Proceedings of the 2022 6th CAA International Conference on Vehicular Control

and Intelligence (CVCI), Nanjing, China, October 28-30, 2022

A Lithium-ion Battery SOC Estimation Method

Involving Battery Internal Temperature
Yuntong Liu Haosen Chen Wei-Li Song Hangcheng Han
School of Integrated Circuits and Institute of Advanced Structure Institute of Advanced Structure School of Integrated Circuits and
Electronics Technology Technology Electronics
Beijing Institute of Technology Beijing Institute of Technology Beijing Institute of Technology Beijing Institute of Technology
Beijing, China Beijing, China Beijing, China Beijing, China
[email protected] [email protected] [email protected] [email protected]
2022 6th CAA International Conference on Vehicular Control and Intelligence (CVCI) | 978-1-6654-5374-5/22/$31.00 ©2022 IEEE | DOI: 10.1109/CVCI56766.2022.9964684

Jihua Lu Shujuan Hou Lei Sun Shaoqi Wang

School of Integrated Circuits and School of Information and School of Integrated Cirsuits and School of Integrated Circuits and
Electronics Electronics Electronics Electronics
Beijing Institute of Technology Beijing Institute of Technology Beijing Institute of Technology Beijing Institute of Technology
Beijing, China Beijing, China Beijing, China Beijing, China
[email protected] [email protected] [email protected] [email protected]

Abstract—Accurate estimating the state of charge(SOC) of To show the connection between the ambient temperature
lithium-ion batteries is important for battery management parameter and the states of lithium-ion batteries, Wang et.al.[3]
systems. The SOC estimation generally considers ambient conducted overcharge and discharge experiments with respect
temperature instead of internal temperature parameter due to to various ambient temperatures. Wang et.al.[4] studied the
the difficulty of capturing internal temperature. Thanks to a relationship between battery capacity and temperature, and the
recent work capturing internal temperature, this work proposes relationship between charge-discharge cycle time and
a method of SOC estimation for lithium-ion batteries based on charge-discharge rate. Belta et.al.[5] studied the energy and
internal temperature which has not been investigated in the
power variation characteristics of the battery at different
existing studies. Three kinds of tests are carried out at 0℃, 15℃, temperatures. It is shown that the energy and power of the
25 ℃ , 35 ℃ and 45 ℃ , including maximum capacity test, open battery would decay with the decrease of temperature.
circuit voltage test, and dynamic stress test (DST). Subsequently,
the second-order RC equivalent circuit model of the battery is Due to the importance of temperature parameter over the
established. And the parameters of the battery model are running of battery, the temperature parameter was also used in
identified by the forgetting factor recursive least squares (FFRLS) the estimation of Battery states. He et.al.[6] proposed a
algorithm considering the ambient temperature and the internal Thevenin equivalent circuit model considering the temperature
temperature. Finally, the SOC of the lithium-ion battery is parameter. Song et.al.[7] and Luo et.al.[8] showed that the
estimated using the DST test data at 0 ℃ , 25 ℃ and 45 ℃ , the maximum capacity of the battery and the relationship between
model parameter identification results at the two temperatures, open circuit voltage (OCV) and SOC were different at different
and the EKF algorithm. The experimental results show that the temperatures. Xing et.al.[9] studied the impact of temperature on
SOC estimation accuracy involving the internal temperature of OCV and proposed a temperature associated correction
the battery is higher than the SOC estimation accuracy coefficients to improve their estimation model.
considering the ambient temperature.
Regarding the battery SOC estimation, one type of
Keywords—Lithium-ion battery, Internal temperature, methods is implemented based on Kalman Filter (KF) or its
Forgetting Factor Recursive Least Squares, Extended Kalman variations. For example, Chin et.al.[10] studied the Extended
Filter, SOC Kalman Filtering (EKF) method considering the ambient
temperature to estimate the SOC of a lithium-ion battery. It was
Ⅰ. INTRODUCTION shown that the EKF based SOC estimation accuracy is highly
State of charge (SOC) estimation of Lithium-ion batteries related to the battery model parameters. It is also interesting to
plays an important role particularly in the application of get an on-line estimation where Recursive Least Squares (RLS)
electrical vehicles. Inaccurate SOC estimation may lead to is studied but with a time lag postponed by processing[11]. The
mismatch operations such as battery overcharging or time lag problem is alleviated by introducing a forgetting factor
over-discharging, and even worse safety risks[1]. Battery states into RLS which provides a possible solution for in-situ SOC
are generally estimated via a set of battery parameters. estimation[12].
Temperature is such an important parameter which is directly However, most of the existing in-situ battery SOC
related to the capacity and status of a battery[2]. The role of estimation addressed the ambient temperature parameter,
temperature and its impact on the battery SOC estimation have instead of the battery inner temperature parameter. One reason
attracted more and more research interests. is the difficulty of capturing the inner temperature. Recently,
the authors’team proposed one solution capturing battery inner

978-1-6654-5374-5/22/$31.00 ©2022 IEEE

Authorized licensed use limited to: UNIVERSITAS GADJAH MADA. Downloaded on July 26,2024 at 14:07:12 UTC from IEEE Xplore. Restrictions apply.
temperature[13]. It thus provides an option discussing the impact top-right side in Fig.2. The major instruments in our
of inner temperature on the battery SOC estimation. experiments include a charge and discharge tester and an
incubator.
In the following, the battery platform and its settings are
shown in Section 2 to get a quick overview of the battery SOC B. Regular Tests
platform. The adopted battery Thevenin modeling is introduced To get an overview of the soft pack battery used in
in Section 3 followed by our proposed online parameter experiments, regular capacity and OCV tests are implemented
identification method using the combination of FFRLS and in this subsection.
EKF. Experimental results are shown in Section 4 and our
work is concluded in Section 5. 1) Capacity Test
The capacity test is implemented with respect to variations
Ⅱ. TEST BENCH SETTINGS AND REGULAR TEST of temperatures. Particular, 5 temperature values 0, 15, 25, 35
A. Test bench and 45 are selected in this test. The attainable capacity values
are plotted in Fig.3. From this figure, it is seen that the
The battery used in this work, without loss of generality, is capacity values are variant with respect to different
a nickel-cobalt-manganese ternary lithium-ion soft pack temperature values. The capacity values at 0℃ and 15℃ are
battery. The battery appearance is shown in Fig.1 and its major relatively small. It seems that the capacity value at temperature
operation parameters are listed in Table 1. This soft pack 25 gets the maximum capacity value in this test. This result is
battery integrates an internal temperature sensor which in accordance with other experiments[4] and evidenced the
transmit the battery internal temperature outside the battery capacity connection with the variation of temperature as well.
pack periodically. This temperature sensing device was
proposed in literature [13] which is referred to interested
readers.

Fig.3 Maximum discharge capacity at five temperatures

Fig.1 nickel-cobalt-manganese ternary lithium-ion soft pack battery 2) OCV Test

Table 1. Parameters of nickel-cobalt-manganese ternary lithium-ion soft Since there is a connection among OCV, SOC and the
pack battery temperature character, it is intuitive to see the performance
surface over the three values.
Parameter name Numerical value
Rated capacity 2Ah The principle of this experiment is as follows. Assuming
Charge cut-off voltage 4.2V
that the maximum capacity of the battery at a certain
temperature is 2Ah, and using a current of 1/2C rate to charge
Discharge cut-off voltage 2.5V
and discharge the battery, it takes 2 hours to charge to the upper
Nominal voltage 3.7V cut-off voltage of 4.2V (SOC=100%) or to discharge to the
Positive electrode material Lithium nickel cobalt manganate lower cut-off voltage of 2.5V (SOC=0%). Therefore, the
battery needs to be fully charged (SOC=100%) first. Then, the
battery is discharged at a rate of 1/2C for 12 min (SOC
decreased by 10%), and left for 1h. The terminal voltage
measured at this time is the OCV corresponding to SOC=90%.
Cycle 10 times until the battery is discharged to the lower
cut-off voltage of 2.5V.
The experiment obtained the corresponding OCV, internal
temperature value and ambient temperature under different
SOC values. Then fit the obtained data into OCV_SOC_Tout
and OCV_SOC_Tin surfaces respectively by using the cftool
tool in MATLAB, as shown in Fig.4 and Fig.5.
As seen from the two figures, there are differences in the
OCV_SOC relation curves at different temperatures. In this
paper, the functional relationship is fitted from the 1 × 1. After
Fig.2 Physical map of battery test platform
many experiments, the effect is better when the 4 × 2 order
There is a sensing data receiver outside the battery pack fitting is used. The obtained functional relationship of the two
and the received data are formatted in stream as shown in the three-dimensional surfaces is

Authorized licensed use limited to: UNIVERSITAS GADJAH MADA. Downloaded on July 26,2024 at 14:07:12 UTC from IEEE Xplore. Restrictions apply.
 OCV  3.336  2.132  SOC  0.06696  Tout  6.14  SOC 2 Where, U0, U1 and U2 represent the voltages across R0, R1 and
2
 0.2198  SOC  Tout  0.001294  Tout  8.755  SOC 3 R2.
2
 0.1946  SOC  Tout  0.004009  SOC  Tout
2 (1)
4 3
 3.953  SOC  0.03566  SOC Tout
2
 0.002874  SOC 2  Tout

OCV  3.614  1.233  SOC  0.07925  Tin  5.296  SOC 2

2
 0.2565  SOC  Tin  0.001376  Tin  8.426  SOC 3
2
 0.2175  SOC  Tin  0.004272  SOC  Tin
2 (2)
4 3
 3.816  SOC  0.03258  SOC Tin
2
 0.003067  SOC 2  Tin

Where, Tout denotes the ambient temperature and Tin denotes

the internal temperature. The root mean square errors of the
two fitted relations are 0.06341 and 0.06279, both reaching
95% confidence intervals.

Fig.4 OCV_SOC_Tout curve considering ambient temperature

Fig.6 Flowchart of the FFRLS algorithm

Fig.5 OCV_SOC_Tin curve considering internal temperature

Ⅲ. METHODOLOGY
A. FFRLS online identification
The EKF is widely used in the battery SOC estimation.
However, EKF itself does not have such a battery model
parameter scheme. In this work, a combination of FFRLS and
EKF is proposed to do the SOC estimation. The FFRLS
algorithm is firstly adopted to estimate a second-order RC
circuit parameter [14]. Fig.6 shows the calculation flow of the
FFRLS algorithm. The second-order RC equations are [14]
 U1 dU 1
I  C1
 R 1 dt
 U2 dU 2 (3)
I  C2 Fig.7 FFRLS+EKF joint estimation process
 R2 dt
 These estimated parameters are subsequently used in the
U  U OC  U 0  U 1  U 2 EKF SOC estimation. The flowchart of this combination is

listed in Fig.7.

Authorized licensed use limited to: UNIVERSITAS GADJAH MADA. Downloaded on July 26,2024 at 14:07:12 UTC from IEEE Xplore. Restrictions apply.
B. EKF algorithm to get the linearized observation equation
In particular, the EKF state space equation is listed as y k  C k X k  Dk u k   k
follows. U  (14)
 d( f (SOC k , T k ))   1, k 
 ΔT ΔT
  1  1   U 2,k   R 0,k I k   k
 d(SOC)  SOC 
U1 (k 1)  (1 )U1 (k)  I (k)  k 
 R C
1 1 C1
 ΔT ΔT (4) Once the parameters in Ak, Bk, Dk obtained from FFRLS, it is
U 2 (k 1)  (1 )U 2 (k)  I (k)
 R C
2 2 C2 possible to do SOC estimation by EKF.

U (k 1)  U OC (k ) U1 (k ) U 2 (k )  I (k )R0 Ⅳ. RESULTS AND DISCUSSION

Δt A. Parameter identification result
SOC( k  1)  SOC( k )  I (k ) (5)
3600 C Fig.8 to 10 show the parameter identification results of the
battery equivalent circuit model considering the ambient
where I denote the input current and the state input is temperature and internal temperature at three temperature
 U1  points of 0°C, 25°C and 45°C.
X   U2  (6) It can be seen from the figures that there is little difference
SOC
  between the two R0 at 0℃. The difference between the two R0
at 25℃ and 45℃ becomes larger towards the end of the
Based on this state, the next state is discharge. The change trends of the other four parameters at
 U1,k 1  the three temperatures are the same, and the difference of the
X k 1   U 2,k 1  identification results under the internal and external
  temperatures also increases with the increase of the discharge
SOCk 1 
   t  time. In summary, there are significant differences in the
t (7)
1 0 0    identification results at the two temperatures, indicating that it
 R1, k C1, k  U   C1,k  is meaningful to consider the internal temperature for SOC
 t  1,k   t 
 0 1 0  U 2,k    C  I k  ωk estimation.
 R2,k C2,k SOCk   2,k 
   t 
 0 0 1   3600C 
   

And the observation equation is

y k  U k  U OC,k  U 1,k  U 2,k  R0,k I k  υk (8)

Let u k  Ik ,

 U 1,k 
X k   U 2, k  (9)
 
SOCk 
 t 
1  0 0 
 R1,k C1,k  (10)
 t  Fig.8 On-line parameter identification results at 0°C inside and outside
Ak   0 1 0 
 R 2, k C 2, k  temperature
 
 0 0 1 
 
 t 
 
 C 1, k  (11)
 t 
Bk   
 C 2,k 
 t 
 3600 C 
 

Since U OC, k  f(SOC ,T ) is a nonlinear function which can

be linearized by
U k  d( f (SOC k , T k )) 
Ck    1  1  (12)
X  d(SOC)  Fig.9 On-line parameter identification results at 25°C inside and outside
D k  [  R 0,k ] (13) temperature

Authorized licensed use limited to: UNIVERSITAS GADJAH MADA. Downloaded on July 26,2024 at 14:07:12 UTC from IEEE Xplore. Restrictions apply.
 Fig.13 Comparison of estimated terminal voltages at 45°C at internal and
external temperatures
Fig.10 On-line parameter identification results at 45°C inside and outside
temperature In order to better illustrate the influence of the two
temperatures on the accuracy of the battery model, this paper
B. Model accuracy verification uses RMSE and MAE to describe the accuracy of the model.
The estimated value of terminal voltage can be obtained by Table 2 shows the comparison of model accuracy at two
substituting the parameter identification results into the battery temperatures.
model. Fig.11 to 13 show the comparison of the estimated Table 2. Comparison of RMSE and MAE of Terminal Voltage Error at
terminal voltage with the actual value and the error Internal and External Temperatures
comparison considering the two temperatures at three Temperature Index Ambient Internal
Reduction
temperature points. temperature temperature
0℃ RMSE 0.0693 0.0650 6.2%
MAE 0.0501 0.0467 6.79%
25℃ RMSE 0.0534 0.0525 1.69%
MAE 0.0361 0.0356 1.39%
45℃ RMSE 0.0627 0.0542 13.55%
MAE 0.0496 0.0407 17.94%
It can be seen that when considering the internal
temperature, the terminal voltage estimation error decreases
the most at 45°C, the RMSE decreases by 13.55%, and the
MAE decreases by 17.94%. At 0°C, the RMSE decreases by
6.2%, and the MAE decreased by 6.79%. At 25℃ RMSE
decreased by 1.69% and MAE decreased by 1.39%. In
summary, the model accuracy is higher when the internal
temperature is considered.
C. SOC estimation results based on FFRLS+EKF joint
algorithm
Fig.11 Comparison of estimated terminal voltages at 25°C at internal and
external temperatures Fig.14 to 16 show the comparison of the estimated SOC
with the true value considering ambient and internal
temperature at three temperature points.

Fig.12 Comparison of estimated terminal voltages at 25°C at internal and Fig.14 Comparison of SOC estimates at 0°C at internal and external
external temperatures temperatures

Authorized licensed use limited to: UNIVERSITAS GADJAH MADA. Downloaded on July 26,2024 at 14:07:12 UTC from IEEE Xplore. Restrictions apply.
 capacity test, open circuit voltage test and DST test were
carried out at 0°C, 15°C, 25°C, 35°C, and 45°C, and the
internal temperature of the battery was monitored during the
test. The maximum capacity values at different temperatures
were obtained via the maximum capacity test, which were
used to solve the Bk in the EKF estimation. The fitting curves
OCV_SOC_Tout and OCV_SOC_Tin were obtained through the
open circuit voltage test, which were used to solve the Ck in
the EKF. The results showed that the battery model
considering the internal temperature was more accurate and
the terminal voltage error was smaller. It thus evidenced that
the SOC estimation accuracy considering the internal
temperature is significant where the estimation error is smaller
Fig.15 Comparison of SOC estimates at 25°C at internal and external
than the compared methods. In short, the SOC estimation
temperatures involving battery internal temperature parameter could be a
promising solution to improve the SOC estimation accuracy.
REFERENCES
[1] Q. Ouyang, R. Ma, Z. X. Wu, et al. Adaptive square-root unscented
Kalman filter-based state-of charge estimation for lithium-ion batteries
with model parameter online identification[J]. Energies, 2020, 13(18):
1-14.
[2] L. Luo, W. J. Song, S. L. Lin, et al. Research progress on effects of
temperature on SOC and its estimation for LPF battery[J]. Advances in
New and Renewable Energy, 2015, 3(1):59-69.
[3] H. W. Wang, S. Deng，H. Q. Xiao，H. F. Zhao, and C. Wang.(2014)
Research of overcharge/overdischarge of power Li-ion battery under
different ambient temperatures. J.Chinese Journal of Power Sources ，
138(3):431-433.
[4] J. Wang, P. Liu, J. Hicks-Garner,et al. (2011) Cycle-Life model for
graphite-LiFePO4 cells. J. Journal of Power Sources, 196(8):3942-3948.
[5] J. R. Belta, C. D. Hoa, T. J. Millerb, ea tl. (2005) The effect of
Fig.16 Comparison of SOC estimates at 45°C at internal and external temperature on capacity and power in cycled lithium ion batteries. J.
temperatures Journal of Power Sources, 142(1-2):354-360.
In order to better illustrate the influence of the two [6] Y. He, Q. Li, X. X. Zheng, ea tl. (2021) Equivalent hysteresis model
temperatures on the accuracy of SOC estimation, this paper based SOC eatimation with variable paremeters considering temperature.
J. Journal of Power Electronics, 21:590-602.
uses RMSE and MAE to describe the accuracy of SOC
[7] K. Song, X. Chen, A. G. Cheng. (2019) Modeling of lithium battery
estimation. Table 3 shows the comparison of the SOC characteristics considering the influence of temperature and hysteresis
estimation results at the two temperatures. effect. J. Automotive Engineering, 41(3):334-339.
Table 3. RMSE and MAE Comparison of SOC Estimation Errors at [8] M. Luo, Y. Z. Gao, J. Q. Kang, et al. (2018) Ternary-material
Internal and External Temperatures lithium-ion battery SOC estimation under various ambient temperature. J.
Temperature Index Ambient Internal Ionics, 24(7):1907-1917.
Reduction
temperature temperature [9] Y. J. Xing, W. He, M. Pecht, et al. (2014) State of charge estimation of
0℃ RMSE 0.0693 0.0651 6.06% lithium-ion batteries using the open-circuit voltage at various ambient
MAE 0.0291 0.0277 4.81% temperature. J. Applied Energy, 113:106-115.
25℃ RMSE 0.0534 0.0525 1.68% [10] C. S. Chin, Z. C. Gao, C. Z. Zhang, et al. Nonlinear
MAE 0.0279 0.0260 6.81% temperature-dependent state model of cylindrical LiFePO4 battery for
open-circuit voltage, terminal voltage and state-of-charge estimation
RMSE 0.0627 0.0542 13.56%
45℃ with extended Kalman filter[J]. Energies, 2018, 11(9):2467.
MAE 0.0465 0.0377 18.92% [11] J. L. Sun, X. Zou, H. T. Gu, et al. State of charge estimation for
It can be seen that when considering the internal lithium-ion battery based on FFRLS-EKF joint algorithm[J]. Automotive
temperature, the SOC estimation error decreases the most at Engineering, 2022, 44(4):505-513.
45°C, with RMSE decreasing by 13.56% and MAE decreasing [12] B. H. Du, Y. Zhang, T. Z. Wu, et al. An online identification method for
equivalent model parameters of aging lithium-ion batteries[J]. Energy
by 18.92%. At 0°C, the RMSE decreased by 6.06% and the Storage Science and Technology, 2021，10(1):342-348.
MAE decreased by 4.81%. At 25°C, the RMSE decreased by
[13] L. Yang, N. Li, L. K. Hu, et al. Internal feld study of 21700 battery based
1.68% and the MAE decreased by 6.81%. In summary, the on long- life embedded wireless temperature sensor[J]. Acta Mechanica
SOC estimation accuracy considering the internal temperature Sinica, 2021, 37(6):895-901.
of the battery is higher. [14] H. D. Liu, L. Y. Kang, C. S. Lu, et al. A Least Squares Model Parameter
Identification Method for Lithium Battery Model with Forgetting
Ⅵ. CONCLUSION Factor[P]. Guang Dong:CN107367692A,2017-11-21.
In order to study the relationship between the battery
internal temperature and its gains to the SOC estimation, an
experimental platform was built in this work. The maximum

Authorized licensed use limited to: UNIVERSITAS GADJAH MADA. Downloaded on July 26,2024 at 14:07:12 UTC from IEEE Xplore. Restrictions apply.