batteries

Article
State of Charge Estimation of LiFePO4 in Various
Temperature Scenarios
Mingzhu Wang, Guan Wang, Zhanlong Xiao, Yuedong Sun and Yuejiu Zheng *

College of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China
* Correspondence: yuejiu.zheng@usst.edu.cn

Abstract: The state estimation of a battery is a significant component of a BMS. Due to the poor
temperature performance and voltage plateau phase in LiFePO4 batteries, the difficulty of state
estimation is greatly increased. At the same time, the ambient temperature in which the battery
operates is changeable, and its parameters will vary with the temperature. Therefore, it is extremely
challenging to estimate the state of LiFePO4 batteries under variable temperatures. In an effort to
accurately estimate the SOC of LiFePO4 batteries at different and variable temperatures, as well as its
capacity at low temperature, the characteristics of LiFePO4 batteries at different temperatures are first
tested. In addition, a variable temperature OCV experiment is designed to obtain the OCV of the full
SOC range. Then, the ECM considering temperature is established and all parameters are identified
by PSO. Finally, an improved EKF algorithm is presented to accurately estimate the SOC of LiFePO4
batteries at different and variable temperatures. Meanwhile, the battery capacity at low temperature
is further estimated based on the estimated SOC result. The results show that SOC estimation errors
at variable temperature are all within 3%, and the capacity estimation errors at low temperature are
all within 1%.

Keywords: variable temperature condition; LiFePO4 battery; least squares; extended Kalman filter;
state estimation

1. Introduction
Citation: Wang, M.; Wang, G.; Xiao,
Due to energy shortages and environmental pollution that are becoming progressively
Z.; Sun, Y.; Zheng, Y. State of Charge
more serious, electric vehicles are favored by the market due to their green characteristics.
Estimation of LiFePO4 in Various
The rapid growth in the electric vehicle market and large-scale power grid applications have
Temperature Scenarios. Batteries 2023,
strongly promoted the development of LiFePO4 batteries, which have been extensively
9, 43. https://doi.org/10.3390/
batteries9010043
used in electric vehicles and energy storage systems [1–4]. Among them, LiFePO4 batteries
have become the main power provider in electric vehicles due to their advantages such
Academic Editor: Hirotoshi Yamada as environmental friendliness, low cost and long cycle life [5]. Accurate state of charge
Received: 13 November 2022
(SOC) estimation is a prerequisite for batteries to work more efficiently and safely [6].
Revised: 24 December 2022 A common SOC estimation method is to obtain battery SOC based on the relationship
Accepted: 3 January 2023 between battery SOC and open circuit voltage (OCV) [7]. However, due to the long plateau
Published: 6 January 2023 period and OCV hysteresis of LiFePO4 batteries, an accurate SOC cannot be obtained by
this method. According to a study by Zheng et al. [8], when the OCV error reaches up to 1
mV, it can generate a 5% SOC estimation error for LiFePO4 batteries. Therefore, it is difficult
to accurately estimate the SOC of LiFePO4 batteries [9]. Meanwhile, the low temperature
Copyright: © 2023 by the authors. performance of LiFePO4 batteries is poor [10], which further affects the accuracy of SOC
Licensee MDPI, Basel, Switzerland. and capacity estimation. Consequently, it is necessary to devise a method to accurately
This article is an open access article estimate the SOC and capacity at low temperatures.
distributed under the terms and The basis for estimating battery state with high precision is an appropriate battery
conditions of the Creative Commons
model [11]. Common battery models include the equivalent circuit model (ECM) [12], the
Attribution (CC BY) license (https://
electrochemical mechanism model and the data-driven model [13–15]. Battery dynamic
creativecommons.org/licenses/by/
characteristics have been described by Dai [16] using a first-order RC model. Hu [17]
4.0/).

Batteries 2023, 9, 43. https://doi.org/10.3390/batteries9010043 https://www.mdpi.com/journal/batteries

Batteries 2023, 9, 43 2 of 17

compared the accuracy, complexity and robustness of twelve common ECMs. He believed
that the terminal voltage estimation accuracy for a LiFePO4 battery could be achieved by
using the first-order RC model and the hysteresis of a single state. Other researchers [18–20]
have achieved improvements in ECMs, but the influence of temperature has not been taken
into consideration. The dynamic properties of a battery were described by Johnson et al.
by using the Rint model, which verified that the OCV and ohmic internal resistance are
functions of the SOC and temperature [21]. Nevertheless, there was a problem with the
structure of this model, since it cannot describe the polarization phenomenon of batteries.
He et al. simulated the battery terminal voltage based on the Thevenin model considering
temperature [22], but they ignored the influence of OCV which is a crucial model parameter.
Xing et al. established a battery model based on a OCV–SOC curve to study the influence
of temperature on OCV [23]. They proposed a correction coefficient to improve the model,
but the influence of temperature on the internal resistance and other parameters of the
model was not considered. Leo et al. considered the variation in battery capacity and
internal resistance with temperature [24,25] but neglected to consider other parameters as
well. Xu et al. established a temperature-related second-order RC model to simulate an
NCM battery [26], but the change in model parameters was not significant owing to the
narrow temperature range.
The parameter identification of battery models is generally realized by intelligent
optimization algorithms which include the genetic algorithm (GA) and particle swarm
optimization (PSO) [27,28]. The optimal parameters can be found by setting reasonable
upper and lower limits of the parameters. The combination of ECM and various algorithms
is the most common method of SOC estimation [29]. In recent years, Kalman filters (KFs)
have been universally applied, benefiting from a balance of accuracy, robustness and com-
puting complexity [30,31]. Chen et al. [32] combined the Rint model and KF algorithm to
estimate the SOC of a battery; the estimation accuracy was significantly improved when
compared with the OCV-corrected ampere-hour integration method. Cui et al. [33] pro-
posed a square-root volumetric KF with temperature correction rules to achieve an accurate
estimation of the SOC in the battery management system (BMS) of an on-board embed-
ded microcontroller. Sangwan et al. [10] optimized the model parameters by considering
temperature, adopting three adaptive filtering algorithms based on recursive Bayesian
filtering to estimate the battery SOC. In addition, they compared the efficiency of mass and
computation time. Despite the aforementioned methods, which consider the influence of
temperature and have a relatively good SOC estimation accuracy, most of them ignore the
dynamic load with simultaneous changes in temperature and current.
Thereby, in this paper, we propose a modified extended Kalman filter (EKF) algorithm.
The LiFePO4 battery is selected to verify the precision of SOC estimation results under
various temperature scenarios and capacity estimation results under low temperature.
Firstly, a second-order RC model considering temperature is established. Identifying all
parameters of the model is performed by the PSO algorithm. Then, a new SOC definition
is put forward according to temperature variation, which redefines the SOC cutoff point
under different ambient temperatures. Furthermore, the OCVs at different temperatures are
estimated based on the forgetting factor recursive least square (FFRLS) method, and the gain
of EKF is adjusted between the plateau phase and non-plateau phase of the OCV curve. The
SOC can subsequently be accurately estimated at different temperatures. In the meantime,
through a discharging experiment under variable temperature conditions, the accuracy of
the proposed method and the reliability of the battery model are verified. Finally, on the
premise of the accurate SOC estimation of LiFePO4 batteries at low temperatures, based
on the principle of SOC electric quantity gain method, the iterative weighted least squares
method is used to estimate the capacity of the battery at low temperatures.

2. Variable Temperature State Estimation Method

Estimation of the SOC is commonly carried out using the EKF algorithm, which is a
method combining the ampere-hour integration method and terminal voltage calculation
 2. Variable Temperature State Estimation Method
Estimation of the SOC is commonly carried out using the EKF algorithm, which
method combining the ampere-hour integration method and terminal voltage calcula
Batteries 2023, 9, 43 3 of 17
based on a battery model. The terminal voltage of the battery model is calculated b
on OCV, and OCV is generally obtained through SOC–OCV curve interpolation or
query. However, there are two obvious plateau periods on the SOC–OCV curve o
based on a battery model. The terminal voltage of the battery model is calculated based
LiFePO
on 4 battery.
OCV, and OCV isAs a result,
generally using the
obtained method
through of OCV
SOC–OCV table
curve query, it isordifficult
interpolation table to ob
an accurate
query. However, OCV thereeven
are with an accurate
two obvious plateauSOC. Inaccurate
periods calculation
on the SOC–OCV curve ofofbattery
the term
LiFePO
voltage4 battery.
will As
lead a
to result, using
inaccurate the method
estimation of OCV
of the table
SOC query,
in it
the is difficult
next step.to obtain
Therefore, ho
an accurate OCV even with an accurate SOC. Inaccurate calculation
use the EKF algorithm in the OCV plateau periods of LiFePO4 batteries is a key prob of battery terminal
voltage will lead to inaccurate estimation of the SOC in the next step. Therefore, how to
Moreover, setting a Kalman gain will largely determine the estimation accuracy o
use the EKF algorithm in the OCV plateau periods of LiFePO4 batteries is a key problem.
EKF algorithm.
Moreover, setting aAn improved
Kalman gain willEKF algorithm
largely determine isthe
established
estimation in this section
accuracy of the EKFto estimat
SOC of a LiFePO
algorithm. An improved EKF algorithm is established in this section to estimate the SOC ofof a LiF
4 battery. The steps are as follows: Firstly, the OCV curve
abattery
LiFePO4will be estimated
battery. The steps are online, and Firstly,
as follows: the noise of the
the OCV EKF
curve of will be adjusted
a LiFePO between z
4 battery will
be estimated online, and the noise of the EKF will be adjusted between
according to the plateau and non-plateau phase of the estimated OCV curve. The zones according
to the plateau
vantages of and non-plateau
combining thephase of the estimated
ampere-hour OCV curve.
integration method The advantages
and the batteryof m
combining the ampere-hour integration method and the battery model method of the EKF
method of the EKF algorithm will be used to estimate SOC of the battery at different
algorithm will be used to estimate SOC of the battery at different temperatures. Then, the
peratures.
SOC will be Then, theatSOC
estimated will temperature
variable be estimated at variable
dynamic temperature
conditions. Based ondynamic
the SOC condit
Based on the SOC estimation results at low temperature, the
estimation results at low temperature, the weighted iterative least squares method will weighted iterative
squares
be utilizedmethod will be
for estimating utilized
the for estimating
battery capacity. the battery
The framework capacity.
of this method The framework
is shown in o
Figure 1.
method is shown in Figure 1.

Figure1.1.Framework
Figure Framework diagram.
diagram.

2.1.
2.1.Battery
BatteryModel
Modelandand
Parameter Identification
Parameter Identification
An accurate selection of the battery model is essential to achieve a high precision of
An accurate selection of the battery model is essential to achieve a high precisio
the state estimation. Following a comprehensive consideration of the complexity, accuracy,
the state estimation.
implementation Following
convenience a comprehensive
and practical consideration
value of the chosen model, the RC of the complexity, a
equivalent
racy, implementation
circuit model is generallyconvenience
adopted. The and practical
higher value
the order ofthe
of RC, thehigher
chosen themodel, the RC eq
accuracy.
alent
At the circuit model
same time, is the
since generally adopted.
variations The higherwith
in the parameters the temperature
order of RC,need the to
higher
be the a
considered, the second-order RC model considering temperature is an appropriate
racy. At the same time, since the variations in the parameters with temperature need choice
to simulate thethe
considered, characteristics
second-order of a RC
LiFePO 4 battery.
model The structure
considering of the selected
temperature is anmodel is
appropriate ch
illustrated in Figure 2.
to simulate the characteristics of a LiFePO4 battery. The structure of the selected mod
illustrated in Figure 2.
Batteries 2023, 9, x FOR PEER REVIEW 4 of 18
Batteries 2023, 9, 43 4 of 17

Figure 2. Second-order RC model considering temperature.

Figure 2. Second-order RC model considering temperature.
In the aforementioned model, UOC represents the OCV of the battery, I is the charg-
ing orthe
In aforementioned
discharging current,model,
Ut is U OC represents the OCV of the battery, I is the charging
the terminal voltage and R0 represents ohmic internal
orresistance.
discharging current, Utinternal
Polarization is the terminal voltage
resistances, R1 andand RR0, represents ohmic internal re-
2 are connected in parallel with
sistance. Polarization internal resistances, R 1 and R2, are connected in parallel with polar-
polarization capacitors, C1 and C2 , to form two RC structures. U1 and U2 are the voltages
ization
of the capacitors, C1 and C2respectively,
two RC structures, , to form two which
RC structures.
are usedUto 1 and U2 are the voltages of the
simulate the voltage rise and
two RC structures, respectively, which are used to
fall characteristics of the battery polarization. The measured simulate the voltage
current,rise andcell
I, and fallsurface
char-
acteristics of the battery polarization. The measured current, I, and cell
temperature, T, are the inputs of the established cell model. Meanwhile, Ut represents the surface tempera-
ture,
outputT, are the model.
of the inputs of thestate
The established cell model.
space equations andMeanwhile,
discretization Ut represents the output
equations obtained are
ofasthe model.
follows: The state space equations and discretization equations obtained are as fol-
lows: The equations of state spaces:
The equations of state spaces:
d (U
d(U ) ) 1 1
1 1 = I (t) 1
= I(t) − −
1
U (t) U1 (t) (1)(1)
dtdt C1 (SOC,
C (SOC
1
,T ) TR) (SOC
1
R1 (,T
SOC, T )C1,T
)C (SOC 1
(SOC,
) 1 T)
d(U
d(2U) ) 1 1 1 1
2= I(t) − − U (t) (2)(2)
dt dt = I ( t )
C 2 (SOC ,T ) TR) 2 (SOC ) 2 T ) U2 (t)
C (SOC,
2 R (,T )C 2 (SOC
SOC, 2T )C (,T
SOC, 2
U(t) = Uoc (SOC,T) + I(t) ⋅ R0 (SOC,T) + U1 (t) + U2 (t) (3)(3)
U (t) = Uoc (SOC, T ) + I (t) · R0 (SOC, T ) + U1 (t) + U2 (t)
Equations (1) to (3) are simplified to obtain:
Equations (1)–(3) are simplified to obtain:
−
Δt
−
Δt

U 1 (t ) = U 1 (t − 1)e τ1
+ I(t )R1 ( 1 − e τ1
) (4)
− ∆t − ∆t
U1 (t) = U1 (t − −1τΔ)t e τ1 + I (t) R1−(τΔ1t − e τ
1 ) (4)
U 2 (t ) = U 2 (t − 1 )e 2
+ I(t )R 2 ( 1 − e 2
) (5)
∆t ∆t
U2 (t= + I(t)R−+τ U1 (t) I (t+)U
U(t) −
)U =ocU(t)
2 ( t − 1) e0 2 + R 2 ((t)
1 − e τ2 ) (6)(5)
where τ 1 = R1C1 , τ 2 = R2CU τ Uoc (τt)2 +are
2 ,( t ) 1=and I (t)time
R0 +constants
U1 (t) + Uand 2 (t)
Δt is the sampling in-
(6)
terval.
where τ1 = R1 C1 , τ2 = R2 C2 , τ1 and τ2 are time constants and ∆t is the sampling interval.
Since the parameters in the model cannot be measured directly, experimental data
Since the parameters in the model cannot be measured directly, experimental data
will be utilized to identify these unknown parameters. In this paper, the PSO algorithm is
will be utilized to identify these unknown parameters. In this paper, the PSO algorithm is
selected for parameter identification for its fast calculation speed and low use of resources.
selected for parameter identification for its fast calculation speed and low use of resources.
The
Themodel
modelparameters
parameterstotobebeidentified
identifiedare
areasasfollows:
follows:
θ = [RCha ,RDch ,R1 ,τ1 ,R2 ,τ2 ] (7)
where RCha θ = [ R Cha , R Dch , R 1 , τ1 , R 2 , τ ]
indicates charging ohmic internal resistance and RDch indicates discharging
2 (7)

ohmic
whereinternal resistance.
RCha indicates To evaluate
charging the precision
ohmic internal of the
resistance andbattery model, we
R Dch indicates take ad-
discharging
vantage of the rooted mean squared error (RMSE) between the terminal
ohmic internal resistance. To evaluate the precision of the battery model, we take advantage voltage of the
model and the measured terminal voltage in the battery experiment.
of the rooted mean squared error (RMSE) between the terminal voltage of the model and theTherefore, the adapt-
ability function
measured of PSO
terminal is: in the battery experiment. Therefore, the adaptability function
voltage
of PSO is: vN 2
ˆ θˆ ) = 1u (U − U
F( k  1 Nk
ˆ (θˆ ))
k k
2 (8)
F̂ (θ̂k ) =Nt ∑ k k k
u
k =1
( U − Û ( θ̂ )) (8)
where F( ˆ θˆ ) indicates the RMSE at timeNk,k= N1 is the number of data points and U and
k k

Ûwhere
k
indicate
F̂ (θ̂kthe measured
) indicates theterminal
RMSE atvoltage
time k, at time
N is theknumber
and theof
terminal voltage
data points andestimated
Uk and Ûk
the measured terminal voltage at time k
by the model at time k, respectively. θ̂ k is an estimate of the individual parameters inbythe
indicate and the terminal voltage estimated the
model at time k, respectively. θ̂k is an estimate of the individual parameters in the model.
model.
Batteries 2023, 9, x FOR PEER REVIEW 5

Batteries 2023, 9, 43 5 of 17
2.2. Improved EKF Algorithm
There are two goals that can be achieved with the improved EKF algorithm. O
2.2.
oneImproved
hand, theEKFOCV
Algorithm
curve of the battery can be determined online from the real-time
accurately.
There areOn twothe other
goals thathand,
can beto improve
achieved thethe
with convergence
improved EKF speed of the On
algorithm. algorithm
the an
one hand,tothe
ability OCV curve
change of the battery
parameters can be
during determined online
identification, from the real-time
the forgetting factor,data
λ, is adde
accurately. On the other hand, to improve the convergence speed of the algorithm and
amplify the weight influence of the next iteration. In this way, it can increase the influ
the ability to change parameters during identification, the forgetting factor, λ, is added
of new data in the dynamic system on the identification results. With respect to the e
to amplify the weight influence of the next iteration. In this way, it can increase the
lished model
influence of newincorporating the temperature
data in the dynamic system on thefactor, the FFRLS
identification algorithm
results. is employe
With respect
the
to theOCV whichmodel
established is identified onlinethefor
incorporating the LiFePO
temperature 4 battery.
factor, The algorithm
the FFRLS form of recursive
is
employed for the OCV
squares method basedwhich is identified
on the standardonline
is ykfor= φthe
θ LiFePO
k k
+ ek , and
4 battery. The form
the specific of
process of FF
recursive least squares method based on the standard is yk = φk θk + ek , and the specific
algorithm
process is shown
of FFRLS in Figure
algorithm 3. in Figure 3.
is shown

Figure3. 3.
Figure Process
Process diagram
diagram of theofFFRLS
the FFRLS online identification
online identification of the OCV.
of the OCV.

λλisisthe
theforgetting
forgettingfactor, affecting
factor, the final
affecting theestimation result with
final estimation different
result values. valu
with different
If λ is too small, the increase in the weight of the old data will lead to a fluctuation or
λ is too small, the increase in the weight of the old data will lead to a fluctuation or d
divergence of the identification result. If λ is too large and the weight of the old data is
gence
too small,ofthe
theidentification
identification result.
result If λ
will not is too
track large and
the dynamic the weight
parameters withof theand
time, old data i
small,
the the identification
convergence result
speed will slow willInnot
down. trackλthe
general, dynamic
is set parameters
in the range with
of 0.95 to 1. time, and
After
convergence speed will slow down. In general, λ is set in the range of 0.95 to 1. A
parameter modification, when λ = 0.98, the method can take both stability and convergence
speed into account.
parameter modification, when λ = 0.98, the method can take both stability and con
gence speed online
When the estimation of is OCV finished, the plateau phase and non-plateau
into account.
phase can be distinguished, and then the Kalman gain can be adjusted in the EKF algorithm.
When the online estimation of is OCV finished, the plateau phase and non-pla
For dynamic nonlinear systems, when system noise and measurement noise are considered,
phase
Equations can(9)beanddistinguished,
(10) are generallyand then
used the Kalman
to express the stategain
spacecan be adjusted
model in the EKF
of the system:
rithm. For dynamic nonlinear systems, when system noise and measurement nois
considered, Equations (9) andxk(10) f (xkgenerally
+ 1 =are ,uk ) + wk used to express the state space
(9) mod
the system: y = g( x , u ) + v (10)
k k k k
xk +1 = f (xk ,uk ) + wk
According to the above state space model, taking xk = [SOC, U1 , U2 ] T as the state
= g(xkas
variable, the charging and dischargingykcurrent ) +input
,uthe
k
vk and the estimated terminal
voltage as the output, we obtain the discrete state space expressions:
According to the above state space model, taking xk = [SOC ,U1 ,U 2 ] T
as the
k +1 = A · xk + current
variable, the charging andxdischarging B · Ik + was (11)
k the input and the estimated term
voltage as the output, we obtain the discrete state space expressions:
Ut,k = Uocv − U1,k − U2,k − Ik R0 + vk (12)
xk +1 = A ⋅ xk + B ⋅ I k + wk
Ut ,k = Uocv − U1,k − U2 ,k − I k R0 + vk
The initialization and iterative estimation equations of the EKF algorithm are sh
in Table 1.
Batteries 2023, 9, 43 6 of 17

The initialization and iterative estimation equations of the EKF algorithm are shown
in Table 1.

Table 1. Iteration equations of the EKF algorithm.

Definition:  
1 0 0
∂ f ( xk ,uk )
Âk = ∂xk = 0 exp(−∆t/τ1 ) 0 
0 0 exp(−∆t/τ2 )
h i
∂g( x ,u ) dOCVk (SOCk )
Ĉk = ∂xk k = dSOCk −1 −1
k
Here are the steps:
Initialization: K = 0, initial state vector x0+ = E[ x0 ], initial bias
T
P0+ = E[( x − x̂0+ )( x − x̂0+ ) ]
Iterative calculation:
When K = 1,2 . . .
Status vector time update: x̂k− = f ( x̂k+−1 , uk−1 )
Error covariance matrix time update: ∑− xe,k
= Âk−1 ∑+
xe,k −1
 T k−1 + ∑w
−1
To calculate the Kalman gain: Lk = ∑− T − T
xe,k Ĉk [Ĉk ∑ xe,k Ĉk + ∑v ]
Status vector measurement update: x̂k+ = x̂k− + Lk [yk − g( x̂k− , uk )]
−
Error covariance matrix measurement updated: ∑+ xe,k = ( I − Lk Ĉk ) ∑ xe,k
where Âk and Ĉk are the coefficient matrixes, Lk is the Kalman gain, ∑w and ∑v are the covariance matrixes of the
input measurement noise named wk and the output measurement noise named vk , respectively. The covariance
matrix, ∑ xe,k , indicates the non-determinacy of the state estimation.

The OCV plateau of the LiFePO4 battery is also known as the non-battery model
reliability interval. Within this interval, an appropriate Kalman gain should be selected
so that the EKF algorithm is even more inclined to trust the actual terminal voltage value
measured by the voltage sensor. In this case, it is preferable to calculate the SOC using
the method of ampere-hour integration. In the non-plateau phase of the battery, that is,
the reliability interval of the battery model, it is more preferable to select the Kalman gain
to obtain the terminal voltage from the EKF algorithm. In this case, the OCV table query
method is preferred to estimate the SOC of the battery. Thereby, the key to solving the
problem is to find the specific voltage value that distinguishes the OCV plateau period
from the OCV non-plateau period.

2.3. Capacity Estimation Method

The discharging capacity of a battery will greatly reduce at low temperatures, which
causes a tremendous impact on the pinpoint capacity estimation results at low temperature.
At present, the SOC electric quantity gain method is commonly used to estimate the battery
capacity. The variation in battery capacity is generally calculated by the ampere-hour
integration method. After obtaining the SOC at different temperatures as described above,
the capacity estimation at low temperature can be carried out based on the idea of the
SOC charge gain method by using the result of the low temperature SOC estimation. The
least square method is a common online estimation method of battery capacity by iterative
calculations. Considering that the SOC estimation process will be affected by uncontrollable
noise at low temperatures, the battery capacity is estimated using the weighted iterative
least squares method.
The two-point method is a basis for the SOC electric quantity gain method. When
the ∆Q of two points and the corresponding ∆SOC is calculated the battery capacity can
be calculated from the ratio between the two points. The calculation formula is shown in
Equation (13):
R t2 η I ( t )
∆Q t1 3600 dt
Cnorm = = (13)
∆SOC SOC (t2 ) − SOC (t1 )
Batteries 2023, 9, 43 7 of 17

where SOC (t1 ) and SOC (t2 ) represent SOC at the moments t1 and t2 , respectively. I(t) is
the charging and discharging current at time t and η is the coulomb efficiency, usually
taken as 1.
The capacity estimation model for ordinary least squares (OLS) is:

yi = β 1 + β 2 xi + vi (14)

where β 1 is a constant value, β 2 is a coefficient value (both are to be calculated) and vi is

the noise.
Converting Equation (14) to the vector Formula (15) gives:
     
y1 1 x11 v1
 y2  1 x21    v 
 β1  2
 ..  =  .. ..  β +  ..  (15)
  
 .  . .  2 .
yn 1 xn1 vn

Thereby, the battery capacity in the model can be calculated by Formula (16):

Cnorm = 1/β 2 (16)

The matrix form of Formula (15) is:

Y = X·H+V (17)

where Y is the observation vector corresponding to n*1, X represents the matrix of n*2,
H is the parameter vector, H = [ β 1 , β 2 ] T and V is the error vector of n*1. The capacity
can be calculated using iteratively reweighted least squares (IRLS) using the OLS capacity
estimation model.
n n
SE = ∑ (Yi − β1 − β2 xi )2 = ∑ v2i (18)
i =1 i =1

where vi represents the residuals. If the influence function, ρ, is chosen as the sum of the
squares of the residuals, the objective function is transformed to:
n n
SE = ∑ ρ(Yi − β1 − β2 xi ) = ∑ ρ(vi ) (19)
i =1 i =1

SE deflects from β so that it is 0, obtaining the equation:

n
∑ ϕ(Yi − β1 − β2 xi )xi = 0 (20)
i =1

where ϕ( x ) is the derivative of ρ( x ).

med(|v |)
Introducing S = 0.6745i in the weight function yields the standardized residuals
vi 0.6745·vi
ui = S = med(|e |) , where med represents the absolute median of the dispersion calculation.
i
Formula (20) can then be written as:
n n n n
ϕ ( ui )
∑ ϕ(Yi − β1 − β2 xi )xi = ∑ ϕ(ui )xi = ∑ ui
· u i · x i = ∑ wi · u i · x i = 0 (21)
i =1 i =1 i =1 i =1

ϕ(u )
where wi = u i is the sample weight of the observation of number i. Formula (18) is
i
vectorized and brought into the matrix to obtain:

X T WY = X T WXH (22)
Batteries 2023, 9, 43 8 of 17

Obtaining the parameter estimates as:

−1
Ĥw = ( X T WX ) XT Y (23)

Once the parametric vector Ĥw = [ β 1 , β 2 ] T is obtained, the capacity estimate can be
calculated by Formula (16).

3. Experiments
3.1. Basic Performance Test at Different Temperatures
In this paper, a LiFePO4 battery from Tianjin Lishen Battery Company was selected
to carry out the basic performance experiments at different temperatures. The battery
was placed in a temperature chamber which could achieve a high temperature and low
temperature conversion during the experiment. In addition, the Xinwei battery test system
was used to conduct temperature variation experiments at different temperatures and
different rest times. A brief description of the basic parameters of the battery is provided in
Table 2.

Table 2. Basic performance parameters of experimental battery.

Parameters Numeric Value

Nominal capacity (Ah) 20
Nominal voltage (V) 3.2
Charging cutoff voltage (V) 3.65 ± 0.5
Discharging cutoff voltage (V) 2.0
Standard charging and discharging current (A) 1/3C
Charging temperature (◦ C) 0~45
Discharging temperature (◦ C) −20~60

The standard capacity of the LiFePO4 battery was tested at 45 ◦ C, 25 ◦ C, 5 ◦ C and

−15 ◦ C.In an effort to obtain the standard charging and discharging capacities at different
ambient temperatures, a current of 6.67 A, which is 1/3C, was used to charge and discharge
the battery. In addition, the hybrid pulse power characteristic (HPPC) test was proposed to
obtain the internal resistance at each temperature. Meanwhile, the dynamic working condi-
tion experiments were conducted at four temperatures to provide data for the estimation
of SOC.

3.2. Variable Temperature OCV Experiment

An important tool which can reflect the operating characteristics is the OCV–SOC
relationship curve of one battery. The OCVs of the same battery will be different when
the temperature changes, and there is an OCV plateau in the OCV of a LiFePO4 battery.
Therefore, analyzing the OCV performance of the LiFePO4 battery at different temperatures
was crucial. An original experiment method was proposed to attain the OCV of the full
range of the SOC. Firstly, the battery temperature was initially set at 25 ◦ C and fully charged
in this scenario, followed by 45 ◦ C, 25 ◦ C, 5 ◦ C and −15 ◦ C settings for the temperature
chamber. The corresponding rest time was set to 3 h, 3 h, 4 h and 5 h, respectively, so as to
obtain the OCV of the SOC node at each temperature. After the rest time, the temperature
chamber was adjusted to the temperature of 25 ◦ C, and the next SOC point can be obtained
by constant discharging current. The above steps are repeated to obtain the OCVs in the full
SOC range at different temperatures. Figure 4 shows the flow chart of the OCV experiment
at variable temperatures.
Batteries 2023, 9,
Batteries 2023, 9, 43
x FOR PEER REVIEW 99 of
of 17
18

4. Variable
Figure 4. Variable temperature
temperature OCV
OCV experimental
experimentalflowchart.
flowchart.

3.3. Experiment to Determine the Upper and Lower Limits

Limits of
of SOC
SOC at
at Different
Different Temperatures
Temperatures
The SOCSOC reflects
reflectsthethedischarging
dischargingcapacity
capacityofofa abattery,
battery, which
whichis defined
is defined as the propor-
as the pro-
tion of the remaining discharging capacity to the total capacity.
portion of the remaining discharging capacity to the total capacity. As temperatures and As temperatures and aging
effect
aging the overall
effect discharging
the overall capacity
discharging of a lithium-ion
capacity of a lithium-ionbattery,battery,
the SOC used
the SOC toused
reflecttothe
re-
residual
flect the discharging capacity ofcapacity
residual discharging the battery must
of the also take
battery must into account
also theseaccount
take into factors. these
SOC
values
factors.for
SOC batteries
values are inconsistent
for batteries at different temperatures
are inconsistent when the standard
at different temperatures when the capacity
stand-
test method is
ard capacity used
test to charge
method is usedand to discharge
charge andthe battery to
discharge thethe cutofftovoltage.
battery the cutoffThus, it is
voltage.
necessary to redefine the SOC and define the upper and lower
Thus, it is necessary to redefine the SOC and define the upper and lower limits of the SOC limits of the SOC when
the
when ambient temperature
the ambient temperatureis different. In this
is different. Inpaper, the SOC
this paper, at 25at◦ C25was
the SOC usedused
°C was as the as
benchmark, meaning the range of SOC at 25 ◦ C was 0 to 100%, and the upper and lower
the benchmark, meaning the range of SOC at 25 °C was 0 to 100%, and the upper and
limits
lower of SOCofatSOC
limits otherattemperatures
other temperatures changes as the temperature
changes as the temperature changes. changes.
To
To determine
determine the the two
two limits
limits ofof SOC
SOC atat different
different temperatures,
temperatures, the the following experi-
following experi-
ments were conducted:
ments were conducted:
(1) ◦ C, the battery was charged with 1/3C constant current to charge it to the
(1) At
At 25
25 °C, the battery was charged with 1/3C constant current to charge it to the
charging
chargingcutoff
cutoffvoltage,
voltage,and andthen
thenit it
was charged
was charged to the charging
to the chargingcutoff current
cutoff withwith
current constant
con-
voltage. At this point, the battery SOC was 100%. Then, it was
stant voltage. At this point, the battery SOC was 100%. Then, it was adjusted to rest adjusted to rest for a certain
for a
time at different
certain temperatures,
time at different and thenand
temperatures, discharged
then dischargedto the lower limit
to the lowerof voltage
limit ofat 1/3C.
voltage
In this process, this ratio was the lower limit of SOC at this
at 1/3C. In this process, this ratio was the lower limit of SOC at this temperature, temperature, which is
which the
is
remaining battery capacity divided by the total discharge capacity at 25 C, as shown in ◦
the remaining battery capacity divided by the total discharge capacity at 25 °C, as shown
Equation (24).
in Equation (24). 25 T
25 discha − TCap discha
Cap
SOCTLT =Cap discha − Cap discha (24)
SOC L = Cap
25
25 (24)
Cap dischadischa
T
where SOC
where SOCL TL denotes
denotesthe
thelower
lowerlimit
limitofofthe
theSOC
SOC when
when the
the redefined
redefined temperature
temperature is T,
25
Capdischa
25 represents the discharging capacity at 25 ◦ C, Capdischa
T
T is the actual discharging
Capdischa represents the discharging capacity at 25 °C, Capdischa◦ is the actual discharging
capacity at temperature T and T is the ambient temperature in C.
capacity at 25
(2) At temperature T andwas
◦ C, the battery T ischarged
the ambient
with temperature
1/3C currentin to°C.
the lower limit of voltage
(2) At 25 °C,atthe
for discharging, battery
which timewas
the charged
battery SOCwithwas1/3C current
0%. Then,to the
the lowerwas
battery limitmaintained
of voltage
forother
at discharging, at which
temperatures for atime the time,
certain battery
andSOCthenwas 0%. Then,
a constant the battery
current was maintained
was applied to charge
at other
the temperatures
battery to the upper forvoltage
a certain time,
limit. Inand
this then a constant
process, current
the ratio was
was the applied
upper toof
limit charge
SOC
the battery to the upper voltage limit. In this process, the ratio was the upper
at this temperature, which is the capacity at this temperature, to the battery capacity limit of SOCat
at ◦
this temperature, which is
25 C, as indicated in Equation (25).the capacity at this temperature, to the battery capacity at 25
°C, as indicated in Equation (25).
Cap
T T
T T Cap cha cha
SOCSOC
U U
= = 25 25 (25)
(25)
CapCap
cha cha

where SOCU T is the upper limit of SOC when the redefined temperature is T, Cap25 is the
cha
charging capacity at 25 ◦ C and Capcha
T is the actual charging capacity at temperature T.
 where SOCUT is the upper limit of SOC when the redefined temperature is T, Capcha
25
is
the charging capacity at 25 °C and CapTcha is the actual charging capacity at temperature
Batteries 2023, 9, 43 10 of 17
T.

3.4. Variable Temperature Dynamic Working Condition Experiment

3.4. Variable Temperature Dynamic Working Condition Experiment
Based on the previous dynamic working condition experiments at different temper-
atures,Based on the previous
estimations of the SOC dynamic working
at different condition experiments
temperatures at different
were calculated. However,temper-
the
atures, estimations of the SOC at different temperatures were calculated.
battery works in an environment with variations in temperature, and the accuracy of SOC However, the
battery works in an environment with variations in temperature, and
estimation should be considered in terms of these environmental changes. Thus, this sec- the accuracy of SOC
estimation
tion adjusts should be considered
the ambient temperature in change
terms of forthese environmental
the experiment changes. NEDC
of continuous Thus, dis-
this
section adjusts the ambient temperature change for the experiment
charging condition of the battery. After being fully charged at 25 °C, the battery was putof continuous NEDC
discharging
in condition
the temperature of the and
chamber battery.
the After being fully
temperature wascharged at 25 ◦the
set to imitate C, the battery
ambient was put
tempera-
in the temperature chamber and the temperature was set to imitate
ture in an actual application. Meanwhile, the battery was discharged under continuous the ambient temperature
in an actual
NEDC application.
conditions Meanwhile,
to achieve the battery
the simultaneous was discharged
current changes with under continuous
ambient NEDC
temperature.
conditions to achieve the simultaneous current changes with ambient
Two experiments of a temperature rise and a temperature drop were carried out to temperature.
Two experiments of a temperature rise and a temperature drop were carried out
satisfy the temperature changes. The temperature and voltage curves for the rising tem-
to satisfy the temperature changes. The temperature and voltage curves for the rising
perature scenario are given in Figure 5. The ambient temperature where the battery is
temperature scenario are given in Figure 5. The ambient temperature where the battery
located gradually rises from −18 °C to◦45 °C at a◦ rate of 0.3 °C/min.◦ At this time, the pro-
is located gradually rises from −18 C to 45 C at a rate of 0.3 C/min. At this time,
gram of the temperature chamber stops, and the temperature gradually drops to 35 °C.
the program of the temperature chamber stops, and the temperature gradually drops to
When the battery reaches a low temperature, the terminal voltage is greatly affected by
35 ◦ C. When the battery reaches a low temperature, the terminal voltage is greatly affected
the temperature and changes sharply. At one point, the cutoff voltage of the battery
by the temperature and changes sharply. At one point, the cutoff voltage of the battery
reaches 3.8 V. With a gradual rise in temperature, the battery terminal voltage changes
reaches 3.8 V. With a gradual rise in temperature, the battery terminal voltage changes
gradually stabilize. As shown in Figure 6, the temperature of the temperature chamber
gradually stabilize. As shown in Figure 6, the temperature of the temperature chamber
drops gradually from 45 ◦°C to −12 °C at a rate of 0.3 °C/min. At this time, the program of
drops gradually from 45 C to −12 ◦ C at a rate of 0.3 ◦ C/min. At this time, the program of
the temperature chamber stops, and the
the temperature chamber stops, and the temperature temperaturegradually
graduallyrises
rises
toto
1515 °C.
◦ C. Addition-
Additionally,
ally,
the terminal voltage curve shows that the terminal voltage changes sharply when
the terminal voltage curve shows that the terminal voltage changes sharply when the the
battery
battery is at a low temperature. As the temperature rises gradually, the battery terminal
is at a low temperature. As the temperature rises gradually, the battery terminal
voltage
voltage changes
changes steadily.
steadily.

60 4
(a) (b)
50
40
Temperature/°C

3.5
Voltage/V

30
20
10
3
0
-10
-20 2.5
0 1 2 3 4 0 1 2 3 4
4
Time/10 s Time/104s
Batteries 2023, 9, x FOR PEER REVIEW 11 of 18
Figure 5.
Figure Temperature variation
5. Temperature variation and
and voltage
voltagecurve
curvefrom
fromthe
thetemperature
temperaturerise
riseexperiment. (a)(aTemper-
experiment. ) Tem-
perature curve
ature curve andand
(b)(voltage
b) voltage curve.
curve.

50 4
(a) (b)
40
3.5
Temperature/°C

30
Voltage/V

20
3
10

0
2.5
-10

-20 2
0 1 2 3 4 5 0 1 2 3 4 5
Time/104s Time/104s

Figure
Figure 6.
6. Temperature
Temperature variation
variation and
andvoltage
voltagecurve
curvefrom
fromthe
thetemperature
temperaturedrop
dropexperiment.
experiment.(a)
(a)Tem-
Tem-
perature
perature curve
curve and
and (b)
(b) voltage
voltage curve.
curve.

4. Results
4.1. Experimental Results of Variable Temperature OCV
The variable temperature experiment was conducted at four temperatures, and all of
them discharge at 25 °C to the next SOC point. Therefore, the SOC points at different tem-
peratures are correspond well. In addition, the whole battery electric quantity can be re-
leased at 25 °C; thus, the OCV at all SOC points can be obtained. The OCV curves at the
 Temp

V
0
2.5
-10

-20 2
0 1 2 3 4 5 0 1 2 3 4 5
4
Time/10 s Time/104s
Batteries 2023, 9, 43 11 of 17
Figure 6. Temperature variation and voltage curve from the temperature drop experiment. (a) Tem-
perature curve and (b) voltage curve.

4. Results
4. Results
4.1.
4.1. Experimental Results
Experimental Results of of Variable
Variable Temperature
Temperature OCVOCV
The variable
The variabletemperature
temperature experiment was conducted
experiment was conductedat four temperatures, and all of and all
at four temperatures,
them
of themdischarge at 25at
discharge °C25to ◦the next
C to the SOC point.
next SOCTherefore, the SOC points
point. Therefore, at different
the SOC pointstem-
at different
peratures are correspond
temperatures well. In
are correspond addition,
well. the whole
In addition, thebattery
wholeelectric
batteryquantity
electriccan be re- can be
quantity
leased at 25 °C;◦ thus, the OCV at all SOC points can be obtained. The OCV curves at the
released at 25 C; thus, the OCV at all SOC points can be obtained. The OCV curves at the
four temperatures show low fluctuation and are practically equal, as shown by the results
four temperatures show low fluctuation and are practically equal, as shown by the results
of the variable temperature experiment. As a result, the model established in this study
of the variable temperature experiment. As a result, the model established in this study will
will not take into account the effect of temperature on OCV in order to reduce the com-
not takeofinto
plexity account
the model andthe effect computational
increase of temperature on OCVThe
efficiency. in OCV–SOC
order to reduce the
curve at 25complexity
°C
of the model and increase
will be adopted uniformly. computational efficiency. The OCV–SOC curve at 25 ◦ C will be
adopted
The uniformly.
battery SOC is stable when discharged at 25 °C, which is reflected in the OCV
The battery SOC is stabletemperatures.
when discharged ◦ C, which is reflected in the OCV
variation with SOC at different Figure at 7b25reveals that the difference be-
variation
tween thewith
OCVSOCat 25 at
°Cdifferent
and everytemperatures. Figure
other temperature 7b revealsThereby,
is negligible. that thethedifference
OCVs atbetween
different
the at 25 ◦ C and are
OCV temperatures every considered to be same is
other temperature in negligible.
the whole range of the
Thereby, theSOC in this
OCVs at different
paper.
temperatures are considered to be same in the whole range of the SOC in this paper.

0.03
45°C (b)
0.02 25°C

OCV difference/V
5°C
0.01 -15°C

-0.01

-0.02

-0.03
0 20 40 60 80 100
SOC/%

Figure 7.
Figure 7. OCV
OCVcurves
curvesatatdifferent temperatures.
different (a) OCV–SOC
temperatures. curvescurves
(a) OCV–SOC at individual temperatures
at individual temperatures
and (b) the difference between the variable temperature OCVs and the OCV at 25 °C.
and (b) the difference between the variable temperature OCVs and the OCV at 25 ◦ C.
4.2. Parameter Identification Results
4.2. Parameter Identification Results
Figure 8 displays the parameters identified by the PSO algorithm using data from
HPPC Figure 8 displays
experiments the parameters
conducted at differentidentified by the
temperatures. ThePSO algorithm
results usingthat
demonstrate data from
HPPC experiments conducted at different temperatures. The results demonstrate
changes in temperature and SOC have an impact on the charging and discharging ohmic that
changes in temperature and SOC have an impact on the charging and discharging
internal resistance, two polarization internal resistances, and two time constants. In the ohmic
internal
followingresistance, two process,
SOC estimation polarization internal
the model resistances,
parameters and twoaccording
are determined time constants.
to dif- In the
Batteries 2023, 9, x FOR PEER REVIEWferent SOCSOC
following and estimation
different temperatures,
process, thewhich
modelcan meet the whole
parameters 12 of 18 to dif-
calculation according
are determined require-
ments at different ambient temperatures and SOC conditions.
ferent SOC and different temperatures, which can meet the whole calculation requirements
at different ambient temperatures and SOC conditions.

Figure 8. Identification results of each temperature parameter.

Figure 8. Identification results of each temperature parameter.

4.3. Fitting the SOC Upper and Lower Limit Functions

According to the actual charging and discharging electric quantity, calculations are
performed using the aforementioned defined upper and lower limits of the SOC at each
temperature. The final range of upper and lower limits of SOC at different temperatures
Batteries 2023, 9, 43 12 of 17

Figure 8. Identification results of each temperature parameter.

4.3. Fitting
4.3. Fitting the
the SOC
SOC Upper
Upper and
and Lower
Lower Limit
Limit Functions
Functions
According to
According to the
the actual
actual charging
charging and
and discharging
discharging electric
electric quantity,
quantity, calculations
calculations are
are
performed using
performed using the
the aforementioned
aforementioned defined
defined upper
upper and
and lower
lower limits
limits of
of the
the SOC
SOC at
at each
each
temperature. The
temperature. The final
final range
range of
of upper
upper and
and lower
lower limits
limits of
of SOC
SOC atat different
different temperatures
temperatures
are shown in Figure
are shown in Figure 9. 9.

100
100 98.5 98.2 97.9

80

60

40

20

3 5.7
0 1.6
0
25 15 10 5
Temperature/°C
Figure
Figure 9.
9. Upper
Upper and
and lower
lower limits
limits of
of SOC
SOC at
at different
different temperatures.
temperatures.

The
The upper
upper and
and lower
lower limits
limits of
of the
the SOC
SOC function
function can
can be
be acquired
acquired by by curve
curve fitting
fitting in
accordance with
withthethedetermined
determined upper
upper andand
lowerlower limits
limits of SOC
of SOC at temperatures.
at four four temperatures.
Based
Batteries 2023, 9, x FOR PEER REVIEW
Based
on the on the results
results of the fitting,
of the fitting, the twothe twooflimits
limits of the
the SOC at SOC
otherattemperatures,
other temperatures,
such as 0of◦18
13 such C,
can0 be
as °C,simulated. The fitting
can be simulated. Thecurves
fittingare shown
curves arein Figurein10.
shown Figure 10.
SOC/%

Figure 10. Fitting

Figure 10. Fitting curves.
curves.

SOC
SOC upper
upper limit
limit function:
SOCUT = 0.2671 ⋅ e 0.08937T + 97.51 (26)
SOC T = 0.2671 · e0.08937T + 97.51 (26)
SOC lower limit function: U
SOC TL = 9.899 ⋅ e −0.08702T − 1.115
SOC lower limit function: (27)
Considering this figure, it is apparent that the fitting curves are capable of displaying
SOCLT of
the actual upper and lower limits = SOC · e−0.08702T
9.899with − 1.115
accuracy. Thus, SOC estimations and (27)
ca-
pacity estimations will be carried out based on this result.
Considering this figure, it is apparent that the fitting curves are capable of displaying
the actual upper and lower limits of SOC with accuracy. Thus, SOC estimations and
4.4. Estimation Results of SOC at Different Temperatures and Variable Temperatures
capacity estimations will be carried out based on this result.
The online OCV estimation of the LiFePO4 battery under different ambient tempera-
tures was carried
4.4. Estimation out by
Results the method
of SOC of online
at Different OCV identification.
Temperatures Figure 11 displays the
and Variable Temperatures
estimation result taking 45 °C as an example.
The online OCV estimation of the LiFePO battery under different ambient tempera-
4
tures was carried out by the method of online OCV identification. Figure 11 displays the
estimation
3.5 result taking 45 ◦ C as an example.
(a)
OCV/V

3
Estimation
Reference
2.5
0 1 2 3 4 5
 4.4. Estimation Results of SOC at Different Temperatures and Variable Temperatures
The online OCV estimation of the LiFePO4 battery under different ambient temp
tures was carried out by the method of online OCV identification. Figure 11 displays
Batteries 2023, 9, 43 13 of 17
estimation result taking 45 °C as an example.

3.5
(a)

OCV/V
3
Estimation
Reference
2.5
0 1 2 3 4 5

(b)
Error/mv
0

-100
Error
-200
0 1 2 3 4 5
Time/104 s

Figure11.
Figure OCVestimation
11.OCV estimation result
result andand estimation
estimation at 45 ◦at
error error C. 45
(a) °C.
OCV(aestimation
) OCV estimation
result andresult an
estimation
(b) estimationerror.
error.

The error in the estimation results at the beginning and the end of OCV are relatively
The error in the estimation results at the beginning and the end of OCV are relati
large. However, the purpose of online OCV estimation in this paper is to distinguish
large. However,
the plateau the purpose
and non-plateau OCVofphases
onlineofOCV estimation
a LiFePO in this paper is to distinguish
4 battery. It is worth noting that
plateau
there and non-plateau
is a significant differenceOCVin thephases of OCV
estimated a LiFePO 4 battery.
between It is and
the plateau worth noting that the
non-plateau
a significant
phases. difference
Consequently, duringinthethelowestimated
SOC interval,OCVlargebetween the plateau
OCV estimation and
errors will notnon-pla
have
phases. Consequently, during the low SOC interval, large OCV estimation the
an effect on the accuracy of subsequent SOC estimation results. In Section 4.1, errors wil
OCV
havecurves at different
an effect on the temperatures
accuracy of obtained
subsequent by the variable
SOC temperature
estimation experiment
results. In Section 4.1
were considered to be same, and so is the plateau phase and non-plateau phase of the
OCV curves at different temperatures obtained by the variable temperature experim
OCV curves at different temperatures. Therefore, the OCV curve at 25 ◦ C in the variable
were considered
Batteries 2023, 9, x FOR PEER REVIEWtemperature experimentto be same, uniformly.
is applied and so is ThetheOCV
plateau
curvephase
and theandtwo non-plateau
plateau 14 of 18phase o
periods
are shown in Figure 12. The red rectangles show the two OCV plateau periods (3.290 Vin
OCV curves at different temperatures. Therefore, the OCV curve at 25 °C tothe vari
temperature
3.298 V and 3.331 experiment
V to 3.334 V) is of
applied uniformly.
the LiFePO 4 battery The OCV
range. curve
The rest and regions
of OCV the twoareplateau p
non-plateau
are non-plateau periods.
periods.Thereby,
Thereby, a specific
a value
specific to
value distinguish
ods are shown in Figure 12. The red rectangles show the two OCV plateau
to the
distinguish plateau
the phase
plateau from
phase the
fromperiods (3
non-plateau
the non-plateau
V to phase
3.298 V and of
phase the battery
of the
3.331 V batteryOCV
to 3.334OCV can be obtained.
canthe
V) of be LiFePO
obtained.4 battery range. The rest of OCV reg

OCV-SOC
3.0

2.9
Plateau
2.8

2.7

2.6
0 20 40 60 80 100
SOC/%

Figure 12.OCV–SOC
Figure12. OCV–SOCat
at25 ◦ Cand
25°C andthe
theplateau
plateauperiod
periodcurve.
curve.

Afterdistinguishing
After distinguishingthe theplateau
plateauphase
phasefrom
fromthe
thenon-plateau
non-plateauphase
phasebased
basedononthetheOCV
OCV
estimationresults,
estimation results,the
theNEDC
NEDCexperiment
experimentresults
resultsatatdifferent
differenttemperatures,
temperatures,thetheOCV–SOC
OCV–SOC
curveatata atemperature
curve temperatureofof 2525°C◦ and
C and
thethe discharging
discharging capacity
capacity willwill be used
be used to estimate
to estimate the
the SOC.
SOC. This is This is for
for the the
sake ofsake of validating
validating the fast convergence
the fast convergence of the EKF
of the improved improved EKF
algorithm
algorithm
tracking. On tracking.
the otherOnhand,
the other hand, considering
considering that the
that the capacity capacity
error error the
will affect willSOC
affect the
esti-
SOC estimation, 10% of the original SOC error and 10% of the capacity error
mation, 10% of the original SOC error and 10% of the capacity error are introduced. The are introduced.
The estimation ◦ C, 25 ◦ C, 5 ◦ C and −15 ◦ C are illustrated in
estimation resultsresults and at
and errors errors at25
45 °C, 45°C, 5 °C and −15 °C are illustrated in Figure 13.
Figure 13.
45°C 100
25°C
100
(a) (a)
SOC/%
SOC/%

50 50
Reference Reference
EKF estimation EKF estimation
0 0
0 1 2 3 4 5 0 1 2 3 4 5
r/%
r/%

0 (b) 0 (b)
 estimation results, the NEDC experiment results at different temperatures, the OCV–SOC
curve at a temperature of 25 °C and the discharging capacity will be used to estimate the
SOC. This is for the sake of validating the fast convergence of the improved EKF algorithm
tracking. On the other hand, considering that the capacity error will affect the SOC esti-
Batteries 2023, 9, 43 14 of 17
mation, 10% of the original SOC error and 10% of the capacity error are introduced. The
estimation results and errors at 45 °C, 25 °C, 5 °C and −15 °C are illustrated in Figure 13.

45°C 100
25°C
100
(a) (a)

SOC/%
SOC/%
50 50
Reference Reference
EKF estimation EKF estimation
0 0
0 1 2 3 4 5 0 1 2 3 4 5

SOC error/%
SOC error/% (b) 0 (b)
0

-5 -5
Error Error
Error reference line Error reference line
-10 -10
0 1 2 3 4 5 0 1 2 3 4 5
Time/104s Time/104s

100 -15°C
(a)

SOC/%
SOC/%

50
Reference
EKF estimation
0
0 1 2 3 4 5

SOC error/%
(b)
SOC error/%

-5
Error
Error reference line
-10
0 1 2 3 4 5
Time/104s

13.SOC
Figure 13.
Figure SOCestimation
estimation results andand
results errors at different
errors temperatures.
at different (a) SOC
temperatures. (aestimation result atresult at
) SOC estimation
each temperature
each temperatureand (b)(bSOC
and ) SOCestimation errorerror
estimation at each temperature.
at each temperature.

In the plateau phase, the algorithm tends to be calculated by the ampere-hour in-
In the plateau phase, the algorithm tends to be calculated by the ampere-hour inte-
tegration method. Due to the capacity error, there exists a cumulative error in the SOC
gration method. Due to the capacity error, there exists a cumulative error in the SOC cal-
calculation. In the non-plateau period, the algorithm tends to be calculated by the battery
culation. In the non-plateau
model combined period,
with the improved themethod.
EKF algorithmThetends to be calculated
accumulated by the battery
error is corrected
model combined
quickly and follows with theSOC.
the real improved EKF
The above method.
results The accumulated
demonstrate that the LiFePOerror is corrected
4 battery
quickly
SOC can and follows the
be accurately real SOC.
estimated Theambient
at four above temperatures
results demonstrate
with the that the LiFePO
proposed method,4 battery

and the maximum error is no more than 4%.

In the process of SOC estimation at variable temperatures, the battery model param-
eters for the full SOC range at that temperature are acquired first based on the ambient
temperature. Then, the model parameter combination of the SOC is found according to
the SOC estimated value at this time. This cycle is iterated to achieve the application of
accurate model parameters at different ambient temperatures and SOC. For verifying the
reliability of the algorithm, an error of 5% was set for both the SOC and capacity initial
values. In addition, noise was added to the current and voltage in the input to simulate
an actual sensor. These two variable temperature experiments produced results which are
given in Figures 14 and 15. The SOC errors fluctuate around 0%, and the estimated SOC
of the two experiments with variable temperature can quickly converge to the real SOC.
Under the condition that the temperature and current change simultaneously, the final
accuracy of SOC estimation is within 3%.
In conclusion, with the conditions of SOC initial error, capacity error and current
and voltage drift, using the improved EKF algorithm, the SOC estimation accuracy and
robustness are maintained at variable temperatures.
 an actual sensor. These two variable temperature experiments produced results which
an actual
given sensor. These
in Figures 14 andtwo
15. variable
The SOCtemperature experiments
errors fluctuate produced
around 0%, results
and the whichS
estimated
given
of the in
twoFigures 14 and with
experiments 15. The SOC errors
variable fluctuate
temperature canaround 0%,
quickly and theto
converge estimated
the real SS
of
Under the condition that the temperature and current change simultaneously,real
the two experiments with variable temperature can quickly converge to the theSf
Batteries 2023, 9, 43
Under
accuracy theofcondition that theis temperature
SOC estimation within 3%. and current change simultaneously,
15 of 17
the f
accuracy of SOC estimation is within 3%.
100
100 (a)

SOC/%
(a)

SOC/%
50
Reference
50 EKF estimation
Reference
0 EKF1estimation 2 3 4

error/%
0 1 2 3 4 (b)
0
error/%
0 (b)
-2
SOC

-2 Error
-4 Error reference line
SOC

Error
-4 0 Error1reference line 2 3 4
Time/104 s
0 1 2 3 4
Time/104 s
Figure 14. SOC estimation result and error of the temperature rise experiment. (a) SOC estima
result
Figure 14.SOC
Figure and
14. (b) estimation
SOC SOC result
estimation
estimation and error
error.
result of the of
and error temperature rise experiment.
the temperature (a) SOC estimation
rise experiment. (a) SOC estima
result and (b) SOC estimation error.
result and (b) SOC estimation error.
100
(a)
100
SOC/%

50 (a)
SOC/%

50 Reference
EKF estimation
0 Reference
0 EKF1estimation 2 3 4 5
0
20
error/%

1 2 3 4 5 (b)
20
error/%

(b)
0
-2
SOC

Error
-2
-4 Error reference line
SOC

Error
-4 0 Error reference line
1 2 3 4 5
4
0 1 Time/10
2 3s 4 5
Time/104s
Figure SOC
15.SOC
Figure 15. estimation
estimation result
result and error
and error of the of the temperature
temperature drop experiment.
drop experiment. (a) SOC estima
(a) SOC estimation
result
result and
and (
(b)b )
SOCSOC estimation
estimation error.
error.
Figure 15. SOC estimation result and error of the temperature drop experiment. (a) SOC estima
result and (b) SOC estimation error.
4.5. Capacity Estimation
In conclusion, with the conditions of SOC initial error, capacity error and current
To
In confirm the accuracy
conclusion, with ofconditions
the IRLS algorithm for estimating the LiFePO 4 battery
voltage drift, using the the
improved EKFofalgorithm,
SOC initial error,
the SOCcapacity error
estimation and current
accuracy and
capacity at low temperatures, NEDC experiments were carried out on fresh and aged
voltage
bustness drift, using
are maintained
LiFePO4 batteries
the improved
atbattery
at 0 ◦ C. The variable SOC
EKF algorithm, the SOC estimation accuracy and
temperatures.
at 0 ◦ C was first estimated with the improved
bustness are maintained at variable temperatures.
EKF algorithm. Afterwards, the IRLS algorithm was used to estimate the capacity by
4.5. Capacity
iterative Estimation
calculations. The results of the capacity estimation are presented in Table 3. Due to
4.5. Capacity
the precise SOC
To confirmEstimation
estimation result at 0of◦ C,
the accuracy thetheIRLS
capacity error estimated
algorithm by the IRLS the
for estimating algorithm
LiFePO4 bat
◦
at 0 C is no more than 1%. As a result, battery capacity estimation by the IRLS algorithm
To confirm
capacity the accuracy ofNEDC
at low temperatures, the IRLS algorithmwere
experiments for estimating
carried outtheonLiFePO 4 bat
fresh and a
in the whole range of SOC can maintain a high accuracy at low temperature. Meanwhile,
capacity
LiFePO at low
the estimation errortemperatures,
4 batteries atstable
is 0 °C.for
The NEDC
battery
both experiments
SOC
fresh and were
at 0batteries.
aged °C was carried
first out on
estimated withfresh and a
the impro
LiFePO 4 batteries
EKF algorithm. at 0 °C. Thethe
Afterwards, battery SOC at 0 °C
IRLS algorithm was
was first
used toestimated with
estimate the the impro
capacity by
EKF
Table algorithm.
ative Afterwards,
3.calculations. theof
Theparameters
Basic performance results ofIRLS algorithm
thebattery.
the capacity was usedare
estimation to estimate
presentedthe
incapacity
Table 3. by
Du
ative calculations. The results of the capacity Capacity
estimation are presented in Table 3. Du
Capacity Estimates Capacity Errors
Experimental Values
Fresh battery 20.50 Ah 20.53 Ah 0.15%
Aged battery 19.56 Ah 19.43 Ah 0.67%

5. Conclusions
For the purpose of achieving an accurate estimation of the SOC of a LiFePO4 battery
at different and variable temperatures, an improved EKF algorithm is proposed. Firstly,
with the forgetting factor, the OCV curve of the battery is identified online using the
least square method. Then, the OCV values of the plateau phase and non-plateau phase
Batteries 2023, 9, 43 16 of 17

are distinguished by the OCV curve at uniform temperature, and the Kalman gain is
adjusted between the zones to accurately estimate the SOC of the full range. In addition,
for estimating the variable temperature SOC, we drew upon a variable temperature NEDC
experiment. The results show that all SOC errors are within 3%. On account of the
pinpoint estimation of low temperature battery SOC, the low temperature battery capacity
is estimated through the use of the IRLS algorithm. A high degree of accuracy is observed
in the capacity estimation results for both fresh and aged batteries at low temperature, with
errors all within 1%.

Author Contributions: Y.Z.: methodology, data curation, writing—review and editing. M.W.: soft-
ware, data curation, validation, writing—original draft. G.W.: data curation, writing—review and
editing. Z.X.: supervision, writing—review and editing. Y.S.: supervision, writing—review and
editing. All authors have read and agreed to the published version of the manuscript.
Funding: This research was supported by the National Natural Science Foundation of China (NSFC)
under the Grant number of 52277222 and the Shanghai Science and Technology Development Fund,
22ZR1444500.
Data Availability Statement: Not applicable.
Acknowledgments: The authors thank the National Natural Science Foundation of China (NSFC)
under the Grant number of 52277222 and the Shanghai Science and Technology Development Fund
22ZR1444500. Furthermore, thanks to Yuejiu Zheng, Guan Wang, Zhanlong Xiao and Yuedong Sun
for their support in the realization of the test and the method.
Conflicts of Interest: The authors declare no conflict of interest.

Abbreviations
BMS Battery management system
ECM Equivalent circuit model
SOC State of charge
OCV Open circuit voltage
PSO Particle swarm optimization
HPPC Hybrid pulse power characteristic
NEDC New European Driving Cycle
GA Genetic algorithm
KF Kalman filter
EKF Extended Kalman filter
FFRLS Forgetting factor recursive least square
OLS Ordinary least squares
IRLS Iteratively reweighted least squares
RMSE Rooted mean squared error

References
1. You, H.; Zhu, J.; Wang, X.; Jiang, B.; Sun, H.; Liu, X.; Wei, X.; Han, G.; Ding, S.; Yu, H.; et al. Nonlinear health evaluation for
lithium-ion battery within full-lifespan. J. Energy Chem. 2022, 72, 333–341. [CrossRef]
2. Xu, Z.; Wang, B.; Zhao, S.; Li, H.; Yu, H. Oxygen anionic redox activated high-energy cathodes: Status and prospects.
eTransportation 2021, 8, 100118.
3. Hu, G.; Huang, P.; Bai, Z.; Wang, Q.; Qi, K. Comprehensively analysis the failure evolution and safety evaluation of automotive
lithium ion battery. eTransportation 2021, 10, 100140. [CrossRef]
4. Qin, P.; Sun, J.; Yang, X.; Wang, Q. Battery thermal management system based on the forced-air convection: A review.
eTransportation 2021, 7, 100097. [CrossRef]
5. Lu, L.; Han, X.; Li, J.; Hua, J.; Ouyang, M. A review on the key issues for lithium-ion battery management in electric vehicles.
J. Power Sources 2013, 226, 272–288. [CrossRef]
6. Liu, K.; Li, K.; Peng, Q.; Zhang, C. A brief review on key technologies in the battery management system of electric vehicles.
Front. Mech. Eng. 2019, 14, 47–64. [CrossRef]
7. Coleman, M.; Lee, C.K.; Zhu, C.; Hurley, W.G. State-of-Charge Determination from EMF Voltage Estimation: Using Impedance,
Terminal Voltage, and Current for Lead-Acid and Lithium-Ion Batteries. IEEE Trans. Ind. Electron. 2007, 54, 2550–2557. [CrossRef]
Batteries 2023, 9, 43 17 of 17

8. Zheng, Y.; Lu, L.; Han, X.; Li, J.; Ouyang, M. LiFePO4 battery pack capacity estimation for electric vehicles based on charging cell
voltage curve transformation. J. Power Sources 2013, 226, 33–41. [CrossRef]
9. Zahid, T.; Li, W. A Comparative Study Based on the Least Square Parameter Identification Method for State of Charge Estimation
of a LiFePO4 Battery Pack Using Three Model-Based Algorithms for Electric Vehicles. Energies 2016, 9, 720. [CrossRef]
10. Sangwan, V.; Kumar, R.; Rathore, A.K. State-of-Charge estimation of Li-ion battery at different temperatures using particle filter.
J. Eng. 2019, 2019, 5320–5324. [CrossRef]
11. Wang, Y.; Wang, L.; Li, M.; Chen, Z. A review of key issues for control and management in battery and ultra-capacitor hybrid
energy storage systems. eTransportation 2020, 4, 100064. [CrossRef]
12. Tang, X.; Wang, Y.; Zou, C.; Yao, K.; Xia, Y.; Gao, F. A novel framework for Lithium-ion battery modeling considering uncertainties
of temperature and aging. Energy Convers. Manag. 2019, 180, 162–170. [CrossRef]
13. Wang, C.; Xing, B.; Zhou, J. A review of traction battery model and parameter identification in electric vehicle. In E3S Web of
Conferences; EDP Sciences: Les Ulis, France, 2020; Volume 185.
14. Yang, H.; Wang, P.; An, Y.; Shi, C.; Sun, X.; Wang, K.; Zhang, X.; Wei, T.; Ma, Y. Remaining useful life prediction based on denoising
technique and deep neural network for lithium-ion capacitors. eTransportation 2020, 5, 100078. [CrossRef]
15. Su, L.; Wu, M.; Li, Z.; Zhang, J. Cycle life prediction of lithium-ion batteries based on data-driven methods. eTransportation 2021,
10, 100137. [CrossRef]
16. Jiang, B.; Dai, H.; Wei, X.; Zhu, L.; Sun, Z. Online Reliable Peak Charge/Discharge Power Estimation of Series-Connected
Lithium-Ion Battery Packs. Energies 2017, 10, 390. [CrossRef]
17. Hu, X.; Li, S.; Peng, H. A comparative study of equivalent circuit models for Li-ion batteries. J. Power Sources 2012, 198, 359–367.
[CrossRef]
18. Feng, T.; Yang, L.; Zhao, X.; Zhang, H.; Qiang, J. Online identification of lithium-ion battery parameters based on an improved
equivalent-circuit model and its implementation on battery state-of-power prediction. J. Power Sources 2015, 281, 192–203.
[CrossRef]
19. Xiong, R.; Sun, F.; He, H.; Nguyen, T.D. A data-driven adaptive state of charge and power capability joint estimator of lithium-ion
polymer battery used in electric vehicles. Energy 2013, 63, 295–308. [CrossRef]
20. Waag, W.; Fleischer, C.; Sauer, D.U. On-line estimation of lithium-ion battery impedance parameters using a novel varied-
parameters approach. J. Power Sources 2013, 237, 260–269. [CrossRef]
21. Johnson, V.H. Battery performance models in ADVISOR. J. Power Sources 2002, 110, 321–329. [CrossRef]
22. He, Y.; Li, Q.; Zheng, X.; Liu, X. Equivalent hysteresis model based SOC estimation with variable parameters considering
temperature. J. Power Electron. 2021, 21, 590–602. [CrossRef]
23. Xing, Y.; He, W.; Pecht, M.; Tsui, K.L. State of charge estimation of lithium-ion batteries using the open-circuit voltage at various
ambient temperatures. Appl. Energy 2014, 113, 106–115. [CrossRef]
24. Leo, W.; Markus, L. Quantifiability of inherent cell-to-cell variations of commercial lithium-ion batteries. eTransportation 2021,
9, 100129.
25. Song, Z.; Yang, X.G.; Yang, N.; Delgado, F.P.; Hofmann, H.; Sun, J. A study of cell-to-cell variation of capacity in parallel-connected
lithium-ion battery cells. eTransportation 2021, 7, 100091. [CrossRef]
26. Xu, Y.; Hu, M.; Fu, C.; Cao, K.; Su, Z.; Yang, Z. State of Charge Estimation for Lithium-Ion Batteries Based on Temperature-
Dependent Second-Order RC Model. Electronics 2019, 8, 1012. [CrossRef]
27. Sun, F.; Xiong, R.; He, H.; Li, W.; Aussems, J.E.E. Model-based dynamic multi-parameter method for peak power estimation of
lithium–ion batteries. Appl. Energy 2012, 96, 378–386. [CrossRef]
28. Fan, J.; He, H.; Xiong, R. Evaluation of Lithium-Ion Battery Equivalent Circuit Models for State of Charge Estimation by an
Experimental Approach. Energies 2011, 4, 582–598.
29. Dai, H.; Xu, T.; Zhu, L.; Wei, X.; Sun, Z. Adaptive model parameter identification for large capacity Li-ion batteries on separated
time scales. Appl. Energy 2016, 184, 119–131. [CrossRef]
30. Yang, F.; Xing, Y.; Wang, D.; Tsui, K.L. A comparative study of three model-based algorithms for estimating state-of-charge of
lithium-ion batteries under a new combined dynamic loading profile. Appl. Energy 2016, 164, 387–399. [CrossRef]
31. Dai, H.; Wei, X.; Sun, Z.; Wang, J.; Gu, W. Online cell SOC estimation of Li-ion battery packs using a dual time-scale Kalman
filtering for EV applications. Appl. Energy 2012, 95, 227–237. [CrossRef]
32. Cheng KW, E.; Divakar, B.P.; Wu, H.; Ding, K.; Ho, H.F. Battery Management System (BMS) and SOC Development for Electrical
Vehicles. IEEE Trans. Veh. Technol. 2011, 60, 76–88. [CrossRef]
33. Cui, X.; He, Z.; Li, E.; Cheng, A.; Luo, M.; Guo, Y. State-of-charge estimation of power lithium-ion batteries based on an embedded
micro control unit using a square root cubature Kalman filter at various ambient temperatures. Int. J. Energy Res. 2019, 43,
3561–3577. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.