Energy 176 (2019) 505e520

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

A new approach to consider the influence of aging state on Lithium-

ion battery state of power estimation for hybrid electric vehicle
M.J. Esfandyari a, b, V. Esfahanian a, b, *, M.R. Hairi Yazdi a, H. Nehzati a, b, O. Shekoofa c
a
School of Mechanical Engineering, University of Tehran, Tehran, Iran
b
Vehicle, Fuel and Environment Research Institute (VFERI), University of Tehran, Tehran, Iran
c
Satellite Research Institute, Iranian Space Research Center, Tehran, Iran

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

Article history: Battery State of Power (SoP) estimation is one of the most crucial tasks of the battery management
Received 15 October 2018 system in electric and hybrid electric vehicles. The inevitable error in estimates of battery State of Charge
Received in revised form (SoC) and State of Health (SoH) is a cause of inaccuracies towards estimating the SoP for an aged battery.
20 March 2019
To overcome this, the present study aims to propose a new approach for predicting an aged cell SoP in
Accepted 29 March 2019
Available online 1 April 2019
which no a priori knowledge of battery SoH is required and the estimation method is robust to inac-
curacies of SoC estimates. Accordingly, a combined reference mode of constant-current and constant-
voltage is utilized to estimate fresh cell SoP which is then adapted to various aging states using a
Keywords:
Fuzzy logic-based model-less control
model-less control system. The control system, which belongs to a class of fuzzy logic-based controllers,
Battery aging state benefits form a closed-loop framework leading to a more reliable and accurate SoP estimate. For veri-
State of power estimation fication, an experimental setup comprised of fresh and aged LiFePO4 cell samples is designed and the
Battery management system (BMS) extracted data are utilized in a Model-in-the-Loop simulation for a hybrid electric vehicle. The results
demonstrate the improved accuracy and robustness of SoP estimation while achieving a guaranteed safe
operation of battery.
© 2019 Elsevier Ltd. All rights reserved.

1. Introduction [13,14]. Some PID controller based methods have been introduced
in which the battery current or power is limited when the SOA
Due to high energy density and specific power, Lithium-ion limits are exceeded. This type of protection has two major draw-
batteries have become a popular candidate for Electric Vehicles backs. First, the limitation of battery current or power is started
(EVs) and Hybrid Electric Vehicles (HEVs) energy storage system only when the SOA limits are exceeded and this may damage the
[1e3]. The Lithium-ion batteries are very sensitive to operating battery. Second, the reduction of battery power is occurred sud-
conditions so that their operation beyond the safe region, which is denly and this may lead to dangerous driving conditions [15e17].
commonly provided by manufacturer, reduces the battery life cycle Battery State of Power (SoP) is defined as the maximum avail-
considerably and in some cases is very dangerous [4e6]. able power which can be drawn from or received by the battery in
The Battery Management System (BMS) in EVs and HEVs is some upcoming seconds [18,19]. The SoP estimation can be used in
responsible for monitoring, management and protection of the the Vehicle Control Unit (VCU) in order to limit the battery within
batteries [7e10]. When the required power from the battery is high its specifications thereby achieving a safe operation [20e22]. The
(e.g. during acceleration, regenerative braking, etc.) a rapid change SOA is characterized by current, voltage, temperature and State of
in the battery current and voltage is occurred and the battery may Charge (SoC) limits. However, due to the slow changes, the battery
reach the boundary of its Safe Operating Area (SOA) [11,12]. The SoC and temperature are not the major limiting factors for battery
transient change in battery voltage will also be intensified as the SoP estimation when the prediction horizon is less than ten sec-
battery gets aged due to increasing in battery internal resistance onds [23e25]. Besides the introduction of methods based on
characteristic maps and machine learning techniques, in which the
adaptation of models to various operational conditions and aging
* Corresponding author. School of Mechanical Engineering, University of Tehran, levels requires high computational cost and storage capacity
Tehran, Iran. [13,26,27], the methods based on battery Equivalent Circuit Model
E-mail address: [email protected] (V. Esfahanian).

https://doi.org/10.1016/j.energy.2019.03.176
0360-5442/© 2019 Elsevier Ltd. All rights reserved.
506 M.J. Esfandyari et al. / Energy 176 (2019) 505e520

(ECM) have been widely applied for SoP estimation due to the platform containing a hybrid electric city bus feedforward model is
consideration of battery dynamics and less computational cost used. This systematic evaluation of the SoP estimation method has
[11,23]. A simple model including an electromotive force with a rarely been investigated in the literature. Also, an experimental
series resistance is employed by Plett to estimate the battery peak setup comprised of fresh and aged samples of LiFePO4 cell is
power capability [19]. Assuming the battery SoC is known, the developed and the results are then used for verification of the
presented method uses both the voltage and SoC as the limiting proposed SoP estimation method. Structure of the paper is as fol-
factors for estimating the battery SoP. Sun et al. investigated the lows. In section 2, battery ECM is developed using the results of
polarization effects on the ECM where both the RC network pa- experimental tests. The proposed SoP estimation method is
rameters are functions of SoC [28]. The dependence of the charge described in section 3 and the driving cycle results are presented in
transfer resistance on battery current has been developed using section 4.
some approximations in the Butler-Volmer equation [29e32], and
Waag et al. considered this dependency in battery available power 2. Battery model
prediction [33]. Using a simplified fuzzy-based ECM, the particle
filter method has been used for SoP estimation in which the SoC is 2.1. Model selection
used as the input parameter. Also the dependency of the charge
transfer resistance to both the current and the SoC has been studied As the first step for estimation of battery SoP, a dynamic model
[34]. Wik et al. presented an adaptive SoP estimation method based for the fresh battery is developed. The model, as depicted in Fig. 1,
on an incremental analysis which benefits from easy imple- consists of the open circuit voltage Voc, internal resistance R0 and a
mentation and low computational cost. The power capability of RC circuit describing the charge-transfer phenomena which con-
Lithium-ion battery decreases as the battery getting aged and this is tains the charge transfer resistance Rct and double-layer capaci-
mainly due to increasing of battery impedance parameters. tance Cdl. The governing equations for the proposed model in
Therefore, it is essential to adapt the SoP estimation with battery discrete time are given in Eq. (1).
aging level. A joint estimator is developed for calculation of both

the SoC and SoP of the battery considering different known health R TsC  Ts
Vc ððk þ 1ÞTs Þ ¼ Vc ðkTs Þe ct dl þ Rct 1  e Rct Cdl IðkTs Þ
conditions [35,36]. The ECM parameters are simply updated in a
period of time using the stored data from various health conditions.
Waag et al. presented an available power estimation method in Vt ððk þ 1ÞTs Þ ¼ Voc ððk þ 1ÞTs Þ  R0 Iððk þ 1ÞTs Þ  Vc ððk þ 1ÞTs Þ
which the battery State of Health (SoH) is taken into account
(1)
through the on-line estimation of ECM parameters [37]. Similarly,
Pei et al. considered the dependency of ECM parameters to battery
where k is an integer representing the discrete time and Ts is the
SoH for available power estimation [38]. They used two parallel
sampling interval. Parameters of the ECM model are obtained from
extended Kalman filters one for on-line estimation of ECM pa-
the experimental test results which are explained in the following
rameters and the other for estimation of the polarization current.
section.
As a major challenge, the accuracy of battery SoP prediction is
influenced by the inevitable errors of SoC and SoH estimates.
2.2. Battery experiments
Meanwhile, the co-estimation of these states considering their ef-
fects on each other makes the power capability estimation more
2.2.1. Test bench
complicated. This is why a few researches have been focused on the
Major components of test bench are illustrated in Fig. 2, which
co-estimation of the all three states [39,40]. On the other hand, the
consist of LiFePO4 cell which has nominal capacity of 10 Ah and
existing methods for SoP estimation of an aged battery have
nominal voltage of 3.2 V, AVL DC power supply, thermal chamber,
focused on on-line estimation of battery ECM parameters in order
BMS, CAN to PCMCIA interface, relay for protection against misuse
to consider the influence of various aging states or health condi-
and host computer for monitoring of the measured data and
tions of battery. However, sufficient lab and field validations are
command to the power supply. The BMS reads the cell voltage with
required to prove out the convergence and accuracy of the on-line
errors less than 0.1% (i.e. 5 mV in 0e5 V range). The charging and
parameter estimation algorithms in all possible operating condi-
discharging of the batteries is performed using the AVL DC power
tions. To overcome these issues, a new approach for estimating the
supply which supports currents up to ±300 A. A shunt resistor of
aged battery SoP is presented which consists of two steps. Firstly,
12 mU is used for measuring the battery current with errors less
the power capability for the fresh cell is estimated considering a
than 2%. The BMS transmits the cell voltage and current
combined reference mode of constant-current and constant-
voltage. Secondly, a model-less control system is designed to
compensate for aging state variations. The control system, which
belongs to a class of fuzzy logic based controllers, utilizes the actual
voltage and current values of the battery in a closed-loop frame-
work leading to a more reliable and accurate SoP estimation.
The major contributions of this paper are as follows: (1) Unlike
the aforementioned works, the proposed SoP estimation method
does not require a priori knowledge of battery aging state or health
condition and this eliminates the difficulties involved with accurate
on-line estimation of battery SoH or the ECM parameters. Also, it is
robust to inevitable inaccuracies in the battery SoC estimates; (2)
The closed-loop framework of the proposed SoP estimation method
not only provides more reliable and accurate estimations, but also
guarantees the battery current and voltage remain in the SOA; (3)
In order to evaluate the performance of proposed SoP estimation
method in a driving cycle, a Model-in-the-Loop (MiL) simulation Fig. 1. Battery equivalent circuit model.
 M.J. Esfandyari et al. / Energy 176 (2019) 505e520 507

Fig. 2. Test bench for LiFePO4 cell (1: GND, 2: command to relay from BMS, 3, 4: shunt resistor sense signals).

measurements to the CAN bus and the host PC receives data using indicating the charge-transfer phenomena. Referring to Eq. (1), the
CANcard2 interface from Softing company [41]. Also, an interface voltage response within this interval can be represented by Eq. (3)
software is designed in LabVIEW environment to be used for data in which Ts and N are the data sampling time and total data sam-
monitoring and storage. ples, respectively.
 
Vt ðkÞ ¼ VB þ c1 1  ec2 kTs ; 0  k  N (3)
2.2.2. Fresh cell model
As the first step, the capacity of the fresh cell is extracted using Values of c1 and c2 are identified using MATLAB identification
the following test procedure. The battery is charged with constant toolbox for each pulse response. Therefore, the ECM parameters can
current rate of 0.5C until the battery terminal voltage reaches its be obtained using Eq. (4). The identification results for the fresh cell
maximum value of 3.65 V. Then, the voltage is remained constant at 25  C are listed in Table 1.
until the battery current drops below 0.02C [7,42]. After a rest
period of 20 min, the battery is discharged with the constant cur- VB  VA c 1
Ri ¼ ; Rct ¼ 1 ; Cdl ¼ (4)
rent rate of 0.5C until the cut-off voltage of 2 V is reached. The total i i c2 Rct
Ampere-hour passed from the battery during this test indicates the
battery capacity. For a sample fresh cell, the cell capacity is obtained
to be 10.58 Ah. Accordingly, the fresh cell SoC can be estimated
using Eq. (2) in which h and Q are the coulombic efficiency and 3. SoP estimation method
capacity, respectively.
The battery SoP is defined as the maximum power which can be
hIðkTs ÞTs drawn from or received by the battery in some upcoming seconds
SoCððk þ 1ÞTs Þ ¼ SoCðkTs Þ þ (2) such that its safe operation is ensured [19]. The SoP depends on the
Q
battery load during the prediction horizon which is not typically
The battery ECM parameters are identified off-line using the predictable. Therefore, some reference modes should be introduced
charge and discharge current pulses at different SoCs for the bat- for the battery available power estimation. Schematic of four
tery [43,44]. In the meanwhile, from the fully charged state, the reference modes for each of the charge and discharge cases which
battery is discharged with a current pulse of 0.5C rate for 12 min can limit the power capability are presented in Fig. 4. These cases
and a consequent rest of 20 min duration and this procedure is are described below:
repeated for ten times. The voltage response is presented in Fig. 3
(a). After the end of the discharge test, the same procedure is I Constant current reference mode
performed for the charge case. The battery is charged with ten  Case (a) and Case (e) (see Fig. 4): battery voltage hits its SOA
sequences of charge pulses with subsequent rests until the limit (i.e. vmin or vmax) at the end of the pulse while the current
maximum voltage is reached. The result is depicted in Fig. 3 (b). is constantly remained within its predefined allowable region
For each pulse, values of the cell voltage at points A, B and C, as
(i.e. ichg
min
< i < idis
max ).
represented in a magnified view in Fig. 3 (a), can be used for
identification of the ECM parameters. The effect of the series  Case (b) and Case (f): battery current is equal to its SOA limit
chg
resistance on the response is illustrated by the sudden change in during the pulse (i.e. imin or idis
max ) while the voltage is within its
the terminal voltage between points A and B. The terminal voltage predefined allowable region (i.e. vmin < v < vmax).
behavior between points B and C related to the RC circuit response II Constant voltage reference mode
508 M.J. Esfandyari et al. / Energy 176 (2019) 505e520

Fig. 3. Pulse test for fresh cell, (a) voltage response for discharge test, (b) voltage response for charge test.

Table 1
ECM parameters for the fresh cell at 25  C.

SoC (%) Voc (V) R 0 ( U) Rct (U) Cdl (F)

a
Dis Chg Dis Chg Dis Chg Dis Chg Dis Chg

90.5 17.1 3.316 3.261 0.0076 0.0068 0.014265 0.013064 1137 1614
81.1 26.6 3.315 3.303 0.0078 0.0074 0.015557 0.014233 1348 1200
71.7 36.0 3.318 3.304 0.0072 0.007 0.018479 0.014602 2333 1357
62.4 45.5 3.303 3.305 0.0084 0.0076 0.017382 0.015545 2221 1371
53.2 54.9 3.285 3.312 0.0068 0.006 0.015777 0.016542 1668 2742
44.0 64.4 3.279 3.339 0.0086 0.0076 0.016774 0.014733 1425 1182
34.7 73.8 3.28 3.339 0.0084 0.0074 0.01944 0.01617 1576 1251
25.6 83.3 3.271 3.341 0.0064 0.0056 0.02157 0.018593 1778 1307
16.5 92.7 3.233 3.337 0.0092 0.0094 0.022664 0.024759 1394 1057
a
Dis ¼ Discharge, Chg ¼ Charge.

 Case (c) and Case (g): battery voltage is equal to its SOA limit operation. Therefore, in cases (a), (c), (e) and (g) the voltage is the
during the pulse (i.e. vmin or vmax) while the current is within limiting factor while in other cases the current is the limiting factor
chg for SoP estimation. In some cases, the maximum power capability is
its predefined allowable region (i.e. imin < i < idis
max ).
achieved following a combination of constant current and constant
 Case (d) and Case (h): battery current hits its SOA limit (i.e. ichg
min voltage modes. In this regard, as depicted in Fig. 5 (a) and Fig. 5 (b),
or idis
max ) at the beginning of the pulse while the voltage is the battery undergoes a constant current mode followed by a
constantly remained within its predefined allowable region constant voltage mode. The reference mode change occurs as soon
(i.e. vmin < v < vmax). as the voltage limit is reached at a hypothetical time DT* during the
prediction horizon. However, all of the reference cases described in
The battery power capability is limited by the current or voltage Fig. 4 can be a special case for the combined reference mode of
depending on whose SOA limit is reached sooner during the battery Fig. 5. In this regard, we have:
 M.J. Esfandyari et al. / Energy 176 (2019) 505e520 509

Fig. 4. Reference cases for SoP estimation, (a)e(d) discharge case, (e)e(h) charge case.

closed-loop control system is designed to compensate for various

aging states of the battery and inaccuracies in SoC estimations.

3.1. Model-based estimation for the fresh cell

The estimation of SoP can be based on cell voltage limit, current

limit or both of them. In the first case, the battery minimum/
maximum allowable voltage acts as the limiting factor for
computing the SoP and in the second case, the maximum/mini-
mum discharge/charge current is considered as the limiting factor
for calculation of the SoP. The formulations for each case are pre-
sented in the following sections.

3.1.1. Voltage-limited estimation

Fig. 5. Combination of constant current and constant voltage modes. In this case, the terminal voltage during the prediction horizon
is supposed to be equal to vmin and vmax for the discharge and
charge cases, respectively. To formulate this, Eq. (1) is first written
in state space form as given by Eq. (5).
8
< constant current mode; DT * ¼ DT

 Ts  Ts
constant voltage mode; DT * ¼ 0 xðk þ 1Þ ¼ e Rct Cdl xðkÞ þ Rct 1  e Rct Cdl uðkÞ
: |fflfflffl{zfflfflffl}
combined mode; 0 < DT * < DT |fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
A B (5)
In the previous studies, the constant current mode is usually
yðkÞ ¼ ð  1Þ xðkÞ þ ð  R0 Þ uðkÞ
used as the reference case for SoP estimation regardless of whether
|fflfflffl{zfflfflffl} |fflfflfflffl{zfflfflfflffl}
the current or the voltage acts as the limiting factor. The combined C D
reference mode for estimation of peak power was first presented by
Pei et al. [38]. In the present methodology, the battery SoP is esti- where the state variable, input and output are defined as xðkÞ ¼
mated in two steps. First, the SoP for the fresh cell is estimated Vc ðkÞ; uðkÞ ¼ IðkÞ; yðkÞ ¼ Vt ðkÞ  Voc ðkÞ. For the upcoming time
using the ECM parameters developed in the previous section window, the predicted output can be represented by Eq. (6).
assuming a combined reference mode. Second, a model-less
510 M.J. Esfandyari et al. / Energy 176 (2019) 505e520

yðk þ 1Þ ¼ Cxðk þ 1Þ þ Duðk þ 1Þ ¼ CðAxðkÞ þ BuðkÞÞ þ Duðk þ 1Þ

¼ CAxðkÞ þ CBuðkÞ þ Duðk þ 1Þ
yðk þ 2Þ ¼ Cxðk þ 2Þ þ Duðk þ 2Þ ¼ CðAxðk þ 1Þ þ Buðk þ 1ÞÞ þ Duðk þ 2Þ
¼ CA2 xðkÞ þ CABuðkÞ þ CBuðk þ 1Þ þ Duðk þ 2Þ
« (6)
s¼j h
X i
yðk þ jÞ ¼ CAj xðkÞ þ CAjs Buðk þ s  1Þ þ Duðk þ jÞ
s¼1
«

From Eq. (6), the input value can be obtained as below: predefined limits of charge or discharge (i.e. I chg
min
or I dis
max ) during the
prediction horizon and the corresponding terminal voltage at each
" s¼j 
# time sample is calculated. Using Eq. (6), we have:
1 X 
j js
uðk þ jÞ ¼ yðk þ jÞ  CA xðkÞ  CA Buðk þ s  1 Þ ; j " ! #
D s¼1 X
s¼j
j js
yðk þ jÞ ¼ CA xðkÞ þ CA B þ D ulim (10)
¼ 1; 2; :::; N s¼1
(7)
where ulim is C. Thus, substituting the values of A, B, C and D
where N is the number of data samples during the prediction ho- from Eq. (5) the terminal voltage at each time sample k is computed
rizon and values of A, B, C and D are given by Eq. (5). Therefore, at by Eq. (11).
each time sample k, assuming the terminal voltage equals the
predefined limits the maximum/minimum discharge/charge cur-
rent during the upcoming time window DT is calculated by Eq. (8).

" s¼j 
)#
1  Ts j X  Ts
T ðjsÞ
s DT
I dis
lim ðk þ jÞ ¼ vmin  Voc ðkÞ þ e Rct Cdl Vc ðkÞ þ Rct ð1  e Rct Cdl Þe Rct Cdl Iðk þ s  1Þ ; j ¼ 1; 2; :::;
R0 s¼1
Ts
" )# (8)
Xs¼j 
1  Ts j  Ts s
T ðjsÞ
DT
I chg
lim
ðk þ jÞ ¼ vmax  Voc ðkÞ þ e Rct Cdl Vc ðkÞ þ Rct ð1  e Rct Cdl Þe Rct Cdl Iðk þ s  1Þ ; j ¼ 1; 2; :::;
R0 s¼1
Ts

"( s¼j

R TsCj X R TsC TRs ðjsÞ DT
V dis
lim ðk þ jÞ ¼ Voc ðkÞ  e ct dl Vc ðkÞ  idis
max Rct ð1  e ct dl Þe C
ct dl Þ þ R0 ; j ¼ 1; 2; :::;
s¼1
Ts
"( (11)
s¼j

X 
chg R TsCj chg R TsC  Rs
T ðjsÞ
DT
V lim ðk þ jÞ ¼ Voc ðkÞ  e ct dl Vc ðkÞ  imin Rct ð1  e ct dl Þe ct Cdl Þ þ R0 ; j ¼ 1; 2; :::;
s¼1
Ts

In this case, the available power during the prediction horizon

can be obtained using Eq. (12).
Therefore, the available power for constant voltage mode during
the prediction horizon can be estimated using Eq. (9). DT
SoP idis ðk þ jÞ ¼ idis dis
max V lim ðk þ jÞ; j ¼ 1; 2; :::;
Ts
DT (12)
SoP vdis ðk þ jÞ ¼ vmin I dis
lim ðk þ jÞ; j ¼ 1; 2; :::; chg chg DT
Ts SoP ichg ðk þ jÞ ¼ imin V lim ðk þ jÞ; j ¼ 1; 2; :::;
(9) Ts
DT
SoP vchg ðk
chg
þ jÞ ¼ vmax I lim ðk þ jÞ; j ¼ 1; 2; :::;
Ts

3.1.3. Combined voltage and current-limited estimation

For the combined reference mode, the time at which the refer-
3.1.2. Current-limited estimation ence mode changes from constant current to constant voltage
In this case, the input current is assumed to be equal to the mode should first be determined (i.e. DT* in Fig. 5). To do so, the
 M.J. Esfandyari et al. / Energy 176 (2019) 505e520 511

current is assumed to be equal to its maximum/minimum value at system, the fresh cell-based estimated SoP is modified before
discharge/charge. In this case, value of DT* is the time at which the reaching the battery current and voltage SOA limits to compensate
terminal voltage equals the predefined limit of Vmin (or Vmax) at battery various aging states and the SoC estimate inaccuracies.
discharge (or charge). Since the input current is assumed to be Accordingly, the actual voltage and current values are used in a
constant, the terminal voltage can be written in the form given by closed-loop framework to make the SoP estimates adaptive to
Eq. (13). varying states of health and robust to inevitable inaccuracies pre-


 sent in ECM parameters and SoC values.
*T *
RK s  K Ts Owing to the high flexibility and easy implementation, the fuzzy
vmin ¼ Voc ðkTs Þ  Vc ðkTs Þe ct Cdl  R0 þ Rct 1  e Rct Cdl idis
max
logic controller is an appropriate candidate for control of nonlinear
*T


*
 systems [45e47]. In particular, for a battery cell with complex dy-
RK s  K Ts chg
vmax ¼ Voc ðkTs Þ  Vc ðkTs Þe ct Cdl  R0 þ Rct 1  e Rct Cdl imin namic behavior, a fuzzy logic controller as a model-less control
approach can eliminate the dependency of battery SoP estimation
(13) accuracy together with states of charge and health estimations
accuracy, thereby achieving a more reliable and safer battery
where K* denotes the time sample which corresponds with DT*.
operation in sense of voltage and current SOAs. To achieve this, the
Accordingly, the DT* can be computed using Eq. (14).
battery fresh cell-based estimated SoP is multiplied by the output
2 3 of a Fuzzy Logic Based Controller (FLBC), named “charge/discharge
6 7 correction coefficient”. The details of FLBC design considering
6 dis 7
6 v  V ðkT Þ þ ðR þ R ct max 7
Þi current-based modifications, voltage-based modifications and
DT *dis ¼ K *dis Ts ¼ Rct Cdl Ln6 7
min oc s 0
6 dis 7 combination of both are explained in the following sections.
6 Rct imax  Vc ðkTs Þ 7
4|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} 5
ddis
2 3 3.2.1. Current-based FLBC
As a first step, the input/outputs of the FLBC should be specified.
6 7
6 7 In the discharge case, three operating regions have been defined
6vmax  Voc ðkTs Þ þ ðR þ Rct Þichg 7
6 0 min 7 which are depicted in Fig. 6 (a). The maximum available discharge
DT *chg ¼ K *chg Ts ¼ Rct Cdl Ln6 7
6 chg
Rct imin  Vc ðkTs Þ 7 current reported by the manufacturer is defined by idis max. Therefore,
6 7
4|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} 5 Region III indicates the unsafe discharge current region. In order to
dchg
prevent the discharge current exceeding the SOA, control of the
current flow should be started before reaching the SOA limit, i.e.
(14)
idis dis
max . Therefore, a secondary current limit is defined, i.e. is , from
Different values of parameters ddis and dchg in Eq. (14) will lead to which the SoP modification process may be started. However, if the
one of the three reference modes which are described by Eq. (15). load varies such a way that the battery current remains in Region II,
8 no modification is required. Consequently, in addition to the
>
>  DT
*
operating region (i.e. u1 in Fig. 6 (a)) and the current value (i.e. u2) as
>
> constant current: d  e Rct Cdl 0DT * ¼ DT
>
< the inputs of the controller, another input should be defined which
constant voltage: d > 10DT * ¼ 0 (15) can be the instant rate of change for battery current (i.e. u3). As
>
>
>
> DT *
shown in Fig. 6 (a), at points “a” and “b”, the rate of change for
>
: combined mode: eRct Cdl < d  10DT * ¼ Rct Cdl LnðdÞ current are low and therefore no modification should be per-
formed. Due to high positive rate of change for the current at point
According to the obtained value of DT* at the time sample k, a “c”, the modification process may be started gradually.
sequence of available power values are calculated for the upcoming According to what explained above, the structure of the FLBC
time samples during the prediction horizon using Eq. (9) and Eq. designed for current-based SoP modification at discharge is illus-
(12). Then, the SoP can be estimated by selecting the power with trated in Fig. 6 (b). The normalized inputs are defined by Eq. (17)
the lowest magnitude for each of the charge and discharge cases as which cover the three inputs explained in Fig. 6 (a).
represented by Eq. (16).


 edis idis
max  i
DT edis
~i ¼
i
¼
SoPdis ðkÞ ¼ min SoPdis ðk þ 1Þ;SoPdis ðk þ 2Þ;:::;SoPdis k þ idis dis
idis dis
Ts max  is max  is


 ! (17)
DT dedis
SoPchg ðkÞ ¼ max SoPchg ðk þ 1Þ;SoPchg ðk þ 2Þ;:::;SoPchg k þ edis
~ dis _dis
i ¼ K ei ei ¼ K dis
ei
i
Ts dt
(16)
In Fig. 6 (b), the parameters K dis dis
ei and K ui are input and output
scaling factors, respectively and t0 indicates the time at which the
battery current equals to idis
s , i.e. the time at which the SoP modi-
fication might began. The output udis i shows the percent of the fresh
3.2. Model-less compensator design cell-based estimated SoP which should be applied to remain in the
SOA of battery discharge current.
As mentioned in the prior sections, the power capability de- When the output udis i is 100%, no modification on the battery
creases as the battery gets aged; therefore, the battery aging state estimated SoP is required. The output should be decreased from
should also be taken into consideration to estimate the battery SoP. 100 if the battery current exceeds idiss (see Fig. 4 (a)) and the rate of
However, the inevitable error in estimates of SoC and SoH as well as change of battery discharge current is large enough in amount (see
the complexity and the high computational cost of concurrent Fig. 4 (b)). Defining the membership functions of the input/outputs
battery states estimation is still a challenge towards accurate SoP as represented in Fig. 7, this can be translated into the following
estimations. To overcome this issue, using a model-less control rules:
512 M.J. Esfandyari et al. / Energy 176 (2019) 505e520

Fig. 6. (a) Operating regions for battery discharge current (the positive current indicates the discharge case), (b), (c) structure of the FLBC for SoP modification at discharge for
current-based and voltage-based estimations, respectively.

dis 3.2.2. Voltage-based FLBC

edis
1. IF ~ ~_ is NS THEN udis
i is PB and ei i is Z
dis
For the battery voltage-based SoP modification at discharging,
edis
2. IF ~ ~_ dis
i is PB and ei is NM THEN ui is NS structure of the control system is presented in Fig. 6 (c). The inputs
dis
~_ is NB THEN udis of the fuzzy controller in this case are given by Eq. (19). The rule
edis
3. IF ~i is PB and ei i is NM
base and membership functions in this case are the same as the
current-based FLBC (Table 2 and Fig. 7).
The rule base for different input values is developed and pre-
sented in Table 2. Starting the modification process before reaching
the SOA limit (i.e. idis
max ), two goals can be achieved. First, it would be edis v  vmin
possible to prevent the battery current exceeding the SOA limits edis
~v ¼
s
¼
using the closed-loop communication with the VCU and second, vmin  vdis
s vdis
s  vmin
! (19)
rapid reduction of battery power could be avoided in cases where dis dedis
the power request from the battery is high for a short time, e.g. e_ v ¼ Kedis
~ e_dis
v ¼ Kedis v
v v
dt
during acceleration or regenerative braking.
Due to the use of normalized input/outputs for the FLBC, the Due to the use of normalized input/outputs for the FLBC, the
structure of the controller for battery SoP modification at charging structure of the controller for battery charge control is the same as
is the same as the discharge case. The normalized inputs for the discharge case. The normalized inputs for the charge case can
current-based SoP modification at charging is given by Eq. (18), be calculated by Eq. (20). In the voltage-based FLBC, the scaling
wherein the current is assumed to be negative at charging. The factors Ke and Ku are obtained by trial and error to be 1 and 1.7,
scaling factors Ke and Ku for the current-based FLBC are obtained by respectively.
trial and error to be 0.003 and 1.7, respectively.
! !
echg i  ichg dechg echg vmax  v chg
_ chg ¼ K chg e_chg ¼ K chg dev
chg
e_ i v
chg chg chg chg
~
ei ¼ i
¼ min
;~ ¼ Kei e_i ¼ Kei i
echg
~ ¼ ¼ ; ~
e
v v e v e
ichg  ichg ichg chg dt chg chg v v
s min s  imin vmax  vs vmax  vs dt
(18) (20)
 M.J. Esfandyari et al. / Energy 176 (2019) 505e520 513

Table 2
Rule base for the fuzzy controller.

~
e ẽ
PB PM PS Z NS NM NB
PVB PVB PVB PB Z Z Z Z
PB PVB PB PM Z Z NS NM
PM PVB PB PM Z NS NM NB
PS PVB PB PM PS Z NS NM
Z PB PM PS Z NS NM NB
NS PM PS Z NS NM NB NVB
NM PS Z NS NM NB NVB NVB
NB Z NS NM NB NVB NVB NVB

NVB: Negative Very Big, NB: Negative Big, NM: Negative Medium, NS: Negative
Small, Z: Zero, PS: Positive Small, PM: Positive Medium, PB: Positive Big, PVB:
Positive Very Big.

3.2.3. Combined current and voltage-based FLBC

If both the current and voltage of battery are taken into account
as the limiting factors, the final output for the FLBC, named “charge/
discharge correction coefficient”, can be acquired by Eq. (21) and
Eq. (22). As shown in Fig. 8 (a), the normalized correction coeffi-
cient is multiplied by the model-based estimated SoP of the fresh
battery for each of the charge and discharge cases to obtain the
aged battery power capability.
 
udis ¼ min udis dis
i ; uv (21)

 
uchg ¼ min uchg
i
; uchg
v (22)

As explained in aforementioned, the maximum available battery

power is achieved when either current or voltage SOA limits of the
battery is reached. Therefore, in order to have accurate estimations
of aged cell SoP, any modifications on the fresh cell-based estima-
tion should be performed when battery operates exactly on the SOA
limits. This may happen when the required power from the battery
is temporarily high (e.g. acceleration, regenerative braking, etc.) in
which the battery current or voltage reaches the safe operation
limit. The designed FLBCs prevent the current or voltage (depend-
ing which one acts as the limiting factor for SoP estimation) to
exceed its maximum/minimum allowable value at these conditions
through a closed-loop communication with the VCU. In this case,
two types of modifications are considered to be applied on the fresh
cell-based estimated SoP. If the time interval at which the battery
operates on the SOA limits is less than the prediction horizon (DT),
the SoP modification using the FLBC would be temporal and the
charge/discharge correction coefficient increases gradually as the
required power from the battery decreases. If the battery operation
on the SOA limits lasts equal to or more than DT seconds, the
charge/discharge correction coefficient at the end of the prediction
horizon will be applied permanently until the same condition
occurs.
These two types of modifications for voltage-based SoP esti-
mation at discharge are represented in Fig. 8 (b). At the time in-
terval [t1, t2], the FLBC provides a safe operation of battery through a
temporary modification of the SoP starting at time t1 and ending at
t2. Again, the modification process is started at time t3. But, since
the battery operation near the minimum allowable voltage lasts
more than the prediction horizon, DT, the modified SoP at time t4 is
in fact the actual SoP for the aged battery. Therefore, from now on
the value of udis v at time t4 will be applied as an upper limit for
~ , (b) ~e_ , (c) u
Fig. 7. Membership functions for fuzzy controller, (a) e ~. future values of the correction coefficient until the same condition
occurs. The maximum/minimum current and voltage values are
those provided by the manufacturer. The current and voltage values
chg dis chg
at which the modification process starts (i.e. idis s , is , vs , vs ) are
not fixed variables and can be chosen by the user. Obviously,
514 M.J. Esfandyari et al. / Energy 176 (2019) 505e520

Fig. 8. (a) The aged cell SoP estimation using the FLBC, (b) explanation of temporary and permanent modifications of voltage-based SoP estimation at discharge.

Table 3
Hybrid electric bus main data.

Component Specification

Vehicle Effective mass: 18000 kg

Frontal area: 6.75 m2
Drag coefficient: 0.79
Engine OM904, 130 kW
Generator Permanent Magnet, 130 kW
Traction motor AC Induction, 170 kW
Battery LiFePO4, 18.4 kWh

choosing these threshold values far from the SOA limits narrows Fig. 9. Block diagram of the series hybrid electric vehicle model.
the battery unlimited operation, but may lead to better accelerating
performance of the vehicle.
model. In this research, a MiL simulation platform is used to verify
the proposed SoP estimation method in a driving cycle.
4. Results and discussion

Battery safe operation in HEVs is achieved through an interac- 4.1. Case study: a series hybrid electric city bus
tion between the BMS and VCU. In other words, using the instant
value of battery SoP provided by the BMS, the VCU safely distributes A series hybrid electric city bus is taken into consideration as a
the power between the battery and the engine-generator. Accord- case study with specifications summarized in Table 3. Fig. 9 illus-
ingly, the performance of the battery SoP estimation method trates the components of feedforward simulation model of the
should be evaluated in a simulation platform containing the vehicle vehicle and their interactions. As depicted, the modified SoP is
 M.J. Esfandyari et al. / Energy 176 (2019) 505e520 515

estimated using the actual current and voltage values in the BMS Without using the FLBC (i.e. assuming the correction coefficients to
and is used in the VCU to regulate the power drawn from or be a hundred percent), a considerable violation from the SOA is
received by the battery. The control strategy developed by Lee et al. occurred and this proves the importance of considering the aging
[22] has been implemented in the VCU model which consists of a state in battery SoP estimation.
thermostatic control strategy combined with some additional Fig. 10 (c) represents the battery SoP using both the modified
logics taking into account the battery power capability. and non-modified estimations. As shown in Fig. 10 (c) and Fig. 10
(d), during the first 1000 s of the cycle, no modification is per-
4.2. Verification method formed on the SoP estimation since the battery operates within the
SOA. However, the first modification on the estimated SoP on
In order to verify the proposed method for SoP estimation, the charging is applied after 960 s where the battery reaches the SOA
actual capacity and ECM parameters for an aged cell sample are also limit (in this case, minimum allowable current). On the other hand,
extracted using the same test procedures as the fresh cell. The total the discharge power capability is first modified at time 1000 where
Ampere-hour passed from the aged LiFePO4 cell sample is obtained the cell voltage reaches its lower limit. As depicted in detail in
to be 9.84 Ah. Referring to Fig. 8 (a), the actual current and voltage Fig. 10 (e), the modification process starts at the time t1 before
values of the aged cell together with the ECM parameters of the reaching the lower limit of 2.1 V (i.e. vmin) and the discharge
fresh cell are utilized for power capability estimation. A common correction coefficient (udis) decreases from 100% (Fig. 10 (f)).
way to verify the battery SoP estimation is to apply constant voltage Through a closed-loop interaction with the VCU, the FLBC prevents
or constant current pulses and check whether the available power the voltage violating its minimum allowable value while modifying
corresponds with the battery current or voltage operation on the the estimated SoP. At the time t3, i.e. just after the cell voltage has
SOA limits or not [37,38]. Concerning this, the SoP estimation is been remained constant for DT seconds on the SOA limit, a per-
accurate if at least one of the following is accomplished: manent modification on the SoP estimation is established (see
Fig. 10 (f)) which in fact compensates the battery aging state.
1) The cell voltage is equal to (refer to cases (c) and (g) in Fig. 4) or Similarly, from the time t3 onward necessary modifications on the
hits (refer to cases (a) and (e) in Fig. 4) the specified limits during estimated SoP occur when reaching the SOA limits.
the pulse (i.e. vmin or vmax) while the current is within its pre- In Fig. 11 (a) and Fig. 11 (b), two time intervals of the cycle are
magnified to reflect a detailed performance of the estimation
defined allowable region (i.e. ichgmin
< i < idis
max ). method when the available power is limited by the battery voltage.
2) The current is equal to (refer to cases (b) and (f) in Fig. 4) or hits This occurs during battery discharge where the cell voltage reaches
(refer to cases (d) and (h) in Fig. 4) the specified limits during the its lower limit of 2.1 V. The results are again compared with the case
pulse (i.e. ichg
min
or idis
max ) while the cell voltage is within its pre- in which the battery aging state is ignored in SoP estimation (i.e. the
defined allowable region (i.e. vmin < v < vmax). correction coefficients are assumed to be a hundred percent). When
the battery power is limited by the estimated available discharge
Therefore, when the voltage acts as the limiting factor, the de- power, the cell voltage does not violate the lower limit of 2.1 V and
viation between the cell voltage and the predefined charge or the battery current is reduced to maintain the cell voltage operating
discharge limit can approximately reflect the inaccuracy of SoP on the SOA limit, as can be seen in Fig. 11 (a) and Fig. 11 (b).
estimation. Similarly, for the current pulse, the deviation between Meanwhile, without the FLBC, the power capability is over-
the actual current with the design limit reflects the inaccuracy of estimated as a result of neglecting battery aging state and there-
estimation. Accordingly, the relative estimation error RE is defined fore, the SOA is violated. Fig. 11 (c) and Fig. 11 (d) represent two time
by Eq. (23) [35,37]. intervals of the cycle in which the power capability is limited by the

 battery current. This case happens during the regenerative braking
Ik  Ilim process in which the battery charge current reaches its allowable
REI ð%Þ ¼ 100  max
Ik limit of 30 A. As depicted, the absolute maximum power received

 (23)
by the battery corresponds with the current operating on its SOA
vk  vlim
REv ð%Þ ¼ 100  max limit.
vk
To evaluate the accuracy of SoP estimation, the relative error for
the current and voltage pulses during the acceleration or regener-
ative braking are computed since the start of modification process
4.3. Aged cell results using Eq. (23) and the results are represented in Fig. 11 (c). Since the
discharge power capability is limited by the battery voltage, relative
In this section, MiL simulation results for the aged cell in Tehran errors for the voltage pulses indicate the accuracy of SoP estimation
city bus driving cycle are presented [48]. Lower and upper limits of at discharge. Similarly, for the charge case, relative errors of current
current and voltage for the LiFePO4 cell are given in Table 4. The pulses reflect the accuracy of SoP estimation. Magnitude of the
prediction horizon (DT) is assumed to be 1 s. The aged cell current average relative error for the voltage and current pulses are 0.24%
and voltage during the driving cycle are shown in Fig. 10 (a) and and 0.28%, respectively, and the maximum absolute relative error is
Fig. 10 (b). As can be seen, no violation from the SOA limits has obtained to be 3.77%. The high accuracy of pulse test results reveals
occurred during the cycle; therefore, the safe operation for the that the maximum available power is drawn from or received by
battery is achieved for the aged cell. The results are compared with the battery on discharging and charging, respectively, leading to
the case in which no modification on the estimated SoP is applied. battery operation on the SOA limits of current and voltage. In other
words, the close correspondence between the maximum available
battery power with its operation on the SOA limits of current and
Table 4 voltage is indicated.
Relevant SOA limits for the LifePO4 cell.

Cell parameter idis

max ichg vmin vmax 4.4. Robustness verification
min

Value 150 A 30 A 2.10 V 3.65 V

To show robustness of the proposed SoP estimation method
516 M.J. Esfandyari et al. / Energy 176 (2019) 505e520

Fig. 10. Aged cell driving cycle results, (a) cell voltage, (b) current, (c) SoP, (d) correction coefficients, (e), (f) magnified view of the cell voltage and discharge correction coefficient
during the first modification process on the estimated SoP at discharge.
 M.J. Esfandyari et al. / Energy 176 (2019) 505e520 517

Fig. 11. Magnified view of two time intervals of the cycle in which the SoP is limited by the battery voltage ((a), (b)) and current ((c), (d)), (e) relative error for the current and
voltage pulses throughout the driving cycle since the start of modification.

against wrong SoC and SoH estimations, the results assuming values during the driving cycle are shown in Fig. 12. According to
under-estimation and over-estimation of battery capacity are pre- Eq. (2), the SoC estimates will gradually diverge from the true one
sented. In this regard, the capacity of the fresh cell is assumed to be as a result of wrong capacity values. In this case, the relative error is
under-estimated and over-estimated by 10% below and above the again computed for the current and voltage pulses since the start of
actual capacity. The actual and the over/under-estimated SoC modification process to evaluate the SoP estimation accuracy. As
518 M.J. Esfandyari et al. / Energy 176 (2019) 505e520

In Fig. 13 (b), the SoP estimates considering the true and under-
estimated fresh cell SoCs are compared and the relative difference
between the two estimations are presented in Fig. 13 (c). As
depicted, the average relative difference between the two estima-
tions is about 0.41% for the charge case and 0.01% for the
discharge case. For the case of SoC over-estimation, the results are
represented in Fig. 14. As can be seen from Fig. 14 (a), the magnitude
of average relative error is 0.38% for the charge case and 0.71% for
the discharge case. Also, the two SoP estimates considering the true
and over-estimated fresh cell SoCs are compared in Fig. 14 (b) and
Fig. 14 (c). The average relative differences between the two esti-
mates are about 0.25% and 0.31% for the charge and discharge cases,
respectively. The results demonstrate that utilizing only the ECM
parameters of the fresh battery, the power capability of the aged
battery is estimated accurately although the aging state or the SoH
is assumed to be unknown and the SoC estimates for the fresh cell
are inaccurate.

5. Conclusion

Fig. 12. True and wrong SoC values for the fresh battery in the robustness verification
A new approach is conducted to estimate battery SoP in hybrid
test.
electric vehicles. The estimation method consists of two steps; the
fresh cell power capability is first estimated assuming a combined
constant-current and constant-voltage reference mode. Secondly, a
represented in Fig. 13 (a), when the fresh cell SoC is under- model-less fuzzy logic based control system is designed to
estimated, the magnitude of the average relative errors for the compensate for battery aging state variations. Due to the use of
current and voltage pulses are 0.45% and 0.58% indicating the battery actual voltage and current values in a closed-loop frame-
relative accuracy of charge and discharge power capabilities, work, experimental and MiL simulation results indicate that not
respectively. only the accuracy and robustness of SoP estimation is improved,

Fig. 13. SoP estimation accuracy against under-estimated fresh cell SoC, (a) relative estimation error (b) SoP estimation comparison considering true and under-estimated SoC
values for the fresh battery, (c) relative SoP estimation difference considering true and under-estimated SoCs of the fresh battery.
 M.J. Esfandyari et al. / Energy 176 (2019) 505e520 519

Fig. 14. SoP estimation accuracy against over-estimated fresh cell SoC, (a) relative estimation error (b) SoP estimation comparison considering true and over-estimated SoC values
for the fresh battery, (c) relative SoP estimation difference considering true and over-estimated SoCs of the fresh battery.

but also battery operation within the SOA is guaranteed. In addi- [7] Campestrini C, Horsche MF, Zilberman I, Heil T, Zimmermann T, Jossen A.
Validation and benchmark methods for battery management system func-
tion, the active communication between the BMS and VCU has been
tionalities: state of charge estimation algorithms. J. Energy Storage 2016;7:
investigated in a driving cycle and the results show the effective- 38e51.
ness of the proposed SoP estimation method towards obtaining a [8] Crego J, Villarreal I, Berecibar M, Garmendia M. State of health estimation
safe performance for the Lithium-ion battery in HEV. The presented algorithm of LiFePO 4 battery packs based on differential voltage curves for
battery management system application. Energy 2016;103:784e96.
study will be extended in several directions. This includes the pack- [9] Ren H, Zhao Y, Chen S, Wang T. Design and implementation of a battery
level estimations considering cell-to-cell inconsistencies, the in- management system with active charge balance based on the SOC and SOH
clusion of battery temperature variation and hardware-in-the-loop online estimation. Energy 2019;166:908e17.
[10] Jung M, Schwunk S. High end battery management systems for renewable
simulation for the proposed method. energy and EV applications. Green 2013;3(1):19e26.
[11] Waag W, Fleischer C, Sauer DU. Critical review of the methods for monitoring
of lithium-ion batteries in electric and hybrid vehicles. J Power Sources
Acknowledgments 2014;258:321e39.
[12] Hu M, Li Y, Li S, Fu C, Qin D, Li Z. Lithium-ion battery modeling and parameter
The authors would like to acknowledge the financial supports identification based on fractional theory. Energy 2018;165:153e63.
[13] Fleischer C, Waag W, Bai Z, Sauer DU. On-line self-learning time forward
provided from Vehicle, Fuel and Environment Research Institute
voltage prognosis for lithium-ion batteries using adaptive neuro-fuzzy infer-
(VFERI) of the University of Tehran. ence system. J Power Sources 2013;243:728e49.
[14] Zhang X, Wang Y, Wu J, Chen Z. A novel method for lithium-ion battery state
of energy and state of power estimation based on multi-time-scale filter. Appl
References Energy 2018;216:442e51. September 2017.
[15] McGee R, Butcher J, Czubay J, Syed F, Kuang M. Closed loop control of battery
[1] Suri G, Onori S. A control-oriented cycle-life model for hybrid electric vehicle power limits based on voltage. Google Patents Mar-2007;27e.
lithium-ion batteries. Energy 2016;96:644e53. [16] Grabowski JR, Degner MW. Battery current limiter for a high voltage battery
[2] Zheng F, Jiang J, Sun B, Zhang W, Pecht M. Temperature dependent power pack in a hybrid electric vehicle powertrain. Google Patents Apr-2004;27e.
capability estimation of lithium-ion batteries for hybrid electric vehicles. [17] Andrea D. Battery management systems for large Lithium ion battery packs.
Energy 2016;113:64e75. Artech House; 2010.
[3] Mahmoudzadeh Andwari A, Pesiridis A, Rajoo S, Martinez-Botas R, [18] Zhang W, Shi W, Ma Z. Adaptive unscented Kalman filter based state of energy
Esfahanian V. A review of Battery Electric Vehicle technology and readiness and power capability estimation approach for lithium-ion battery. J Power
levels. Renew Sustain Energy Rev 2017;78(May):414e30. Sources 2015;289:50e62.
[4] Soares FJ, et al. The STABALID project: risk analysis of stationary Li-ion bat- [19] Plett GL. High-performance battery-pack power estimation using a dynamic
teries for power system applications. Reliab Eng Syst Saf 2015;140:142e75. cell model. Veh. Technol. IEEE Trans. 2004;53(5):1586e93.
[5] Stroe A-I, Knap V, Stroe D-I. Comparison of lithium-ion battery performance at [20] Kim Y, Salvi A, Stefanopoulou AG, Ersal T. Reducing soot emissions in a diesel
beginning-of-life and end-of-life. Microelectron Reliab 2018;88(90):1251e5. series hybrid electric vehicle using a power rate constraint map. IEEE Trans
[6] Galeotti M, Cina L, Giammanco C, Cordiner S, Di Carlo A. Performance analysis Veh Technol 2015;64(1):2e12.
and SOH (state of health) evaluation of lithium polymer batteries through [21] Lee T-K, Kim Y, Stefanopoulou A, Filipi ZS. Hybrid electric vehicle supervisory
electrochemical impedance spectroscopy. Energy 2015;89:678e86.
520 M.J. Esfandyari et al. / Energy 176 (2019) 505e520

control design reflecting estimated lithium-ion battery electrochemical dy- batteries. Appl Energy 2016;161:349e63.
namics. Proc. 2011 Am. Control Conf. 2011:388e95. [35] Sun F, Xiong R, He H. Estimation of state-of-charge and state-of-power
[22] Lee T-K, Kim Y, Rizzo DM, Filipi ZS. Battery power management in heavy-duty capability of lithium-ion battery considering varying health conditions.
HEVs based on the estimated critical surface charge. Int J Veh Des 2013;61(1): J Power Sources 2014;259:166e76.
108e27. [36] Xiong R, He H, Sun F, Liu X, Liu Z. Model-based state of charge and peak power
[23] Farmann A, Uwe D. A comprehensive review of on-board State-of-Available- capability joint estimation of lithium-ion battery in plug-in hybrid electric
Power prediction techniques for lithium-ion batteries in electric vehicles. vehicles. J Power Sources 2013;229:159e69.
J Power Sources 2016;329:123e37. [37] Waag W, Fleischer C, Uwe D. Adaptive on-line prediction of the available
[24] Wik T, Fridholm B, Kuusisto H. Implementation and robustness of an power of lithium-ion batteries. J Power Sources 2013;242:548e59.
analytically based battery state of power. J Power Sources 2015;287:448e57. [38] Pei L, Zhu C, Wang T, Lu R, Chan CC. Online peak power prediction based on a
[25] Wang Y, Pan R, Liu C, Chen Z, Ling Q. Power capability evaluation for lithium parameter and state estimator for lithium-ion batteries in electric vehicles.
iron phosphate batteries based on multi-parameter constraints estimation. Energy 2014;66:766e78.
J Power Sources 2018;374:12e23. January. [39] Shen P, Ouyang M, Lu L, Li J, Feng X. The Co-estimation of state of charge, state
[26] Bohlen O, Fh D, Bohlen OS. Impedance-based battery monitoring. 2008. of health and state of function for lithium-ion batteries in electric vehicles.
[27] Fleischer C, Waag W, Bai Z, Sauer DU. Adaptive on-line state-of-available- IEEE Trans Veh Technol 2017;9545(c):1e12.
power prediction of lithium-ion batteries. J. Power Electron. 2013;13(4): [40] Zheng L, Zhang L, Zhu J, Wang G, Jiang J. Co-estimation of state-of-charge,
516e27. capacity and resistance for lithium-ion batteries based on a high-fidelity
[28] Sun F, Xiong R, He H, Li W, Eric J, Aussems E. “Model-based dynamic multi- electrochemical model. Appl Energy 2016;180:424e34.
parameter method for peak power estimation of lithium e ion batteries. [41] Ag S. Softing CAN Layer2 manual software description. 2009. no. February.
Appl Energy 2012;96:378e86. [42] Nejad S, Gladwin DT, Stone DA. A systematic review of lumped-parameter
[29] Waag W, Fleischer C, Sauer DU. On-line estimation of lithium-ion battery equivalent circuit models for real-time estimation of lithium-ion battery
impedance parameters using a novel varied-parameters approach. J Power states. J Power Sources 2016;316:183e96.
Sources 2013;237:260e9. [43] Plett GL. Battery management systems: volume I, battery modelling. Artech
[30] Fleischer C, Waag W, Heyn H-M, Sauer DU. On-line adaptive battery imped- House; 2015.
ance parameter and state estimation considering physical principles in [44] Dong G, Wei J, Chen Z. Kalman filter for onboard state of charge estimation
reduced order equivalent circuit battery models part 2. Parameter and state and peak power capability analysis of lithium-ion batteries. J Power Sources
estimation. J Power Sources Sep. 2014;262:457e82. 2016;328:615e26.
[31] Fleischer C, Waag W, Heyn H-M, Sauer DU. On-line adaptive battery imped- [45] Mohammadi E, Montazeri-Gh M, Khalaf P. Metaheuristic design and optimi-
ance parameter and state estimation considering physical principles in zation of fuzzy-based gas turbine engine Fuel controller using hybrid invasive
reduced order equivalent circuit battery models. J Power Sources Aug. weed optimization/particle swarm optimization algorithm. J Eng Gas Turbines
2014;260:276e91. Power Nov. 2013;136(3):031601.
[32] Jiang J, Liu S, Ma Z, Wang LY, Wu K. Butler-Volmer equation-based model and [46] Mohammadi E, Montazeri-Gh M. A fuzzy-based gas turbine fault detection
its implementation on state of power prediction of high-power lithium tita- and identification system for full and part-load performance deterioration.
nate batteries considering temperature effects. Energy 2016;117:58e72. Aerosp. Sci. Technol. Oct. 2015;46:82e93.
[33] Waag W, Fleischer C, Sauer DU. Adaptive on-line prediction of the available [47] Wang L-X. A course in fuzzy systems. USA: Prentice-Hall press; 1999.
power of lithium-ion batteries. J Power Sources 2013;242:548e59. [48] Montazeri M, Arefian M. Development of the first driving cycle for tehran city
[34] Burgos-Mellado C, Orchard ME, Kazerani M, Ca rdenas R, Sa ez D. Particle- buses using Markov chain and transition analysis. In: First Conference of
filtering-based estimation of maximum available power state in Lithium-Ion environmental engineering; 2007.