Applied Energy xxx (2017) xxx–xxx

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

A novel method on estimating the degradation and state of charge of

lithium-ion batteries used for electrical vehicles
Ruixin Yang a, Rui Xiong a,⇑, Hongwen He a, Hao Mu a, Chun Wang a,b
a
Collaborative Innovation Center of Electric Vehicles in Beijing, School of Mechanical Engineering, Beijing Institute of Technology, Beijing 100081, China
b
School of Mechanical Engineering, Sichuan University of Science and Engineering, Zigong 643000, Sichuan, China

h i g h l i g h t s

A three-dimensional response surface-based battery OCV model was constructed.

A genetic algorithm is used to identify the model parameters.

The accuracy and robustness of the proposed method are verified systematically.

The proposed method shows high accuracy and robustness during the entire battery life.

The proposed method can effectively improve the efficiency of on-line computation.

a r t i c l e i n f o a b s t r a c t

Article history: The accurate determination of the capacity degradation path and state of charge (SoC) is very important
Received 9 February 2017 for the battery energy storage systems widely used in electric vehicles. This research can be summarized
Received in revised form 26 May 2017 as follows. First, a three-dimensional response surface-based SoC-open circuit voltage (OCV) capacity
Accepted 27 May 2017
method covering the entire lifetime of a battery has been constructed, which can be used to describe
Available online xxxx
the battery capacity degradation characteristics and determine the corresponding SoC. Second, in order
to capture the battery health state and energy state, a genetic algorithm (GA) is applied to identify the
Keywords:
battery capacity and initial SoC based on a first-order RC model. Finally, to verify the proposed method,
Electric vehicles
Battery
six experimental cases, including batteries with different aging states and with different data calculation
Capacity durations, are considered. The results indicate that the maximum capacity and SoC estimation errors are
State of charge less than 5.0% and 2.1%, respectively, for batteries with different aging states, which points to the high
Degradation accuracy, stability and robustness of the proposed GA-based battery capacity and initial SoC estimator
Recognition during the entire battery lifespan.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction battery capacity diminishes due to the modification of the positive

and negative structural properties, growth on the solid electrolyte
Lithium-ion batteries are important energy storage devices that interface (SEI), and variation of the electrolyte chemical composi-
have been widely applied to various types of electric vehicles (EVs). tion or reduction in active materials [1–3]. It is necessary to obtain
Battery health assessment and efficient energy management are an accurate estimation of battery capacity and SoC for the battery
basic principles pertaining to battery application. As an electric during its service life [4]. Otherwise, battery abuse including over-
vehicle core technology, accurate estimations of battery capacity charging, over-discharging, or even thermal runaway may occur.
and state of charge (SoC) are very important. Battery SoC is residual Therefore, accurate and high-efficiency estimation of battery
capacity and usually described by percentage. Battery capacity, capacity and SoC can prolong battery service life and improve over-
defined as the maximum energy in ampere-hours that the battery all battery performance.
can hold, is generally referred to as the maximum available capac-
ity of the battery at the current aging level. After repeated cycling,
1.1. Review of existing estimation methods for battery capacity and
⇑ Corresponding author at: Department of Vehicle Engineering, School of SoC
Mechanical Engineering, Beijing Institute of Technology, No. 5 South Zhongguancun
Street, Haidian District, Beijing 100081, China. With increased popularity and application of Li-ion
E-mail address: rxiong@bit.edu.cn (R. Xiong). batteries, the estimation of battery SoC has received much

http://dx.doi.org/10.1016/j.apenergy.2017.05.183
0306-2619/Ó 2017 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Yang R et al. A novel method on estimating the degradation and state of charge of lithium-ion batteries used for electrical
vehicles. Appl Energy (2017), http://dx.doi.org/10.1016/j.apenergy.2017.05.183
2 R. Yang et al. / Applied Energy xxx (2017) xxx–xxx

attention, resulting in improved estimation precision. The SoC esti- current charging and discharging conditions. Consequently, they
mation methods are summarized and evaluated in Refs. [1,2]; cannot be used in EVs directly.
these methods are often divided into four types: conventional
methods [3–5] (OCV, ampere-hour counting, and others), adaptive 1.2. Contributions of this study
filter algorithms, learning algorithms [6] and non-linear observers
[7]. The OCV based method is a simple approach to obtain the bat- Both the capacity and the SoC are important parameters for the
tery SOC, which is not suitable for EVs for batteries are required to battery management system (BMS) in EVs. Their accurate estima-
have long time resting in order to reach balance. The ampere-hour tions have recently been the focus of research, particularly the
counting method is also simple and general way for low-cost sen- capacity estimation. The accessible capacity is the base of accurate
sors and computing, which is often combined with other model battery health assessment and efficient energy management. In
based estimation approaches. But it has accumulated error and is this paper, the three-dimensional response surface (TRS)-based
hard to calibrate the initial error. The learning algorithms can be SoC-OCV capacity estimator has been constructed covering the
employed for SoC estimation. The disadvantages of this method entire lifetime of a battery, based on a GA estimator for accurate
are a great number of training data are needed and a lot of compu- estimation of battery capacity and SoC. As a state-of-the-art opti-
tations are required. The method of non-linear observers is not mal algorithm, the GA method is able to obtain the global optimum
complex but when the system is observable the method can be very efficiently. Except for the SoC-OCV-capacity response surface
used. Among these SoC estimation methods, adaptive filter algo- constructed in the beginning, limited on-board data is needed to
rithms are more suitable for EV application, and the methods pro- accurately determine the battery capacity and initial SoC, even
posed by the authors in Refs. [8–10] result in a SoC estimation error during the entire battery application lifespan. Additionally,
precision within 5%. However, most of the battery SoC estimation because of its high reliability and portability, the GA algorithm is
methods mentioned previously are based on a known capacity. expected to be more applicable in practice compared with tradi-
Because battery capacity diminishes and performance degrades tional advanced estimators such as Kalman filters. For electric
in an unpredictable and random manner, also the degradation vehicles application, we can send a certain period of time data of
paths of batteries are difficult to capture accurately. SoC estimation driving cycles stored in BMS on-line to a powerful computer off-
using a known capacity has obvious limitations to practical appli- line or even a cloud computing platform to compute the battery
cations. Therefore, it is important that calibrate battery capacity capacity and SoC. Then the calculation results are returned to the
on-line to estimate SoC accurately for practical applications. BMS via a SIM card or other mediums. We use an on-line and
Another important battery parameter is battery capacity. off-line computing manner in combination to reduce computa-
Capacity estimation techniques can be divided into two categories tional cost of BMS.
[11–14]: SoC-correlative and SoC-independent. The methods from
the SoC-correlative category consider that battery capacity estima-
tion is concerned with SoC. Ref. [15] uses the ratio of ampere-hour 1.3. Organization of the paper
accumulation and SoC variation to estimate the battery capacity
for the current aging condition. The relationship between battery Section 2 first describes the flow diagram of the optimization
OCV and SoC is used to calculate battery capacity in Refs. [16– algorithm. Next, the specific computation procedure for the capac-
18]. The advantage of this method is that fewer parameters are ity and SoC is presented. Section 3 introduces the configuration of
needed in the battery capacity estimation, because only the OCV- the battery test bench, the schedule and the results. A battery with
SoC relationship is used. The accuracy of this method is high for four different aging states and two different operating conditions is
different aging states during the entire battery application lifespan analyzed in the six case studies in Section 4 before conclusions are
if the cell inconsistency of the OCV-SoC relationship is very small. drawn in Section 5.
The disadvantage of this method is that accurately measured OCV
is difficult to obtain during the entire battery service life. Therefore, 2. Description of the capacity and SoC estimation approach
this method is difficult to use on-line. A joint estimation or dual
estimation method is used to estimate battery capacity and SoC 2.1. An equivalent circuit model of a battery
concurrently, and in Refs. [3,19–23], the extended Kalman filter
(EKF), in Ref. [24], the unscented Kalman filter (UKF), in Ref. [25], Lithium-ion battery characteristics can be described by an
the particle filter (PF), in Ref. [26], the unscented particle filter equivalent circuit model for different operating conditions. Plett
(UPF) and in Ref. [27], the recursive least square (RLS) algorithms [27] and Hu [35] have proposed several equivalent circuit models,
are employed. These methods can potentially estimate battery including the simple, one-state hysteresis, enhanced self-
capacity accurately as the battery is modeled, and SoC is thereby correcting, first-order RC, first-order RC with one-state hysteresis
estimated [2]. However, these methods performance could models, etc., and Hu has presented a comparative study of the 12
increase when the system and observation noise satisfy the Gaus- equivalent circuit models. Based on their previous research experi-
sian distribution and the model parameters are accurate. The dis- ence, we consider the first-order RC model, also called a Thevenin
advantage of these methods is that the algorithms suffer from model, as the best choice in this paper for its high model accuracy
high complexity, high computational cost and instability. and low computational effort. The schematic diagram of the Theve-
The methods that use the SoC independently can avoid the need nin model is presented in Fig. 1. The model equations can be
for a complex model. In Ref. [28], the change in battery voltage is expressed by:
observed to estimate the battery capacity. This estimation method


is suitable for identical charging conditions, but electric vehicles U D;kþ1 ¼ expðDt=sD ÞU D;k þ RD ½1  expðDt=sD ÞIk
ð1Þ
subject to dynamic operating conditions could be charged incon- U k ¼ U oc;k  Ro Ik  U D;k
sistently with different current. Battery impedance can also reveal
capacity fade of the battery current state condition [29]. In Refs. where U k , U oc;k and Ik are the battery terminal voltage, open circuit
[30–34], the authors use incremental capacity analysis (ICA) and voltage and load current at time equal to k, U D;kþ1 and U D;k denote
differential voltage analysis (DVA) techniques to investigate the the voltage of the RC network at time equal to k + 1 and k, respec-
behavior of a battery and estimate the battery capacity. These tively, sD and RD represent the time constant and polarization resis-
methods can estimate battery capacity accurately only for constant tance of the RC network, and Dt represents the sampling time.

Please cite this article in press as: Yang R et al. A novel method on estimating the degradation and state of charge of lithium-ion batteries used for electrical
vehicles. Appl Energy (2017), http://dx.doi.org/10.1016/j.apenergy.2017.05.183
 R. Yang et al. / Applied Energy xxx (2017) xxx–xxx 3

of individuals and selection is made according to a certain proba-

bility. To generate new a population, duplication, crossover and
mutation are necessary. If the individuals of the new population
satisfy the end conditions, the calculation is terminated and the
result of the optimization problem identified.

2.3. The capacity and SoC joint estimator

The proposed battery capacity and SoC estimator can be

described with four steps, as shown in Fig. 3.
Fig. 1. The schematic diagram of the first-order RC model.
Step 1. Build the TRS-based OCV model.

The open circuit voltage (OCV) is the voltage source of the entire At a certain temperature (e.g. 25 °C), the battery is repeatedly
circuit, and the resistor Ro represents the contact resistance of the cycled. The battery OCV, SoC corresponded to OCV and battery
electrode material, electrolyte, separator and other parts. RD and C D capacity are recorded at different aging states. With the complete
of the RC network describe the dynamic voltage performance of the battery OCV data for different SoC and capacity, the TRS-based
battery which includes the diffusing effects and polarization char- OCV model can be built. It is described by the following equation:
acteristics. The time constant sD of the RC network can be
expressed as sD ¼ RD C D . U oc ¼ f ðz; Q Þ ð2Þ

where f is the battery OCV function, z is the battery SoC and Q is the
2.2. Description of the optimization algorithm
battery capacity. A lithium-ion battery cell OCV versus SoC and bat-
tery capacity three-dimensional response surface is presented in
In consideration of the convergence rate, computational mem-
Fig. 4. We hold the opinion that the TRS-based OCV model can cover
ory and convenience of operation, a genetic algorithm (GA) is
the entire battery lifespan of this lithium-ion batteries type and is
selected. The GA concept was introduced by Holland and his team
determined concurrently by the battery capacity and SoC.
in the 1960s and 1970s. Based on the theory of biological evolution
Note that the temperature factor has been ignored in the OCV
in nature, strong and fit species tend to survive, passing their genes
model because (1) in general, the battery system has a good bat-
to the next generation by natural selection, while weak and unfit
tery thermal management system and the temperature deviation
species have less opportunity to pass their genes to future offspring
is maintained within ±5 °C, and (2) Fig. 7 (Section 3) shows that
and are faced with extinction. The main steps in a GA are presented
when the battery SoC is between 20–90%, temperature has little
in Fig. 2. The figure shows that the population is normally ran-
influence on the relationship between OCV and SoC.
domly initialized with a given size and each individual of the pop-
ulation is a possible solution to the optimization problem. If the
Step 2. Establish the equivalent circuit model of the battery.
results do not meet the terminal condition, we calculate the fitness

The Thevenin equivalent circuit model is built to describe the

battery charge and discharge characteristics. It includes three
Initial population parts, the voltage source OCV, the ohmic resistance Ro and the RC
network, as shown in Fig. 1.

Step 3. Estimate battery capacity Q and initial SoC z0.

Yes
Obtaining End
During a continuous operating time T, the measured terminal
results conditions ?
voltage and (dis)charging load current are sampled. Based on the
No ampere-hour integral method, the battery SoC z, the initial SoC z0
and the battery capacity Q are calculated by the following
Fitness equation:
End
caltulation Rt
I d
0 k
s
z ¼ gðz0 ; Q Þ ! gðz0 ; QÞ ¼ z0  ð3Þ
Q

Selection Next, OCV-Uoc, battery capacity Q and initial SoC z0 are corre-
lated as given in

U oc ¼ f ðz0 ; Q Þ ð4Þ
When battery capacity, initial SoC and model parameters are
Duplication Crossover Mutation known, we can use the model to predict battery behavior and pro-
vide an accurate SoC. These parameters are considered optimiza-
tion variables v in the GA optimization algorithm, where
v ¼ ½Ro ; RD ; sD ; Q ; z0 . The measured terminal voltage U and sam-
pled (dis)charging load current I of the battery are substituted into
New population
the model Eq. (1), and the optimization variables v are identified
using the GA optimization algorithm method. The battery capacity
Fig. 2. Flow diagram of a GA. Q, the initial SoC z0 and other battery parameters are identified

Please cite this article in press as: Yang R et al. A novel method on estimating the degradation and state of charge of lithium-ion batteries used for electrical
vehicles. Appl Energy (2017), http://dx.doi.org/10.1016/j.apenergy.2017.05.183
4 R. Yang et al. / Applied Energy xxx (2017) xxx–xxx

Off-line Data Measurement Algorithm Framework

Start
Loading UDDS profiles Building equivalent
circuit model

Loading profiles
Characteristic test Ageing cycle test

A three-dimensional response Battery parameter

Load
surface based OCV model identification
Lithium-ion battery cell

Modeling based Constraint condition

battery test data
A three-dimensional response Objective
surface based OCV model function
Battery capacity and initial
SoC estimation results

Fig. 3. A general diagram of the battery capacity and SoC estimation-based battery OCV model.

4.5

4
OCV (V)

3.5

2.5
28
26 1
0.8
24 0.6
22 0.4
0.2
20 0
Battery capacity Q (Ah) Battery SOC

Fig. 4. A TRS-based battery OCV model.

concurrently. In the process of identification, the objective function of the TRS-based OCV model. The constraint conditions of the opti-
F is the sum of the least square error between the estimated and mization algorithm are shown as
measured voltage, as given in 8
>
> 80%C N 6 Q 6 C max
( ) >
>
Xn >
> 0:1 6 z0 6 1
2 >
>
F ¼ min ðU k  U L Þ ð5Þ >
>
k¼1
>
< 0:1 m X 6 Ro 6 10 mX
s:t: jmin 6 j 6 jmax ð6Þ
>
>
where Uk represents the estimated terminal voltage and UL repre- >
> Hmin 6 H 6 Hmax
>
>
sents the measured terminal voltage. The sum of the least square >
>
> 0:1 mX 6 RD 6 10 mX
>
error between the estimation and measurement is correlated to >
:
the optimum matching relationship between the OCV and the SoC
50 s 6 sD 6 600 s

Please cite this article in press as: Yang R et al. A novel method on estimating the degradation and state of charge of lithium-ion batteries used for electrical
vehicles. Appl Energy (2017), http://dx.doi.org/10.1016/j.apenergy.2017.05.183
 R. Yang et al. / Applied Energy xxx (2017) xxx–xxx 5

The ranges of computational parameters in the constraint con- which records test data for a battery from the beginning of service
dition are based on the characteristics of the battery parameters to the end of its service life in an EV. This database is a basis for
themselves and technical parameters provided by the battery man- building an OCV model.
ufacturers. For example, in EV applications, batteries usually reach
the end of service life once their capacity is reduced to 80% of their
3.1. The lithium-ion battery test system
nominal capacity and therefore, the low end of the optimization
range for variable Q is 80% of the battery’s nominal capacity. In
A 25 Ah NCM material lithium-ion battery cell is tested system-
addition, the minimum value for the SoC is generally not less than
atically and the cell specification of it is shown in Table 1. The bat-
20%, otherwise, the service life of the battery will be reduced. Con-
tery test bench is shown in Fig. 5. It consists of a battery testing
sidering the working conditions of electric vehicles, the low end of
system (Arbin BT2000) for charging or discharging batteries, a
the optimization variable z0 range is 10%. Other model parameter
thermal chamber for programmable temperature control and a
values are based on several battery experiments we have per-
computer for test data storage and human-computer interaction.
formed previously and some parameter values shown in Section 3.
The range of Arbin BT2000 is 5 V–100 A, allowing it to charge or
discharge a 25Ah battery cell at the maximum rate of 4C. Its sam-
Step 4. Estimate the battery SoC.
pling interval is 1 s, and the auxiliary temperature channel can
gather temperature information. The programmable thermal
The battery SoC z at any time during the battery’s life can be
chamber can control the temperature range from 50 °C to
obtained based on the battery capacity Q and the initial SoC z0
120 °C with a precision of ±0.5 °C. Using the host computer, we
using the ampere-hour integral method, as given in
can control an Arbin BT2000 test system using the MITS Pro soft-
Rt
I d
0 k
s ware and adjust the temperature of the thermal chamber using
z ¼ z0  ð7Þ temperature control software.
Q
where battery capacity Q and initial SoC z0 are initially known, and
3.2. The lithium-ion battery test schedule
Ik is the sampling current data.
The battery capacity tests, OCV-SoC tests, urban dynamometer
3. Experiments driving schedule tests (UDDS) and dynamic stress tests (DST) are
executed at different battery aging states. The batteries are also
An equivalent circuit model for a battery can be based upon repeatedly cycled. The battery OCV versus SoC and battery capac-
data obtained from a variety of battery experiments. In this section, ities are recorded at different aging states for a specific tempera-
the configuration and basic principles of the battery test bench are ture. The battery test schedule is shown in Fig. 6.
introduced. A lithium-ion battery cell database is established, The aging cycle test results of a lithium-ion battery cell are pre-
sented for eight cycle states of 0, 100, 200, 300, 400, 480, 550 and
600 cycles (referred to as cyc00, cyc100, cyc200, . . ., and so on). The
Table 1
Cell specification of the NCM battery. aging cycle test results for values lower than 400 cycles is equal to
that of 100 cycles, but after 400 cycles, the battery capacity dimin-
Items Specification
ishes to less than 90% of its nominal capacity. Considering the bat-
Cathode material Li(NiCoMn)O2 tery service life and experimental safety, the number of cycles
Anode material Graphite
tested was reduced. When 600 cycles were reached, the battery
Nominal capacity 25 Ah
Allowed voltage range 2.5–4.2 V capacity diminished to 20.75 Ah and the experiment was
terminated.

Fig. 5. Configuration of the battery test bench.

Please cite this article in press as: Yang R et al. A novel method on estimating the degradation and state of charge of lithium-ion batteries used for electrical
vehicles. Appl Energy (2017), http://dx.doi.org/10.1016/j.apenergy.2017.05.183
6 R. Yang et al. / Applied Energy xxx (2017) xxx–xxx

lithium-ion battery cell at different aging states are given in

Characteristic Test (10ć) Table 2.
Static Capacity Test As Table 2 shows, with increased cycling times, the battery
capacity continues to decay, dropping to nearly 80% of the nominal
capacity when 600 cycles are reached. At this point, the battery
Characteristic Test (25ć) OCV-SoC Test
should not continue to be cycle-tested and the test should be
terminated.
At a temperature of 25 °C, a battery cell is tested using an OCV-
Loading profiles Test SoC test for different aging states, and the resulting relationship
Characteristic Test (40ć)
between OCV and SoC is shown in Table 3. We list only the SoC
ranges between 30–80%.
As shown in Fig. 4, the OCV model is established based on the
Aging Cycle Test (10ć) battery capacity data and the OCV-SoC test results at the tempera-
ture of 25 °C at different aging states. As the figure shows, the bat-
Fig. 6. The battery test schedule. tery cell OCV is determined concurrently by the battery capacity Q
and the SoC. For that reason, the battery capacity Q and SoC can be
3.3. The battery test results used to describe the model equation and are two optimization
variables in the GA algorithm.
At temperatures of 10 °C, 25 °C and 40 °C, a 25Ah nominal To illustrate the influence of temperature on the relationship
capacity battery is tested with a static capacity test, an OCV-SoC between battery cell OCV and SoC, the OCV-SoC test is imple-
test, a loading profiles test and an aging cycle test. For the static mented at temperatures of 10 °C and 40 °C for different aging
capacity test (10 °C, 25 °C, 40 °C), values for the capacity of the states. The battery test results are shown in Tables 4 and 5.

Table 2
Results of a battery cell capacity test (10 °C, 25 °C, 40 °C) at different aging states.

Cycle number cyc00 cyc100 cyc200 cyc300 cyc400 cyc480 cyc550 cyc600
Capacity at 10 °C (Ah) 25.57 24.79 24.34 24.15 23.56 22.10 20.68 20.07
Capacity at 25 °C (Ah) 27.34 25.68 25.20 24.87 24.31 22.74 21.36 20.75
Capacity at 40 °C (Ah) 27.89 25.92 25.38 25.03 24.61 23.07 21.74 20.73

Table 3
The battery cell OCV-SoC test result (25 °C) for different aging states.

cyc00 cyc100 cyc200 cyc300 cyc400 cyc480 cyc550 cyc600

SOC OCV SOC OCV SOC OCV SOC OCV SOC OCV SOC OCV SOC OCV SOC OCV
80% 3.92 80% 3.93 80% 3.93 80% 3.93 80% 3.93 80% 3.94 80% 3.94 80% 3.94
70% 3.83 70% 3.83 70% 3.83 70% 3.83 70% 3.83 70% 3.84 70% 3.85 70% 3.85
60% 3.73 60% 3.74 60% 3.74 60% 3.74 60% 3.74 60% 3.76 60% 3.77 60% 3.78
50% 3.67 50% 3.68 50% 3.68 50% 3.69 50% 3.69 50% 3.70 50% 3.71 50% 3.72
40% 3.64 40% 3.65 40% 3.65 40% 3.65 40% 3.65 40% 3.66 40% 3.67 40% 3.67
30% 3.61 30% 3.62 30% 3.61 30% 3.61 30% 3.62 30% 3.62 30% 3.62 30% 3.62

Table 4
The battery OCV-SoC test results (10 °C) for different aging states.

cyc00 cyc100 cyc200 cyc300 cyc400 cyc480 cyc550 cyc600

SOC OCV SOC OCV SOC OCV SOC OCV SOC OCV SOC OCV SOC OCV SOC OCV
80% 3.91 80% 3.92 80% 3.92 80% 3.93 80% 3.93 80% 3.93 80% 3.94 80% 3.93
70% 3.83 70% 3.83 70% 3.83 70% 3.83 70% 3.83 70% 3.84 70% 3.85 70% 3.84
60% 3.75 60% 3.74 60% 3.74 60% 3.74 60% 3.74 60% 3.76 60% 3.77 60% 3.77
50% 3.68 50% 3.68 50% 3.68 50% 3.69 50% 3.69 50% 3.70 50% 3.71 50% 3.71
40% 3.65 40% 3.65 40% 3.65 40% 3.65 40% 3.65 40% 3.66 40% 3.67 40% 3.67
30% 3.62 30% 3.62 30% 3.62 30% 3.62 30% 3.62 30% 3.62 30% 3.62 30% 3.62

Table 5
The battery OCV-SoC test results (40 °C) for different aging states.

cyc00 cyc100 cyc200 cyc300 cyc400 cyc480 cyc550 cyc600

SOC OCV SOC OCV SOC OCV SOC OCV SOC OCV SOC OCV SOC OCV SOC OCV
80% 3.92 80% 3.93 80% 3.93 80% 3.93 80% 3.94 80% 3.95 80% 3.95 80% 3.95
70% 3.82 70% 3.83 70% 3.83 70% 3.84 70% 3.84 70% 3.84 70% 3.85 70% 3.86
60% 3.73 60% 3.74 60% 3.74 60% 3.74 60% 3.75 60% 3.76 60% 3.77 60% 3.78
50% 3.67 50% 3.69 50% 3.69 50% 3.69 50% 3.69 50% 3.71 50% 3.72 50% 3.72
40% 3.64 40% 3.65 40% 3.65 40% 3.65 40% 3.65 40% 3.66 40% 3.67 40% 3.67
30% 3.61 30% 3.61 30% 3.61 30% 3.61 30% 3.61 30% 3.62 30% 3.62 30% 3.62

Please cite this article in press as: Yang R et al. A novel method on estimating the degradation and state of charge of lithium-ion batteries used for electrical
vehicles. Appl Energy (2017), http://dx.doi.org/10.1016/j.apenergy.2017.05.183
 R. Yang et al. / Applied Energy xxx (2017) xxx–xxx 7

Fig. 7. The OCV versus SoC for different aging states at different temperatures.

To compare the influence of different temperatures on the different cycles-cyc00, cyc100, etc. For different numbers of
relationship between OCV and SoC for different numbers of cycles, the influence of temperature on the relationship between
cycles, the OCV versus SoC for 10 °C, 25 °C and 40 °C are shown OCV and SoC is very small, especially for battery states of charge
in Fig. 7. Parts A-H of Fig. 7 represent the OCV versus SoC for 8 that exceed 20%.

Please cite this article in press as: Yang R et al. A novel method on estimating the degradation and state of charge of lithium-ion batteries used for electrical
vehicles. Appl Energy (2017), http://dx.doi.org/10.1016/j.apenergy.2017.05.183
8 R. Yang et al. / Applied Energy xxx (2017) xxx–xxx

Table 6 tery cell is subject to the operating conditions of the DST test and
Description of specific details for the six cases considered. the duration of calculating and storing data is 7200 s, while the
Case number Data length Cycle number Operating condition aging states are 550 and 600 cycles, respectively. Table 6 presents
01 3600 s 100 UDDS a clear view of the six cases considered.
02 7200 s 100 UDDS We used six cases to verify stability and robustness of the bat-
03 3600 s 400 UDDS tery capacity and SoC estimation method proposed in this paper
04 7200 s 400 UDDS with different aging states and two operation conditions (UDDS
05 7200 s 550 DST
06 7200 s 600 DST
and DST). In addition, estimation accuracy influenced by different
durations of data calculation is discussed using two different cases
(3600 s and 7200 s).
The measured and estimated voltages versus time for the six
4. Case studies cases are shown in Fig. 8. The figures feature a close-up view to
enlarge the differences between the estimated and measured val-
This section presents six cases to demonstrate the effectiveness ues. It is apparent that the estimated voltages trace the measured
of the proposed methods for battery capacity and SoC estimation voltages closely.
for different aging states (nearly fresh battery, lightly cycled, heav- The battery capacity and initial SoC estimation results for the
ily cycled and near end of service life) and different operating con- six cases are presented in Fig. 9. The results indicate that the bat-
ditions. In case01 and case02, the battery cell has an aging level of tery capacity estimation error is less than 5% and the battery initial
100 cycles and an operating condition identical to that in the UDDS SoC estimation error is less than 3%, both of which are beneficial
test. The duration for calculating and storing data is 3600 s for for initial SoC calibration.
case01 while the duration is 7200 s for case02. In a similar manner, Results from Fig. 9, showing the number of cycles for cases 01–
in case03 and case04 the battery cell is at an aging level of 400 06 ranging from cyc100 to cyc600, indicate the proposed method
cycles and the same operating condition and data calculation dura- can accurately estimate battery capacity and initial SoC for differ-
tion as in case01 and case02 apply. In case05 and case06, the bat- ent aging states. From the estimation results, we can see that the

4 4.1 3.9
Voltage (v)

Voltage (v)
3.9 4 3.8
4
3.8
1600 1800 2000 3.9 3.7
Voltage (v)

Voltage (v)

3.9 Time (s) 3000 3200 3400

Time (s)
3.8
3.8
3.7

3.7 Measured
Measured 3.6
Estimated A Estimated
B
3.5
0 500 1000 1500 2000 2500 3000 3500 4000 0 1000 2000 3000 4000 5000 6000 7000 8000
Time (s) Time (s)
4.1 4 4.1 3.9
Voltage (v)
Voltage (v)

3.9
4 4 3.8

3.8
3.9 3.7
Voltage (v)

1600 1800 2000 3000 3200 3400

Voltage (v)

3.9 Time (s) Time (s)

3.8
3.8
3.7
3.7 Measured
3.6 Measured

3.6
Estimated
C Estimated D
0 500 1000 1500 2000 2500 3000 3500 4000 3.5
0 1000 2000 3000 4000 5000 6000 7000 8000
Time (s) Time (s)
3.8 3.8
4 4
Voltage (v)
Voltage (v)

3.7 3.7

3.9 3.9 3.6

3.6

3.8 3.5
Voltage (v)

Voltage (v)

3.8 3.5
3200 3300 3400 3500 3600 3700
3200 3400 3600
Time (s)
Time (s)
3.7
3.7
3.6
3.6
3.5 Measured
3.5 Measured
Estimated
E 3.4
Estimated F
3.4
0 1000 2000 3000 4000 5000 6000 7000 8000 0 1000 2000 3000 4000 5000 6000 7000 8000
Time (s) Time (s)

Fig. 8. Voltage versus time for six cases: A. 100 cycles and 3600 s duration (UDDS); B. 100 cycles and 7200 s duration (UDDS); C. 400 cycles and 3600 s duration (UDDS); D.
400 cycles and 7200 s duration (UDDS); E. 550 cycles and 7200 s duration (DST); F. 600 cycles and 7200 s duration (DST).

Please cite this article in press as: Yang R et al. A novel method on estimating the degradation and state of charge of lithium-ion batteries used for electrical
vehicles. Appl Energy (2017), http://dx.doi.org/10.1016/j.apenergy.2017.05.183
 R. Yang et al. / Applied Energy xxx (2017) xxx–xxx 9

Fig. 9. Battery capacity and initial SoC estimation results for 6 cases.

estimation approach applies well to a variety of operation condi- robustness, stability and high accuracy for different aging states
tions. Comparing the results between case01 and case02 and and different operation conditions, and the estimation accuracy
between case03 and case04, we see how the estimation results can also be improved by increasing the time duration of data
are influenced by different durations of calculating data. From calculation.
the case01 and case02 estimation results in Fig. 9, the battery
capacity estimation error is reduced 0.6% from 3.23% to 2.63%
while the battery initial SoC estimation error is enhanced 0.1% Acknowledgments
from 1.07% to 1.17%. The reducing and enhancing of the estimation
error for case03 and case04 is similar to that of case01 and case02. This work was supported in part by the National Natural
In general, the battery capacity estimation accuracy is improved Science Foundation of China (Grant No. 51507012), Beijing Nova
more than the battery SoC estimation error is increased. It is appar- Program (Grant No. Z171100001117063), the National High
ent that increasing the data calculation duration is beneficial to the Technology Research and Development Program of China
estimation accuracy. (2015BAG01B01) and Joint Funds of the National Natural Science
Foundation of China (Grant No. U1564206). The systemic experi-
ments of the lithium-ion batteries were performed at the Advanced
5. Conclusion Energy Storage and Application (AESA) Group, Beijing Institute of
Technology.
A battery capacity and SoC estimation approach is proposed in
this paper. To verify the effectiveness of the estimation approach,
many battery experiments were implemented. From the battery References
experiments, a battery cell test database is established, and a [1] Cuma MU, Koroglu T. A comprehensive review on estimation strategies used in
three-dimensional response surface-based battery OCV versus hybrid and battery electric vehicles. Renew Sustain Energy Rev
SoC battery capacity model is established for different aging states. 2015;42:517–31.
[2] Xiong R, Tian JP, Mu H, Wang C. A systematic model-based degradation
The battery capacity and initial SoC which are taken as battery
behavior recognition and health monitor method of lithium-ion batteries. Appl
parameters are estimated using a genetic algorithm (GA) based Energy 2017. http://dx.doi.org/10.1016/j.apenergy.2017.05.124.
on the Thevenin equivalent circuit model. The battery SoC and [3] Kong SN, Moo CS, Chen YP, Hsieh YC. Enhanced coulomb counting method for
capacity are estimated. In Section 4, six cases were used to validate estimating state-of-charge and state-of-health of lithium-ion batteries. Appl
Energy 2009;86(9):1506–11.
the reliability of the estimation approach. From the six cases, the [4] Xiong R, Yu QQ, Wang LY, Lin C. A novel method to obtain the open circuit
results show that the estimation approach proposed has good voltage for the state of charge of lithium ion batteries in electric vehicles by

Please cite this article in press as: Yang R et al. A novel method on estimating the degradation and state of charge of lithium-ion batteries used for electrical
vehicles. Appl Energy (2017), http://dx.doi.org/10.1016/j.apenergy.2017.05.183
10 R. Yang et al. / Applied Energy xxx (2017) xxx–xxx

using H infinity filter. Appl Energy 2017. http://dx.doi.org/10.1016/j. [20] Xiong R, Sun F, Chen Z, He H. A data-driven multi-scale extended Kalman
apenergy.2017.05.136. filtering based parameter and state estimation approach of lithium-ion
[5] Zheng Y, Ouyang M, Lu L, Li J, Han X, Xu L, et al. Cell state-of-charge polymer battery in electric vehicles. Appl Energy 2014;113(C):463–76.
inconsistency estimation for LiFePO4 battery pack in hybrid electric vehicles [21] Lee S, Kim J, Lee J, Cho BH. State-of-charge and capacity estimation of lithium-
using mean-difference model. Appl Energy 2013;111:571–80. ion battery using a new open-circuit voltage versus state-of-charge. J Power
[6] Charkhgard M, Farrokhi M. State-of-charge estimation for lithium-Ion batteries Sources 2008;185(2):1367–73.
using neural networks and EKF. Ind Electron IEEE Trans 2011;57(12):4178–87. [22] Pérez G, Garmendia M, Reynaud JF, Crego J, Viscarret U. Enhanced closed loop
[7] Quoilin S, Kawadias K, Mercier A, Pappone I, Zucker A. Quantifying self- State of Charge estimator for lithium-ion batteries based on Extended Kalman
consumption linked to solar home battery systems: statistical analysis and Filter. Appl Energy 2015;155:834–45.
economic assessment. Appl Energy 2016;182:58–67. [23] Hu C, Youn BD, Chung J. A multiscale framework with extended Kalman filter
[8] Sun F, Xiong R, He H. A systematic state-of-charge estimation framework for for lithium-ion battery SOC and capacity estimation. Appl Energy
multi-cell battery pack in electric vehicles using bias correction technique. 2012;92:694–704.
Appl Energy 2016;162:1399–409. [24] Zhang F, Liu G, Fang L. Battery state estimation using Unscented Kalman Filter.
[9] Wang Y, Zhang C, Chen Z. A method for state-of-charge estimation of LiFePO4 IEEE Int Conf Robot Autom 2009:1863–8.
batteries at dynamic currents and temperatures using particle filter. J Power [25] Wang Y, Zhang C, Chen Z. A method for joint estimation of state-of-charge and
Sources 2015;279:306–11. available energy of LiFePO4 batteries. Appl Energy 2014;135(C):81–7.
[10] Zhang Y, Xiong R, He H, Shen W. A lithium-ion battery pack state of charge and [26] He Y, Liu X, Zhang C, Chen Z. A new model for State-of-Charge (SOC)
state of energy estimation algorithms using a hardware-in-the-loop validation. estimation for high-power Li-ion batteries. Appl Energy 2013;101(1):808–14.
IEEE T Power Electr 2017;32(6):4421–31. [27] Plett GL. Sigma-point Kalman filtering for battery management systems of
[11] Lu L, Han X, Li J, Hua J, Ouyang M. A review on the key issues for lithium-ion LiPB-based HEV battery packs : Part 2: Simultaneous state and parameter
battery management in electric vehicles. J Power Sources 2013;226 estimation. J Power Sources 2006;161(2):1369–84.
(3):272–88. [28] Zheng Y, Lu L, Han X, Li J, Ouyang M. LiFePO4 battery pack capacity estimation
[12] Barré A, Deguilhem B, Grolleau S, Gérard M, Suard F, Riu D. A review on for electric vehicles based on charging cell voltage curve transformation. J
lithium-ion battery ageing mechanisms and estimations for automotive Power Sources 2013;226:33–41.
applications. J Power Sources 2013;241(11):680–9. [29] Waag W, Käbitz S, Sauer DU. Experimental investigation of the lithium-ion
[13] Berecibar M, Gandiaga I, Villarreal I, Omar N, Van Mierlo J, Van den Bossche P. battery impedance characteristic at various conditions and aging states and its
Critical review of state of health estimation methods of Li-ion batteries for real influence on the application. Appl Energy 2013;102(2):885–97.
applications. Renew Sustain Energy Rev 2015;56:572–87. [30] Han X, Ouyang M, Lu L, Li J. A comparative study of commercial lithium ion
[14] Farmann A, Waag W, Marongiu A, Sauer DU. Critical review of on-board battery cycle life in electric vehicle: capacity loss estimation. J Power Sources
capacity estimation techniques for lithium-ion batteries in electric and hybrid 2014;268(4):658–69.
electric vehicles. J Power Sources 2015;281:114–30. [31] Dubarry M, Truchot C, Liaw BY, Gering K, Sazhin S, Jamison D, et al. Evaluation
[15] Waag W, Sauer DU. Adaptive estimation of the electromotive force of the of commercial lithium-ion cells based on composite positive electrode for
lithium-ion battery after current interruption for an accurate state-of-charge plug-in hybrid electric vehicle applications. Part II. Degradation mechanism
and capacity determination. Appl Energy 2013;111(4):416–27. under 2 C cycle aging. J Power Sources 2011;196(23):10328–35.
[16] Waag W, Fleischer C, Sauer DU. Critical review of the methods for monitoring [32] Weng C, Cui Y, Sun J, Peng H. On-board state of health monitoring of lithium-
of lithium-ion batteries in electric and hybrid vehicles. J Power Sources ion batteries using incremental capacity analysis with support vector
2014;258(14):321–39. regression. J Power Sources 2013;235(4):36–44.
[17] Pop V, Bergveld HJ, Danilov D, Regtien Paul PL, Notten Peter HL. Battery [33] Feng X, Li J, Ouyang M, Lu L, Li J, He X. Using probability density function to
management systems: accurate state-of-charge indication for battery- evaluate the state of health of lithium-ion batteries. J Power Sources 2013;232
powered applications. London: Springer Verlag; 2008. (18):209–18.
[18] Kessels JTBA, Rosca B, Bergveld HJ, van den Bosch PPJ. On-line battery [34] Bloom I, Walker LK, Basco JK, Abraham DP, Christophersen JP, Ho CD.
identification for electric driving range prediction. In: IEEE Vehicle Power and Differential voltage analyses of high-power lithium-ion cells. 4. Cells
Propulsion Conference (VPPC); 2011. p. 1–6. containing NMC. J Power Sources 2010;195(3):877–82.
[19] Hu C, Youn BD, Chung J. A multiscale framework with extended Kalman filter [35] Hu X, Li S, Peng H. A comparative study of equivalent circuit models for Li-ion
for lithium-ion battery SOC and capacity estimation. Appl Energy 2012;92 batteries. J Power Sources 2012;198(198):359–67.
(4):694–704.

Please cite this article in press as: Yang R et al. A novel method on estimating the degradation and state of charge of lithium-ion batteries used for electrical
vehicles. Appl Energy (2017), http://dx.doi.org/10.1016/j.apenergy.2017.05.183