See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/352000264

STATE OF CHARGE ESTIMATION OF HIGH-POWER LITHIUM-ION BATTERIES

WITH IMPROVED EQUIVALENT CIRCUIT MODELING AND ADAPTIVE
EXTENDED KALMAN FILTERING ALGORITHM.

Article  in  Theoretical & Applied Science · May 2021

DOI: 10.15863/TAS.2021.05.97.49

CITATIONS READS

0 505

7 authors, including:

Etse Dablu Bobobee Shunli Wang

Southwest University of Science and Technology Southwest University of Science and Technology
49 PUBLICATIONS   144 CITATIONS    5,087 PUBLICATIONS   2,729 CITATIONS   

SEE PROFILE SEE PROFILE

Chunmei Yu Emmanuel Appiah

Southwest University of Science and Technology Southwest University of Science and Technology
396 PUBLICATIONS   1,174 CITATIONS    34 PUBLICATIONS   38 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Thesis View project

Prize_Haotian Shi View project

All content following this page was uploaded by Etse Dablu Bobobee on 02 June 2021.

The user has requested enhancement of the downloaded file.

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

QR – Issue QR – Article
SOI: 1.1/TAS DOI: 10.15863/TAS
International Scientific Journal
Theoretical & Applied Science
p-ISSN: 2308-4944 (print) e-ISSN: 2409-0085 (online)

Year: 2021 Issue: 05 Volume: 97

Published: 21.05.2021 http://T-Science.org

Etse Dablu Bobobee
School of Information Engineering, Southwest University of Science and Technology (SWUST)
B. Ed IT, M. Eng. Info. & Com., PhD. student
Mianyang 621010, Sichuan China
bobosonj@gmail.com

Shunli Wang
School of Information Engineering (SWUST)
Dr., Professor Head of DTLab
497420789@qq.com

Chun-Mei Yu
School of Information Engineering (SWUST)
Dr., Professor

Chuanyun Zou
School of Information Engineering (SWUST)
Dr., Professor

Emmanuel Appiah
School of Information Engineering (SWUST)
PhD. student

Cong Jiang
School of Information Engineering (SWUST)
Master’s student

Mingfang He
School of Information Engineering (SWUST)
Master’s student

STATE OF CHARGE ESTIMATION OF HIGH-POWER LITHIUM-ION

BATTERIES WITH IMPROVED EQUIVALENT CIRCUIT MODELING
AND ADAPTIVE EXTENDED KALMAN FILTERING ALGORITHM
Abstract: This paper focuses on the accurate estimation of the state of charge of lithium-ion batteries through
the establishment of an equivalent model, experimentation, simulation, and the use of an adaptive extended Kalman
filtering algorithm. Several models have been used in the creation of the high-power lithium-ion battery and as it is
difficult to estimate the state of charge of the lithium-ion battery accurately numerous methods and techniques are
employed. A Thevenin equivalent circuit model is designed to include two resistor-capacitors in series for easy
parameterization and estimation of the state of charge of the battery. An experimental approach is adopted and data
from the open-circuit voltage and the hybrid pulse power characteristic tests are used for parameterization. The
battery is modeled and simulated in Simulink/MATLAB with inputs from the results and calculations from the
experimental data. An improved adaptive extended Kalman filtering algorithm was used to accurately estimate the
state of charge. The main idea of using the improved adaptive algorithm is to update the statistical noise covariance
parameters and to improve the estimation performance and accuracy. This reduced the interference of system noise

Philadelphia, USA 248

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

effectively and minimized estimation error to the smallest value. An extended Kalman filtering algorithm was
employed alongside the adaptive extended Kalman filtering algorithm to verify the effectiveness of the adaptive
algorithm. Results and computations from the experiment and simulation are compared and the results show that the
improved adaptive extend Kalman filtering algorithm has good convergence speed, is more stable, and has a high
precision of accuracy in the estimation of the state of charge. The maximum estimation error realized with the use of
the extended Kalman filtering algorithm was 4.97%, and the maximum estimation error based on the use of the
improved adaptive extended Kalman filtering algorithm was 1.85%. The results, therefore, show that the adaptive
algorithm adopted in this paper can be used efficiently and effectively for the accurate state of charge estimation of
the high-power lithium-ion battery.
Key words: high-power lithium-ion batteries; battery management system; Thevenin equivalent circuit model;
state of charge; adaptive extended Kalman filtering algorithm.
Language: English
Citation: Bobobee, E. D., et al. (2021). State of charge estimation of high-power lithium-ion batteries with
improved equivalent circuit modeling and adaptive extended Kalman filtering algorithm. ISJ Theoretical & Applied
Science, 05 (97), 248-268.
Soi: http://s-o-i.org/1.1/TAS-05-97-49 Doi: https://dx.doi.org/10.15863/TAS.2021.05.97.49
Scopus ASCC: 2208.

Introduction electric vehicles [8], and artificial intelligence applied

Currently, lithium-ion batteries are perhaps the for the battery state estimation [9]. SOC estimation
most important new energy source in the 21st century, has therefore drawn the attention of many researchers,
and research in the area of improving and enhancing and many different methods have been proposed [10,
the performance of the technology through various 11]. The OCV method, a full charge detector/dynamic
techniques and methods is key to unearthing greater load observer, and the CC method with robust
potential. The key to improving, controlling, extended Kalman filtering algorithm (REKF) are
monitoring, and managing the lithium-ion battery is combined in [12]. It is difficult to determine the
the battery management system (BMS) [1], and the specific approach when such methods are used,
estimation of the state of charge (SOC), state of health however, based on the classification made in [12, 13],
(SOH), state of power (SOP), and other battery two categories; direct and indirect methods, and
parameters is a very important research field that several subcategories that summarize trends in SOC
ensures the safety and reliability of electronic devices estimation are mentioned and aligned appropriately.
that use these batteries as a source of power. Research Battery models are used to study the relationship
on the improvement of the function, reliability, and between the external characteristics and the internal
performance of lithium-ion battery technology is states of a battery by establishing a mathematical
important and any breakthrough in the area would go model. Models in a discrete-time state-space form are
a long way to improve upon the technology [2]. The also used for SOC estimation [14, 15]. Current
BMS is responsible for measuring the states of the literature makes mention of the use of equivalent
battery accurately and also ensure safe operation and circuit models (ECMs) as being widely used as a
prolong the battery life [3]. An improved Thevenin foundation for model-based estimation and control
equivalent circuit model was proposed, designed, and [16]. Generally, equivalent circuit models including
implemented through experimentation and the Rint model, the Thevenin model, the RC model,
simulation. The model was achieved by adding an and the partnership for a new generation of vehicle
extra RC branch to the Thevenin model, making it a (PNGV) model [17] are selected. The first-order RC,
second-order resistor-capacitor. The 2RC Thevenin second-order RC, and third-order RC are all employed
model has better accuracy, stability, robustness, and is to accurately model the battery [18, 19]. Among them,
very effective for SOC estimation [4]. This model was the Thevenin model is widely used, but not accurate
used to study and record parameters and estimate enough since all of its elements can change, depending
relationships between voltage, current, SOC, and the on the condition and state of the battery.
charging/discharging characteristics. According to [20, 21], to maximize parameter
Research concerning battery management identification, new designs are being developed using
systems (BMS) from a global perspective includes charging/discharging, and [22, 23] combined SOC
those which display an entire BMS design adopting a estimation methods as a means of estimating SOC in
distributed structure to reach better scalability and the presence of unknown or time-varying battery
portability [5]. Different approaches to designing a parameters. Research in the area either assumes an
BMS depend on the functionalities desired for the accurate SOC/OCV relationship or seeks to impose
specific application, but most of them focus on key approximations such as OCV being piece-wise linear
functions such as SOC estimation [6] and the with SOC [24] or constant during discharging [25].
balancing process [7]. The improvement is towards According to [11, 26], RC parameters are determined
the design of intelligent BMS’s for electric and hybrid through analyzing the transients in a battery voltage

Philadelphia, USA 249

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

response under certain excitations such as constant or filtering algorithm for SOC estimation is the
pulse current experiments. The voltage source in an unscented Kalman filtering algorithm, which is used
ECM typically represents the battery's open-circuit in [48, 49] to improve estimation accuracy. A UKF
voltage, which depends on the SOC [27]. A algorithm is implemented in [50] to estimate SOC
relationship between SOC and OCV can be identified using an improved ECM with a resistance and a
by charging or discharging the battery using a small capacitor correction factor. This was done to first,
current [28]. Parameter identification based on measure the effect of different current rates and the
current-voltage data is addressed in [21], by a method SOC estimation on the battery internal resistance, and
that reduces the problem of solving a set of high-order secondly to identify the impact of different current
polynomial equations into solving several linear rates and temperatures on the battery’s capacity [51].
equations and a single-variable polynomial equation. The works presented in [52, 53] use a multi-model
A Thevenin ECM is used in [14] for every single cell approach that adopts a bank of EKFs to estimate the
in an array of more than 90 series-connected cells, to SOC of the battery. A robust extended Kalman
identify the internal resistance of each cell. In [29] a filtering algorithm is implemented in [54] to be more
Thevenin model using two different branches for robust to uncertainties in the system, and
charge and discharge is connected in series n times to measurement equations, and noise covariances. A
represent n cells in a series. According to [30, 31], SOC estimation approach that uses an improvement in
there are three different ECMs of lithium-ion batteries the measurement noise treatment is proposed in [55]
widely adopted because of their excellent dynamic and, by correcting the covariance matrix error in the
performance and postulates that the second-order depicted EKF, establishes an adaptive Kalman
ECM is the most accurate and has the best dynamic filtering algorithm that can reduce the estimation
performance, but it is also the most complex. The error. To deal with the variation of battery parameters
Thevenin model and second-order ECM were used for due to temperature changes [56], an online approach
SOC estimation and compared with three other is proposed for SOC estimation and parameter
enhanced ECMs in [14, 32] and the difference updating using a dual square root UKF based on unit
between these models is the way the SOC equations spherical unscented transform.
are calculated. The parameters of the second-order To obtain a more accurate and reliable SOC, an
ECM can be calculated with different datasets improved Thevenin equivalent model is proposed and
depending on the purpose, like in [33]. In [34], a its parameters are identified. Experimental data results
comparison between continuous-time and discrete- and simulated results are compared and analyzed to
time equations of the second-order ECM is made and further appreciate the effectiveness of the improved
concludes that discrete-time identification methods adaptive extended Kalman filtering (AEKF)
are less robust due to undesired sensitivity issues in algorithm used. The use of the AEKF algorithm in this
the transformation of discrete domain parameters. research is to accurately and diligently estimate the
According to [35], SOC can be directly SOC and eliminate or reduce errors by updating the
calculated by the transformation of model equations noise covariance matrix. The use of this adaptive
with the deployment of a simple Thevenin ECM. In extended Kalman filtering algorithm is an innovation
[36], a Thevenin ECM is used to obtain the OCV, and in this work coupled with the 2RC Thevenin
then, a linear fitting of a portion of the OCV–SOC equivalent model for successful implementation and
curve is used to obtain the SOC. A similar process is SOC estimation. According to the results realized, it is
used in [37] to obtain the SOC with a simple ECM that worth noting that, the SOC estimation using the AEKF
considers just a voltage source and an internal algorithm is more accurate and reliable comparatively
resistance. A Kalman filter (KF) is introduced in [38] than the EKF algorithm. This paper is organized in
to extract an OCV value in a noisy environment, and sections as follows; The mathematical analysis is
then to estimate SOC based on the OCV–SOC conducted in section 2, experimental analysis in
mapping. In addition to models, there are different section 3, and the last section is section 4 which coves
types of closed-loop methods for SOC estimations, conclusion and further research plan.
such as direct feedback [13], extended Kalman Filters 2. Mathematical Analysis
[39], unscented Kalman Filters (UKF) [40, 41], and The basic concepts of SOC estimation, battery
neural network [42, 43]. These methods are used to modeling, and the Kalman filtering algorithm are first
further improve the accuracy of the SOC estimation introduced in this section. The improved Thevenin
since they can recursively estimate the uncertainty of equivalent model including the 2RC’s added as well
system state estimations and adapt Kalman gain to as the description of corresponding parameters to be
achieve optimal estimation in the next time step of the identified through experimentation are presented, the
iteration process. It is common to find a combination EKF and the improved AEKF algorithms are
of CC or OCV methods with the KF method like in introduced in detail to emphasize the improvement in
[44, 45] and as proposed in [46, 47], the EKF method the SOC estimation accuracy.
is also used in combination with CC and/or OCV.
Another common improvement to the Kalman

Philadelphia, USA 250

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

2.1 The structure of the improved 2RC Where Qt is the residual capacity of the battery
Thevenin equivalent model at time t and Q0 is the rated capacity. The value of SOC
An appropriate equivalent circuit model is the ratio between the current residual capacity and
established and described, and the input, output, and the capacity in the fully charged state when the battery
state variables are determined. The modeling and is fully static at a certain time.
simulation are done in Simulink/MATLAB and the 2.1.1 Electrical circuit model
parameters identified from the experimental data are The improved Thevenin equivalent circuit model
used for real-time correction of variables like the is suggested for experimental and simulation purposes
internal resistor, capacitors, and current. Simulations to model the behaviour of the high-power lithium-ion
are carried out and the results are verified to make sure battery. The dynamic system of the Thevenin model
the proposed method works perfectly for the accurate describes the fixed parameters as variables that vary
estimation of the SOC of the battery. The state of with the state of charge and temperature, enabling a
charge of the high-power lithium-ion battery is more accurate description of the performance of the
defined as the remaining capacity of the battery and is battery. This method was adopted because the
written mathematically as, the ratio of remaining topology generally proposes a tradeoff between
capacity to the maximum available capacity and battery cell computational requirements and the
expressed as shown in Eq. (1). approximation of voltage precision. The model is as
Qt shown in (figure 1).
SOCt = 100%
Q0 (1)

It R1 R2

U1 U2
R0
+
Uoc C1 UL
C2
-

Figure 1: The 2RC Thevenin Equivalent Circuit Model

Where Uoc is the Open-Circuit Voltage, R0 is the 

U = U SOC − i t R − U − U
oc ( ) () 0 1 2
ohmic resistance of the battery, It is the
 L
charge/discharge current flowing from the voltage 
 dU1 U i (2)
source and UL represents the terminal voltage of the  =− 1 +
 dt R C C
battery. R1 and R2 denote the electrochemical 1 1 1
 dU 2 U2 i
polarization resistance and concentration polarization  =− +
resistance, respectively. C1 and C2 denote the 
 dt R2 C2 C2
electrochemical polarization capacitance and Where [SOC, U1, U2] is selected as the state
concentration polarization capacitance respectively. variables which need to be realized and after
2.1.2 State-space description discretization of Eq. (2) and considering the definition
The state-space representation is a mathematical of SOC as stated earlier, its state space equation can
model of a physical system as a set of input, output, be written as shown in Eq. (3)
and state variables related to the first-order   SOCk +1  1 0 0 
differential equations or difference equations.   
Concerning (figure 1) and the application of  U1, k +1  = 0 1 − T  1 0 
   1 − T  2 
(3)
Kirchhoff's law, Eq. (2) is obtained and written as: U 2, k +1  0 0
U
 L , k +1 = U oc ( SOC , k + 1) − U1 − U 2 − IR0

In the above equation, parameters that the model

needs to identify include ohmic internal resistance Ro,
open circuit voltage Uoc, polarization internal
resistance R1, R2, and polarization capacitor C1 and C2

Philadelphia, USA 251

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

which will lead to identifying U1 and U2. appropriate parameters are identified. This can be
2.2 Iterative calculation algorithms achieved as shown in Eq. (7).
2.2.1 Extended Kalman filtering
The extended Kalman filter algorithm is used for
( )
 Xˆ k = Xˆ k− + Kt U L , k − CXˆ k−
 (7)

predicting the future state of a system based on a ˆ ˆ− ˆ−
 Pk |k −1 =Pk |k −1 − Kt CPk |k −1

previous state. Kalman filter is a linear unbiased
The moment of state estimation xk and Pk|k-1 is
recursive filter, which is constantly "predicted and
realized after this and the Kalman filter algorithm is
corrected" in the calculation process. Whenever new
completed in one iteration, and an iterative estimation
data is observed, new predicted values can be
is carried out for each observation, with good real-
calculated at any time, which is very convenient for
time performance.
real-time processing. Due to the discharge rate,
2.2.2 Adaptive extended Kalman filtering
temperature, and complex internal chemical reaction,
algorithm
the battery presents a nonlinear state. Based on
Currently, the SOC estimation method is mainly
Kalman, the Jacobian matrix is obtained by using the
based on the equivalent model combined with the
Taylor formula for linearization, and the extended
Kalman filter algorithm and its extended algorithm as
Kalman filter algorithm is obtained. The extended
well as fuzzy logic and neural network-related
Kalman filter consists of two equations; state
algorithms. The Kalman filter (KF) algorithm is one
equations and observation equations, which include A,
of the most widely used intelligent algorithms, and it
B, C, and D matrices that can be realized using Ro, R1,
is usually used in practical situations, such as path
R2, C1, and C2. xk is the system state matrix that
planning, target tracking, and SOC estimation of
captures the system dynamics and one of the matrix
lithium-ion batteries. The basic principle of the
values represents SOC. The input of the system is uk
algorithm is to take the minimum mean square error
which is a control variable matrix that is known or can
as the best estimation criterion, and by establishing a
be measured. wk, is the noise of the system state
state equation and an observation equation model, a
equation. Eq.. (4) shows the state equation and the
state-space model of signals and noise is used to
observation equation.
introduce the relationship between the state variables
 X k = AX k −1 + BI L , k −1 + w k and the observed variables. Time estimates and
 (4)
U L , k = CX k + DI L , k −1 + v k observations of the current time, update the estimates
Where, xk represents the system state variable at of the state variables. The Kalman filter is a
time k, yk is the system observed variable at time k, uk mathematical function that provides estimating states
is the system input which is used as the control with iterative steps, in a way to minimize the mean
variable; Ak is the transfer matrix of state x from k-1 to squared error. This technique has been providing
k, and Bk is the input matrix. Ck is the measurement performance efficiency in the field of parameter
matrix, Dk is the feedforward matrix; ωk is the noise of estimation and state transition. The improved
the system state equation. Kalman filter algorithm is Thevenin equivalent circuit model of the lithium-ion
used for state prediction and estimation. Kalman filter battery can be simplified as shown in Eq. (8).
is mainly composed of five equations, which can be  x =f ( x , u ) + w ......=A x +B u + w
 k +1 k k k k k k k k
divided into the prediction stage and correction stage. 
The recursive relationship between the estimated  yk = h( xk , uk ) + vk ....... = Ck xk + Dk uk + vk (8)
value of state and covariance in the prediction stage
(Time Update): is shown in Eq. (5). The functions f (*) and h (*) are nonlinear
 ˆ− ˆ
 X k =X k −1 + BI L, k −1 + w k (5)
equations and the first equation is the state equation,
 − where xk is the n-dimensional system state vector at
ˆ ˆ
 Pk |k −1 =APk −1 A + Qw

T
time point k, and v is the n-dimensional system noise
According to the model, the last moment of the vector. The function f (xk, uk) is a non-linear state
state estimate of k-1, Xk-1 and its covariance matrix transition function. The second equation is an
Pk-1 directly calculates the forecast of this moment, Xk- observation equation, where y is an observation
and its covariance matrix Pk|k-1. Qw is the covariance vector, and v is a multi-dimensional system
matrix of process noise wk. The estimated values of interference vector at time point k. The function h (xk,
Kalman gain are calculated as shown in Eq. (6). uk) is a non-linear measurement function. The above
function can be explored by using the Tailor method
(
K k = Pˆk−|k −1C T CPˆk−|k −1C T + vk ) (6) on the prior estimation point xk of the state xk + 1. The
higher-order components of the process can be
The state correction stage is then performed for ignored, and linear approximations of off (*) and h (*)
further computations to arrive at an appropriate can be used as shown in Eq. (9).
equation that can be used to effectively make sure the

Philadelphia, USA 252

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

 f ( xk , uk )
 f ( xk , uk )  f ( x k|k −1 , uk ) + ( xk − x k|k −1 )
 xk xk = x k|k −1
 (9)
 h( xk , u )
 k
 h ( x , u )  h ( x k|k −1
, u ) + ( xk − x k|k −1 )
xk
k k k
 xk = x k|k −1


The estimation process of the Kalman filter xk = xk |k −1 + K k ( yk − h( xk |k −1 , uk )) (14)

algorithm includes a time update and measurement
update. The time update process is also known as the 6. The state covariance matrix consists of the
forecast process. It is a one-step prediction of the variances associated with each of
current state variable and provides a prior estimation the state estimates obtained and the correlation
process for the next moment. The measurement between the errors in the state estimates. The
update process is the process of feeding back update state covariance matrix is given as:
observations and correcting deviations. The EKF Pk =(I − K k H ) Pk |k −1 (15)
algorithm equations are as follows. In the above equations, xk|k-1is the direct time
1. The filter initial conditions given in terms of the estimate at time k, xk-1 is the optimal estimate state
one-step prediction means the first state value at the last moment. Pk is the covariance update
prediction has the same statistics as the initial of xk, Qk is the covariance of process noise w, Kk is the
condition of the system. The initial condition of Kalman gain coefficient. Rk is the covariance of
the filter equation is given as: observation noise v. Since the covariance matrix Pk is
decomposed, at least it is guaranteed that Pk is always
x 0
= E ( x), P 0
= Var ( x) (10) non-negative definite, which can overcome the filter
divergence caused by the limited word length of the
2. When EKF is used to estimate the SOC of the computer. Sage-Husa algorithm adaptively updates
lithium-ion battery, the SOC is a component of the noise variables by comparing the final and initial
the state vector, the current is used as control estimated values. The estimator-related quantities are
quantity in the input parameters and the output is calculated as shown in Eq. (16).
terminal voltage. The state vector estimation time
update is given as: 
xk |k −1 = f ( xk −1 , uk −1 ) (11)  yk = yk − h( xk , uk ) − Rk -1

The state covariance time update process predicts  1 k −1
Qk =  ( K k yk yk K k + Pk − FPk |k −1 F )
3. T T T (16)
the current state variables by updating them and
providing a prior estimate of the next time. State  k i =0
 1 k −1
 Rk =  ( yk yk − HPk / k −1 H )
covariance update time update is given as: T T

Pk |k −1 = FPk −1 F T + Qk (12)  k i =0
4. The Kalman gain is the relative weight given to
the measurements and current state estimate and To make the estimation of noise more accurate
can be manipulated to achieve a particular and to avoid the influence of the observed value on the
performance. The calculation of the Kalman gain estimated value, this paper considers the noise at the
coefficient is given as: previous moment and the current moment at the same
time. In practice, the smaller the value of b, the smaller
Kk = Pk |k −1 H T ( HPk |k −1 H T + Rk ) (13) the impact at the previous moment; if the value of b is
5. The measurement update process, also known as small, the estimated noise will oscillate, so it can be
the correction process, is a process of feedback on determined according to the specific situation. Then
the observed values and the correction of the the calculation of the noise matrix is as shown in Eq.
deviation. The state vector measurement update is (17).
given as:

Qk = (1 − d k −1 ) Qk −1 + d k −1 ( K k yk ykT K kT + Pk − FPk / k −1 F T )

(17)

 Rk = (1 − d k −1 ) Rk −1 + d k −1 ( yk yk − HPk / k −1 H )
T T

2.2.3 Iterative calculation process Thevenin equivalent circuit model, the state space
After obtaining the main parameters in the 2RC equation is obtained using the relationship between

Philadelphia, USA 253

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

voltage and current as shown in Eq. (18). In performing the mathematical calculation for
the algorithm, there are a few steps to follow to
 achieve this. The first step is known as the state
 E (t) = U + R I (t ) + u (t ) prediction stage. The predicted value at time k is
 L 1
calculated as shown in Eq. (22).
 u (t ) du
 I (t ) = +C x(k|k − 1) = Ak −1 x(k − 1) + Bk −1ik −1
R2 dt (18) (22)
 The second step in this calculation is the
 1 t
 SOC (t ) = SOC (t0 ) − t  i(t )dt prediction of the covariance matrix. This is done by
 Q0 0 calculating the estimation error of x(k|k-1), the
The three equations above are combined and covariance matrix of the corresponding x(k|k-1) is
discretized to obtain the state equation needed to obtained as shown in Eq. (23).
perform further calculations leading to the conversion ˆ AT + Q
P(k | k − 1) = Ak −1 P (23)
k −1 k −1 k
of non-discretized parameters into discretized ones as The Kalman gain is calculated in the third step to
shown in Eq. (19). further improve the computation and arrive at the
value for a specific time. The Kalman gain at time k is
 obtained as shown in Eq. (24).

 K k = Pk Ck T (Ck Pk Ck T + Rk )−1 (24)
 x(k |k − 1) = Ak x(k − 1) + Bk ik −1 + wk The fourth step is the update of the status. The
 optimal estimated value of the existing state is
 1 0  estimated from the real-time measured/obtained open-
 Ak =  − t / 
 0 e  circuit voltage value Uoc(k) as shown in Eq. (25).
  t  (19)
Xˆ k = x(k|k − 1) + K U oc (k ) − Ck * x(k |k − 1) (25)

=  −Q  k

 k 
B 0  In the fifth step, the noise covariance is updated.
  R (1 − e−t / )  The noise covariance is updated according to the
  2 
Kalman gain. This helps to avoid errors or at least
It is important to also consider the observation minimize them and leads to accurate estimation. The
equation for further successful computation and noise covariance of the previous moment is shown in
attainment of accurate results. The next thing to do Eq. (26).
therefore is the observation equation which is shown Pˆk = (1 − Kk Ck ) Pk (26)
in Eq. (20).
In the calculation, the five steps are continually
cycled in a loop, and the estimated state is
yk = h ( xk −1 , ik −1 ) + vk = U oc − R1ik − uk + vk (20) continuously updated so that the estimated value is
closer to the true value during the update process. The
The parameters of the battery are initially non- basic iterative calculation process for SOC is as shown
linear variables and have to be linearized. After in (figure 2).
linearization, using the first-order Taylor series with
the values of Ak, Bk, and Ck, the formulas are obtained
as shown in Eq. (21).

 1 0 
 Aˆk = 
− t /
 0 e 
  t 
  −
 Bˆk = Q0  (21)
 
  R (1 − e−t / ) 
  2 
  u 
Ck =  oc , −1
  soc 

Philadelphia, USA 254

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

Begin initial condition

State vector estimation

time update

State covariance
update time update Update the
process noise
Calculate Kalman
gain coefficient

State vector
measurement update

Update state
covariance matrix

Figure 2: The SOC estimation process using the AEKF algorithm

2.3 Model building and Realization the codes for calculations based on the algorithms and
2.3.1 Simulink/MATLAB Simulation model to prevent the battery from overcharge and over-
After obtaining the required circuit model discharge. In this work, the influence of temperature
parameters, the simulation model of the lithium-ion change on the output voltage of the lithium-ion battery
battery is established in Simulink/MATLAB. The is ignored. For the time-domain ordinary differential
simulation model has mainly composed of the SOC equation of the 2RC Thevenin equivalent circuit, the
calculation module, the circuit parameter updating corresponding voltage response equation is solved and
module, and the terminal output voltage calculation discretization as it is required before modeling to
module. The SOC calculation module is based on the obtain the discretization state-space equation of the
extended Kalman filtering algorithm and the model. The logical structure of the circuit model is
Improved Adaptive extended Kalman filtering shown in (figure 3).
algorithm. SOC values are obtained through the use of

SOC Voltage

Current Battery Model

Power
Temperature
Figure 3: Logical structure of the circuit model

In the battery management system, both current circuit structure is directly used to build the module,
and terminal voltage are input, as the battery including an ohmic internal resistance, two RC
simulation model is to verify the accuracy of the parallel structures, a controllable voltage source, and
parameter settings in the model, the current is input controllable current source, voltage, and current
and terminal voltage is the output. The simulation sensor, and input and output interface, as shown in
module can be built and the second-order RC internal (figure 4).
circuit is the core part of the whole module, and the

Philadelphia, USA 255

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

Figure 4: The Simulation Model of the 2RC Thevenin equivalent model

(Figure 5) shows the internal component of the inputs and outputs are labeled and other components
model in (figure 4) representing the proposed 2RC are duly presented in the figure based on the proposed
Thevenin equivalent circuit model. The necessary model.

Figure 5: The internal circuit of the 2RC Thevenin equivalent model

Controllable voltage source and controllable transducers, converting physical interfaces into signal
current source are the signal interface in SIMULINK, interfaces. The extended Kalman filtering algorithm
which can turn the signal into a material port. Convert and the Adaptive extended Kalman filtering algorithm
the external input voltage source into a voltage and are coded into the simulation model and the complete
current source that the circuit can connect to. Voltage diagram is shown in (figure 6).
sensors and current sensors are also signal

Philadelphia, USA 256

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

Figure 6: Integration of algorithms into the simulation diagram

3. Experimental Analysis
3.1 Test platform construction
The GTK 3.7V 40Ah high-power lithium-ion
battery was used for this experiment with a rated
capacity of 40 Ampere hour (Ah), a charge cut-off
voltage of 4.2V, and a discharge cut-off voltage of
2.75V. The test equipment is the sub-source BTS 750-
200-100-4, with a maximum charge-discharge power
of 750W, a maximum voltage of 200V, and a
maximum current of 100A. The basic properties of
the battery are shown in table 1.

Table 1: Basic technical parameters of the battery

Factor Specification
Size: length * width * 148×27×92
height/ mm
Rated voltage/V 3.7
Maximum load current /A 5C
Rated capacity/Ah 40
Charge cut-off voltage/V 4.2
Discharge cutoff voltage/V 2.75

Setting up the experiment requires the battery for the experiment to commence. The software
connection of the battery to the test machine which is on the computer is then programmed to follow a
also connected to a computer. A specific terminal is logical algorithm to accomplish the task.
chosen from the 16 available, and connect to the

Philadelphia, USA 257

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

BTS 750-200-100-4 Computer system

Program settings on
computer screen

Lithium-ion Battery
Experimental Data

Figure 7: Experimental Setup

3.2 Results, verification, and comparison to measure the amount of energy stored in the battery.
3.2.1 Parameter identification and extraction The relationship between the battery OCV and the
The Open-Circuit Voltage test: battery SOC can be attained through this experiment.
The open-circuit voltage (OCV) of the battery is There are a few ways to obtain the open-circuit
the stable voltage value of the battery when the battery voltage of a battery and include the stationary method,
is left in the open circuit condition. The test was also known as the direct method which is relatively
performed on the Lithium-ion battery to acquire data more accurate.
for parameter identification. Using the Ampere hour The program that controls the experiment which
(Ah) method, the real capacity of the battery is resides on a computer connected to the battery testing
determined and used in the programming during the machine (BTS 750-200-100-4) is then set to follow
experiment. After charging the battery, the battery conditions to obtain a flow chart of the processes
terminal voltage will gradually decline to a stable involved in the experiment. The test consists of simple
value when it is left in the open circuit condition and steps and follows a logical sequence with a loop that
after discharge, the battery terminal voltage will ensures continuity and data capture at relevant SOC
gradually rise to a stable value when the load is points. The first step is the capacity test that calibrates
removed. The electromotive force of the battery is the capacity of the battery and the subsequent steps
equal to the open-circuit voltage of the battery. The capture the OCV of the battery at specific SOC points.
battery electromotive force is one of the metrics used

Philadelphia, USA 258

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

START

CCCV charging 4.2V (1C)

Shelving for 1hour

N=0

CC discharge 1C

Shelving for 40min N++

N<10&UL>3V
YES

NO

END

Figure 8: OCV experiment process

In the test, the measured voltages at the end of various levels of SOC. These values were extracted
each standby stage are regarded as the final open- and used for the curve fitting that reveals the
circuit voltage. All the values of the open-circuit relationship between OCV and SOC and further used
voltages at different SOC’s are measured and to obtain the polynomial equation for further
recorded. Table 2 shows the OCV data obtained at the calculations.

Table 2: Values from the SOC/OCV test

SOC 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1
OCV 4.1914 4.0628 3.9521 3.8529 3.7602 3.6852 3.6393 3.6188 3.5645 3.4886

The OCV/SOC values were imported and using of OCV with SOC obtained through the experimental
the curve fitting tool, a relationship was realized method is shown in (figure 9).
through the use of a polynomial fitting. The variation

4.2

4.0
OCV

3.8

3.6

3.4
0.0 0.2 0.4 0.6 0.8 1.0
SOC
Figure 9: OCV/SOC Curve fitting

Philadelphia, USA 259

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

The capacity and Hybrid Pulse Power discharge capacity of each measurement not
Characterization (HPPC) Test: to exceed 2%, and then the measured
The experiment is performed as a method for capacity is deemed to be the actual capacity
parameter identification of the Thevenin model and of the battery.
then data from the test conducted is analyzed and used 2. The battery is left in the open-circuit
in equations to calculate the parameters. Taking the condition to rest for 40 minutes to achieve
GTK 3.7V 40Ah high-power lithium-ion battery as electrochemical and heat equilibrium. After
the research object, the battery test equipment was is performing step (1) on the lithium-ion
BTS750-200-100-4. The HPPC experiment was battery, SOC = 1.
conducted on the lithium-ion battery according to the 3. Discharge the experimental battery at a
"American Freedom CAR battery experiment constant current of 1C for 10s, leave it for
manual". The single HPPC working step was to take 40s, and charge at 1C for 10s, then leave it.
1C (current 40A) constant current 10s, shelve for 40s, 4. Discharge the battery with a current value of
and 1C (current 40A) constant current charge 10s, and 40A for 6 minutes to decrease the battery
then shelve. The Capacity and HPPC experiments SOC to the next SOC point and leave it for
performed on the high-power lithium-ion battery were 30 minutes.
conducted according to the following steps. 5. Repeat steps (3) (4) 9 times to obtain the
1. The battery is fully charged using a constant complete data for the test.
current of 1C followed by a constant voltage The HPPC test is carried out using simple steps
of 4.2V. Then, the battery is discharged with in order to obtain the necessary values needed for
a constant current to its discharge cut-off parameterization and further computation as far as the
voltage of 2.75V. The experiment is repeated research is concerned. The experimental process of
three (3) times for the difference between the the HPPC test can be seen in (figure 10).

START

N=0

CCCV charging
till fully charged

Shelve for
40min

CC discharge
1C for 10s

Shelve for 40s

CC charge 1C N++
for 10s

Shelve for
5min

CC charge 1C
for 6min

NO
N=9

YES

END

Figure 10: HPPC test process

Through parameter identification, the functional is completed. Through direct comparison of the
relations of resistance, capacitance, voltage, and SOC curves, the error in parameter identification can be
were obtained. Then, the circuit module was built in observed and manipulated by changing the value of
Simulink/MATLAB. The module contains other parameter input, and the optimal simulation model can
submodules that assist in the processing, calculation, be obtained by modifying the functions.
and optimization of the interface for better simulation. The experimental results from the OCV and
The EKF and the improved AEKF algorithms were HPPC tests are used as the basis for parameterization.
incorporated in this module as well. The parameters The test was conducted at 0.1 SOC intervals from 1.0
identified in the experiment and subsequent use in to 0.1. When a current I is loaded, a voltage rise or
SOC estimation are compared with the results from drop appears when there is a pulse charge or discharge
the simulated. This can be obtained by simulating the and this can be used to calculate the parameters.
construction and the design of function to operate on (Figure 11) shows the test data for SOC at 0.9.

Philadelphia, USA 260

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

U6

4.05 U5

U7

U (V)
4.00 U0
U4

U3
3.95
U1

U2
3.90

0 20 40 60 80
t (s)
Figure 11: Voltage curve for a specific SOC

The labels U0 to U7 represent values for various make it stable and achieve possible desired outcomes.
segments of the curve corresponding to the battery Whenever the potential of an electrode is forced away
during the experiment. U0-U3 and U4-U7 show the from its value at the open-circuit, R1 is calculated.
discharge characteristics and charging characteristics This causes current to flow
of the battery respectively. through electrochemical reactions that occur at the
1. The U0-U1 segment shows the ohmic internal electrode surface. The value of R1 can be determined
resistance Ro, which depicts the rapid drop of the using the voltage response of the battery cell to a
voltage at the moment of discharge of the battery. discharging current pulse and voltage and can be
2. U1-U2 shows the steady drop in terminal voltage calculated using the difference in voltage (U) over the
due to the polarization capacitor which is due to current I. The build-up of solutes on the membrane
the polarization effect of the battery as it is being surface due to convective-diffusive transport in the
charged. This polarized capacitor is the zero- boundary layer which can be linked to R2 can be
state response of the 2RC series loop. determined and calculated using the difference in
3. U2-U3 shows the resting of the battery and the voltage (U) over the current I. A capacitor is formed
disappearance of the load current, the ohmic when two conducting plates are separated by a non-
voltage disappears, and the rapid rise of the conducting media called the dielectric. The value of
terminal voltage. the capacitance depends on the size of the plates, the
4. U3-U4 shows the steady rise of the terminal distance between the plates, and the properties of the
voltage as the polarization capacitance dielectric. The part of the polarization of the
discharges through the polarization resistance, battery resulting from changes in the electrolyte
forming a zero-input response of the 2RC series concentration due to the passage of current through
loop. the electrode C2 is determined.
The opposition to the flow of the current offered With these equations, the information can be
by the battery itself resulting in the generation of heat used to derive values of the parameters. This data can
is measured in Ohms. Based on the voltage drop from then be simulated in Simulink MATLAB for
U0 to U1 for each of the SOC’s, the ohmic resistance verification and validity of the values and the
R0 can be deduced. The value of (R1, R2, C1, C2) as modeling of the 2RC Thevenin equivalent circuit.
seen in the Thevenin equivalent circuit diagram is to Table 3, shows the data acquired after the calculations.

Table 3. The model Parameter calculation result

SOC OCV/V Ro/mΩ R1/mΩ R2/mΩ C1/103 F C2/103 F
1.0 4.1914 0.0324 0.0642 0.0578 20.3215 19.2359
0.9 4.0628 0.0307 0.0532 0.0635 20.7558 19.2560
0.8 3.9521 0.0474 0.0508 0.0657 21.3673 20.2565
0.7 3.8529 0.0474 0.0635 0.0732 21.6135 20.6354
0.6 3.7602 0.0542 0.0676 0.0770 22.6417 19.6580
0.5 3.6852 0.0568 0.0730 0.0793 23.0131 22.0827
0.4 3.6393 0.0603 0.0725 0.0840 22.1528 21.8318
0.3 3.6188 0.0603 0.0832 0.0872 22.3564 20.2624
0.2 3.5645 0.0604 0.0841 0.0765 21.0251 20.4528
0.1 3.4886 0.0661 0.0762 0.0818 23.2540 20.5187

Philadelphia, USA 261

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

The values of the internal Ohmic resistance R0, electrochemical polarization capacitance C1 and the
the electrochemical polarization resistance R1, and concentration polarization capacitance C2 calculated
concentration polarization resistance R2 calculated from the 2RC Thevenin equivalent circuit are
from the proposed 2RC Thevenin equivalent circuit compared in (figure 12(b)).
are compared in (figure 12(a)). The values of the
0.10
Ro 23 C1
R1 C2
0.08
R2
22
Ro (Ω)

C (F)
0.06
21

0.04 20

0.2 0.4 0.6 0.8 1.0 19

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
SOC% SOC%
(a) Resistance identification result (b) Capacitance identification result
Figure 12: comparison of the identified parameters

According to the result, it can be observed from 3.2.2 Capacity test result
(figure 12(a)) that, the internal resistance R0 decreases The capacity experiment was conducted for
steadily as state-of-charge increases. The calibration of the capacity, current, energy, and
electrochemical polarization resistance R1 first voltage of the battery. According to the capacity
increases and then decreases with an increasing SOC experiment, these various parameters can be deduced
and then increases again as SOC approaches the and compared with the information provided by the
highest point. The concentration polarization manufacturer to compare and see if the experiment
resistance R2 first decreases then increases sharply and was successful or not. (Figure 13(a)) is the Capacity
gently decreases as SOC increases. From (figure variation curve with time and it can be deduced from
12(b)), the results show the variation in the curves of the figure that the Capacity of the battery is
the two capacitances are somewhat similar and seem approximately 40Ah. Three maximum values
to be fluctuating as SOC increases and finally obtained in the experiment are 39.1360Ah,
decreases when SOC is above 0.7 till it reaches the 39.2072Ah, and 39.2323Ah.
highest point.
40 39.1360Ah 39.2072Ah 39.2323Ah 150000 139951.2741mWh 139879.0435mWh
139534.5534mWh

125000
30
E (mWh)
C (Ah)

100000
20 75000
50000
10
25000
0 0
0 10000 20000 30000 40000 50000 60000 0 10000 20000 30000 40000 50000 60000
t (s) t (s)
(a) Capacity variation curve (b) Energy variation curve
20 20.0086A 4.1998V

4.0
I (A)

0
U (V)

3.5

-20
3.0

2.7495V
-40 -40.0049A
2.5
0 10000 20000 30000 40000 50000 60000 0 10000 20000 30000 40000 50000 60000
t (s) t (s)
(c) Current variation curve (d) Voltage variation curve
Figure 13: capacity test results

Philadelphia, USA 262

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

(Figure 13(b)) shows the energy variation curve shows the voltage variation curve and displays the
and displays three maximum energy values obtained maximum and minimum values obtained in the
in the experiment as 139.534Wh, 139.951Wh, and experiment as 4.1998V which is approximately the
139.879Wh. The energy of the battery is therefore maximum voltage of the battery stated as 4.2V and
approximately 140Wh. The energy variation curve 2.7495V approximately 2.75V stated earlier as the
fluctuates in the whole process since its main purpose minimum voltage of the battery.
is to charge the lithium-ion battery when the constant- The comparison of the current and voltage curves
current charge-discharge interval is carried out. from the experiment and how the variation is depicted
(Figure 13(c)) shows the current variation curve of the to achieve the capacity of the battery. The current flow
capacity test and displays the maximum and minimum and the voltage can be seen and analyzed from (figure
current values obtained in the experiment as 14).
20.0086A, and -40.0049A respectively. (Figure 13(d))

20
4.0

I (A)
U (V)

3.5

-20
3.0
I
U
-40
2.5
0 10000 20000 30000 40000 50000 60000
t (s)
Figure 14:Voltage/Current discharge variation curve

3.2.3 HPPC test results current curves with time respectively and illustrates
The result from the HPPC test was used to the mechanism of the HPPC experiment throughout
identify each OCV at specific SOC points. The the test. The figure also emphasizes one of the SOC
voltage curve is also useful in the identification of points for reference and shows the single charge and
battery parameters. (Figure 15(a)) and (b) shows the discharge variation in the experiment.
schematic diagram of the HPPC terminal voltage and

40

4.0
0
I (A)
U (V)

3.5
-40
40

3.7
20

3.0 -80 0
3.6

-20

3.5

2.5
19640 19660 19680 19700 19720
-120 -40

19640 19660 19680 19700 19720

5000 10000 15000 20000 25000 5000 10000 15000 20000 25000 30000
t (s) t (s)
(a) Voltage variation curve (b) Current variation curve
Figure 15: HPPC test result

From the result, it can be observed that as the shows the comparison of the voltage and current
number of cycles increases, the battery terminal curves from the HPPC experiment and the variation
voltage assumes a downward trend. The current with time. The overlaying curves shown in the figure
variation reveals that, as the number of cycles depict the real or actual occurrence with the battery at
increases, the discharge current increases. (Figure 16) specific times.

Philadelphia, USA 263

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

60
U
I
4.0 40

I (A)
20

U (V)
3.5

0
3.0
-20

2.5 -40
5000 10000 15000 20000 25000
t (s)
Figure 16: Voltage/Current discharge variation curve

3.2.4 SOC estimation result gradual decrease and fluctuating trend which is
The experimental results and parameterization because there is a process of alternating charge and
did then used for SOC estimation. (Figure 17) shows discharge during the experiment with the discharge
the estimation depicting a downward slope with time time longer than the charging time.
as SOC reduces from 1 to 0.1. The result shows a

1.0

0.8
SOC (%)

0.6
0.72
0.4 0.70

0.68

0.2 0.66

0.64
5300 5400 5500 5600 5700 5800 5900
0.0
0 2000 4000 6000 8000 10000 12000
t (s)
Figure 17: SOC variation curve

3.3 Verification of research results experiment and the computations done thereafter. The
3.3.1 Voltage characteristics value of current (I) in the experimental data obtained
The 2RC Thevenin equivalent circuit model is by the test equipment is taken as the input condition,
established through parameterization with the HPPC and the simulation terminal voltage is obtained
experimental results. To verify the validity of the SOC through the simulation model and the experimental
estimated values in the simulation, the results are terminal voltages are compared to obtain the results as
compared with and the results from the HPPC shown in (figure 18).

Philadelphia, USA 264

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

U1
4.0 U2

U (V)
3.5

3.0

2.5
5000 10000 15000 20000 25000
t (s)
Figure 18: Comparison of voltage variation curves

Where U1 is the change curve of real terminal designed and the effectiveness of the algorithm for
voltage data obtained through the HPPC test while U2 accurate SOC estimation was compared with the
is the output terminal voltage curve obtained through extended Kalman filtering algorithm. SOC estimation
the simulation model. The figure shows the variation results from the proposed model and the use of the two
trend of both the experiment and simulation curves is algorithms; the extended Kalman filtering algorithm
similar to that of the actual test curve. This means that and the adaptive extended Kalman filtering algorithm
the results can be verified and authenticated as are then compared to assess and verify the validity and
appropriate for use in any calculation toward the efficient performance of the model and algorithms
accurate estimation of SOC. implemented. The three methods are compared as
3.3.2 Comparison and Verification of SOC shown in (figure 19). It can be observed from the
estimation result figure that the three methods all follow the same trend
To illustrate the adaptability and performance of and have a good convergence. The same figure also
the improved adaptive Kalman filtering algorithm, for shows the difference in error margin comparing the
SOC estimation, Beijing Bus Dynamic Stress Test use of the EKF and the AEKF estimation curves with
(BBDST) working condition experiment was that of the true SOC curve.

1.0 SOC1 0.04

SOC2
SOC3 0.03
0.8
SOC (%)

SOC (%)

0.02
0.6
0.71
0.01
0.70

0.4 0.69

0.68 0.00 E1
0.2
0.67 E2
5000 5100 5200 5300 5400
-0.01
2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000
t (s) t (s)
(a). comparison of SOC estimation curves (b). SOC estimation Error
Figure 19: SOC variation curves for three different methods

(Figure 19(a)) is the result of SOC estimation for less than 1.85% and has a strong correction function.
three different methods. SOC1 is the true SOC value, In recent years, researchers have proposed some
and SOC2 is the SOC estimate using the extended new estimation methods to help correct the errors with
Kalman algorithm and SOC3 is the SOC estimation especially the traditional SOC estimation methods.
using the adaptive extended Kalman filter. (Figure The use of these improved and adaptive methods
19(b)) is the error curve obtained through the would reduce the estimation errors as much as
difference in SOC values of the two algorithms. The possible and improve the estimation accuracy. In the
error of SOC estimation using the extended Kalman is actual battery management system, the methods to
around 4.97%, and the error of SOC estimation using estimate the SOC of the battery are all traditional, and
the improved adaptive extended Kalman algorithm is most of these new methods are only in the theoretical

Philadelphia, USA 265

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

research or simulation stage. compared to the extended Kalman algorithm. The

4. Conclusions and Further Research Plan results are of great instructional significance to the
In the quest to improve the accuracy and application in practical control systems for the
reliability of the lithium-ion battery state of charge equivalent circuit modeling of batteries. The improved
estimation in this study, the 2RC Thevenin model was adaptive extended Kalman filter algorithm has good
established and the parameters were obtained. This convergence speed, higher estimation accuracy, and
model was used due to its advantages of low error, stability and is appropriate and convenient for SOC
long-term testing, and accounting for polarization estimation. Therefore, for stringent applications such
effects and transient analysis for power battery as automotive and high-power demanding devices, the
charging and discharging. The steps used in the second-order RC model could be the preferred choice,
experiment were simple and convenient, and the and the use of an adaptive algorithm for SOC
algorithm complexity, very moderate making the estimation would be better for accurate and timely
experimental results obtained accurate and information.
appropriate for parameterization. The maximum For further and future research, it is important to
relative errors of the 2RC Thevenin equivalent models perform several experiments to acquire necessary
using both algorithms were all less than 4.97%, which experimental data for analysis. It is also important to
can generally satisfy the precision requirements for use several methods and techniques in the estimation
practical engineering calculation, such as algorithms of SOC. The use of algorithms is also an important
based on ECM for advanced BMSs. The choice of an aspect of research in this direction because it helps to
adaptive law for the process noise covariance matrix improve the accuracy of SOC estimation. These
shows an improvement in estimation performances. In algorithms have been improved over the years and
terms of maximum estimation error, the EKF result more improvement and research leading to the
was 4.97%, while with AEKF this range reduced up to establishment of more effective and efficient
1.85%. The extended Kalman filter algorithm used algorithms must be promoted to better the function of
effectively reduces the influence of nonlinear Battery management systems for estimations.
equations and successfully realizes the SOC
estimation of the battery. The improved adaptive Acknowledgments
extended Kalman filter algorithm for SOC estimation The work was supported by the National Natural
used in this work can effectively and accurately Science Foundation of China (No. 61801407), and China
estimate the battery SOC and has high precision Scholarship Council (No. 201908515099)

References:

1. Yang, J., et al. (2018). Online state-of-health of electric vehicles. Frontiers of Mechanical
estimation for lithium-ion batteries using Engineering, 14(1): pp. 47-64.
constant-voltage charging current analysis. 6. Cacciato, M., et al. (2017). Real-Time Model-
Applied Energy, 212: pp. 1589-1600. Based Estimation of SOC and SOH for Energy
2. Hesse, H.C., et al. (2017). Lithium-ion battery Storage Systems. IEEE Transactions on Power
storage for the grid—A review of stationary Electronics, 32(1): pp. 794-803.
battery storage system design tailored for 7. Beirão, M.D., et al. (2016). Balancing
applications in modern power grids. Energies, management system for improving Li-ion
10(12): p. 2107. batteries capacity usage and lifespan. in 2016
3. Ali, M.U., et al. (2019). Towards a smarter IEEE 16th International Conference on
battery management system for electric vehicle Environment and Electrical Engineering
applications: A critical review of lithium-ion (EEEIC).
battery state of charge estimation. Energies, 8. Ali, U.M., et al. (2019). Towards a Smarter
12(3): p. 446. Battery Management System for Electric
4. Lai, X., Zheng, Y., & Sun, T. (2018). A Vehicle Applications: A Critical Review of
comparative study of different equivalent circuit Lithium-Ion Battery State of Charge Estimation.
models for estimating state-of-charge of lithium- Energies, 12(3).
ion batteries. Electrochimica Acta, 259: pp. 566- 9. Feng, Y., et al. (2016). Intelligent battery
577. management for electric and hybrid electric
5. Liu, K., et al. (2019). A brief review on key vehicles: A survey. in 2016 IEEE International
technologies in the battery management system Conference on Industrial Technology (ICIT).

Philadelphia, USA 266

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

10. Tang, X., Liu, B., & Gao, F. (2017). State of 24. Writer, B. (2019). Models, SOC, Maximum,
Charge Estimation of LiFePO4 Battery Based on Time, Cell, Data, Parameters, in Lithium-Ion
a Gain-classifier Observer. Energy Procedia, Batteries: A Machine-Generated Summary of
105: pp. 2071-2076. Current Research, B. Writer, Editor. Springer
11. Wei, X., Yimin, M., & Feng, Z. (2019). Lithium- International Publishing: Cham. (pp. 195-247).
ion Battery Modeling and State of Charge 25. Serhan, H.A., & Ahmed, E.M. (2018). Effect of
Estimation. Integrated Ferroelectrics, 200(1): the different charging techniques on battery life-
pp. 59-72. time: Review. in 2018 International Conference
12. Rivera-Barrera, P.J., Muñoz-Galeano, N., & on Innovative Trends in Computer Engineering
Sarmiento-Maldonado, O.H. (2017). SoC (ITCE).
Estimation for Lithium-ion Batteries: Review 26. Jarraya, I., et al. (2019). An online state of charge
and Future Challenges. Electronics, 6(4). estimation for Lithium-ion and supercapacitor in
13. Venugopal, P. (2019). State-of-Health hybrid electric drive vehicle. Journal of Energy
Estimation of Li-ion Batteries in Electric Vehicle Storage, 26: p. 100946.
Using IndRNN under Variable Load Condition. 27. Barai, A., et al. (2015). A study of the open
Energies, 12(22). circuit voltage characterization technique and
14. Blaabjerg, F., Dragicevic, T., & Davari, P. hysteresis assessment of lithium-ion cells.
(2019). Applications of Power Electronics. Journal of Power Sources, 295: pp. 99-107.
Electronics, 8(4). 28. Zhang, C., et al. (2016). A Generalized SOC-
15. Zhang, K., et al. (2019). Parameter Identification OCV Model for Lithium-Ion Batteries and the
and State of Charge Estimation of NMC Cells SOC Estimation for LNMCO Battery. Energies,
Based on Improved Ant Lion Optimizer. 9(11).
Mathematical Problems in Engineering, 2019: p. 29. Gao, Z.C., et al. (2017). State-of-Charge
18. Estimation and Active Cell Pack Balancing
16. Su, J., et al. (2019). An equivalent circuit model Design of Lithium Battery Power System for
analysis for the lithium-ion battery pack in pure Smart Electric Vehicle. Journal of Advanced
electric vehicles. Measurement and Control, Transportation, 2017: p. 14.
52(3-4): pp. 193-201. 30. Wang, Q., et al. (2017). Correlation between the
17. Qin, D., et al. (2019). Modeling and Simulating model accuracy and model-based SOC
a Battery for an Electric Vehicle Based on estimation. Electrochimica Acta, 228: pp. 146-
Modelica. Automotive Innovation, 2(3): pp. 169- 159.
177. 31. Yang, Q., et al. (2017). A simplified fractional
18. Tian, Y., et al. (2019). Performance analysis and order impedance model and parameter
modeling of three energy storage devices for identification method for lithium-ion batteries.
electric vehicle applications over a wide PloS one, 12(2): p. e0172424-e0172424.
temperature range. Electrochimica Acta, 2019: 32. Guangming, L., et al. (2014). A comparative
p. 135317. study of equivalent circuit models and enhanced
19. Saldaña, G., et al. (2019). Analysis of the equivalent circuit models of lithium-ion batteries
Current Electric Battery Models for Electric with different model structures. in 2014 IEEE
Vehicle Simulation. Energies, 12(14). Conference and Expo Transportation
20. Mendoza, S., et al. (2016). Optimization and Electrification Asia-Pacific (ITEC Asia-Pacific).
experimental validation of a thermal cycle that 33. Yang, J., et al. (2017). Improved Battery
maximizes entropy coefficient fisher Parameter Estimation Method Considering
identifiability for lithium iron phosphate cells. Operating Scenarios for HEV/EV Applications.
Journal of Power Sources, 308: pp. 18-28. Energies, 10(1).
21. Rothenberger, M.J., et al. (2015). Genetic 34. Xia, B., et al. (2016). Accurate Lithium-ion
optimization and experimental validation of a battery parameter estimation with continuous-
test cycle that maximizes parameter time system identification methods. Applied
identifiability for a Li-ion equivalent-circuit Energy, 179: pp. 426-436.
battery model. Journal of Energy Storage, 4: pp. 35. Jiang, B., et al. (2017). Online Reliable Peak
156-166. Charge/Discharge Power Estimation of Series-
22. Zhang, C., et al. (2018). Online estimation of Connected Lithium-Ion Battery Packs. Energies,
battery equivalent circuit model parameters and 10(3).
state of charge using decoupled least squares 36. Parvini, Y., Vahidi, A., & Fayazi, S.A. (2018).
technique. Energy, 142: pp. 678-688. Heuristic Versus Optimal Charging of
23. Xu, Y., et al. (2020). State of charge estimation Supercapacitors, Lithium-Ion, and Lead-Acid
for lithium-ion batteries based on adaptive dual Batteries: An Efficiency Point of View. IEEE
Kalman filter. Applied Mathematical Modelling, Transactions on Control Systems Technology,
77: pp. 1255-1272. 26(1): pp. 167-180.

Philadelphia, USA 267

 ISRA (India) = 6.317 SIS (USA) = 0.912 ICV (Poland) = 6.630
ISI (Dubai, UAE) = 1.582 РИНЦ (Russia) = 0.126 PIF (India) = 1.940
Impact Factor: GIF (Australia) = 0.564 ESJI (KZ) = 9.035 IBI (India) = 4.260
JIF = 1.500 SJIF (Morocco) = 7.184 OAJI (USA) = 0.350

37. Raszmann, E., et al. (2017). Modeling stationary scalable battery management system for lithium-
lithium-ion batteries for optimization and ion batteries. in 2017 Twelfth International
predictive control. in 2017 IEEE Power and Conference on Ecological Vehicles and
Energy Conference at Illinois (PECI). Renewable Energies (EVER).
38. Lin, N., Ci, S., & Wu, D. (2016). A novel low- 48. Peng, S., et al. (2017). State of Charge
cost online state of charge estimation method for Estimation of Battery Energy Storage Systems
reconfigurable battery pack. in 2016 IEEE Based on Adaptive Unscented Kalman Filter
Applied Power Electronics Conference and With a Noise Statistics Estimator. IEEE Access,
Exposition (APEC). 5: pp. 13202-13212.
39. Huria, T., Ludovici, G., & Lutzemberger, G. 49. Partovibakhsh, M., & Liu, G. (2015). An
(2014). State of charge estimation of high power Adaptive Unscented Kalman Filtering Approach
lithium iron phosphate cells. Journal of Power for Online Estimation of Model Parameters and
Sources, 249: pp. 92-102. State-of-Charge of Lithium-Ion Batteries for
40. Xiao, T., et al. (2019). Comparative Study of Autonomous Mobile Robots. IEEE Transactions
EKF and UKF for SOC Estimation of Lithium- on Control Systems Technology, 23(1): pp. 357-
ion Batteries. in 2019 IEEE Innovative Smart 363.
Grid Technologies - Asia (ISGT Asia). 50. Tian, Y., et al. (2014). A modified model based
41. Liu, F., Ma, J., & Su, W. (2019). Unscented state of charge estimation of power lithium-ion
Particle Filter for SOC Estimation Algorithm batteries using unscented Kalman filter. Journal
Based on a Dynamic Parameter Identification. of Power Sources, 270: pp. 619-626.
Mathematical Problems in Engineering, 2019: p. 51. Barai, A., et al. (2018). A study of the influence
14. of measurement timescale on internal resistance
42. Chemali, E., et al. (2018). State-of-charge characterisation methodologies for lithium-ion
estimation of Li-ion batteries using deep neural cells. Scientific reports, 8(1): pp. 21-21.
networks: A machine learning approach. Journal 52. Nejad, S., Gladwin, D.T., & Stone, D.A. (2016).
of Power Sources, 400: pp. 242-255. On-chip implementation of Extended Kalman
43. Jiménez-Bermejo, D., et al. (2018). Using Filter for adaptive battery states monitoring. in
Dynamic Neural Networks for Battery State of IECON 2016 - 42nd Annual Conference of the
Charge Estimation in Electric Vehicles. IEEE Industrial Electronics Society.
Procedia Computer Science, 130: pp. 533-540. 53. Xia, X., & Wei, Y. (2016). Lithium-ion batteries
44. Ma, Y., et al. (2016). Fractional modeling and State-of-charge estimation based on interactive
SOC estimation of lithium-ion battery. multiple-model Extended Kalman filter. in 2016
IEEE/CAA Journal of Automatica Sinica, 3(3): 22nd International Conference on Automation
pp. 281-287. and Computing (ICAC).
45. Jung-Hyo, B., & Bai, Z. (2016). The 54. Zhou, W., & Hou, J. (2019). A New Adaptive
development of technology on reduces the SOC High-Order Unscented Kalman Filter for
error rate using Hybrid Kalman Filter(HKF) Improving the Accuracy and Robustness of
and the demonstration its performance using Target Tracking. IEEE Access, 7: pp. 118484-
BMS platform. in 2016 IEEE Transportation 118497.
Electrification Conference and Expo, Asia- 55. Zhang, Z., et al. (2017). SOC Estimation of
Pacific (ITEC Asia-Pacific). Lithium-Ion Batteries With AEKF and Wavelet
46. Nejad, S., Gladwin, D.T., & Stone, D.A. (2015). Transform Matrix. IEEE Transactions on Power
Enhanced state-of-charge estimation for Electronics, 32(10): pp. 7626-7634.
lithium-ion iron phosphate cells with flat open- 56. Aung, H., Low, K.S., & Goh, S.T. (2015). State-
circuit voltage curves. in IECON 2015 - 41st of-Charge Estimation of Lithium-Ion Battery
Annual Conference of the IEEE Industrial Using Square Root Spherical Unscented Kalman
Electronics Society. Filter (Sqrt-UKFST) in Nanosatellite. IEEE
47. Liu-Henke, X., Scherler, S., & Jacobitz, S. Transactions on Power Electronics, 30(9): pp.
(2017). Verification oriented development of a 4774-4783.

Philadelphia, USA 268

View publication stats