energies

Article
Integrated Approach Based on Dual Extended
Kalman Filter and Multivariate Autoregressive Model
for Predicting Battery Capacity Using Health
Indicator and SOC/SOH
Jinhyeong Park 1 , Munsu Lee 2 , Gunwoo Kim 1 , Seongyun Park 1 and Jonghoon Kim 1, *
1 Department of Electrical Engineering, Chungnam National University, Daejeon 34134, Korea;
pig25t@o.cnu.ac.kr (J.P.); qiqu2771@o.cnu.ac.kr (G.K.); pb7008@o.cnu.ac.kr (S.P.)
2 Department of Energy Science, Sungkyunkwan University, Suwon 16419, Korea; munsu@skku.edu
* Correspondence: whdgns0422@cnu.ac.kr




Received: 28 March 2020; Accepted: 23 April 2020; Published: 29 April 2020


Abstract: To enhance the efficiency of an energy storage system, it is important to predict and
estimate the battery state, including the state of charge (SOC) and state of health (SOH). In general,
the statistical approaches for predicting the battery state depend on historical data measured via
experiments. The statistical methods based on experimental data may not be suitable for practical
applications. After reviewing the various methodologies for predicting the battery capacity without
measured data, it is found that a joint estimator that estimates the SOC and SOH is needed to
compensate for the data shortage. Therefore, this study proposes an integrated model in which the
dual extended Kalman filter (DEKF) and autoregressive (AR) model are combined for predicting the
SOH via a statistical model in cases where the amount of measured data is insufficient. The DEKF
is advantageous for estimating the battery state in real-time and the AR model performs better for
predicting the battery state using previous data. Because the DEKF has limited performance for
capacity estimation, the multivariate AR model is employed and a health indicator is used to enhance
the performance of the prediction model. The results of the multivariate AR model are significantly
better than those obtained using a single variable. The mean absolute percentage errors are 1.45%
and 0.5183%, respectively.

Keywords: Li-ion battery; battery degradation; statistical model; dual extended Kalman filter

1. Introduction
Due to the Paris Agreement for preventing catastrophic climate change, the current energy system
requires a rapid global shift toward decarbonization in all sectors, such as industry, transportation,
and residential and commercial buildings, through the use of renewable energy [1]. The generation
of renewable energy must be increased to achieve decarbonization of power and energy systems.
Because renewable energy resources have intermittent characteristics, an energy storage system
(ESS) is essential for the transition to a sustainable energy system with flexibility and reliability [2].
Among different types of battery, the Lithium-ion battery is effective for high power and high energy
density applications and is also widely used from small-to-large scale electronics, electric vehicles, and
utility-scale storages [3].
In the global market, the installation of ESSs is expected to increase exponentially, from 9 GW/17
GWh deployed as of 2018 to 1095 GW/2850 GWh expected by 2040 [4]. Studies have indicated that
the ability to store and release electricity nearly instantaneously offers multiple benefits in the power
system not only for the integration of diverse renewable energy sources but also for grid reliability [5].

Energies 2020, 13, 2138; doi:10.3390/en13092138 www.mdpi.com/journal/energies

Energies 2020, 13, 2138 2 of 20

To enhance grid efficiency, it is important to consider both the battery cost and the performance
of the ESS over its lifespan [6]. The cost of battery packs is decreasing by an average of 13% per
year and is expected to reach $156/kWh in 2019 and $100/kWh in 2023 [7]. Because improvements
in battery performance can reduce operation and management costs, the importance of the battery
management system (BMS) is gradually increasing to manage and monitor the battery state more
efficiently. Accurate estimation and prediction of the state of charge (SOC) and the state of health
(SOH) can ensure the BMS prolong the cycle life of energy management and reduce the utilization cost
for replacing the battery [8].
The SOC represents the remaining capacity that is compared with the stored capacity [9]. The SOC
varies between 0% to 100%, which corresponds to the fully discharged and charged state, respectively.
However, if the battery is aged, the maximum value of the SOC decreases because of the capacity
degradation. When the capacity is reduced to 80% of the initial value, the SOC region of the battery is
reduced by 20%. Thus, accurate capacity estimation is required for the efficient and safe operation
of the battery [10]. Research on estimating the SOC and SOH, according to the variability of the
battery characteristics, is crucial for developing an advanced BMS. Among the estimation strategies,
the Kalman filter is advantageous in the sense that it has a very flexible coordinator to handle battery
characteristic changes [8,10].

2. Literature Review
The SOH is the ratio of the characterization parameters (e.g., capacity and internal resistance).
The remaining useful life (RUL) is the number of available cycles from the SOH to the failure threshold.
Among research studies regarding battery, it is a necessity for efficient utilization to identify the aging
mechanism and predict the capacity loss based on SOH, according to the different perspectives [11,12].
From an economic viewpoint, because the battery wear cost can be calculated according to the
achievable cycle count, depth of discharge, battery size, and round-trip efficiency [13], the SOH can be
a reference to determine the battery replacement cost and economic utilization cost for maximizing the
profitability [14–16]. From a technical perspective, the battery cycle life and SOH can be utilized to
optimize the energy management strategy by determining the threshold for energy capability [17].
In the long term, SOH management improves the lifecycle by assessing the impact of battery aging
using a health indicator (HI), which tracks the degradation of batteries [17].
In recent studies, estimation methods of the battery capacity have employed the correlation
between the characterization parameters and aging. This seems to be an important issue that the
estimation method for the HI should be selected properly to reflect the nonlinear pattern of degradation.
The integrated HI is defined in the aging model by incorporating a certain period of capacity and
resistance. Xiong et al. [18] investigated the correlation between the partial charge capacity obtained in
a certain region of the voltage and the capacity for extracting the effective HI in the charging mode.
A linear aging model was constructed using the moving-window-based method considering the HI.
Zheng et al. [19] utilized the HI to determine the optimal charge region of the fixed voltage window
and proposed an online capacity estimation method that incorporates both the discrete Arrhenius
aging model and extended Kalman filter. Liu et al. [20] proposed a relevance vector machine using
the optimized HI to enhance the accuracy and stability of RUL prediction. Zhou et al. [21] optimized
the HI considering the different times of voltage drops in the discharge curve and used the Box-Cox
transformation to stabilize the variance. Even though these methods are widely used for estimating
the SOH because the capacity is highly dependent on the charge/discharge region, the HI extraction is
limited to the charge/discharge region. If the battery voltage does not reach a certain region, the HI
cannot be updated and the prediction performance is unreliable. Therefore, aging characterization
parameters based on real-time methods are needed to estimate the SOH accurately.
To select the appropriate HI in real-time, resistance is the most influential candidate for reflecting
the aging condition of the battery. The resistance is related to the solid-electrolyte interphase (SEI)
layer in the physicochemical mechanism of the battery. One of the most influential factors affecting the
Energies 2020, 13, 2138 3 of 20

capacity loss with aging is the SEI layer growth [22]. According to cell size, capacity, cell design, and
types of active material in the battery, thermal abuse tolerance of the battery can be varied from the SEI
formation [23]. Even though it has a strong relationship with the capacity loss, it is difficult to identify
the multi-physical behavior and thermodynamics of the SEI layer in real-time applications because of
the model’s complexity and a large number of parameters [24]. The SEI growth for the prediction of
the SOH can be simplified by using electrochemical impedance spectroscopy (EIS) and an equivalent
circuit model (ECM). The resistance of the SEI layer can be extracted from the high-frequency region
of the EIS results and the parameters from EIS can be applied to the ECM [25]. This method has the
advantage of quickly extracting the parameters to be implemented in the BMS board [26]. However, it
is too complex to conduct computations for fitting a model. Additionally, the EIS is difficult to apply
for real-time estimators and applications because it requires the high cost for the experiment [27].
Due to this limitation, the method for estimating the SOH in real-time applications needs to define
the relationship between the internal resistance based on the direct-current signal and the capacity.
Pan et al. [28] proposed an online SOH estimation method using machine learning and the ECM
parameters, such as the enlarged ohmic and polarized internal resistances. Because these parameters
are intuitive values and are flexible for different types of dynamic load profiles, they can be measured
by an online parameter identification algorithm. However, the machine learning method is difficult to
apply to online applications because of complex computation.
In recent years, to enhance the flexibility of the capacity model, the prediction model has been
merged with the real-time estimation method without measured data. Qie et al. [29] proposed the
RUL prediction method based on the SOC/SOH joint estimator. The SOC/SOH was estimated using a
multiscale hybrid Kalman filter and the estimated SOH was used to update the RUL prediction model.
Xue et al. [30] presented an integrated algorithm. These studies revealed it is difficult to construct
the prediction model and the relationship between the capacity loss and other factors because of
the nonlinear characteristics of the battery capacity according to the cycle and historical conditions.
The studies indicated the limitation of conventional prediction methods that use the measured capacity
for constructing the aging model. The measured capacity may not apply to real-time applications
because the capacity in a real application has a different trend. Therefore, researchers should consider
methods for estimating the SOH or optimization methods of the HI.
To address the need for a capacity model and HI selection, an integrated model is proposed herein
for predicting the battery capacity using a dual extended Kalman filter (DEKF) and a multivariate
autoregressive (AR) model due to the limited data availability for the measured capacity. The capacity
model was developed using the DEKF based on the ECM and SOC-open-circuit voltage (OCV)
relationship, which reflects the aging conditions and battery characteristics. The multiple HIs are
analyzed using the correlation coefficient and extracted using a parameter identification method based
on the battery model. To verify the proposed method, this study addresses the effect of the multiple
HIs to the performance of the prediction model by dividing the cases into relatively better and worse
estimation performance of the DEKF.
The remainder of this paper is organized as follows. Section 3 presents the experimental setup, the
algorithm for the battery parameters, state estimation, and the prediction model. Section 4 presents the
simulation results for capacity prediction compared with the conventional method and the integrated
model. Section 5 presents the conclusions and directions for future research.

3. Methodologies

3.1. Experimental Conditions for a Battery Aging Test

To validate the proposed method and accumulate the aging data, the experimental testbench
shown in Figure 1 was used. The battery test-bench was set up to collect the experimental data and
consisted of a battery charge/discharge regulator (Maccor 4300K), a thermal chamber, and a personal
computer (PC). The battery regulator used a charging/discharging battery. The computer controlled
Energies 2020, 13, x FOR PEER REVIEW 4 of 20
Energies 2020, 13, 2138 4 of 20
consisted of a battery charge/discharge regulator (Maccor 4300K), a thermal chamber, and a personal
computer (PC). The battery regulator used a charging/discharging battery. The computer controlled
the battery regulator and recorded the voltage and current data. The thermal chamber was used to
the battery regulator and recorded the voltage and current data. The thermal chamber was used to
maintain a constant temperature of 25 ◦ C. The battery used in the experiment was a nickel manganese
maintain a constant temperature of 25 °C. The battery used in the experiment was a nickel
cobalt-oxide (NMC) battery with a rated discharge capacity of 3.3 Ah and a rated voltage of 3.56 V, as
manganese cobalt-oxide (NMC) battery with a rated discharge capacity of 3.3 Ah and a rated voltage
represented in Table 1.
of 3.56 V, as represented in Table 1.

Figure 1.
Figure 1. Experimental
Experimental platform
platform of
of the
the battery.
battery.

Table 1. Battery
Table1. Battery specification.
specification.

Item
Item Specification
Specification
Battery
Batterytype
type INR
INR 21700 (LiNiMnCoO
21700 (LiNiMnCoO2 )2)
Standard
Standarddischarge
discharge capacity
capacity 3.3
3.3 Ah
Ah
Nominalvoltage
Nominal voltage 3.56 V
3.56 V
4.2 V
4.2 V
Charging method Constant voltage and constant current
Charging method Constant
(CC: 3.3 A CV: 4.2 Vand
voltage constant
cut-off current
current: 0.05C)
(CC: 3.3 A CV:Constant
4.2 V cut-off
currentcurrent: 0.05C)
Discharge method
(CC: 3.3
Constant A)
current
Discharge method
C-rate 1C-rate: 3.3 A
(CC: 3.3 A)
C-rate 1C-rate: 3.3 A
The short-period profile is insufficient for capturing the change in capacity. To obtain the variable
The with
capacity short-period profile
charge and is insufficient
discharge cycles, afor capturing
cycling the change
test was performed,in capacity.
as shown Toinobtain
Figurethe2.
variable capacity with charge and discharge cycles, a cycling test was performed, as
The overall profile exhibited repeated charging and discharging, as shown in Figure 2a. Figure 2b shown in Figure
2. The overall
indicates profile
that the exhibited
battery was agedrepeated
in thecharging andwith
experiment discharging,
charging as andshown in Figure
discharging at 2a.
3.3 Figure 2b
A. In this
indicates
study, thatthe
since thecoulomb
battery was aged inofthe
efficiency theexperiment
NMC battery withischarging
almost 100%and discharging at 3.3 A.
within the useful In this
life, the
study, since
discharge the coulomb
capacity efficiency
is used for of the NMC
the prediction modelbattery
[31]. is almost 100% within the useful life, the
discharge capacity is used for the prediction model [31].
Energies 2020, 13, 2138 5 of 20
Energies 2020, 13, x FOR PEER REVIEW 5 of 20

4.5
(a)

Voltage (V)
3.5

0 1 2 3 4 5 6 7 8
Time (s) 5
10

Figure 2. Profiles obtained

4 in the cycling test: (a) entire voltage profile and4.5 (b) voltage and current
profile at one cycle.
2 4

Voltage (V)
Current (A)

3.2. Parameter Definition and0 Identification 3.5

-2 3
3.2.1. Battery Equivalent Circuit Model
(b)
-4 2.5
The ECM corresponding 0 to2000the battery
4000 6000 is essential
8000 10000 for estimating
12000 14000 the internal state. The ECM
16000
Time (s)
consists of the SOC, OCV, ohmic resistance (R0), diffusion resistance (R1), and diffusion capacitance
(C1), Figure
as shown in Figure
2. Profiles 3. The
obtained OCV
in the is SOC
cycling test:and (a) time-dependent
entire voltage profile andand it is (b)defined
voltage as
anda current
function of
Figure 2. Profiles obtained in the cycling test: (a) entire voltage profile and (b) voltage and current
SOC.profile
The SOC iscycle.
at one the ratio of the remaining capacity of the battery to the nominal capacity (Cn). The
profile at one cycle.
SOC is calculated using the Ampere hour (Ah) counting method, as follows.
3.2. Parameter Definition and Identification
3.2. Parameter Definition and Identification SOCk +1 = SOCk + (Δt / Cn )ik (1)
3.2.1. Battery Equivalent Circuit Model
3.2.1. Battery
where Equivalentthe
SOC0 represents Circuit
initialModelSOC, ∆t represents the sampling time of the experimental setup, ik
The ECM
represents corresponding
the current applied to to the battery model,
the battery is essential and C for estimating the internal state. The ECM
n is the nominal capacity of the battery.
consistsThe ECM corresponding to the battery is essential for estimating1the internal state. The ECM
Theof the SOC,voltage
diffusion OCV, ohmic
(V1) to resistance
the parallel (Rcircuit
0 ), diffusion
of the resistance
resistance (R and), capacitance
and diffusion capacitance
is represented
consists
(C of the SOC, OCV, ohmic resistance (R 0 ), diffusion resistance (R 1 ), and diffusion capacitance
as:1 ), as shown in Figure 3. The OCV is SOC and time-dependent and it is defined as a function of SOC.
(C1),SOC
The as shown in Figure
is the ratio of the3.remaining
The OCVcapacity is SOC and of the time-dependent
battery to the nominal and it iscapacitydefined(Casn ).a The
function
SOC of is
SOC. The SOC is the
calculated using the Ampere ratio
V of the
= remaining
exp( −Δ
1, k +1 hour (Ah) countingt / τ ) capacity
⋅ V + Rof
1, kmethod,
1 ⋅ the
(1 − battery
exp( −Δ
as follows. to
t / τ
the)) nominal
⋅ ik
capacity (Cn ). The
(2)
SOC is calculated using the Ampere hour (Ah) counting method, as follows.
where τ is the time constant of the battery. SOCk+1 = SOCk + (∆t/Cn )ik (1)
The terminal voltage (Vt,k) of the battery SOC= k +1 pack + (∆t / Cn )ik as:
SOCisk represented (1)
where
where SOC
SOC00 represents
represents the
the initial Vt,∆t
initial SOC,
SOC, ∆t represents
represents the
the sampling
sampling time
k = OCVk (SOCk ) + ik R0 +V1
time of
of the
theexperimental
experimentalsetup, i
(3) ikk
setup,
represents the current applied to the battery model, and C is the nominal capacity of the
represents the current applied to the battery model, and Cnn is the nominal capacity of the battery.battery.
The diffusion voltage (V1) to the parallel circuit of the resistance and capacitance is represented
as: R1,k
ik R0,k
V1,=
k +1 exp(−∆t / τ ) ⋅ V1, k + R1 ⋅ (1 − exp(−∆t / τ )) ⋅ ik (2)
C1,k
where τ is the time constant of the
+ Vbattery.
0,k - +
The terminal voltage (Vt,k) of the battery pack is+ represented
V1,k - as:
OCVk(SOCk) Vt,k
=Vt ,k OCVk ( SOCk ) + ik R0 + V1 (3)
-

R1,k
Figure 3.
Figure 3. Battery
Battery equivalent
equivalent circuit model.
circuit model.
ik R0,k
The diffusion voltage (V 1 ) to the parallel circuit of Cthe
1,k resistance and capacitance is represented as:
+ V0,k - +
V1,k+1 = exp(−∆t/τ) · V1,k ++RV1 1,k · (-1 − exp(−∆t/τ)) · ik (2)
OCVk(SOCk) Vt,k
where τ is the time constant of the battery.
-

Figure 3. Battery equivalent circuit model.

Energies 2020, 13, 2138 6 of 20

The terminal voltage (Vt,k ) of the battery pack is represented as:

Vt,k = OCVk (SOCk ) + ik R0 + V1 (3)

TheEnergies
purpose
Energies 2020,13,
2020, 13,x xthe
of FOR
FORPEER
PEERREVIEW
OCV test is to identify the electrical potential capability and parameters
REVIEW 6 6ofof2020
of
the ECM, accordingThe to the SOC, as shown in Figure 4. The OCV test is conducted to acquire of the OCV
Thepurpose
purposeofofthe theOCV
OCVtesttestisistotoidentify
identifythe theelectrical
electricalpotential
potentialcapability
capabilityand andparameters
parameters of
according the to
theECM,
the SOC.
ECM,according
The
accordingtotothe
full charged
theSOC,
battery
SOC,asasshown
shownininFigure
(SOC 100%)
Figure4.4.TheTheOCV
is discharged
OCVtesttestisisconducted
with constant
conductedtotoacquire
acquirethe
current
theOCVOCV
(CC)
and we according
settle the to the SOC.
discharging The full
amountchargedto battery
SOC 5%. (SOC The100%)
OCV is discharged
points
according to the SOC. The full charged battery (SOC 100%) is discharged with constant current (CC) were with constant
extracted current
to a 5% (CC)
interval in
and
the SOC range we
and weof settle the
0–100%.
settle discharging
The applied
the discharging amount
amount to
current SOC
to SOC 5%.
value The OCV
of the
5%. The OCV points
pulse
points were
test extracted
is set
were to a
to 1 C-rate,
extracted 5% interval
to a 5%which
interval in
refers
in to a
multiplethe of SOC
the range
SOCrated
the ofof0–100%.
rangecapacity.0–100%. The
Theapplied
Figure applied current
4 representscurrentvaluevalue ofofthe
the method thepulse
pulse test
testisisset
of parameter settotoidentification.
1 1C-rate,
C-rate,which
whichrefers
refersOCV is
The
to a amultiple ofofthe
therated
ratedcapacity. Figure 4 4represents the method of parameter identification. The
obtained to when multiple
the region that capacity.is
current Figure
not appliedrepresents
rests,the asmethod
shownofinparameter
Figure 5.identification. The
The ohmic resistance
OCV is obtained when the region that current is not applied rests,
OCV is obtained when the region that current is not applied rests, as shown in Figure 5. The ohmic as shown in Figure 5. The ohmic
(R0 ), diffusion
resistance resistance (R1 ), resistance
and capacitance
(R1), and(C 1 ) are acquired
(C1) arewhen thewhen
current the applies to the battery
resistance(R(R 0), diffusion capacitance acquired
0), diffusion resistance (R1), and capacitance (C1) are acquired when the current applies
current applies
totothe
as follows: thebattery
batteryasasfollows:
follows: ∆V0 ∆V10s τ
R0 = ,R
∆ΔV 1 = ∆Δ VV , C1 = τ (4)
= I= = V τ
1 = =
RR 0
0, R
,R I 10s, C, C1 = R1
10s
(4)
0
0 II 1 II 1 RR (4)
1
where ∆V 0 is the voltage drop in OCV, ∆V 10s is the voltage drop1 during the discharge at 10 s, and τ is
the timewhere
whereΔV
constant
ΔV0 is the voltage drop in OCV, ΔV10s is the voltage drop during the discharge at 10 s, and τ is
0as shown
is the in drop
voltage Figure 5. ΔV10s is the voltage drop during the discharge at 10 s, and τ is
in OCV,
the
thetime
timeconstant
constantasasshown
shownininFigure
Figure5.5.
5 4.5
Voltage
Current
4

4
3

2
3.5

Voltage (V)
Current (A)

0 3

-1

2.5
-2

-3

2
0 1 2 3 4 5 6 7 8
Time (s) 4
10

FigureFigure
4. The4.4.
Figure The
Thebattery
battery state
state
battery of of
state ofcharge
charge (SOC)
(SOC)-- open
(SOC)
charge opencircuit
-open circuitvoltage(OCV)
circuit test
voltage(OCV)
voltage(OCV) testprofile.
test profile.
profile.

Figure 5. Method of parameter identification.

Energies 2020, 13, 2138 7 of 20

3.2.2. Multiple Adaptive Forgetting Factor-Recursive Least Square (MAFF-RLS) Algorithm

During the battery operation, the battery parameters are varied according to the number of the
cycle. The simplified battery model is used as shown in Figure 3. By identifying these parameters, it
is possible to indirectly estimate the battery capacity reduction. To estimate the battery parameters
and merge with the time-series model, the autoregressive exogenous (ARX) model is used to apply
the recursive method to the battery terminal, as indicated by Equation (5) [32]. The input vector (Φk )
and parameter vector (εk ) are defined as shown in Equations (6) and (7), respectively. The battery
parameter can be rearranged as Equation (8).

Vt,k = εTk · φk (5)

h i
φk = Ik ; Ik−1 ; (OCVk−1 − yk−1 ); 1 (6)
h i
εk = b0,k ; b1,k ; a1,k ; OCVk (7)

b1 − a1 · b0 ∆t
R0 = b0 , R1 = , C1 = (8)
1 + a1 b1 − a1 · b0
The conventional RLS is set as the forgetting factor for one variable. However, because the battery
state is changed and multivariate parameters as the optimal forgetting factor can be changed, according
to the parameters. In particular, each parameter such as the OCV, resistance, and capacitance have
different characteristics and change, according to the aging condition. Therefore, a single and constant
forgetting factor cannot maintain the optimal estimation performance. To reflect the aging condition, in
this study, we estimated various parameters using the MAFF-RLS method reported in Reference [32].
In the first step, the initial value is defined. The multiple forgetting factors (λi,k ) are calculated
as follows.
−1 −1
λi,k = 1 − [1 + ζi (φTi,k · Ei,k−1 · φi,k ) ] (9)

where ζk is the constant parameter for forgetting factor and Ei,k is the error covariance of parameters.
The multiple adaptive gains (Li,k ) and the error covariance are calculated as Equations (10) and
(11), respectively. The adaptive gain is updated using Equation (12) and the parameter vector is
calibrated as shown in Equation (13).

−1
Li,k = Pi,k−1 · φi,k [λi,k + φTi,k · Ei,k−1 · φi,k ] (10)

Ei,k = λ−1
i,k
(1 − Li,k · φTi,k )Ei,k−1 (11)

 λ1,k · E1,k−1 · φ1,k

 −1 

 λ · E2,k−1 · φ
 −1 
 X4
Lk =  2,k 2,k
[1 + λ−1 · Ei,k−1 · φ2i,k ] (12)
 
 λ−1 · E3,k−1 · φ i,k
 3,k 3,k  i=1
λ4,k · E4,k−1 · φ4,k
 −1 

θk = θk−1 + Lk (Vt,k − φTk · θk−1 ) (13)

3.2.3. Parameter Estimation Results and Relationship with Capacity

Figure 6 is the comparison of the SOC-OCV curve, which is extracted from Figure 4. The fresh
value is the initial condition and aged when the battery is charged and discharged at 700 cycles.
The tendency of OCV-SOC is very similar and the average difference is about 0.014 V. OCV has little
change in the SOC 100% region, but OCV in SOC = 0%, which can be selected as a candidate for HI
because the difference is largest in the entire SOC region. Therefore, since the OCV is highly dependent
on the SOC than the number of cycles, it is not suitable to define the OCV for HI.
Energies 2020,
Energies 2020, 13,
13, xx FOR
FOR PEER
PEER REVIEW
REVIEW 88 of
of 20
20

because
Energies
because the13,difference
2020,
the difference
is largest
2138
is largest in
in the
the entire
entire SOC
SOC region.
region. Therefore,
Therefore, since
since the
the OCV
OCV is
is highly
highly
8 of 20
dependent on
dependent on the
the SOC
SOC than
than the
the number
number of
of cycles,
cycles, it
it is
is not
not suitable
suitable to
to define
define the
the OCV
OCV for
for HI.
HI.

Figure 6. Comparison
Figure Comparison of the
the SOC-OCV curve
curve between the
the fresh and
and aging condition.
condition.
Figure 6.
6. Comparison of
of the SOC-OCV
SOC-OCV curve between
between the fresh
fresh and aging
aging condition.

Using the
Using the Multiple
the Multiple
Adaptive
MultipleAdaptive
Forgetting
AdaptiveForgetting
Factor-Recursive
ForgettingFactor-Recursive
Factor-Recursive
Least
Least Square
Square
Least Square
(MAFF-RLS)
(MAFF-RLS)
(MAFF-RLS)
method,
method,
method,
R00
R
and
and
R
R 1 were identified according to time-series data. These parameters were utilized as HIs to
R1 Rwere identified according tototime-series
time-seriesdata.
data.These
Theseparameters
parameterswere wereutilized
utilizedasas HIs
HIs to
0 and 1 were identified according
supplement the
supplement the nonlinear
nonlinear relationship
relationship of of variables
variables in
in the
the prediction
prediction model.
model. As As the
the cycle
cycle progressed,
progressed,
As the cycle progressed,
the battery
the battery capacity
capacity increased,
increased, along
along with
with the
the ohmic
ohmic resistance
resistance and
and diffusion
diffusion resistance,
resistance, as
as shown
shown inin
Figure 7.
Figure 7. In In the entire cycle, when the capacity decreases, both ohmic (R 0) and diffusion resistance
In the entire cycle, when the capacity decreases, both ohmic (R00) and diffusion resistance
(R11)) increase,
(R increase, according
according to
to the
the cycles,
cycles, as
as shown
shown in in Figure
Figure 7a
7a and Figure
Figure 7b. Therefore,
Therefore, those
those results
results
1) the cycles, as shown in Figure 7a,b.and
Therefore,7b.
those results show that the
show that
show that the
the capacity
capacity and
and the
the two
two resistances
resistances have
have aa relationship
relationship for each other.
capacity and the two resistances have a relationship for each other. for each other.

(a) Relationship
(a) Relationship between
between ohmic
ohmic resistance
resistance and
and capacity
capacity

Figure 7. Cont.
Energies 2020, 13, 2138 9 of 20
Energies 2020, 13, x FOR PEER REVIEW 9 of 20

(b) Relationship between diffusion resistance and capacity.

Figure 7. Relationship
Relationship between capacity and resistance.

The Pearson correlation analysis (PCA) was used to confirm the linear relationship between the
estimated resistance and capacity.
capacity. The correlation
correlation coefficient
coefficient (r)
(r) is
is calculated
calculated by
by the
the equation
equationbelow.
below.

r r== q
i=1 ((XX i −− XX)()(YYi −− YY))
Pn n
i =1 i i
(14)
(14)
 i=1 (Yi − Y)
n n
q
2
Pn ( X − X )2 Pn (Y − Y ) 2 2
i = 1 ( Xi − X )
i =1 i i =1 i

where n is the sample size, Xi and Yi are the individual sample points with i, X and Y are the
where n is the sample size, Xi and Yi are the individual sample points with i, X and Y are the sample
sample mean values of two variables. When the two variables are not distributed normally, the
mean values of two variables. When the two variables are not distributed normally, the Spearman
Spearman rank correlation (SRC) is used to evaluate the monotonic relationship between Xi and Yi
rank correlation (SRC) is used to evaluate the monotonic relationship between Xi and Yi via the
via the same procedure.
same procedure.
If r is closer to 1 or −1, the relationship between the two variables is more linear. If it is larger
If r is closer to 1 or −1, the relationship between the two variables is more linear. If it is larger
than 0.8, the relationship between the two variables has strong linearity [33]. The results of the
than 0.8, the relationship between the two variables has strong linearity [33]. The results of the
correlation analysis are presented in Table 2. R0 was almost linearly related to the capacity because
correlation analysis are presented in Table 2. R0 was almost linearly related to the capacity because the
the coefficient was close to −1. Even though the R1 coefficient was smaller than the R0 coefficient, it is
coefficient was close to −1. Even though the R1 coefficient was smaller than the R0 coefficient, it is
larger than 0.8. This indicates a strong relationship between the capacity and the resistance, as shown
larger than 0.8. This indicates a strong relationship between the capacity and the resistance, as shown
in Figures 8 and 9. R0 had a stronger linear relationship with the capacity than R1. Therefore, the
in Figures 8 and 9. R0 had a stronger linear relationship with the capacity than R1 . Therefore, the
ohmic and diffusion resistances were selected as the HIs for developing the multivariate prediction
ohmic and diffusion resistances were selected as the HIs for developing the multivariate prediction
model. Considering these two HI values, the prediction performance to improve the nonlinearity of
model. Considering these two HI values, the prediction performance to improve the nonlinearity of
the prediction model, in the case where the estimated capacity is inaccurate when compared with the
the prediction model, in the case where the estimated capacity is inaccurate when compared with the
measured data, was examined.
measured data, was examined.
Table 2. Correlations between the capacity and resistance.
Table 2. Correlations between the capacity and resistance.
Cn and R0 Cn and R1
Cn and R0 Cn and R1
Pearson correlation analysis
Pearson correlation analysis (PCA) −0.9783
−0.9783 −0.8517−0.8517
(PCA)
Spearman rank correlation (SRC) −0.9895 −0.9391
Spearman rank correlation
−0.9895 −0.9391
(SRC)
Energies 2020, 13, 2138 10 of 20

Energies 2020, 13, x FOR PEER REVIEW 10 of 20

Energies 2020, 13, x FOR PEER REVIEW 10 of 20

Figure
Figure 8.
Figure 8.Linear
8. Linearfitting
Linear fitting and
fitting and residual
and residual analysis
residual analysisof
analysis of the
of thecapacity
the capacity and
capacityand ohmic
andohmic resistance.
ohmicresistance.
resistance.

Figure
Figure 9.
9.Linear fitting
fitting and
and residual
residual analysis
analysisof
ofthe
thecapacity
capacityand
and diffusion resistance.
diffusionresistance.
resistance.
Figure 9. Linear fitting and residual analysis of the capacity and diffusion

3.3. Dual
3.3. Dual Extended
Extended Kalman
Kalman Filter
Filter
The
The Dual
Dual Extended
Dual ExtendedKalman
Extended KalmanFilter
Kalman Filter(DEKF)
Filter (DEKF)is
(DEKF) aa recursive
isis arecursive
recursive algorithm
algorithm
algorithm for accurately
forfor accurately
accurately estimating
estimating
estimating the
the
two-state
the of
two-state the
of system
the system by merging
by mergingthe previous
the previousstate and
state the
two-state of the system by merging the previous state and the measurement value. Using a singleand measurement
the measurement value. Using
value. a single
Using a
extended
single Kalman
extended filter
Kalman (EKF)
filter has
(EKF) the hasadvantages
the advantagesof a relatively
extended Kalman filter (EKF) has the advantages of a relatively simple mechanism and a smallof a simple
relatively mechanism
simple mechanism and a small
and a
computational
small computational
computational burden.
burden. Although
burden. Although
Although EKF
EKF EKF is
is widely
is widely
widely accepted,
accepted,
accepted, therethere
there are
areareseveral
severallimitations
several limitationsto
limitations to this
to this
approach. Nikolaos
Nikolaos et
et al.
al. [34]
[34] analyze
analyze the
the performance
performance of
of EKF
EKF
approach. Nikolaos et al. [34] analyze the performance of EKF and DEKF, according to the different and
and DEKF,
DEKF, according to the different
SOH degrees.
SOHdegrees. Although
degrees.Although
Althoughthe the DKEF
theDKEF shows
DKEFshowsshows weaker
weaker
weaker performance
performance
performance than than
thanthethe
EKF
the EKF
EKF in
in thein the
fresh
the fresh condition,
condition,
fresh condition, the
the
the DEKF
DEKF DEKF enhances
enhances
enhances the
the convergence
the convergence
convergence ability
ability of the
ability of the
the observer
ofobserver observer when whenthe SOH
when the
the SOH is
is decreasing.
is decreasing.
SOH decreasing. As aAs aa result,
result,
As the
result,
the
the DEKF has relatively better performance than EKF in the long term. Thus, to merge with the
DEKF DEKF
has has relatively
relatively better better performance
performance than EKFthan in EKF
the in
long the
term. long term.
Thus, to Thus,
merge to
with merge
the with
prediction the
prediction
model
prediction model
and state
model and
and state
estimator,state estimator,
a single EKF ais
estimator, a single EKF
EKF is
not suitable
single not
isfor suitable
suitable for
maintaining
not maintaining
the
for stability andthe
maintaining the stability
accuracy
stability ofandthe
and
accuracy
observer
accuracyin of the
ofthe observer
observer in
thecircumstance in the circumstance
of battery
the of
of battery
aging [34,35].
circumstance battery aging
aging [34,35].
To simultaneously[34,35]. To To simultaneously
identify the SOC and
simultaneously identify
capacity
identify the
the
SOC
of a and capacity
battery and of
improve a battery
the and
stabilityimprove
of an the stability
observer, in of
the an
DEKF,
SOC and capacity of a battery and improve the stability of an observer, in the DEKF, the ECM has observer,
the ECM in the
has DEKF,
two the
EKFs. ECM
The has
two
two
two EKFs.
EKF EKFs. The
observers two
TheruntwoinEKF
EKF observers
parallel run
run in
to estimate
observers parallel
inthe SOCto
parallel and
to estimate
capacity.
estimate the
the SOCSOC and and capacity.
capacity.
Due to the battery parameter observer, the DEKF does not need experimental
Due to
to the
the battery
battery parameter
parameter observer,
observer, the the
DEKF DEKF
does notdoes
need not need
experimental data for composing
experimental data
data for for
composing
the functions. the functions.
The parameter The parameter
observer observer
calibrates calibrates
the
composing the functions. The parameter observer calibrates the OCV, ohmic resistance OCV, the
ohmic OCV, ohmic
resistance resistance
(R 0 ), and (R
(R00),), and
diffusion and
diffusion
diffusion resistance
resistance (R11),), according
(R
(R1 ), according
resistance to the ECM
according to the
the ECM
to voltage ECM voltage
error.
voltage error.
Because
error.the Because
observer
Because the
thecanobserver
reflectcan
observer the reflect
can reflect the
nonlinear the
nonlinear characteristics
characteristics of the battery, of the
the battery,
DEKF can the DEKF
estimate the can SOC
nonlinear characteristics of the battery, the DEKF can estimate the SOC and the capacity estimate
and the the SOC
capacity and
more the capacity
accurately than more
the
more
accurately
offline
accurately than
than the
method. the offline
Because
offline themethod.
observer
method. Because
improves
Because the observer
thethe performance
observer improves
improves of the
the performance
ECM, the DEKF
performance of
of the
canECM,
the estimate
ECM, the
the
DEKF
the SOC can
and estimate
capacity themoreSOC and capacity
accurately than more
the accurately
offline method
DEKF can estimate the SOC and capacity more accurately than the offline method when the ECM than when the offline
the ECM method
accuracy when the
increases. ECM
accuracy
accuracy increases.
increases.

3.3.1.
3.3.1. SOC
SOC Estimation
Estimation
Energies 2020, 13, 2138 11 of 20

3.3.1. SOC Estimation


To estimate the SOC, a nonlinear state space equation represented as f (x k , uk , θk and z(x k , uk , θk )
indicating the process and measurement equations of the battery is needed. The state-space matrix
and the input are expressed as follows.

xk = [ SOCk V1,k ]T ∼ N (x̂k , Pk ) (15)

where x̂k represents the estimated value of the battery state and Pk represents the error covariance
indicating the deviation of the estimated value from the true value. The Pk influences the state
estimation performance because it affects the Kalman gain, as indicated by Equation (22). The system
input is the current, which is represented by the equation below.

uk = ik (16)

The process equation of the ECM indicating the SOCk and V 1 yield is expressed as follows.
" # " # " #
1 0 SOCk −∆t/Cn
f (xk , uk , θk ) = × + × i k + wx (17)
0 exp(−∆t/τk ) V1,k R1 (1 − exp(−∆t/τk )

where wx and vx are the state and measurement noise of state filer, respectively.
The measurement function of the ECM is defined as follows.

z(xk , uk , θk ) = OCVk (SOCk ) + ik R0 + V1 + vx (18)

In the first EKF, the state and the system matrices are defined as a Jacobian matrix through
differentiation with respect to xk for linearization, as follows.

∂ f (xk , uk , θ̂− )
" #
k 
 1 0
Ak = = (19)
∂x

 +
0 exp(−∆t/τk )
xk =x̂k


∂z(xk , uk , θ̂− ) 
k 
h ∂OCV 
 i
Hk = = ∂SOC xk =x̂−
1 (20)
∂x

xk =x̂−

k

The state filter recursively progresses as follows.

x̂−
k +1
= f (x̂k+ , uk , θ̂−k+1 )
(21)
Pk+1 = Ak Pk Ak T + Qx
−

x −

x − T − T
 Kk = Pk+1 Hk (Hk Pk+1 Hk + R )


(22)
 x̂+ = x̂− + Kx (zk − ẑk ), P+ = (I − Kk Hk )P−


k +1 k +1 k k +1 k +1

In the first step, the initial values (x0 and P0 ) and noise parameters (Qx and Rx ) of the system are
set. The second step involves predicting when the estimated prior value (x̂− k +1
) and the error covariance
−
(Pk+1 ) are calculated. This step is related to the system model corresponding to the ECM. The final
step is the innovation step in which the Kalman gain (Kkx ) is calculated using the system variables
(Hk and Rx ) from the prediction step. x̂k++1 is estimated and calibrated by adding the prior estimate to
the value obtained by multiplying the Kalman gain and the measurement error. Because the Kalman
gain adjusts the state, the error covariance, which indicates the difference between the estimate and the
true value, is calculated. All the calculation steps were repeated at each sampling time.
Energies 2020, 13, 2138 12 of 20

3.3.2. Capacity Identification

The second EKF observer identifies the ohmic resistance and capacity of the battery. The weight
matrix vector is defined by the formula below.

θk = [ Cn,k R0,k ]T (23)

where θk is the parameter indicating the ohmic resistance and capacity of the battery. The process
equation, which indicates the weight filter and measurement equation of the second EKF, is as follows.

θ̂k+1 = θ̂k + wθ (24)

z(xk , uk , θk ) = OCVk (SOCk ) + ik R0 + V1 + vθ (25)

where wθ and vθ are the state and measurement noise of the weight filer, respectively.
For the successful operation of the DEKF, the linearization of the weight filter in the innovation
step of EKFθ is essential. The Jacobian matrix is defined by the equation below.

dz(x̂− , u , θ) 

k +1 k +1
Ck =  (26)
dθ 
θ=θ̂− k +1

The Ck identification is more complex than the Hk identification. The relationship between the
estimated terminal voltage of the ECM and θk is not determined directly. To obtain the Jacobian
matrix, we use the relationship between the partial difference and the total difference, by decomposing
Equation (26) as follows.

∂z(x̂− ∂z(x̂−

dz(x̂− ,u ,θ) 
k +1 k +1
,u ,θ)
k +1 k +1
,u ,θ) dx̂−
k +1 k +1 k +1
dθ
 = ∂θ
+ ∂ x̂ − dθ
θ=θ̂− k +1
k +1 (27)
dx̂−k +1
∂ f (x̂k+ ,uk ,θ) ∂ f (x̂k+ ,uk ,θ) dx̂k+
dθ = ∂θ
+ ∂x̂k + dθ

dx̂−
k +1 ∂ f (x̂k+ , uk , θ) ∂ f (x̂k+ , uk , θ) dx̂k+
= + (28)
dθ ∂θ ∂x̂k+ dθ

dx̂k+ dx̂−
k
dz(x̂−
k
, uk , θ)
= − Kkx (29)
dθ dθ dθ
The calculation of Kalman gain and calibration procedure is the same with Equations (21) and (22)
as follows.
θ̂−
k +1
= θ̂k+
(30)
Sk+1 = Sk + Qθ
−

θ θ −

− T − T
 Kk = Sk+1 Ck (Ck Sk+1 Ck + R )


(31)
+
 θ̂n,k = θ̂−n,k+1 + Kkθ (zk − ẑk ), Sk++1 = (I − Kk Hk )S−k+1


+1

where Sk is the error covariance of θk .

The DEKF procedure is implemented as follows.

1. Initial value setting and adjustment: According to the measured voltage and SOC-OCV
relationship, the initial values of the system variables (OCV 0 , H0 , and P0 ) are automatically
determined. The OCV is applied to the ECM and the initial value of the Hk is calculated using
the SOC-OCV curve based on the experimental data, as shown in Figure 6. According to the
OCV-SOC curve for the initial condition, the error covariance (P0 ) is approximated as follows.
 
P0 = SOCtable (OCVtable ) − SÔC0  (32)
Energies 2020, 13, 2138 13 of 20

where SOCtable (OCVtable ) represents the SOC-OCV relationship extracted from the experimental
data, and SÔC0 represents the initial SOC.
2. Adaptive model: In the adaptive battery model, the ohmic resistance, diffusion resistance, and
OCV are estimated via MAFF-RLS. According to the estimated parameter, the adaptive ECM
estimates the terminal voltage, Hk , and Ck for calculating the Kalman gain. The estimated
terminal voltage is applied to the proposed DEKF through an ECM error with the measured
terminal voltage.
3. SOC and capacity estimation: The DEKF estimates the SOC and capacity in parallel structure, as
shown in Figure 10.

2) Parameter identification

Measure input current and

terminal voltage

3) The adaptive dual Kalman filter

1) Off-line data input Set noise parameter & Set noise parameter &
Initial Covariance Initial Covariance
Calculate initial OCV Calculate initial error covariance (P0)

OCV-SOC curve Time update for the state Time update for the parameter
SOCk − , 𝑃𝑘 (−) Cn,k − , 𝑆𝑘 (−)

SOC-OCV curve
𝐶
Calculate Kalman gain ( Kk𝑆𝑂𝐶 ) Calculate Kalman gain ( K k 𝑛 )

Measurement update Measurement update

Initial terminal voltage (Vt) Initial error covariance (P0) for the state for the parameter
SOCk + 𝑃𝑘 (+) Cn,k + 𝑆𝑘 (+)

Figure 10. Flowchart of the dual extended Kalman filter (DEKF).

In this study, the DEKF was not only used for estimating the state of the battery (SOC and capacity)
but also applied to the prediction model in real-time (except for the initial value). The proposed
algorithm identifies the parameters according to the MAFF-RLS. These parameters are applied to the
DEKF and prediction models.

3.4. Autoregressive (AR) Model

International Journal of Geom. Methods in Mod. Physics ...
The time-series approach helps predict the capacity loss without the complete aging data because
Quantum model
the prediction Relativ.for
and Gravit. data assumes that the predicted data are related to past data.
time-series
The AR model represents the current variable using the linear relationship of the previous time variable,
Fortschr.
as follows.
x(t) = a1 xt−1 + a2 xt−2 + · · · + ap xt−p + εt (33)

where p represents the model order, which indicates the number of past data. ap is a coefficient and ε
represents white noise with zero mean.
The conventional AR model cannot predict nonlinear trends. Long et al. [36] used an AR model
based on particle swarm optimization to determine the optimal order. Because the AR model is affected
by the previous trend, the prediction result for the capacity fading is characterized by a linear trend.
To overcome the limitation of the AR model, Liu et al. [37] proposed an AR model that considers
two capacity degradation trends fitting different aging curves. However, in conventional studies,
the capacity value was not feasible because it was measured via an under restricted experimental
condition [12,26]. These findings indicate the following. First, the statistical model requires a large
amount of experimental data to fit the capacity fading and additional nonlinear factor. Second, the
measurement value in the experiment is not useful in real applications because the capacity changes,
according to the SOC, depth of discharge (DOD), cycle, C-rate, etc. Lastly, if the variability of the
battery capacity is nonlinear, it is difficult to predict using a linear relationship.
Energies 2020, 13, 2138 14 of 20

3.5. Integrated Model

For the previously mentioned AR model, the estimation and capacity prediction results were
dependent on the accuracy of the measured data and the selection of the HI. In this study, the capacity
was predicted using the multivariate AR model and the HI. The capacity for the next 500 cycles was
predicted according to the capacity in the previous 200 cycles. The multivariate AR model combining
R0 and Rdiff prediction data, which was expected to have strong correlation with capacity, was used for
the AR model. The equation for predicting the capacity was as follows.
p
X
Cn,t = at Cn,t−τ + b · R0,t + c · R1,t + εt (34)
τ=1

where t represents the prediction time, Cn,t−τ is the index value of the previous 200 cycles data from the
prediction time, and p represents the number of previous load values used. The prediction for 500 cycles
was performed by increasing the prediction time through the iterative multivariate regression model
and the previous predicted value was used when there was no previous measured load value.

4. Results and Discussion

The capacity prediction result was compared with the measured capacity and the estimated
capacity from the SOC and SOH joint estimator (DEKF). These simulations were implemented for two
cases: the measured capacity and estimated capacity of the DEKF. To validate the multivariate AR
model and integrated model, the capacity prediction results were compared according to the variable
numbers used for defining the AR model. Then, the performance for the two cases was evaluated
using the mean absolute percentage error (MAPE), which was defined as follows.

N
1 X |Rk − Fk |
MAPE = × 100 (35)
N |Rk |
k =1

where Rk is the reference value and Fk is the forecast value.

As shown in Figure 11, the simulation results depended on which variable was applied to the AR
model. The red line represents the measured capacity for fitting the data using 200 cycles that were
obtained from the experimental data. The blue line represents the prediction results and the green line
represents the reference. Figure 11a shows the prediction results obtained using only capacity data
for the AR model. The prediction results exhibited a linear trend because the approximate capacity
trend was almost linear in the 200 cycles. The MAPE for these results was 4.4006%. Figure 11b
shows the results obtained using the capacity and ohmic resistance (R0 ). Although the MAPE was
1.0606%, the prediction error increased exponentially with the cycle number. According to Figure 11c,d,
the prediction results were more reliable, as the MAPE was 0.5324% and 0.4165%, respectively.
This indicates that, in addition to the capacity and R1 , Rdiff reflects the nonlinear characteristics of
capacity prediction.
The prediction results of the AR model utilizing the estimated capacity from the DEKF are shown
in Figure 12. The red and purple lines represent the estimated capacity. The average MAPE was under
1%. The prediction result in Figure 12a is similar to that in Figure 12a obtained using the estimated
capacity only. The MAPE of the prediction results was 1.4518%, and the performance of prediction
model was higher than that achieved using the measured capacity because the overall trend between
50 and 200 cycles tended to decrease, but the estimated capacity increased and decreased repeatedly.
When R0 and R1 were applied to the AR model, the prediction results were superior to those obtained
using the capacity only, as shown in Table 3.
approximate capacity trend was almost linear in the 200 cycles. The MAPE for these results was
4.4006%. Figure 11b shows the results obtained using the capacity and ohmic resistance (R0).
Although the MAPE was 1.0606%, the prediction error increased exponentially with the cycle number.
According to Figures 11c and 11d, the prediction results were more reliable, as the MAPE was 0.5324%
and 0.4165%, respectively. This indicates that, in addition to the capacity and R1, Rdiff reflects the
Energies 2020, 13, 2138 15 of 20
nonlinear characteristics of capacity prediction.

3.3
3.3
Pre data
Pre data
Prediction
3.2 Prediction
3.2 Reference
Reference

3.1

3.1

Energies
3
2020, 13, x FOR PEER REVIEW 15 of 20

Capacity (Ah)
Capacity (Ah)

3
2.9
3.3 3.3

2.8 Pre data Pre data

2.9
Prediction Prediction
3.2 Reference 3.2 Reference
2.7

2.8

2.6
3.1 3.1

2.7
2.5
Capacity (Ah)

Capacity (Ah)
0 100 200 300 400 500 600 700 800
0 100 200 300 400 500 600 700 800
3 3
Cycle Cycle

Energies 2020, 13, x FOR PEER REVIEW 15 of 20

2.9
(a) Conventional AR model using Cn 2.9
(b) Multivariate AR model using Cn and R0
3.3 3.3
2.8 2.8
Pre data Pre data
Prediction Prediction
3.2 Reference 3.2 Reference
2.7 2.7
0 100 200 300 400 500 600 700 800 0 100 200 300 400 500 600 700 800
Cycle Cycle
3.1 3.1

(c) Multivariate AR model using Cn and R1 (d) Multivariate AR model using Cn, R0, and R1
Capacity (Ah)

Capacity (Ah)

3 3

Figure 11. Capacity prediction results obtained using the autoregressive (AR) model and the
2.9
measured capacity.
2.9

The prediction results of the AR model utilizing

2.8 2.8 the estimated capacity from the DEKF are
shown in Figure 12. The red and purple lines represent the estimated capacity. The average MAPE
was under 1001%. 200
The prediction result600in Figure 12a is 2.7
similar to that in 300
Figure 12a500obtained using the
2.7
0 300 400 500 700 800 0 100 200 400 600 700 800

estimated capacity only. The MAPE of the prediction results was 1.4518%,Cycle
Cycle
and the performance of
prediction model was
(c) Multivariate AR higher thanCthat
model using n andachieved
R1 using the measured
(d) Multivariate AR capacity because
model using Cn, R0, the
and overall
R1
trend between 50 and 200 cycles tended to decrease, but the estimated capacity increased and
Figure 11. Capacity prediction results obtained using the autoregressive (AR) model and the
decreased Figure 11. Capacity
repeatedly. Whenprediction
R0 andresults
R1 were obtained
appliedusingto the
theautoregressive
AR model, the (AR) model andresults
prediction the were
measured capacity.
measured capacity.
superior to those obtained using the capacity only, as shown in Table 3.

The prediction results of the AR model utilizing 3.3 the estimated capacity from the DEKF are

shown in Figure 12. The red and purplePrelines represent the estimated capacity. The average MAPE
3.3 Pre data

data Estimation

was3.2under 1%. The prediction result in Estimation

Figure 12a is similar
3.2
to that in Figure 12a obtainedPrediction
using
Reference
the
Prediction
estimated capacity only. The MAPE of Reference
the prediction results was 1.4518%, and the performance of
prediction model was higher than that achieved using the measured capacity because the overall
3.1
3.1

trend3 between 50 and 200 cycles tended to decrease, but the estimated capacity increased and
Capacity (Ah)

decreased repeatedly. When R0 and R1 were applied to the AR model, the prediction results were
Capacity (Ah)

superior
2.9
to those obtained using the capacity only, as 2.9 shown in Table 3.
2.8
3.3
2.8
3.3 Pre data
2.7 Estimation
Pre data
Estimation 3.2 Prediction
2.7
3.2 Reference
Prediction 0 100 200 300 400 500 600 700 800
2.6
Reference Cycle
0 100 200 300 400 500 600 700 800
3.1
3.1 Cycle

(a) Conventional AR model using Cn (b) Multivariate AR model using Cn and R0

Capacity (Ah)

3 3
Capacity (Ah)

2.9 Figure 12. Cont.

2.9

2.8
2.8

2.7

2.7
0 100 200 300 400 500 600 700 800
2.6
0 100 200 300 400 500 600 700 800 Cycle

Cycle
Energies 2020,
Energies 13, 2138
2020, 13, x FOR PEER REVIEW 16
16 of 20
of 20
Energies 2020, 13, x FOR PEER REVIEW 16 of 20
3.3 3.3
Pre data Pre data
3.3 Estimation 3.3 Estimation
3.2 Prediction
Pre data 3.2 Prediction
Pre data
Reference
Estimation Reference
Estimation
3.2 Prediction 3.2 Prediction
3.1 Reference 3.1 Reference

3.1 3.1
(Ah)

(Ah)
3 3
CapacityCapacity

CapacityCapacity
(Ah)

(Ah)
3 3
2.9 2.9

2.9 2.9
2.8 2.8

2.8 2.8
2.7 2.7
0 100 200 300 400 500 600 700 800 0 100 200 300 400 500 600 700 800

2.7 Cycle 2.7 Cycle

0 100 200 300 400 500 600 700 800 0 100 200 300 400 500 600 700 800

(c) Multivariate AR model

Cycle using Cn and R1 Cycle using Cn, R0, and R1
(d) Multivariate AR model
(c) Multivariate AR model using Cn and R1 (d) Multivariate AR model using Cn, R0, and R1
Figure 12. Capacity prediction results obtained using the AR model and dual extended Kalman
Figure 12. Capacity prediction resultsfilter (DEKF)
using(case
using the1).
Figure 12. Capacity prediction results obtained the AR model and dual extended Kalman filter
obtained AR model and dual extended Kalman
(DEKF) (case 1).
filter (DEKF) (case 1).
Table 3. Comparison of the mean absolute percentage error (MAPE).
Table 3. Comparison of the mean absolute percentage error (MAPE).
Table 3. Comparison Cnof the mean absolute
Cn + R0 percentage Cnerror
+ R1 (MAPE). Cn + R0 + R1
Cn Cn + R0 Cn + R1 Cn + R 0 + R 1
Experiment (%) 4.4006
Cn C1.0606
n + R0 C0.5324
n + R1 Cn +0.4165
R0 + R1
Case Experiment
1 (%)
Experiment (%)1 (%)
(%) 4.4006
1.4518
4.4006
1.0606 0.7302 0.53240.4566 0.4165 0.5183
Case 1.4518 0.7302 1.0606 0.45660.5324 0.5183 0.4165
Case 12 (%)
Case (%)
Case 2 (%)
3.6067
1.4518
3.6067
3.0266
3.0266 0.7302
9.3908
9.39080.4566 2.5599
2.5599
0.5183
Case 2 (%) 3.6067 3.0266 9.3908 2.5599
The underestimated capacity of DEKF is selected in order to determine the HI, as shown in
The
Figure underestimated
13.underestimated
The underestimated capacity of DEKF
capacity has ais is selected
more in order
nonlinear trend to determine the HI, as shownand in
The capacity of DEKF selected in order to because
determine thisthevalue
HI,fluctuates
as shown in
Figure 13.
the accuracyThe underestimated capacity has a more nonlinear trend because this value fluctuates and
Figure 13. Thedecreases more than
underestimated Figure
capacity has 12.a In Figure
more 13a and
nonlinear 13b,because
trend the prediction
this valueresults are almost
fluctuates and
the accuracy
linear and MAPE decreases more
is 3.6067% than Figure
and Figure
3.0266%, 12. In Figure 13a,b, the prediction results are almost linear
the accuracy decreases more than 12. respectively.
In Figure 13aThus, and 13b,it seems that the other
the prediction resultsparameter
are almost is
and MAPEfor
required is 3.6067%
predicting andthe3.0266%, respectively.
nonlinear Thus, itand
characteristics seems that the other
enhancing the parameter isperformance.
prediction required for
linear and MAPE is 3.6067% and 3.0266%, respectively. Thus, it seems that the other parameter is
predicting
When thefor the nonlinear
diffusion characteristics
resistance and enhancing
is applied the prediction
to the multivariate performance.
AR model, When theresult
diffusion
required predicting the nonlinear characteristics and enhancing the the prediction
prediction performance. has
resistance
weakening is applied
performance to the multivariate
than other AR model,
results,toastheshown the prediction
in Figure AR result
13c, model, has weakening
which represents performance
the result
When the diffusion resistance is applied multivariate the prediction resultofhasthe
than other results,
multivariate AR as shown
model and in Figure
the MAPE 13c,
is which represents
9.3908%. If two the result
resistances (R of the multivariate AR model
0 and R1) are applied to the AR
weakening performance than other results, as shown in Figure 13c, which represents the result of the
and
modeltheatMAPE
theAR is 9.3908%.
same time, If two resistances (R0 and R1 ) are In
applied to the AR canmodel at the same
multivariate model andthetheMAPE
MAPE is improved
is 9.3908%. to 2.55%.
If two resistances particular,
(R0 and Rit be seen that
1) are applied to the AR
the
time, the
prediction MAPE is improved
performance to 2.55%. In particular, it can be seen that the prediction performance is
model at the same time,is theimproved
MAPEby is changing
improvedthe toslope
2.55%.fromIn about 500 cycles,
particular, it canasbeshown
seen in Figure
that the
improved
13d. by changing the slope from about 500 cycles, as shown in Figure 13d.
prediction performance is improved by changing the slope from about 500 cycles, as shown in Figure
13d.3.3
3.3
Pre data Pre data
3.3 Estimation 3.3 Estimation
3.2 Prediction
Pre data 3.2 Prediction
Pre data
Reference
Estimation Reference
Estimation
3.2 Prediction 3.2 Prediction
3.1 Reference 3.1 Reference

3.1 3.1
(Ah)
(Ah) (Ah)

3 3
Capacity
Capacity

(Ah)

3 3
Capacity
Capacity

2.9 2.9

2.9 2.9
2.8
2.8

2.8
2.8
2.7
2.7
0 100 200 300 400 500 600 700 800
0 100 200 300 400 500 600 700 800
2.7 Cycle
2.7 Cycle
0 100 200 300 400 500 600 700 800
0 100 200 300 400 500 600 700 800
AR model using Cn
(a) Conventional Cycle (b) Multivariate AR model
Cycle
using Cn and R0
(a) Conventional AR model using Cn Figure 13. Cont.
(b) Multivariate AR model using Cn and R0
Energies 2020, 13, 2138 17 of 20
Energies 2020, 13, x FOR PEER REVIEW 17 of 20

3.4
3.3
Pre data
Pre data
3.2 Estimation
Estimation
Prediction Prediction
3.2
Reference Reference
3

3.1
2.8
Capacity (Ah)

Capacity (Ah)
2.6 3

2.4
2.9

2.2

2.8
2

2.7
1.8
0 100 200 300 400 500 600 700 800
0 100 200 300 400 500 600 700 800
Cycle
Cycle

(c) Multivariate AR model using Cn and R1 (d) Multivariate AR model using Cn, R0, and R1

Figure 13.
Figure Capacityprediction
13. Capacity predictionresults
resultsobtained
obtained using
using the
the AR
AR model
model and
and DEKF
DEKF (case
(case 2).
2).

As indicated by the foregoing simulation results, the capacity prediction results depend on the
As indicated by the foregoing simulation results, the capacity prediction results depend on the
accuracy of the capacity data and the multivariate HI, which indicates the nonlinear trend of the
accuracy of the capacity data and the multivariate HI, which indicates the nonlinear trend of the
prediction results. Thus, the prediction performance is highly dependent on selecting the appropriate
prediction results. Thus, the prediction performance is highly dependent on selecting the appropriate
experimental data and HI. The limitation of the AR model that the prediction result is linear can be
experimental data and HI. The limitation of the AR model that the prediction result is linear can be
resolved by applying another HI. However, the measured capacity cannot be used in real applications
resolved by applying another HI. However, the measured capacity cannot be used in real applications
because the capacity is only measured under restricted experimental conditions. Therefore, in this
because the capacity is only measured under restricted experimental conditions. Therefore, in this
study, the AR model was integrated with the joint estimator to estimate the SOC and SOH.
study, the AR model was integrated with the joint estimator to estimate the SOC and SOH.
5. Conclusions
5. Conclusions
In practical applications, it is difficult to determine the appropriate HI and capacity because of
In practical applications, it is difficult to determine the appropriate HI and capacity because of
the fluctuation of the battery parameters with the aging of the battery. The battery cycle is difficult to
the fluctuation of the battery parameters with the aging of the battery. The battery cycle is difficult to
determine according to the operating range of applications. Hence, in this study, capacity data were
determine according to the operating range of applications. Hence, in this study, capacity data were
obtained from a real-time estimator for predicting the capacity loss. The multivariate AR model and
obtained from a real-time estimator for predicting the capacity loss. The multivariate AR model and
DKEF were integrated for real-time applications. The HI was estimated via MAFF-RLS, which was
DKEF were integrated for real-time applications. The HI was estimated via MAFF-RLS, which was
synchronized with the DEKF. Two resistances of the battery were applied to the AR model to reflect the
synchronized with the DEKF. Two resistances of the battery were applied to the AR model to reflect
nonlinear characteristics of capacity reduction. These resistances were estimated by the ARX model
the nonlinear characteristics of capacity reduction. These resistances were estimated by the ARX
and MAFF-RLS to be utilized for the HI. Additionally, to overcome the limitations of existing methods
model and MAFF-RLS to be utilized for the HI. Additionally, to overcome the limitations of existing
predicted using measurement data, integration of the multivariate AR model and DEKF was necessary
methods predicted using measurement data, integration of the multivariate AR model and DEKF
to estimate the capacity values in real-time.
was necessary to estimate the capacity values in real-time.
According to the simulation results, the capacity prediction performance depends on aging factors
According to the simulation results, the capacity prediction performance depends on aging
related to the battery capacity. If the accuracy of the capacity value is increased, better performance can
factors related to the battery capacity. If the accuracy of the capacity value is increased, better
be achieved. Therefore, the capacity is the factor with the greatest effect on the prediction performance.
performance can be achieved. Therefore, the capacity is the factor with the greatest effect on the
However, for practical applications, the AR model should utilize the capacity estimated by a real-time
prediction performance. However, for practical applications, the AR model should utilize the
estimator. Future research plans are as follows. First, a significant HI, such as the discharging rate or
capacity estimated by a real-time estimator. Future research plans are as follows. First, a significant
temperature, will be applied to the AR model since the lifespan is affected by these factors. Second,
HI, such as the discharging rate or temperature, will be applied to the AR model since the lifespan is
data-driven methods will be utilized for optimizing both the prediction model and HI. Our technique
affected by these factors. Second, data-driven methods will be utilized for optimizing both the
can be applied to a wide range of data-driven methods by adopting the adaptive controller that can
prediction model and HI. Our technique can be applied to a wide range of data-driven methods by
estimate the optimal battery parameters in real-time. Third, since the proposed method utilized the
adopting the adaptive controller that can estimate the optimal battery parameters in real-time. Third,
adaptive controller and parameter identification, it has the potential to contstruct the new model for
since the proposed method utilized the adaptive controller and parameter identification, it has the
the next generation battery, which has different characteristics.
potential to contstruct the new model for the next generation battery, which has different
characteristics.
Author Contributions: Data curation, G.K.; Methodology, J.P. and M.L.; Supervision, J.K.; Validation, J.P.;
Visualization, J.P.; Writing—original draft, J.P.; Writing—review & editing, M.L., G.K., S.P. and J.K. All authors
Author Contributions:
have read and agreed toData curation, version
the published G.K.; Methodology, J.P. and M.L.; Supervision, J. K.; Validation, J.P.;
of the manuscript.
Visualization, J.P.; Writing – original draft, J. P.; Writing – review & editing, M.L., G.K., S.P. and J.K.
Funding: This research received no external funding.
Funding: This research received no external funding.
Energies 2020, 13, 2138 18 of 20

Acknowledgments: The Korea Electric Power Corporation (R17XA05-55) and Korea Institute of Energy Technology
Evaluation and a grant (17TLRP-C135446-01, Development of Hybrid Electric Vehicle Conversion Kit for Diesel
Delivery Trucks and its Commercialization for Parcel Services) from the Transportation & Logistics Research
Program (TLRP) funded by the Ministry of Land, Infrastructure, and Transportation of the Korean government
supported this work.
Conflicts of Interest: The authors declare no conflict of interest.

Abbreviations
AR Autoregressive
ARX Autoregressive Exogenous
BMS Battery Management System
DEKF Dual Extended Kalman Filter
DOD Depth of Discharge
ECM Equivalent Circuit Model
EIS Electrochemical Impedance Spectroscopy
EKF Extended Kalman Filter
ESS Energy Storage System
HI Health Indicator
MAFF Multiple Adaptive Forgetting Factor
MAPE Mean Absolute Percentage Error
NMC Nickel Manganese Cobalt Oxide
OCV Open Circuit Voltage
PCA Pearson Correlation Analysis
RLS Recursive Least Square
RUL Remaining Useful Life
SEI Solid–Electrolyte Interphase
SOC State of Charge
SOH State of Health
SRC Spearman Rank Correlation

References
1. IRENA. Electricity Storage and Renewables: Costs and Markets to 2030; International Renewable Energy Agency:
Abu Dhabi, United Arab Emirates, 2017.
2. International Renewable Energy Agency (IRENA). Global Energy Transformation: A Roadmap to 2050, 2019 ed.;
IRENA: Abu Dhabi, United Arab Emirates, 2019.
3. Mauger, A.; Julien, C.M.; Paolella, A.; Armand, M.; Zaghib, K. Building better batteries in the solid state:
A review. Materials 2019, 12, 3892. [CrossRef] [PubMed]
4. Bloomberg New Energy Finance. Energy Storage Outlook 2019; Bloomberg New Energy Finance Publication:
Paris, France, 2019.
5. Dunn, B.; Kamath, H.; Tarascon, J.M. Electrical energy storage for the grid: A battery of choices. Science 2011,
334, 928–935. [CrossRef] [PubMed]
6. World Energy Council. Energy Storage Monitor Latest Trends in Energy Storage; World Energy Council: London,
UK, 2019; p. 15.
7. Bloomberg New Energy Finance. Available online: https://about.bnef.com/blog/behind-scenes-take-lithium-
ion-battery-prices/ (accessed on 5 May 2019).
8. Ren, H.; Zhao, Y.; Chen, S.; Wang, T. Design and implementation of a battery management system with
active charge balance based on the SOC and SOH online estimation. Energy 2019, 166, 908–917. [CrossRef]
9. Garche, J.; Karden, E.; Moseley, P.T.; Rand, D.A. (Eds.) Lead-Acid Batteries for Future Automobiles; Elsevier:
Amsterdam, The Netherlands, 2017; pp. 133–146.
10. Kwiecien, M.; Schröer, P.; Kuipers, M.; Sauer, D.U. Current research topics for lead–acid batteries. In Lead-Acid
Batteries for Future Automobiles; Elsevier: Amsterdam, The Netherlands, 2017; pp. 133–146.
11. Wu, J.; Wang, Y.; Zhang, X.; Chen, Z. A novel state of health estimation method of Li-ion battery using group
method of data handling. J. Power Sources 2016, 327, 457–464. [CrossRef]
Energies 2020, 13, 2138 19 of 20

12. Wu, J.; Zhang, C.; Chen, Z. An online method for lithium-ion battery remaining useful life estimation using
importance sampling and neural networks. Appl. Energy 2016, 173, 134–140. [CrossRef]
13. Kang, B.O.; Lee, M.; Kim, Y.; Jung, J. Economic analysis of a customer-installed energy storage system for
both self-saving operation and demand response program participation in South Korea. Renew. Sustain.
Energy Rev. 2018, 94, 69–83. [CrossRef]
14. Gruber, P.W.; Medina, P.A.; Keoleian, G.A.; Kesler, S.E.; Everson, M.P.; Wallington, T.J. Global lithium
availability: A constraint for electric vehicles? J. Ind. Ecol. 2011, 15, 760–775. [CrossRef]
15. Lucu, M.; Martinez-Laserna, E.; Gandiaga, I.; Camblong, H. A critical review on self-adaptive Li-ion battery
ageing models. J. Power Sources 2018, 401, 85–101. [CrossRef]
16. Maheshwari, A.; Paterakis, N.G.; Santarelli, M.; Gibescu, M. Optimizing the operation of energy storage
using a non-linear lithium-ion battery degradation model. Appl. Energy 2020, 261, 114360. [CrossRef]
17. Sun, Y.; Hao, X.; Pecht, M.; Zhou, Y. Remaining useful life prediction for lithium-ion batteries based on an
integrated health indicator. Microelectron. Reliab. 2018, 88, 1189–1194. [CrossRef]
18. Xiong, R.; Zhang, Y.; Wang, J.; He, H.; Peng, S.; Pecht, M. Lithium-ion battery health prognosis based on
a real battery management system used in electric vehicles. IEEE Trans. Veh. Technol. 2018, 68, 4110–4121.
[CrossRef]
19. Zheng, Y.; Qin, C.; Lai, X.; Han, X.; Xie, Y. A novel capacity estimation method for lithium-ion batteries using
fusion estimation of charging curve sections and discrete Arrhenius aging model. Appl. Energy 2019, 251,
113327. [CrossRef]
20. Liu, D.; Zhou, J.; Liao, H.; Peng, Y.; Peng, X. A health indicator extraction and optimization framework for
lithium-ion battery degradation modeling and prognostics. IEEE Trans. Syst. Man Cyber. Syst. 2015, 45,
915–928.
21. Zhou, Y.; Huang, M.; Chen, Y.; Tao, Y. A novel health indicator for on-line lithium-ion batteries remaining
useful life prediction. J. Power Sources 2016, 321, 1–10. [CrossRef]
22. Verma, P.; Maire, P.; Novák, P. A review of the features and analyses of the solid electrolyte interphase in
Li-ion batteries. Electrochim. Acta 2010, 55, 6332–6341. [CrossRef]
23. Jung, B.; Lee, B.; Jeong, Y.C.; Lee, J.; Yang, S.R.; Kim, H.; Park, M. Thermally stable non-aqueous ceramic-coated
separators with enhanced nail penetration performance. J. Power Sources 2019, 427, 271–282. [CrossRef]
24. Sarkar, A.; Shrotriya, P.; Chandra, A.; Hu, C. Chemo-economic analysis of battery aging and capacity fade in
lithium-ion battery. J. Energy Storage 2019, 25, 100911. [CrossRef]
25. Krewer, U.; Röder, F.; Harinath, E.; Braatz, R.D.; Bedürftig, B.; Findeisen, R. dynamic models of Li-Ion
batteries for diagnosis and operation: A review and perspective. J. Electrochem. Soc. 2018, 165, A3656–A3673.
[CrossRef]
26. Pastor-Fernández, C.; Uddin, K.; Chouchelamane, G.H.; Widanage, W.D.; Marco, J. A comparison between
electrochemical impedance spectroscopy and incremental capacity-differential voltage as Li-ion diagnostic
techniques to identify and quantify the effects of degradation modes within battery management systems.
J. Power Sources 2017, 360, 301–318. [CrossRef]
27. Gomez, J.; Nelson, R.; Kalu, E.E.; Weatherspoon, M.H.; Zheng, J.P. Equivalent circuit model parameters
of a high-power Li-ion battery: Thermal and state of charge effects. J. Power Sources 2011, 196, 4826–4831.
[CrossRef]
28. Pan, H.; Lü, Z.; Wang, H.; Wei, H.; Chen, L. Novel battery state-of-health online estimation method using
multiple health indicators and an extreme learning machine. Energy 2018, 160, 466–477. [CrossRef]
29. Qiu, X.; Wu, W.; Wang, S. Remaining useful life prediction of lithium-ion battery based on improved cuckoo
search particle filter and a novel state of charge estimation method. J. Power Sources 2020, 450, 227700.
[CrossRef]
30. Xue, Z.; Zhang, Y.; Cheng, C.; Ma, G. Remaining useful life prediction of lithium-ion batteries with adaptive
unscented kalman filter and optimized support vector regression. Neurocomputing 2020, 376, 95–102.
[CrossRef]
31. Yang, F.; Wang, D.; Zhao, Y.; Tsui, K.L.; Bae, S.J. A study of the relationship between coulombic efficiency and
capacity degradation of commercial lithium-ion batteries. Energy 2018, 145, 486–495. [CrossRef]
32. Duong, V.H.; Bastawrous, H.A.; Lim, K.; See, K.W.; Zhang, P.; Dou, S.X. Online state of charge and model
parameters estimation of the LiFePO4 battery in electric vehicles using multiple adaptive forgetting factors
recursive least-squares. J. Power Sources 2015, 296, 215–224. [CrossRef]
Energies 2020, 13, 2138 20 of 20

33. Zilberman, I.; Ludwig, S.; Jossen, A. Cell-to-cell variation of calendar aging and reversible self-discharge in
18650 nickel-rich, silicon–graphite lithium-ion cells. J. Energy Storage 2019, 26, 100900. [CrossRef]
34. Wassiliadis, N.; Adermann, J.; Frericks, A.; Pak, M.; Reiter, C.; Lohmann, B.; Lienkamp, M. Revisiting the
dual extended Kalman filter for battery state-of-charge and state-of-health estimation: A use-case life cycle
analysis. J. Energy Storage 2018, 19, 73–87. [CrossRef]
35. Plett, G.L. Extended Kalman filtering for battery management systems of LiPB-based HEV battery packs:
Part 3. State and parameter estimation. J. Power Sources 2004, 134, 277–292. [CrossRef]
36. Long, B.; Xian, W.; Jiang, L.; Liu, Z. An improved autoregressive model by particle swarm optimization for
prognostics of lithium-ion batteries. Microelectron. Reliab. 2013, 53, 821–831. [CrossRef]
37. Liu, D.; Luo, Y.; Peng, Y.; Peng, X.; Pecht, M. Lithium-ion battery remaining useful life estimation based
on nonlinear AR model combined with degradation feature. In Proceedings of the Annual Conference of
the Prognostics and Health Management Society, Minneapolis, MN, USA, 23–27 September 2012; Volume 3,
pp. 1803–1836.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).