 396
two-stage extended Kalman filter. Automatica 2009, 45, 1819-1827. 397
21. Simon, D. Optimal state estimation: Kalman, H infinity, and nonlinear approaches. John Wiley 398
&Sons: Hoboken, New Jersey, USA; 2006. 399
22. Ignagni, M.B. Separate-bias Kalman estimator with bias state noise. IEEE Trans. on Automat. 400
Contr. 1990, 35, 338-341. 401
23. Deng, Z. Self-tuning filtering theory with applications: modern time series analysis method. 402
Harbin Institute of Technology Press: Harbin, China, 2003. 403
24. Frankin, G.F.; Powell, J.D.; Workman, M.L. Digital control of Dynamic Systems. Addison- 404
Wesley: Boston, USA, 1998. 405
Appendix A 406
The robust state estimation combined EKF based on separate bias estimator can be derived as 407
follows. 408
In the absence of bias error the vector b
k
is zero, the estimates of x using EKF are given by 409
1 1 1 1 k k k k k
x A x F u
+

= + (A.1) 410
, 1 , 1 1 , 1
T
x k k x k k x k
P A P A Q
+

= + (A.2) 411
, , ,
( )
x k x k k x k
P I K C P
+
= (A.3) 412
,
( ( , , 0))
k k x k k k k
x x K y h x u
+
= + (A.4) 413
with initial condition (0) (0)
x x
P P = . x is the bias-free estimate of x,
x
P is the error covariance matrix of 414
x ,
x
K is the Kalman gain matrix,
x
Q is the process noise covariance matrix of x , R is the measurement 415
noise covariance matrix. 416
Energies 2012, 5


16
The bias vector estimator is given by: 417
1

k k
b b
+

= (A.5) 418
, , 1 , 1 b k b k b k
P P Q
+

= + (A.6) 419
1
, , , ,
( )
T T T
b k b k k k b k k k x k k k
K P S S P S C P C R

= + + (A.7) 420
, , ,
( )
b k b k k b k
P I K S P
+
= (A.8) 421
,

[ ( , , 0) ]
k k b k k k k k k
b b K y h x u D b
+
= + (A.9) 422
where the weighted matrices U, S, and V are defined by: 423
1 k k k k
U AV B

= + (A.10) 424
k k k k
S C U D = + (A.11) 425

, k k x k k
V U K S = (A.12) 426
Considering the estimates of x and

b , the adjusted estimates of x are defined by: 427

k k k k
x x U b

= + (A.13) 428

k k k k
x x V b
+ + +
= + (A.14) 429
where the present estimate x is adjusted with the current estimate of the bias vector

b . 430
2012 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article 431
distributed under the terms and conditions of the Creative Commons Attribution license 432
(http://creativecommons.org/licenses/by/3.0/). 433