applied

sciences
Article
State of Charge Estimation for Lithium-Ion Power
Battery Based on H-Infinity Filter Algorithm
Lan Li 1 , Minghui Hu 1, *, Yidan Xu 1 , Chunyun Fu 1, * , Guoqing Jin 2 and Zonghua Li 2
1 State Key Laboratory of Mechanical Transmissions, School of Automotive Engineering,
Chongqing University, Chongqing 400044, China; lilan@cqu.edu.cn (L.L.); 20173202022t@cqu.edu.cn (Y.X.)
2 Chongqing Changan Automobile Co., Ltd., Chongqing 400023, China; jingq@changan.com.cn (G.J.);
Lizh3@changan.com.cn (Z.L.)
* Correspondence: hu_ming@cqu.edu.cn (M.H.); fuchunyun@cqu.edu.cn (C.F.)




Received: 20 July 2020; Accepted: 10 September 2020; Published: 13 September 2020


Abstract: To accurately estimate the state of charge (SOC) of lithium-ion power batteries in the event
of errors in the battery model or unknown external noise, an SOC estimation method based on the
H-infinity filter (HIF) algorithm is proposed in this paper. Firstly, a fractional-order battery model
based on a dual polarization equivalent circuit model is established. Then, the parameters of the
fractional-order battery model are identified by the hybrid particle swarm optimization (HPSO)
algorithm, based on a genetic crossover factor. Finally, the accuracy of the SOC estimation results
of the lithium-ion batteries, using the HIF algorithm and extended Kalman filter (EKF) algorithm,
are verified and compared under three conditions: uncertain measurement accuracy, uncertain
SOC initial value, and uncertain application conditions. The simulation results show that the SOC
estimation method based on HIF can ensure that the SOC estimation error value fluctuates within
±0.02 in any case, and is slightly affected by environmental and other factors. It provides a way to
improve the accuracy of SOC estimation in a battery management system.

Keywords: lithium-ion power batteries; fractional-order model; state of charge estimate; H-infinity
filter; hybrid particle swarm optimization algorithm

1. Introduction
The state of charge (SOC) estimation is the most basic function of battery status monitoring.
SOC represents the remaining power of the battery, and accurate SOC estimations are of great
significance in improving the battery efficiency and safety performance [1,2]. Currently, domestic and
foreign scholars have been researching different SOC estimation methods for lithium-ion batteries,
which can be divided into four major categories [3]: estimation methods based on characteristic
parameters of lithium-ion batteries, estimation methods based on ampere-hour integration, data-driven
estimation methods, and model-based estimation methods.
The derivation process of SOC estimation for lithium-ion batteries involves many characteristic
parameters, such as open circuit voltage (OCV), residual capacity, and electrochemical impedance.
There is a certain mapping relationship between these characteristic parameters and SOC. The latter
can be estimated by establishing this mapping relationship. The method of obtaining the battery
residual capacity by a discharge test is considered to be the most direct method to determine the SOC
value of lithium-ion batteries [4]. In addition to using the current residual capacity as a parameter,
M. Einhorn et al. [5] used a linear interpolation method to estimate the battery’s SOC based on the
functional relation curve between the SOC and OCV. However, this type of method can only be
applied under specific environmental constraints, such as laboratories, since it is impossible to carry
out long-term constant current discharges or static states during the motion of electric vehicles.

Appl. Sci. 2020, 10, 6371; doi:10.3390/app10186371 www.mdpi.com/journal/applsci

Appl. Sci. 2020, 10, 6371 2 of 18

Therefore, characterization parameters such as residual capacity and OCV cannot be determined [6,7].
A. J. Salkind et al. [8] optimized the parameters of the established fuzzy model through experimental
data. This fuzzy model can fit the relationship between the battery charge state and electrochemical
impedance spectroscopy (EIS) values, and the estimated error of the SOC can be controlled within
5%. However, online measurement of EIS is not easy, and varies with battery type and experimental
conditions. The exact relationship between EIS and SOC and the repeatability of the online measurement
of EIS needs to be studied further.
The ampere-hour integration [9], also known as the coulomb counting method, incurs a low
calculation cost, as well as enables fast calculation. It is the estimation method of the SOC currently
adopted by most electric vehicles. However, the estimation method based on ampere-hour integration
requires high accuracy requirements from the initial parameters of the battery model. It is difficult to
meet the sampling accuracy and frequency requirements of the sensor during the actual application of
the battery.
Data-driven estimation methods can be divided into three categories according to different
algorithms: support vector machine (SVM) [10], neural network (NN) [11], and the fuzzy control
method [12]. These methods are similar to the black box model in the battery modeling process.
In order to achieve high estimation accuracy and strong adaptability, a large amount of data is usually
required for training. Moreover, the selection of training data is very strict, while requiring high
precision in data fitting, as well as meeting the applicability of non-linear models. The data that does
not meet the requirements will make it difficult to train the model, which will affect the accuracy and,
consequently, the SOC estimation to a great extent.
The model-based SOC estimation method is self-adjusted by the difference between the battery’s
measured voltage and the model’s output voltage. This is done to overcome the error caused due to
uncertainty, and achieve the purpose of improving the estimation accuracy of the SOC. It is a robust
closed-loop estimation algorithm. Common methods include the impedance method [13], particle filter
(PF) [14], Kalman filter (KF) [15], and improved algorithm [16–18]. Among them, the PF algorithm can
provide better results in a non-Gaussian white noise system, but its operation intensity is higher than
that of the KF algorithm. The KF algorithm is widely used, because it can find the optimal solution for
a linear Gaussian system. A simple linear system can use the ordinary KF algorithm [19], whereas
for complex nonlinear models, the extended Kalman filter (EKF) algorithm has been studied and
applied extensively [20–22]. Based on the equivalent circuit model, Plett [23,24] used Kalman filter
series algorithms (KF, EKF, unscented Kalman filter (UKF)) to achieve the identification of model
parameters and SOC estimation at the same time. Sun et al. [25] improved the traditional EKF algorithm,
and realized that when the input and output of the system changed, the noise covariance matrix in
the algorithm responded accordingly to self-correct the noise. Thus, the influence of noise on the
accuracy of the traditional EKF algorithm is overcome; however, the EKF algorithm requires that the
input noise of the model be white noise with known statistical characteristics. It is greatly affected
by the measurement accuracy of the sensor during vehicle operation, and has not been applied in
real vehicles.
To solve the above problems, the H-infinity filter (HIF) algorithm based on the minimum-maximum
criterion is adopted in this study instead of the traditional EKF algorithm. This algorithm takes into
account the time-varying element of battery parameters, and does not require the details of process noise
and measurement noise. Therefore, the HIF can expand the scope of application of the SOC estimation
algorithm and improve the robustness of the algorithm under real vehicle operating conditions.
Chen et al. [26] used the HIF algorithm to estimate the SOC of lithium-ion batteries, and the method
was tested by real-time experimental data of batteries. However, it ignores the measurement noises
and random disturbances. Shu et al. [27] proposed an adaptive H-infinity filter (AHIF) algorithm to
predict the SOC of lithium-ion batteries. The proposed algorithm is comprehensively validated within
a full operational temperature range of battery, and with different aging status. Even so, it ignores
the effects of application conditions on SOC estimation. In [28–30], the HIF is also applied to the
 Appl. Sci. 2020, 10, 6371 3 of 18

Appl. Sci. 2020, 10, x FOR PEER REVIEW 3 of 18

model parameters identification and state estimation. The results show that it has better performance
96 estimation. The results show that it has better performance than UKF and EKF. However, none of
than UKF and EKF. However, none of the above algorithms comprehensively analyze the influence
97 the above algorithms comprehensively analyze the influence of various uncertain factors (i.e., noises,
of various uncertain factors (i.e., noises, SOC initial value, and application conditions) on the SOC
98 SOC initial value, and application conditions) on the SOC estimation results of lithium-ion batteries.
estimation results of lithium-ion batteries.
99 To explore the influence of different uncertainties (i.e., noises, SOC initial value, and application
To explore the influence of different uncertainties (i.e., noises, SOC initial value, and application
100 conditions) on the estimation results when HIF is used for SOC estimation, the following chapters of
conditions) on the estimation results when HIF is used for SOC estimation, the following chapters
101 this paper are arranged as follows. In the second part, a fractional-order battery model of the
of this paper are arranged as follows. In the second part, a fractional-order battery model of the
102 lithium-ion battery is established. The hybrid particle swarm optimization (HPSO) algorithm based
lithium-ion battery is established. The hybrid particle swarm optimization (HPSO) algorithm based on
103 on a genetic crossover is used to identify the parameters of the built model to obtain the initial values
a genetic crossover is used to identify the parameters of the built model to obtain the initial values of
104 of the model parameters. The accuracy of the model parameter identification under different
the model parameters. The accuracy of the model parameter identification under different working
105 working conditions is also verified. In the third part, the HIF based on the minimum-maximum
conditions is also verified. In the third part, the HIF based on the minimum-maximum criterion is
106 criterion is proposed to estimate the SOC of the lithium-ion battery. The robustness of the algorithm
proposed to estimate the SOC of the lithium-ion battery. The robustness of the algorithm is verified
107 is verified under three conditions: uncertain measurement accuracy, uncertain initial value of SOC,
under three conditions: uncertain measurement accuracy, uncertain initial value of SOC, and uncertain
108 and uncertain application conditions. Hereafter, we compared the SOC estimation results obtained
application conditions. Hereafter, we compared the SOC estimation results obtained using the EKF
109 using the EKF algorithm. The fourth part presents the conclusion of the article.
algorithm. The fourth part presents the conclusion of the article.
110 2.2. Establishment
Establishment of
of Fractional-Order
Fractional-OrderModel
Modeland
andParameter
ParameterIdentification
Identification
111 Theestablishment
The establishmentofofanan effective
effective battery
battery model
model is theispremise
the premise and foundation
and foundation for studying
for studying battery
112 battery SOC. Lower model accuracy will directly reduce the precision of the
SOC. Lower model accuracy will directly reduce the precision of the SOC estimation algorithm, SOC estimation
and
113 algorithm, and even directly lead to the divergence of the estimation algorithm in serious
even directly lead to the divergence of the estimation algorithm in serious cases. From the perspective cases.
114 From
of EIS, the perspective
a circuit composed of of
EIS, a circuit
fractional composed can
components of better
fractional components
fit the impedancecan better fit the
characteristics of
115 impedance characteristics of the battery, so that they can be applied in battery principle
the battery, so that they can be applied in battery principle analysis, model establishment, and analysis,
state
116 model establishment,
estimation [31]. and state estimation [31].

117 2.1. Establishment

2.1. Establishment of
of Fractional-Order
Fractional-OrderModel
Model
118 According to
According to the
the detailed
detailed analysis
analysis of
of EIS
EIS inin [32],
[32], this
this study
study established
established the
the fractional-order
fractional-order
119 equivalent circuit model of a lithium-ion battery, as shown in Figure 1. R
equivalent circuit model of a lithium-ion battery, as shown in Figure 1. R0 represents the Ohmic
0 represents theinternal
Ohmic
120 internal resistance
resistance of the
of the battery, R1 battery,
and R2 areR1 the
andpure
R2 are the pure
resistive resistiveCPE
impedance, impedance,
stands forCPE stands for
the constant the
phase
121 constant phase element, and W represents the Warburg element. The terminal
element, and W represents the Warburg element. The terminal voltage and OCV of the battery are voltage and OCV of
122 the battery are
represented by Urepresented
d and Uocv , by Ud and Uocvand
respectively, , respectively, and value
I is the current I is theofcurrent value of the circuit.
the circuit.

R1 R2
R0

W
CPE1 CPE2
+ Ud
- Uocv

123
Figure 1. Schematic diagram of fractional-order equivalent circuit model.
124 Figure 1. Schematic diagram of fractional-order equivalent circuit model.

125 In this circuit, the impedance of the fractional-order element can be described by the following
126 equation:
Appl. Sci. 2020, 10, 6371 4 of 18

In this circuit, the impedance of the fractional-order element can be described by the
following equation:
ZCPE1 = C 1sα



 1
ZCPE2 = C 1sβ

(1)


 2

 Z = 1γ

W Ws

where, C1 , C2 , and W indicate the parameters of fractional-order elements; α and β represent the
fractional order of CPE1 and CPE2 components. γ is the fractional order of the Warburg components.
The fractional differential equation of the parallel circuit of CPE and resistance can be expressed as:

1 1
Dα UCPE1 (t) = − UCPE1 (t) + I (t) (2)
R1 C1 C1

1 1
Dβ UCPE2 (t) = − UCPE2 (t) + I (t) (3)
R2 C2 C2
The fractional differential equation of the Warburg element is shown in Equation (4):

1
Dγ UW (t) = − I (t) (4)
W
According to the definition, the system state of the SOC as a fractional-order model is presented
in Equation (5) as follows:
η
D1 SOC(t) = − I (t) (5)
Cn
where Cn is the battery rated capacity, η is the coulombic efficiency, and η is desirable 1 for
lithium-ion batteries.
According to Kirchhoff’s law of current and voltage, we obtain

Ud = Uocv − I (t)R0 − UCPE1 − UCPE2 − UW (6)

where Uocv is the OCV of the battery, which is a monotonic function of the SOC.
In summary, the fractional-order equivalent circuit model of a lithium-ion battery can be established
by Equations (1)–(6). A complex lithium-ion battery system can be described by the simple structure
and limited parameters of the model. The parameters to be identified in the fractional model are shown
in Equation (7), below:
θ = [R0 R1 C1 R2 C2 W α β γ] (7)

2.2. Lithium-Ion Battery Test Experiment

Battery test experiments are the premise of battery model establishment and state estimation, and
are an indispensable step in battery research. In this paper, the battery test system is shown in Figure 2.
The test object is a universal A123, ternary lithium-ion soft packet battery cell. The main parameters
are listed in Table 1.
 Appl. Sci. 2020, 10, 6371 5 of 18
Appl. Sci. 2020, 10, x FOR PEER REVIEW 5 of 18
Appl. Sci. 2020, 10, x FOR PEER REVIEW 5 of 18

Computer Battery
Computer Battery

Ethernet
Ethernet

Test
Test
cable
BTS-5V100A Incubator
BTS-5V100A cable Incubator
148
148
149 Figure 2. Software
Figure 2. Software and
and hardware
hardware system
system for
for battery
battery test
test experiment.
experiment.
149 Figure 2. Software and hardware system for battery test experiment.
Table 1. Main performance parameters of universal A123 battery.
150 Table 1. Main performance parameters of universal A123 battery.
150 Table 1. Main performance parameters of universal A123 battery.
Nominal Nominal Charging Cut-Off Discharge Cut-Off Charging Cut-Off
Nominal
Capacity (Ah) Nominal
Voltage (V) Charging Cut-Off
Voltage (V) Discharge Cut-Off Charging
Voltage Cut-Off
(V) Cut-Off
Current (A)
Nominal Nominal Charging Cut-Off Discharge Charging Cut-Off
Capacity（Ah） Voltage（V） Voltage（V） Voltage（V） Current（A）
28
Capacity（Ah） 3.7
Voltage（V） 4.2
Voltage（V） 2.5
Voltage（V） Current（A）
1.25
28 3.7 4.2 2.5 1.25
28 3.7 4.2 2.5 1.25
151 In
In order
order to
to identify
identify the
the model
model parameters,
parameters, aa series
series of
of experiments
experiments on
on lithium-ion
lithium-ion batteries
batteries were
were
151 In order
conducted to identify the model parameters, a series of experiments on lithium-ion batteries were
152 conducted with reference to relevant national standards and the USABC Electric Vehicle Battery Test
with reference to relevant national standards and the USABC Electric Vehicle Battery Test
152 conducted
Procedures with reference to relevant national standards and the USABC Electric VehicleFigure
Battery Test
153 Procedures Manual.
Manual. TheThe battery
battery test
test flow
flow chart
chart of
of the
the relevant
relevant experiments
experimentsisisshown
showninin Figure3.3.
153 Procedures Manual. The battery test flow chart of the relevant experiments is shown in Figure 3.

Start
Start
Peak power
① Constant current and constant voltage Peaktest
power
① Constant
chargingcurrent andcut-off
to charge constant voltage
voltage. test
Static capacity
②charging
Constanttocurrent
chargedischarging
cut-off voltage.
to discharge Static test
capacity
② Constant current discharging to discharge
cut-off voltage. test Refer to the USABC
cut-off voltage. DST Refertest
to the USABC
manual
① Constant current discharging to discharge DST test manual
① Constant current Battery
cut-off voltage. discharging to discharge
shelved for 1 h. Double pluse
cut-off voltage. Battery shelved for 1 h. Double
testpluse
② Charging with duration of 10% of battery test
② Charging
capacity. with duration
Battery shelved forof 10%
2 min.of battery FUDS
capacity. Battery shelved for 2 min. FUDS
③ Repeat step ② until the voltage reaches
③ Repeat step cut-off
the charge ② untilvoltage.
the voltage reaches
Battery shelved HPPC
the
for charge
12 h. cut-off voltage. Battery shelved HPPC
for 12 h.
④ Discharging with duration of 10% of End
④ Discharging
battery with
capacity. duration
Battery of 10%
shelved for 2ofmin. End
battery capacity. Battery shelved for 2 min.
⑤ Repeat step ④ until the voltage reaches
⑤ Repeat step ④cut-off
the discharge until the voltage reaches
voltage.
154 the discharge cut-off voltage.
154
155 Figure 3. Battery
Figure 3. Battery test
test flow
flow chart.
chart.
155 Figure 3. Battery test flow chart.

156 According to the experimental data obtained by the double-pulse test, the relationship curve
156 According to the experimental data obtained by the double-pulse test, the relationship curve
157 between OCV and SOC can be obtained by [33] using the empirical Equation (8), as shown in Figure
157 between OCV and SOC can be obtained by [33] using the empirical Equation (8), as shown in Figure
158 4.
158 4.
 Appl. Sci. 2020, 10, 6371 6 of 18

Appl. Sci. 2020, 10, x FOR PEER REVIEW 6 of 18

According to the experimental data obtained by the double-pulse test, the relationship curve
between OCV and SOC can be obtained by [33] using the empirical Equation (8), as shown in Figure 4.
1
U OCV ( SOC ) = C0 + C1SOC + C2 1 + C3 ln( SOC ) + C4 ln(1 − SOC ) (8)
UOCV (SOC) = C0 + C1 SOC + C2 SOC + C3 ln(SOC) + C4 ln(1 − SOC) (8)
SOC
159
4.2

4.1
Experimental data
4
Fitted curve
3.9
OCV (V)

3.8

3.7

3.6

3.5

3.4
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
160 SOC

161 Figure 4. Open

Open circuit
circuit voltage (OCV)–state of charge (SOC) fitting curve.

162 2.3. Model Parameter

2.3. Model Parameter Identification
Identification Based
Based on
on HPSO
HPSO
163 The
The parameter identification link
parameter identification link isis very
very important
important before
before the
the model
model isis successfully
successfully applied.
applied.
164 The parameter identification results directly affect the accuracy of the model. Because of
The parameter identification results directly affect the accuracy of the model. Because of the obvious the obvious
165 time-varying non-linear and
time-varying non-linear and complex
complex structure
structure ofof the
the lithium-ion battery system,
lithium-ion battery system, there
there are
are many
many
166 parameters involved in operation, and most parameter values cannot be directly
parameters involved in operation, and most parameter values cannot be directly measured. measured. Therefore,
167 parameter
Therefore, identification is a difficult problem
parameter identification in theproblem
is a difficult battery modeling process.modeling
in the battery For high-dimensional
process. For
168 complex
high-dimensional complex optimization problems, this paper proposes an HPSOonalgorithm
optimization problems, this paper proposes an HPSO algorithm based a geneticbased
cross
169 factor. The flowchart of the algorithm is shown in Figure 5.
on a genetic cross factor. The flowchart of the algorithm is shown in Figure 5.
Referring to the genetic algorithm (GA), the genetic crossover factor is introduced into the classic
particle swarm optimization (PSO). After sorting the fitness of the particles from high to low, it is used
to select the particles with the fitness of the top half in each generation. As part of the next generation
of updated particle swarms, the particles in the second half of fitness are paired as crossover factors.
The position of the exchange between the particles is called the intersection, which is randomly set.
At the intersection point, the particles pair and exchange with each other to produce two completely
new offspring. By comparing the fitness of the offspring particles with that of their parent particles
and sorting them, the top half of the particles with high fitness values are selected to enter the next
generation of updated particle swarm so as to realize the update of the particle swarm. The introduction
of genetic crossover factors enriches the diversity of population particles, solves the problem that
the classical PSO algorithm is prone to fall into the local optimality, improves the global searching
capability of the parameter identification algorithm, and further improves the fitting degree of the
battery model.
 Appl. Sci. 2020, 10, 6371 7 of 18
Appl. Sci. 2020, 10, x FOR PEER REVIEW 7 of 18

Calculate particle fitness

Sort the fitness of the

particles
Y N

Half-fit particles The latter half of the less fit

particles

Randomly set intersection

points, cross pairing
Go directly to the updated

Calculate the
fitness of the
particle swarm

New progeny New progeny

particles particles parent particles

Compare
with the
parent

Select the first half of

particles with good fitness
Regroup for
updates

The updated new

170 particle swarm

171 Figure 5.
Figure 5. Flow
Flow chart
chart of
of hybrid
hybrid particle
particle swarm
swarm filtering algorithm.
filtering algorithm.

172 2.4. Fractional-Order

Referring to theModel Accuracy
genetic Analysis
algorithm (GA), the genetic crossover factor is introduced into the
173 classic particle swarm optimization (PSO).
In this selection, the dynamic stress test After
(DST)sorting the fitness
condition of the
test data particles
of the fromA123
universal high battery
to low,
174 it is
at 25used to selecttemperature
◦ C ambient the particles iswith theasfitness
used trainingof the
datatop
to half in each
identify the generation.
parametersAs of part of the next
the established
175 generation of updated particle swarms, the particles in the second half of fitness
fractional-order model. Through different identification methods and different simulation conditions, are paired as
176 crossover factors. The position of the exchange between
the accuracy and robustness of the fractional-order model are analyzed.the particles is called the intersection, which
177 is randomly set. At the results
The identification intersection point,
obtained the particles
using pairPSO
the classical andalgorithm
exchange arewithshown
each other to produce
in Table 2.
178 two completely new offspring. By comparing the fitness of the offspring particles with that of their
179 parent particles Table
and sorting them,recognition
2. Parameter the top half of the
results of particles with high
ordinary particle swarmfitness values are selected to
algorithm.
180 enter the next generation of updated particle swarm so as to realize the update of the particle swarm.
R0 R1 C1 R2 C2 W A β γ
181 The introduction of genetic crossover factors enriches the diversity of population particles, solves the
182 0.003 0.0001 960 7.12 1112 500 0.61
problem that the classical PSO algorithm is prone to fall into the local optimality, improves the 0.13 0.64
183 global searching capability of the parameter identification algorithm, and further improves the
184 fitting
Thedegree of the battery
identification model.
results obtained using the HPSO algorithm are shown in Table 3.

185 2.4. Fractional-Order Model

Table Accuracyrecognition
3. Parameter Analysis results of hybrid particle swarm algorithm.
186 InR0this selection,
R1 the dynamic
C1 stress
R2test (DST)Ccondition
2
test data ofαthe universal
W β A123γbattery
187 at 25 °C ambient temperature is used as training data to identify the parameters of the established
0.001 0.05 5431 5.31 4758 2122 0.59 0.22 0.12
188 fractional-order model. Through different identification methods and different simulation
189 conditions, the accuracy and robustness of the fractional-order model are analyzed.
190 The identification results obtained using the classical PSO algorithm are shown in Table 2.
192 The identification results obtained using the HPSO algorithm are shown in Table 3.

193 Table 3. Parameter recognition results of hybrid particle swarm algorithm.

Appl. Sci. 2020, 10, 6371 R0 R1 C1 R2 C2 W α β γ 8 of 18

0.001 0.05 5431 5.31 4758 2122 0.59 0.22 0.12

194 According to
According tothe
theidentification results
identification of the
results ofabove two algorithms,
the above the model
two algorithms, theaccuracy
model is simulated
accuracy is
195 simulated and verified under the three working conditions of the dynamic stress test (DST), driving
and verified under the three working conditions of the dynamic stress test (DST), federal urban federal
196 schedule
urban (FUDS),
driving and hybrid
schedule pulseand
(FUDS), power characteristic
hybrid pulse power(HPPC). The simulation
characteristic (HPPC).results
Theare shown in
simulation
197 Figures 6–8. It can be seen from Figures 6–8 that, under different operating conditions,
results are shown in Figures 6–8. It can be seen from Figures 6–8 that, under different operating the voltage error
198 value output by the fractional-order model fluctuates slightly around zero. When
conditions, the voltage error value output by the fractional-order model fluctuates slightly aroundthe input current
199 has aWhen
zero. large mutation, the voltage
the input current has aerror
largevalue is large,
mutation, the but can also
voltage errorbevalue
controlled atbut
is large, approximately
can also be
200 0.05 V. Therefore, the improved fractional-order model is more accurate and
controlled at approximately 0.05 V. Therefore, the improved fractional-order model is more suitable for different
accurate
201 working
and conditions.
suitable for different working conditions.

4.5
Experimental data
PSO
Voltage (v)

4
HPSO

3.5 3.76

3.72
3
0 500 1000 1500 2000 2500 3000
3.68
1360 1370 1380 1390 1400 Time (s)
0.1
PSO
0.05 HPSO
Error (v)

-0.05

-0.1
0 500 1000 1500 2000 2500 3000
202 Time (s)

203 Figure
Appl. Sci. 2020, 10, x 6.
Figure 6. Model
FOR PEERvoltage
Model REVIEWand
voltage and simulation
simulation error
error under
under dynamic
dynamic stress
stress test
test (DST)
(DST) conditions. 9 of 18

204 4.2
Experimental data
4
Voltage (v)

PSO

3.8 HPSO

3.6
3.76
3.4 3.72

3.68
0 500 1000 1500 2000 2500 3000
3.64
1365 1370 1375 1380
Time (s)

0.1
PSO
0.05
Error (v)

HPSO

-0.05

-0.1
0 500 1000 1500 2000 2500 3000
205 Time (s)

206 Figure 7.7. Model

Modelvoltage andand
voltage simulation errorerror
simulation underunder
federalfederal
urban driving
urban schedule (FUDS) condition.
driving schedule (FUDS)
207 condition.

4.2
Experimental data
4 PSO
Voltage (v)

3.8 HPSO

3.6 3.64

3.4
3.60
3.2
0 5000 10000 15000 20000 25000
3.56
20500 20600 20700 20800 20900 21000 Time (s)
0.05
PSO
0.025 HPSO
ror (v)

0
 -0.05

-0.1
0 500 1000 1500 2000 2500 3000
205 Time (s)

206 Appl. Figure

Sci. 2020,7.
10,Model
6371 voltage and simulation error under federal urban driving schedule (FUDS)9 of 18
207 condition.

4.2
Experimental data
4 PSO

Voltage (v)
3.8 HPSO

3.6 3.64

3.4
3.60
3.2
0 5000 10000 15000 20000 25000
3.56
20500 20600 20700 20800 20900 21000 Time (s)
0.05
PSO
0.025 HPSO
Error (v)

-0.025

-0.05
0 5000 10000 15000 20000 25000
208 Time (s)

209 Figure
Figure 8.8. Model voltage
Model and and
voltage simulation error error
simulation under under
hybrid hybrid
pulse power
pulse characteristic (HPPC)
power characteristic
210 conditions.
(HPPC) conditions.

211 Table 44shows

Table shows thethe
rootroot
meanmean
squaresquare
error (RMSE)
error value
(RMSE)of the model
value of output voltage
the model corresponding
output voltage
212 to the classical to
corresponding PSO thealgorithm
classical and
PSOthe HPSO algorithm
algorithm under algorithm
and the HPSO different working conditions.
under different It can
working
213 be seen from
conditions. It Table
can be4seen
that from
the parameter identification
Table 4 that method
the parameter based on method
identification the HPSO algorithm
based on the further
HPSO
214 improved further
algorithm the accuracy of thethe
improved fractional
accuracymodel.
of the fractional model.

215 Table 4.
Table Rootmean
4. Root meansquare
square error
error under
under different
different working conditions.

RMSE RMSEHPPC HPPC DST DST FUDS FUDS

PSO PSO0.00610.0061 0.00500.0050
0.0057 0.0057
HPSO HPSO0.00360.0036 0.00220.0022
0.0035 0.0035

216 3.
3. SOC
SOCEstimation
EstimationBased
Basedon
on the
the HIF
HIF Algorithm
Algorithm
217 The
The Kalman
Kalman series
series of
of filtering
filtering algorithms
algorithms have
have been
been supported,
supported, widely
widely researched
researched and
and applied
applied
218 owing to their
owing to their high
highaccuracy,
accuracy,lowlow computational
computational complexity,
complexity, andand good
good robustness.
robustness. However,
However, they they
have
219 have two limitations that are difficult to overcome: (1) It is necessary to ensure that the input noise
two limitations that are difficult to overcome: (1) It is necessary to ensure that the input noise is white is
220 white noise, and the statistical characteristics of the noise are approximately known. In practice,
noise, and the statistical characteristics of the noise are approximately known. In practice, the noise the
is often not a white noise that satisfies the Gaussian distribution, and the statistical characteristics
are difficult to obtain. (2) The accuracy of the estimation is very dependent on the accuracy of the
established model. When the accuracy of the model is insufficient or the accuracy of the model
gradually decreases with the dynamic changes of the working process, the filtering error becomes
increasingly larger.
To overcome the Kalman series filtering algorithm’s requirements for white noise and its
dependence on accurate modeling, an HIF with poor model accuracy, uncertain SOC initial value,
and stronger robustness of various colored noise interference is introduced.

3.1. State Space Equation of the Lithium-Ion Battery

The establishment of system state space equations is the basis for controlling the parameters
of dynamic systems in modern control theory. Establish a mathematical model that reflects the
internal structure of the system and characterizes the system through the state equation and the output
equation. Owing to the time-varying non-linearity of the battery system, its state space equation can
be expressed as:
xk+1 = Ak xk + Bk uk + ωk (9)
Appl. Sci. 2020, 10, 6371 10 of 18

yk = Ck xk + Dk uk + υk (10)

This is the first step in the derivation of the state space equation of a lithium-ion battery to
determine the relevant variable parameters of the battery system. According to the established battery
model, the state variable of the system is set as xk = [VCPE1 ,k VCPE2 ,k VW,k SOCk ]. The input of the
system is u = Ik , which is the current state of the battery. The output is y = Ud,k , which is the terminal
voltage of the battery.
After determining the parameters, according to the fractional-order model established in Section 2.1,
the state space equations of Equations (8) and (9) are derived, where Ak , Bk , Ck , and Dk are defined
as follows:  Tα 
 α − Tα R11C1
 
0 0 0   C1 
β 1  Tβ 
 
β
 
 0 − T 0 0

Ak =  R C C
 , Bk =  T2γ 
 2 2

 
 
 0 0 γ 0  −
 W 
 (11)

0 0 0 1
  ηT 
− Cn
d f (SOCk )
Ck = [−1 − 1 − 1 dSOC ] , Dk = −R0
The implementation of SOC estimation is to design the corresponding filter algorithm based
on the above-mentioned state space equations to eliminate the effects of noise ωk and υk as much
as possible. With improvements in the accuracy of the system state estimation, the SOC estimation
achieves the desired effect.

3.2. SOC Estimation Based on HIF Algorithm

To judge the accuracy of the estimated object xk , the HIF algorithm defines a cost function J:

N−1
kxk − x̂k k2S
P
k
k =0
J= (12)
N−1
kx0 − x̂0 k2 −1 (kωk k2Q−1 + kυk k2R−1 )
P
P0
+
k =0 k k

The matrices P0 , Qk , Rk , and Sk in the equation are all symmetric positive definite matrices,
which are set by the designer for specific problems. Designers use these matrices to reflect the degree
to which different types of noise affect the system. In general, we expect to perform similar processing
on the weights of different types of noises to avoid the impact on the accuracy of estimation when a
certain type of noise is ignored.
It can be seen from Equation (11) that the cost function of HIF is a relative proportional value,
and its numerator and denominator are related to the estimated value error and the overall noise of the
system. The significance of establishing the cost function is to reflect the proportional relation between
the estimation error and the noise interference. In order to meet the design requirements, the cost
function has an upper bound, which satisfies the following condition:

J < 1/θ

To solve the above optimization process, the algorithm designer must find a suitable xk method to
minimize the denominator of the cost function J. However, the reality is contrary to the designer’s
goal; it hopes to produce a specific special initialization state x0 , noise ωk , and υk , so as to maximize the
cost function J.
Therefore, this problem becomes a maximum-minimum problem, that is, when ωk , υk , and x0
make the cost function J maximum, the appropriate xk should be selected to minimize the cost function
Appl. Sci. 2020, 10, 6371 11 of 18

J. By re-integrating Equations (11) and (12) and combining it with the output Equation (9) of the state
space, Equation (13) can be obtained:

N−1
1 X 1

− kx0 − x̂0 k2 −1 + kxk − x̂k k2S − (kωk k2 −1 + kyk − Ck xk − Dk uk k2 −1 ) < 0 (13)
θ P0 k θ Qk Rk
k =0

By solving the maximum-minimum problem of the cost function, a recursive relation satisfying
Equation (12) can be obtained, as shown below:
 −1


 Kk = Ak Pk (I − θSk Pk + CTk R−1k
Ck Pk ) CTk R−1 k

x̂k+1 = Ak x̂k + Bk uk + Kk ( yk − Ck x̂k − Dk uk ) (14)



 −1 T
k+1 = Ak Pk (I − θSk + Ck Rk Ck Pk ) Ak + Qk
 P
 T −1

Similar to the KF algorithm, in order to facilitate calculation and application of the SOC estimation
method based on the HIF, the recursive relation (Equation (14)) is divided into two stages: prediction
and update. The specific calculation process is shown in Table 5.

Table 5. Battery SOC estimation algorithm based on H∞ filter algorithm.

The Specific Calculation Process Is as Follows:

xk = f (xk−1 , uk−1 ) + ωk−1
(
Building a nonlinear system:
yk = g(xk , uk ) + υk
Initialization: x̂0 = E(x0 ), P0 = E (x0 − x̂0 )(x0 − x̂0 )T
h i
+

When k ∈ {1, 2, . . . , ∞ }, calculation

First step: prediction stage
+
System status estimation: x̂− k
= f (x̂k−1 , uk−1 )
− +
Error covariance prediction: Pk = Ak−1 Pk−1 ATk−1 + Qk−1
Second step: update stage
Innovation matrix: ek = yk − g(x̂− k
, uk )
−1
Gain matrix: Kk = Ak P−k
(I − θSk P−k + CTk R−1
k
Ck P−
k
) CTk R−1
k
+ −
System state correction: x̂k = x̂k−1 Kk ek
−1
Error covariance correction: Pk+ = Ak P−
k
(I − θSk + CTk R−1
k
Ck P−
k
) ATk + Qk

3.3. SOC Estimation Accuracy Verification

After describing in detail the principle of the HIF and the specific calculation process, in order to
verify the robustness of the algorithm under the three conditions of uncertain measurement accuracy,
initial SOC value uncertainty, and uncertain application conditions, this section uses a file in MATLAB
software to write the HIF and EKF running procedures, and carries out simulation operations in
Simulink. Based on the obtained experimental data, accuracy verification and comparative analysis
are conducted on the estimation results of the SOC of the lithium-ion battery using HIF and EKF.
In the current literature, the reference values used for battery SOC estimation algorithm verification
are all obtained by the ampere-hour integration method [34]. In order to eliminate the measurement
error of the current sensor as far as possible, a high-precision battery testing equipment with a stability
of 0.1% of FS is adopted, and the battery is fully placed before the test. Therefore, this study uses
the approximate SOC calculation value obtained by the ampere-hour integration method as the real
SOC value.

3.3.1. Uncertainty of Measurement Accuracy

Owing to the limitation of sensor acquisition accuracy, there are some acquisition errors in the
current and voltage data acquired during vehicle driving. At present, the acquisition error of the
mainstream current sensor is within 1%, and the voltage acquisition accuracy is within ±5 mV. In order
 Appl. Sci. 2020, 10, 6371 12 of 18

to fully ensure the reliability of the verification, this study evaluates the estimation effect of the HIF
in the event of inaccurate measurement by adding colored noise interference to the current and
voltage data.
Figure 9 shows that when colored noise is added to the current and voltage data, the SOC
estimation results obtained by HIF and EKF have different degrees of deviation. Since the robustness
of HIF algorithm is better than EKF, the SOC estimation results based on HIF are slightly better than
that
Appl. Sci.based onxEKF.
2020, 10, The estimation
FOR PEER REVIEW error of the former is within ±0.02, while that of the latter is13within
of 18
±0.04.

1
0.9 Measured data
0.8 EKF
0.7 H∞
0.8
SOC

0.6
0.5 0.75
0.4
0.3 0.7
0 1500 2000 2500 3000
0.65 Time (s)
1000 1100 1200 1300 1400 1500

0.04
EKF
0.02 H∞
Error

-0.02

-0.04
0 500 1000 1500 2000 2500 3000
299 Time (s)

300 Figure
Figure9. 9.
Comparison
Comparison of of
SOCSOCestimation results
estimation and
results errors
and based
errors onon
based DST conditions
DST (colored
conditions noise
(colored noise
301 added
added to to current
current andand voltage).
voltage).

302 3.3.2.
3.3.2. Uncertainty
Uncertainty ofofSOCSOC Initial
Initial Value
Value
303 InInactual
actualuse,
use,itit is
is difficult
difficulttotoobtain
obtainananaccurate initial
accurate SOC
initial value.
SOC Therefore,
value. the algorithm
Therefore, should
the algorithm
304 have the ability to track the system state when the initial SOC value is inaccurate,
should have the ability to track the system state when the initial SOC value is inaccurate, and correctand correct the initial
305 SOC
the value.
initial SOCTovalue.
evaluate the convergence
To evaluate ability ofability
the convergence the HIFof under
the HIF uncertain initial values,
under uncertain initialthe initial
values,
306 value of the SOC is set to 0.7 (the exact initial value of the SOC is 1), and the results
the initial value of the SOC is set to 0.7 (the exact initial value of the SOC is 1), and the results of itsof its convergence
307 curve and error
convergence curveare
and shown
error inareFigure
shown 10.in Figure 10.
Figure 10 shows that in the case of the inaccurate initial SOC value, the SOC values estimated by
the two algorithms
1 can quickly converge to the accurate value after algorithm calculation. However,
0.9
for the convergence rate, HIF needs approximately 1.5 s to make the SOCMeasured convergedata to the initial value,
0.8
which is 50% higher than the 3 s of the EKF algorithm. Moreover, the estimated EKF
error of HIF fluctuates
0.7 H∞
within ±0.020.6 after1convergence, which proves that the algorithm is not sensitive to the setting of the
SOC

initial value 0.5

of SOC,
0.9
can calculate the true value recursively, and has relatively better robustness.
0.4
0.8
0.3
0 0.7
1500 2000 2500 3000
0 10 20
Time (s)

0.3

0.2 EKF
H∞
Error

0.1

-0.1
0 500 1000 1500 2000 2500 3000
308 Time (s)

309 Figure 10. Comparison of SOC estimation results and errors under inaccurate SOC initial value
310 based on DST conditions.

311 Figure 10 shows that in the case of the inaccurate initial SOC value, the SOC values estimated
 302 3.3.2. Uncertainty of SOC Initial Value
303 In actual use, it is difficult to obtain an accurate initial SOC value. Therefore, the algorithm
304 should have the ability to track the system state when the initial SOC value is inaccurate, and correct
305 the initial SOC value. To evaluate the convergence ability of the HIF under uncertain initial values,
306 the initial
Appl. value
Sci. 2020, of the SOC is set to 0.7 (the exact initial value of the SOC is 1), and the results13ofofits
10, 6371 18
307 convergence curve and error are shown in Figure 10.

1
0.9 Measured data
0.8 EKF
SOC 0.7 H∞
1
0.6
0.5 0.9
0.4
0.8
0.3
0 0.7
1500 2000 2500 3000
0 10 20
Time (s)

0.3

0.2 EKF
H∞
Error

0.1

0
Appl. Sci. 2020, 10,-0.1
x FOR PEER REVIEW 14 of 18
0 500
1500 1000
2000 2500 3000
316
308 (s)
Time value
setting of the initial value of SOC, can calculate the true recursively, and has relatively better
317
309 robustness. 10. Comparison
Figure 10. ComparisonofofSOC
SOC estimation
estimation results
results andand errors
errors under
under inaccurate
inaccurate SOC initial
SOC initial value
value based
310 based
on DSTonconditions.
DST conditions.
318 3.3.3. Uncertainty of Application Conditions
311 3.3.3.Figure
Uncertainty of Application Conditions
319 10 shows
The applicability that
of thein HIF
the case of the inaccurate
algorithm initialoperating
under different SOC value, the SOC values
conditions estimated
was evaluated
312
320 by the
using Thetwo
DST, algorithms
applicability
FUDS, and HPPC canHIF
of the quickly
conditions. converge
algorithm under
It can betoseen
thefrom
accurate
different value
operating
Figures after12algorithm
conditions
11 and wasunder
that calculation.
evaluated
DST using
and
313
321 However,
DST, FUDS, for the
and convergence
HPPC rate,
conditions. HIF
It canneeds
be approximately
seen from Figures 1.5
11 s to
and make
12 the
that SOC
under
FUDS conditions, the current and voltage fluctuations are relatively severe. The noise interference of converge
DST and to
FUDS the
314
322 initial value,
conditions,
the which is
the current
battery model 50% higher
and is
in this case than
voltage
obvious, the 3 s
fluctuationsof the EKF algorithm.
are relatively
and it no longer exhibitssevere. Moreover, the estimated
The noise interference
the characteristics error
of zero mean. of Inof
the
315
323 HIF fluctuates
battery
addition, model within
in this
the model ±0.02
case after
parameters convergence,
is change
obvious, to and it which
a large no proves
longer
extent that
withexhibits
the the algorithm
the
current is Figures
not sensitive
characteristics
mutation. of zero
13 andto14the
mean.
324 In addition,
show that thetheEKF
model parameters
algorithm is change
used totoestimate
a large extent with theSOC
the battery current mutation.
under Figures
the DST and13FUDS
and 14
325 show that The
conditions. the EKF algorithm
simulated is used
output to estimate
voltage the battery
of the battery modelSOC under greatly
deviates the DSTfrom
and the
FUDSrealconditions.
voltage
326 The simulated
value, outputthe
so that when voltage of the
current battery
pulse model the
is large, deviates
SOC greatly from accuracy
estimation the real voltage
is notvalue,
high,soand
that
327 when thewithin
fluctuates pulse is large, the SOC estimation accuracy is not high, and fluctuates within ±0.04.
current±0.04.

60 4.2
Voltage
Current

0 3.9
Voltage (V)
Current (A)

-60 3.6

-120 3.3
0 500 1000 1500 2000 2500 3000
328 Time (s)

329 Figure11.
Figure Current
11.Current and
and voltage
voltage diagram
diagram underDST
under DSTconditions.
conditions.

150 5
Voltage
Current
75 4.5
Voltage (V)
Current (A)

0 4

-75 3.5

-150 3
0 500 1000 1500 2000 2500 3000
330 Time (s)

331 Figure 12. Current and voltage diagram under FUDS condition.

1
0.9 Measured data
0.8 EKF
H∞
 -60-60 3.63.6

V
-120
-120 3.33.3
0 0 500
500 1000
1000 1500
1500 2000
2000 2500
2500 3000
3000
328
328 Time(s)(s)
Time
Appl. Sci. 2020, 10, 6371 14 of 18
329
329 Figure
Figure 11.11. Current
Current and
and voltage
voltage diagram
diagram under
under DST
DST conditions.
conditions.

150
150 55
Voltage
Voltage
Current
Current
7575 4.54.5

Voltage (V)
Current (A)

Voltage (V)
Current (A)

00 44

-75-75 3.53.5

-150
-150 33
00 500
500 1000
1000 1500
1500 2000
2000 2500
2500 3000
3000
330
330 Time(s)(s)
Time

331
331 Figure
Figure 12.12.
Figure 12. Current
Current
Current and
and voltage
voltage
and diagram
diagram
voltage under
under
diagram FUDS
FUDS
under condition.
condition.
FUDS condition.

1 1
0.90.9 Measured
Measured data
data
EKF
0.80.8 EKF
H∞H∞
0.70.7
SOC

0.60.6 0.8 0.8

SOC

0.50.5
0.40.4
0.30.3
0 0 0.7 500500 1000
1000 1500
1500 2000
2000 2500
2500 3000
3000
0.7
10001100
1000 1100 1200
1200 1300
1300 14001500
1400 1500 Time(s)(s)
Time
0.04
0.04
EKF
EKF
0.02
0.02 H∞
H∞
Error
Error

0 0
-0.02
-0.02
-0.04
-0.04
0 0 PEER REVIEW
Appl. Sci. 2020, 10, x FOR 500
500 1000
1000 1500
1500 2000
2000 2500
2500 3000
3000 15 of 18
332
332 Time(s)(s)
Time
333 Figure 13.
Figure 13. Comparison of SOC
Comparison of SOC estimation
estimation results
results and
and errors
errors under
under DST
DST conditions.
conditions.

1
0.9 Measured data
0.8 EKF
0.7 0.8
H∞
SOC

0.6
0.5
0.4
0.3 500 1000
0 0.7 1500 2000 2500 3000
1100 1200 1300 1400 1500
Time (s)
0.04
EKF
0.02 H∞
Error

-0.02

-0.04
0 500 1000 1500 2000 2500 3000
334 Time (s)

335 Figure 14. Comparison of SOC estimation results and errors under
under FUDS
FUDS conditions.
conditions.

336 It can
can be
be seen
seen from
from Figures
Figures 15
15 and
and 16
16 that
that the
the HPPC
HPPC operating
operating conditions
conditions are
are constant
constant current
current
337 pulse discharges,
pulse discharges, and the current fluctuates periodically.
the current fluctuates periodically. The HPPC condition is relatively stable,
338 compared
compared with
with the
the above
above two
two conditions,
conditions, so
so the
the SOC
SOC estimation
estimation accuracy
accuracy is
is relatively
relatively high.
high. The SOC
339 estimation error based on the EKF can be controlled within ±0.02, and the SOC estimation
estimation error based on the EKF can be controlled within ±0.02, and the SOC estimation error based
error
340 based on the HIF can be controlled within ±0.01, which further verifies the applicability of the HIF
341 algorithm under different working conditions.

20 4.5
Voltage
10 Current 4
334 Time (s)

335 Figure 14. Comparison of SOC estimation results and errors under FUDS conditions.

336 It can be seen from Figures 15 and 16 that the HPPC operating conditions are constant current
337 pulse
Appl. discharges, and the current fluctuates periodically. The HPPC condition is relatively stable,
Sci. 2020, 10, 6371 15 of 18
338 compared with the above two conditions, so the SOC estimation accuracy is relatively high. The SOC
339 estimation error based on the EKF can be controlled within ±0.02, and the SOC estimation error
340 on the
based onHIFthe can
HIFbecancontrolled
be controlled ±0.01, ±0.01,
withinwithin which which
furtherfurther
verifiesverifies
the applicability of the HIF
the applicability algorithm
of the HIF
341 under different working conditions.
algorithm under different working conditions.

20 4.5
Voltage
10 Current 4

Voltage (V)
Current (A)

0 3.5

-10 3

-20 2.5

-30 2
0 5000 10000 15000 20000 25000 30000
342 Time (s)

343 Appl. Sci. 2020, 10, x FOR PEER REVIEW

Figure
Figure 15.15. Current–voltage
Current–voltage diagram
diagram under
under HPPC
HPPC conditions.
conditions. 16 of 18

1
0.8 Measured data
EKF
0.6 H∞
SOC

0.4
0.2
0 5000 10000 15000 20000 25000 30000
1550 1600 1650 1700 1750 1800 Time (s)
0.04
EKF
0.02 H∞
Error

-0.02

-0.04
0 5000 10000 15000 20000 25000 30000
344 Time (s)

345 Figure 16.

Figure Comparisonof
16. Comparison ofSOC
SOCestimation
estimation results
results and
and errors
errors under
under HPPC
HPPC conditions.
conditions.

346 4. Conclusions
4. Conclusions

347 Aiming at
Aiming at the
theSOC
SOCestimation
estimation problem
problem of electric vehicle
of electric power
vehicle lithium-ion
power batteries,
lithium-ion this paper
batteries, this
348 proposes an SOC estimation method based on the HIF algorithm. Firstly,
paper proposes an SOC estimation method based on the HIF algorithm. Firstly, based on the EIS based on the EIS of the
of
349 lithium-ion battery, a fractional-order model based on the dual polarization equivalent
the lithium-ion battery, a fractional-order model based on the dual polarization equivalent circuit circuit model
350 was established.
model WithoutWithout
was established. reducing the accuracy
reducing of the model,
the accuracy of theitmodel,
not only simplifies
it not the modelthe
only simplifies structure,
model
351 but also reduces the amount of calculation. Then, the parameters of the fractional-order
structure, but also reduces the amount of calculation. Then, the parameters of the fractional-order battery model
352 are identified
battery model by arethe HPSO algorithm
identified by the HPSO based on the genetic
algorithm based on crossover
the geneticfactor. It solves
crossover the It
factor. problem
solves
353 that the classic PSO algorithm easily falls into the local best, improves the global
the problem that the classic PSO algorithm easily falls into the local best, improves the global search search ability of
354 the parameter identification algorithm, and further improves the fitting degree
ability of the parameter identification algorithm, and further improves the fitting degree of the of the battery model.
355 Finally,model.
battery the accuracy
Finally,ofthetheaccuracy
SOC estimation
of the SOC results of lithium-ion
estimation results ofbatteries
lithium-ion using HIF and
batteries EKFHIF
using are
356 verified and compared under three conditions: uncertain measurement
and EKF are verified and compared under three conditions: uncertain measurement accuracy, accuracy, uncertain SOC initial
357 value, andSOC
uncertain uncertain application
initial value, conditions.
and uncertain The simulation
application results
conditions. Theshow that theresults
simulation SOC estimation
show that
358 method based on HIF can ensure that the SOC estimation error value
the SOC estimation method based on HIF can ensure that the SOC estimation error value fluctuates within ±0.02 in any
fluctuates
359 case, and is slightly affected by environmental factors and other factors. It provides
within ±0.02 in any case, and is slightly affected by environmental factors and other factors. It a way to improve
360 the accuracy
provides a wayof SOC estimation
to improve in a battery
the accuracy management
of SOC estimation system.
in a battery management system.
361 Author Contributions: Conceptualization: M.H.; methodology: M.H.; validation: Y.X.; formal analysis: L.L. and
362 Y.X.; investigation: L.L., G.J., and Z.L.; writing—original draft: L.L.; writing—review and editing: M.H. and
363 C.F.; supervision: M.H. and C.F.; project administration: G.J. and Z.L.; funding acquisition: M.H. and C.F. All
364 authors have read and agreed to the published version of the manuscript.

365 Funding: This work was supported by the National Key R&D Program of China [No. 2018YFB0106102], and the
366 Major Program of Chongqing Municipality [No. cstc2018jszx-cyztzxX0007].

367 Conflicts of Interest: The authors declare no conflicts of interest. The funders had no role in the design of the
Appl. Sci. 2020, 10, 6371 16 of 18

Author Contributions: Conceptualization: M.H.; methodology: M.H.; validation: Y.X.; formal analysis: L.L. and
Y.X.; investigation: L.L., G.J., and Z.L.; writing—original draft: L.L.; writing—review and editing: M.H. and C.F.;
supervision: M.H. and C.F.; project administration: G.J. and Z.L.; funding acquisition: M.H. and C.F. All authors
have read and agreed to the published version of the manuscript.
Funding: This work was supported by the National Key R&D Program of China [No. 2018YFB0106102], and the
Major Program of Chongqing Municipality [No. cstc2018jszx-cyztzxX0007].
Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design of the
study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to
publish the results.

Abbreviations
SOC state of charge
HIF H-infinity filter
HPSO hybrid particle swarm optimization
EKF extended Kalman filter
OCV open circuit voltage
EIS electrochemical impedance spectroscopy
SVM support vector machine
NN neural network
PF particle filter
KF Kalman filter
AUKF adaptive unscented Kalman filter
AHIF adaptive H-infinity filter
CPE constant phase element
GA genetic algorithm
DST dynamic stress test
FUDS federal urban driving schedule
HPPC hybrid pulse power characteristic
RMSE root mean square error

References
1. Xiong, R.; Cao, J.; Yu, Q.; He, H.; Sun, F. Critical review on the battery state of charge estimation methods for
electric vehicles. IEEE Access 2018, 6, 1832–1843. [CrossRef]
2. Chaoui, H.; Ibe-Ekeocha, C.C. State of charge and state of health estimation for lithium batteries using
recurrent neural networks. IEEE Trans. Veh. Technol. 2017, 66, 8773–8783. [CrossRef]
3. Lin, C.; Mu, H.; Xiong, R.; Shen, W. A novel multi-model probability battery state of charge estimation
approach for electric vehicles using h-infinity algorithm. Appl. Energy 2016, 166, 76–83. [CrossRef]
4. Zheng, Y.; Ouyang, M.; Han, X.; Lu, L.; Li, J. Investigating the error sources of the online state of charge
estimation methods for lithium-ion batteries in electric vehicles. J. Power Sources 2018, 377, 161–188. [CrossRef]
5. Einhorn, M.; Conte, F.V.; Kral, C.; Fleig, J. A method for online capacity estimation of lithium ion battery cells
using the state of charge and the transferred charge. IEEE Trans. Ind. Appl. 2011, 48, 736–741. [CrossRef]
6. Li, Z.; Huang, J.; Liaw, B.; Zhang, J. On state-of-charge determination for lithium-ion batteries. J. Power Sources
2017, 348, 281–301. [CrossRef]
7. Hannan, M.A.; Lipu, M.S.H.; Hussain, A.; Mohamed, A. A review of lithium-ion battery state of charge
estimation and management system in electric vehicle applications: Challenges and recommendations.
Renew. Sustain. Energy Rev. 2017, 78, 834–854. [CrossRef]
8. Salkind, A.J.; Fennie, C.; Singh, P.; Atwater, T.; Reisner, D.E. Determination of state-of-charge and
state-of-health of batteries by fuzzy logic methodology. J. Power Sources 1999, 80, 293–300. [CrossRef]
9. Jin, G.; Li, L.; Xu, Y.; Hu, M.; Fu, C.; Qin, D. Comparison of SOC estimation between the integer-order model
and fractional-order model under different operating conditions. Energies 2020, 13, 1785. [CrossRef]
Appl. Sci. 2020, 10, 6371 17 of 18

10. Gupta, D.; Richhariya, B.; Borah, P. A fuzzy twin support vector machine based on information entropy for
class imbalance learning. Neural Comput. Appl. 2018, 31, 7153–7164. [CrossRef]
11. Herzog, S.; Tetzlaff, C.; Wörgötter, F. Evolving artificial neural networks with feedback. Neural Networks
2020, 123, 153–162. [CrossRef] [PubMed]
12. Lv, M.; Baldi, S.; Liu, Z. The non-smoothness problem in disturbance observer design: A set-invariance-based
adaptive fuzzy control method. IEEE Trans. Fuzzy Syst. 2019, 27, 598–604. [CrossRef]
13. Yang, Q.; Cao, B.; Li, X. A simplified fractional order impedance model and parameter identification method
for lithium-ion batteries. PLoS ONE 2017, 12, e0172424. [CrossRef] [PubMed]
14. Ma, Y.; Chen, Y.; Zhou, X.; Chen, H. Remaining useful life prediction of lithium-ion battery based on
gauss–hermite particle filter. IEEE Trans. Control. Syst. Technol. 2019, 27, 1788–1795. [CrossRef]
15. Li, R.; Jan, N.M.; Huang, B.; Prasad, V. Constrained multimodal ensemble Kalman filter based on
Kullback–Leibler (KL) divergence. J. Process. Control. 2019, 79, 16–28. [CrossRef]
16. Waag, W.; Fleischer, C.; Sauer, D.U. Critical review of the methods for monitoring of lithium-ion batteries in
electric and hybrid vehicles. J. Power Sources 2014, 258, 321–339. [CrossRef]
17. Sun, F.; Xiong, R.; He, H. A systematic state-of-charge estimation framework for multi-cell battery pack in
electric vehicles using bias correction technique. Appl. Energy 2016, 162, 1399–1409. [CrossRef]
18. Wang, T.; Chen, S.; Ren, H.; Zhao, Y. Model-based unscented Kalman filter observer design for lithium-ion
battery state of charge estimation. Int. J. Energy Res. 2017, 42, 1603–1614. [CrossRef]
19. Cui, X.; He, Z.; Li, E.; Cheng, A.; Luo, M.; Guo, Y. State-of-charge estimation of power lithium-ion batteries
based on an embedded micro control unit using a square root cubature Kalman filter at various ambient
temperatures. Int. J. Energy Res. 2019, 43, 3561–3577. [CrossRef]
20. Xiong, B.; Zhao, J.; Wei, Z.; Skyllas-Kazacos, M. Extended Kalman filter method for state of charge estimation
of vanadium redox flow battery using thermal-dependent electrical model. J. Power Sources 2014, 262, 50–61.
[CrossRef]
21. Shen, P.; Ouyang, M.; Lu, L.; Li, J.; Feng, X. The co-estimation of state of charge, state of health, and state of
function for lithium-ion batteries in electric vehicles. IEEE Trans. Veh. Technol. 2018, 67, 92–103. [CrossRef]
22. Wang, L.; Lu, N.; Liu, Q.; Liu, L.; Zhao, X. State of charge estimation for lifepo4 battery via dual extended
kalman filter and charging voltage curve. Electrochim. Acta 2019, 296, 1009–1017. [CrossRef]
23. Plett, G.L. Extended Kalman filtering for battery management systems of lipb-based hev battery packs part 1.
background. J. Power Sources 2004, 134, 262–276. [CrossRef]
24. Plett, G.L. Sigma-point Kalman filtering for battery management systems of lipb-based hev battery packs.
J. Power Sources 2006, 161, 1356–1368. [CrossRef]
25. Sun, F.; Xiong, R. A novel dual-scale cell state-of-charge estimation approach for series-connected battery
pack used in electric vehicles. J. Power Sources 2015, 274, 582–594. [CrossRef]
26. Chen, N.; Zhang, P.; Dai, J.; Gui, W. Estimating the state-of-charge of lithium-ion battery using an h-infinity
observer based on electrochemical impedance model. IEEE Access 2020, 8, 26872–26884. [CrossRef]
27. Shu, X.; Li, G.; Shen, J.; Yan, W.; Chen, Z.; Liu, Y. An adaptive fusion estimation algorithm for state of charge
of lithium-ion batteries considering wide operating temperature and degradation. J. Power Sources 2020,
462, 228132. [CrossRef]
28. Gong, X.; Suh, J.; Lin, C. A novel method for identifying inertial parameters of electric vehicles based on the
dual h infinity filter. Veh. Syst. Dyn. 2019, 58, 28–48. [CrossRef]
29. Xiong, W.; Mo, Y.; Yan, C. Lithium-ion battery parameters and state of charge joint estimation using bias
compensation least squares and the alternate algorithm. Math. Probl. Eng. 2020, 2020, 1–16. [CrossRef]
30. Xu, W.; Xu, J.; Lang, J.; Yan, X. A multi-timescale estimator for lithium-ion battery state of charge and state of
energy estimation using dual h infinity filter. IEEE Access 2019, 7, 181229–181241. [CrossRef]
31. Zhang, Q.; Shang, Y.; Li, Y.; Cui, N.; Duan, B.; Zhang, C. A novel fractional variable-order equivalent circuit
model and parameter identification of electric vehicle li-ion batteries. ISA Trans. 2020, 97, 448–457. [CrossRef]
[PubMed]
32. Wang, Q.; He, Y.-J.; Shen, J.-N.; Hu, X.; Ma, Z.-F. State of charge-dependent polynomial equivalent circuit
modeling for electrochemical impedance spectroscopy of lithium-ion batteries. IEEE Trans. Power Electron.
2017, 33, 8449–8460. [CrossRef]
Appl. Sci. 2020, 10, 6371 18 of 18

33. Lin, X.; Perez, H.E.; Mohan, S.; Siegel, J.B.; Stefanopoulou, A.G.; Ding, Y.; Castanier, M.P. A lumped-parameter
electro-thermal model for cylindrical batteries. J. Power Sources 2014, 257, 1–11. [CrossRef]
34. Xu, Z.; Gao, S.; Yang, S. Lifepo4 battery state of charge estimation based on the improved thevenin equivalent
circuit model and Kalman filtering. J. Renew. Sustain. Energy 2016, 8, 24103. [CrossRef]

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).