AN OVERVIEW OF STATE OF CHARGE(SOC) AND STATE OF HEALTH(SOH)

ESTIMATION METHODS OF LI-ION BATTERIES

K. SAQLI(a), H. BOUCHAREB(b), M. OUDGHIRI(c), N.K. M’SIRDI(d)

(a),(b),(c)
Sidi Mohamed Ben Abdellah University - National School of Applied Sciences, Fez, Morocco
(d)
LIS - Informatics and Systems Laboratory (LIS CNRS 7020), Aix Marseille University, CNRS,13397 Marseille Cedex,
France
(a)
[email protected], (b)[email protected], (c)[email protected]
(d)
[email protected]

ABSTRACT battery pack. An accurate SOC can be achieved using

Battery Management System (BMS) is an essential the proper battery model and estimation algorithm.
component for lithium-ion battery-based devices. It
Different estimation approaches were introduced in
provides a variety of functionalities that help improve the
literature in the aim of predicting this parameter, in
overall lifespan of the battery, including states estimation
(Chang, 2013) Wen-Yeau Chang presented a
algorithms. An accurate estimation of the battery State
classification of the different mathematical methods that
Of Health (SOH) and State Of Charge (SOC) is a crucial
were used in literature to predict the battery SOC.
task that an advanced battery management system should
However, this variable only is not enough for proper
perform.
utilization of the battery, since the battery is subject to
This paper aims to outline the most relevant battery different ageing mechanisms, it’s important to track of
model types that were used in literature for Electric the battery health. The battery ageing results in an
Vehicle (EV) applications. An overview of the increment in internal resistance and a decrease in
estimation algorithms that estimate the battery state of capacity, which affects the performance and the ability to
charge and state of health are presented and simulations provide the same energy decreases. Therefore,
of some methods are also illustrated in order to test their estimation of the battery SOH can be achieved by
accuracy. tracking the change of one of these two metrics. This
battery state was the subject of research of different
authors who used different estimation algorithms with
Keywords: Battery management system, State Of
different battery models to provide an accurate
Health, State Of Charge, Lithium-ion.
estimation. (Ungurean, Cârstoiu, & Groza, 2016),
presented a detailed review of the most relevant models,
1. INTRODUCTION
algorithms and commercial devices that were used in
The global need for a clean and renewable energy
literature to estimate the battery Remaining Useful Life
sources that can replace fossil energy is essential to
(RUL) and SOH.
create a more sustainable planet. This global shift favours
the use of electric vehicles as a safe and clean alternative This paper outlines the most relevant methods that were
of fossil fuels in transportation to help reduce air used in literature to estimate the battery SOC and SOH.
pollution. Due to their high power and energy density, First, we present the battery model categories that were
Li-ion batteries are widely used to power electric cars. used for EV applications. Then an overview of battery
However, to ensure a long-life cycle and avoid any risk state of health and state of charge estimation algorithms,
of explosion, a Battery Management System (BMS) is in particular, the coulomb counting method, internal
needed to boost the efficiency and guarantee a safe usage resistance method, voltage-based method, Kalman
of the battery. filtering based methods, sliding mode observer, fuzzy
logic and least squares-based method are presented. an
A smart battery management system uses the required
equivalent circuit model with 2 RC networks is used to
data to estimate the battery states, whether it's the battery
test the accuracy of some methods to determine the
state of charge (SOC), state of health (SOH), or any other
battery SOC and internal resistance.
battery state that can help improve the performance.
The ability to predict the instantaneous battery state and
2. BATTERY MODELING FOR EVS
conditions is a crucial task a BMS should perform with
Building a proper battery model that suits better the
accuracy. The SOC indicates the battery available
target application is a crucial task during BMS design; it
capacity to help avoid overcharging/discharging the
helps capture the battery electric and thermal behaviors
under different operating conditions, to ensure safe and each electrode of the battery by a single spherical particle
fast charging for optimal utilization and secure ignoring thermal effects, approximating spatial and time
discharging of the battery (Kailong, Kang, Qiao, & characteristics at the separator region to 0 and assuming
Cheng, 2019). all unknown states to be scalar and uniform Figure 1.
SPM are simple and they can be adopted for real time
Literature has presented numerous battery models with
applications, however, they lack accuracy at high C-rate.
different complexity and fidelity scales. For EV
applications the battery model needs to be simple, To overcome this limitation an extended version of these
computationally efficient and suitable for high discharge model that incorporates the electrolyte dynamics has
rates. Therefore, two main groups of EV battery models been developed and they were proved to maintain a high
were presented: Equivalent Circuit Models (ECM) and accuracy prediction even at high C-rates conditions
Reduced Order Models (ROM) (A., J., K., & Longo S. (Moura, Argomedo, Klein, & Krstic, 2016).
and Wild, 2016).

2.1. Equivalent Circuit battery Model

Equivalent circuit models on the other hand are widely
used when developing a BMS for EV application thanks
to their speed, simplicity and reasonable accuracy
(Rincon-Mora, 2006).
The model uses a combination of resistors, capacitors
and voltage sources, to describe the battery behavior
under load. Starting by the most basic model that
represents the battery as an ideal voltage source mounted
on series with internal resistance to describe the voltage
polarization, researchers have include different electrical
components, in the interest of building improved
versions which take into account the dependence of the
battery cell on state of charge, temperature, and other
characteristics (Kempera, Li, & Kum, 2015). Figure 1: P2D and single particle battery model
Figure 2 illustrates a typical ECM that was widely used
in literature for battery states estimation. The resistor-
capacitor networks are used to describe the battery Another version that aims to reduce the computational
charge transfer or diffusion processes (Kailong, Kang, burden of the P2D model was presented in the literature
Qiao, & Cheng, 2019). as the Simplified P2D model (SP2D), it describes the
dynamic concentration profiles derived from the P2D
model to help improve the accuracy of the BMS (G., X.,
& M., 2018).
Once the battery model has been decided, it can be used
as an input for the states estimation algorithms whether
it is the SOC, SOH or any other state.

3. BATTERY SOC AND SOH ESTIMATION

3.1. Coulomb Counting

Figure 2 : Second order equivalent circuit model Coulomb counting, also known as Ampere-hour method
is one of the most common techniques that were used to
estimate the battery states and especially the battery state
2.2. Reduced-Order battery Model of charge. As the name suggests, this method calculates
Reduced-order models are simplified version of the accumulated current that flows in or out during the
electrochemical battery model (A., B., & M., 2016) . The charge-discharge process to determine the battery state
governing Partial Differential Equations (PDE) of charge (equation 1) (Fleischer, W., Z., & D.U., 2013)
equations that describe the electrochemical reactions or state of health (equation 2) (Ungurean, Cârstoiu, &
inside the battery are approximated into low order Groza, 2016).
systems of ODEs equations using a set of Model Order
Reduction (MOR) techniques (Fan & Canova, 2017).
𝑡
1
The most basic electrochemical model is known as 𝑆𝑂𝐶(𝑡) = 𝑆𝑂𝐶(𝑡0 ) − ∫ 𝜂(𝑡)𝐼𝑏𝑎𝑡 𝑑𝑡 (1)
Single Particle Model (SPM) (Figure 1), it represents 𝑄𝑟𝑎𝑡𝑒𝑑 𝑡0
Where 𝑆𝑂𝐶(𝑡0 ) represents the initial SOC, 𝑄𝑟𝑎𝑡𝑒𝑑 is the
rated capacity and 𝐼𝑏𝑎𝑡 is the battery current.
𝑡
1 (2)
𝑆𝑂𝐻(𝑡) = ∫ 𝐼(𝑡)𝑑𝑡
𝑄𝑟𝑎𝑡𝑒𝑑 𝑡0

Where I is the discharge current.

Coulomb counting method requires an accurate
estimation of the initial SOC and a precise measurement
of the battery current to be able to estimate the battery
states as correctly as possible. However, in reality, the
measured current includes sensor noise, and it doesn't
take into consideration the self-discharge current and
current losses during charging and discharging, which
makes the measured current different from the true cell
current.
To overcome those limitations, several researchers have
Figure 3 : Internal resistance estimation using DRE
proposed modified versions of the coulomb counting
approach
method. For example, in (Berecibar, et al., 2016) The
initial value of the state charge was first estimated using
a SOC-OCV mapping function, and then a periodic re-
Figure 3 shows the implementation of this approach
calibration of the capacity was performed. The measured
using the same battery model as 3. The estimated results
results have shown a reliable estimation.
slightly converge to the true value of the battery internal
resistance. The implementation of this method requires a
3.2. Direct Resistance Estimation Algorithm smaller memory space since no training data or initial
The life evolution of Li-ion battery cells is affected by battery characterization are needed. Also, it does not
different degradation mechanisms that can be require complex matrix calculation, which makes it
represented by two measurable quantities: Capacity loss suitable for an embedded system (Mathew, Janhunen,
and increment of internal resistance. Consequently, we Rashid, Long, & Fowler, 2018).
can estimate the battery state of health by observing the
changes of these two parameters.
3.3. Open Circuit Voltage (OCV)
By observing the step change in the voltage curve during Another simplistic approach to estimate the battery states
the discharge process, the battery internal resistance can is by measuring the open circuit voltage (OCV) of the
be approximated using the following equation (Yu, et al., cell. Literature has proven a strong dependence between
2017): OCV and SOC of the battery cell. This voltage-based
method gives the equivalent SOC value of the given
voltage value using the OCV vs. SOC discharge curve of
Δ𝑈 (3)
𝑅= the battery.
Δ𝐼
Based on a simplified electrical model we can define the
Where Δ𝐼 represents the current step-change and Δ𝑈 is
battery OCV as follow:
the corresponding voltage to the same step-change.
The estimation of the battery internal resistance using the
𝑈𝑂𝐶𝑉 (𝑆𝑂𝐶(𝑡)) = 𝑈(𝑡) − 𝑅0 𝐼(𝑡) (5)
DRE algorithm gives noisy results due to noise in the
measurement reading. Therefore, Lievre et al. (2016)
have used an Exponentially Weighted Moving Average Where 𝑈𝑂𝐶𝑉 is the battery OCV, 𝑈(𝑡) is the battery
(EWMA)(equation 4) filter to reduce the noise while terminal voltage, 𝑅0 is the internal resistance and I(t) is
maintaining the efficiency of the algorithm. the battery current.
Using the same equivalent circuit model previously
𝐸𝑊𝑀𝐴𝑘 = 𝜆 × 𝑅𝑠 + (1 − 𝜆) × 𝐸𝑊𝑀𝐴𝑘−1 (4) discussed, we can implement this method in MATLAB.
The estimated SOC was filtered using the EWMA
Where, 𝐸𝑊𝑀𝐴𝑘 represents the current filtered resistance approach to reduce noises (see Figure 4).
estimate, 𝜆 is a tunable constant that represents the depth
of memory, 𝐸𝑊𝑀𝐴𝑘−1 is the previous value of the
filtered resistance estimate and 𝑅𝑠 is the raw ohmic
resistance.
 𝐾𝑡 = 𝑃(𝑡|𝑡−1) 𝐻𝑡𝑇 (𝐻𝑡 𝑃(𝑡|𝑡 − 1) 𝐻𝑡𝑇 + 𝑅𝑡 )−1 (9)
𝑃(𝑡|𝑡) = 𝑃(𝑡|𝑡−1) − 𝐾𝑡 𝐻𝑡 𝑃(𝑡|𝑡−1) (10)

Where 𝑥̂ is the estimated state, F is the state transition

matrix, B is the control matrix, u is the input vector, P
and Q are respectively the state and the process variance
matrix, y is the output measurement, K is the Kalman
gain, H is the measurement matrix and R is the
measurement variance matrix.
Since the Kalman filter is limited for linear systems and
has proven a reliable estimation of the states of a process,
researchers have developed different extensions of this
model-based algorithm to adapt it for nonlinear systems.
The extended Kalman Filter is one of the enhanced
versions that deals with non-linear systems such as
battery state estimation. This model-based approach was
Figure 4 : SOC estimation using OCV widely used to estimate the battery SOC. The process of
estimation starts by choosing the battery model that
Figure 4 illustrates the estimate and the true SOC curve describes its response in discrete-time, then the algorithm
of the Li-ion battery cell, and by observing these curves is applied to estimate the state. The state description of
we notice that the predicted values are noisy and that can the battery model has to include the wanted unknown
be explained by the fact that this voltage-based method quantities that should be determined. For this paper, a
doesn't take into consideration the diffusion voltage and second-order ECM (Figure 2) will be used to test the
the hysteresis. performance of this approach to estimate the battery state
of charge.
OCV has been used also to estimate the battery SOH. By
knowing the SOC/OCV relation the estimation of the The authors of (E. Kamal, 2015) and (Sepasi, Ghorbani,
internal resistance can be easily conducted. (Mathew, & Liaw, 2014) used the same battery model to estimate
Janhunen, Rashid, Long, & Fowler, 2018) used a the battery SOC. The nonlinear system is first
combination of OCV and Coulomb counting approach to represented using equations 11 and 12, to represent
estimate the battery SOH. However, others considered respectively the system dynamics expressed in state
this voltage-based method unsuitable for estimating equations and the output equation of the system (O., P.,
SOH, (A., et al., 2016) present some disadvantages of S., & Molinas, 2017).
using this method and tried to eliminate the OCV from
the equations to simplify the computation of the battery
𝑥𝑘+1 = 𝑓(𝑥𝑘 , 𝑢𝑘 ) + 𝑤𝑘 (11)
SOH.

3.4. Kalman filtering 𝑦𝑘+1 = 𝑔(𝑥𝑘 , 𝑢𝑘 ) + 𝑣𝑘 (12)

Kalman filter is a model-based algorithm that uses the
mathematical representation of a linear system to
determine its states. The literature defines this approach Where 𝑓(𝑥𝑘 , 𝑢𝑘 ) represents the nonlinear transition
as a sturdy algorithm that operates in two fundamental function, 𝑔(𝑥𝑘 , 𝑢𝑘 ) represents the nonlinear
steps (Ungurean, Cârstoiu, & Groza, 2016): measurement function, 𝑤𝑘 and 𝑣𝑘 denote respectively
the process and the measurement noise.
• Prediction phase: the system state is estimated
using the following equations: To apply the EKF 𝑓(𝑥𝑘 , 𝑢𝑘 ) and 𝑔(𝑥𝑘 , 𝑢𝑘 ) are linearized
at each time step using the first order of Taylor-series. 13
and equation 14 can be rewritten as follows (E. Kamal,
2015):
𝑥̂(𝑡|𝑡−1) = 𝐹𝑡 𝑥̂(𝑡−1|𝑡−1) + 𝐵𝑡 𝑢𝑡 (6)

𝑃(𝑡|𝑡−1) = 𝐹𝑡 𝑃(𝑡−1|𝑡−1) + 𝑄𝑡 (7) 𝑥𝑘+1 = 𝐴𝑘 𝑥𝑘 + 𝐵𝑘 𝐼𝐿,𝑘 + 𝑤𝑘 (13)

𝑦𝑘 = 𝐶𝑘 𝑥𝑘 + 𝐷𝑘 𝐼𝐿,𝑘 + 𝑣𝑘 (14)
• Update phase: the algorithm updates the
prediction based on the system errors as follow:
𝜕𝑓(𝑥𝑘 ,𝑢𝑘 ) 𝜕𝑓(𝑥𝑘 ,𝑢𝑘 )
Where 𝐴𝑘 = | , 𝐵𝑘 = | ,
𝜕𝑥𝑘 𝑥𝑘 ,𝑢𝑘 𝜕𝑢𝑘 𝑥𝑘 ,𝑢𝑘
𝜕𝑔(𝑥𝑘 ,𝑢𝑘 ) 𝜕𝑔(𝑥𝑘 ,𝑢𝑘 )
𝑥̂(𝑡|𝑡) = 𝑥̂(𝑡|𝑡−1) + 𝐾𝑡 (𝑦𝑡 − 𝐻𝑥̂(𝑡|𝑡−1) ) (8) 𝐶𝑘 = | and 𝐷𝑘 = | .
𝜕𝑥𝑘 𝑥𝑘 ,𝑢𝑘 𝜕𝑢𝑘 𝑥𝑘 ,𝑢𝑘
 3.6. Fuzzy Logic
Fuzzy logic is a useful mathematical concept that allows
modelling non-linear and complex systems using the
appropriate training dataset. It's a non-monotonic logic
that uses crisp sets to categorize the measured data. The
relationship between a member of a set and its degree of
membership is defined using a membership function. In
the case of SOH, the membership function outputs can
be set as healthy, tolerable and not healthy.
Jonghoon Kim ( 2014), used a fuzzy logic-controlled
methodology to predict the battery SOH, first the cell
resistance and maximum capacity were determined
based on voltage, current, temperature and time, then
fuzzy logic approach was applied to estimate the battery
SOH using the resistance and the maximum capacity
Figure 5 : SOC estimation using EKF values.
Burgos et al. proposed a novel fuzzy logic algorithm to
EKF was proven to give accurate results even with noisy predict the battery SOC. A fuzzy model that
input data, the algorithm is much lighter and can be characterizes the relationship between the battery open-
implemented for real-time application. However, when circuit voltage, SOC and the discharge current was used
the system is highly non-linear, linearization error would in combination with an EKF to predict the battery SOC
lead to highly unstable filters because of uncertainties in (Burgos, Saez, Orchard, & Cardenas, 2015).
the first order Taylor series (Sun, Hu, Zou, & Li, 2011).
3.7. Least squares
Many authors have used dual extended Kalman filter to The least square algorithm is a widely used approach that
provide an online estimation of both battery SOH and identifies the best fit function that minimizes the sum of
SOC. As the name suggested, this approach uses two quadratic errors between measured output and system
extended Kalman filters to predict the battery states and response.
update its parameters to give more reliable results. In In (L, 2011), Gregory L.Pett proposed an enhanced
(Wassiliadisa, et al., 2018) the authors provide an version of the LS algorithm called Weighted Total Least
outstanding investigation of the DEKF performance for square to estimate the battery capacity. Since the
SOC and SOH estimation under different dynamics and standard least square approach doesn’t consider the
degradation stages. Compared to a simple EKF, the uncertainties that the input measurement includes, the
DEKF was proven to improve the accuracy of the SOC author used the WTLS algorithm that takes into account
over battery lifetime, while the battery internal resistance the noises of the accumulated ampere-hour
and capacity become inaccurate with advanced measurements and the battery SOC. The estimation
degradation. results are more accurate than a standard least squares
approach, and the algorithm can be used for real-time
3.5. Sliding Mode Observer applications.
Another model-based approach that was used to estimate
the internal states of the battery is known as the sliding- 4. CONCLUSION
mode observer (SMO). This algorithm has the advantage SOC and SOH estimation is of a great importance when
of compensating the modelling errors caused by developing a battery management system, they provide
parameters variation of circuit model and can help an overview of the short- and long-term state of the
overcome some drawbacks that other model-based battery.
methods present. Introduced by Emel'yanov (1959) the The main goal of this paper is to provide a basic
SMO algorithm was adopted to estimate the battery SOC. understanding of the different algorithms, the advantages
A combination of an improved ECM and a SMO were and shortcomings of each to help build an advanced BMS
used by different authors to provide an accurate for EV application. The review shows that there is no
estimation of the battery SOC (Chen, X., Z., & A., 2012) perfect approach to estimate the battery states, and the
(Nacer, Ahmed, & Naamane, 2012). choice should be made based on the complexity of the
In (Ning, Xu, Cao, Wang, & Xu, 2016 ), the authors used system and the target application.
SMO to estimate the battery SOC based on a parameter The battery model types that are used for EV applications
adaptive battery model to reduce the systematic errors. are first presented. Then the most relevant estimation
The estimation result shows a rapid convergence of SOC algorithms that were used in literature to predict the
curve with an estimation error of less than 2%, which battery SOC and SOH are outlined. Certain algorithms
reflects the robustness of the algorithm. were tested using a second order equivalent circuit
battery model to test out their accuracy.
 Kim, J. ( 2014). Fuzzy Logic-Controlled Online State-of-
5. REFERENCES Health (SOH) Prediction in Large Format
A., F., J., A. D., K., P., & Longo S. and Wild, M. (2016). LiMn2O4 Cell for Energy Storage System
A review on electric vehicle battery modelling: (ESS) Applications.
From Lithium-ion toward Lithium–Sulphur. L, P. G. (2011). Recursive approximate weighted total
Renewable and Sustainable Energy Reviews, least squares estimation of battery cell total
56, 1008-1021. capacity. Journal of Power Sources, 2319–
A., J., B., R., & M., D. M. (2016). Review of simplified 2331.
Pseudo-two-Dimensional models of lithium-ion Mathew, M., Janhunen, S., Rashid, M., Long, F., &
batteries. Journal of Power Sources, 327, 44– Fowler, M. (2018). Comparative Analysis of
55. Lithium-Ion Battery Resistance Estimation
A., L., A., S., Venet, P., A., H., M., O.-B., & S., P. (2016). Techniques for Battery Management Systems.
Practical online estimation of lithium-ion Energies, 11, 1490.
battery apparent series resistance for mild Moura, S. J., Argomedo, F. B., Klein, R., & Krstic, A. M.
hybrid vehicles. IEEE Transactions on (2016). Battery State Estimation for a Single
Vehicular Technology, 65, 4505–4511. Particle Model With Electrolyte Dynamics.
Berecibar, M., Gandiaga, I., Villarreal, I., Omar, N., IEEE Transactions on Control Systems
Mierlo, J. V., & Bossche., P. V. (2016). Critical Technology, 25(2), 453–468.
review of state of health estimation methods of Nacer, M., Ahmed, B., & Naamane, A. (2012). Battery
Li-ion batteries for real applications. Renewable Models for Estimation of State of Charge by
and Sustainable Energy Reviews, 56 , 572-587. Sliding Mode Observer. Journal of Smart
Burgos, C., Saez, D., Orchard, M. E., & Cardenas, R. Innovation, Systems and Technologies, 133-
(2015). Fuzzy modelling for the state-of-charge 149.
estimation of lead-acid batteries. Journal of Ning, B., Xu, J., Cao, B., Wang, B., & Xu, G. ( 2016 ).
Power Sources , 355-366. A Sliding Mode Observer SOC Estimation
Chang, W.-Y. (2013). The State of Charge Estimating Method Based on Parameter Adaptive Battery
Methods for Battery: A Review. International Model. Energy Procedia, 619 – 626.
Scholarly Research Notices, 7. O., G., P., V., S., J., & Molinas, M. ( 2017). Battery
Chen, X., X., S. W., Z., C., & A., K. ( 2012). Sliding modeling and Kalman filter-based State-of-
Mode Observer for State of Charge Estimation Charge estimation for a race car application.
Based on Battery Equivalent Circuit in Electric IEEE 14th International Conference on
Vehicles. Australian Journal of Electrical and Networking, Sensing and Control (ICNSC). .
Electronics Engineering, 225–234. Rincon-Mora, E. C. (2006). Accurate electrical battery
E. Kamal, A. E. (2015). State of charge estimation based model capable of predicting runtime and I-V
on extended Kalman filter algorithm for performance. IEEE Transactions on Energy
Lithium-Ion battery. 23rd Mediterranean Conversion, 21(2), 504-511.
Conference on Control and Automation (MED). Sepasi, S., Ghorbani, R., & Liaw, B. Y. (2014). Improved
Fan, G., & Canova, X. L. (2017). A Reduced-Order extended Kalman filterfor state of charge
Electrochemical Model of Li-Ion Batteries for estimation of battery pack. Journal of Power
Control and Estimation Applications. IEEE Sources , 368-376.
Transactions on Vehicular Technology, 67(1), Sun, F., Hu, X., Zou, Y., & Li, S. (2011). Adaptive
76-91. unscented Kalman filtering for state of charge
Fleischer, C., W., W., Z., B., & D.U., S. (2013). Adaptive estimation of a lithium-ion battery for electric
on-line state of available power prediction of vehicles. Energy, 36, 3531-3540.
lithium-ion batteries. Journal Power Electron, Ungurean, L., Cârstoiu, G., & Groza, M. V. (2016).
13, 516–527. Battery state of health estimation: a structured
G., F., X., L., & M., C. (2018). A Reduced-Order review of models, methods and commercial
Electrochemical Model of Li-Ion Batteries for devices. International Journal of Energy
Control and Estimation Applications. IEEE Research, 41(2), 151-181.
Transactions on Vehicular Technology, 67(1), Wassiliadisa, N., Adermanna, J., Frericksb, A., Pakb, M.,
76–91. Reitera, C., Lohmannb, B., & Lienkampa., M.
Kailong, L., Kang, L., Qiao, P., & Cheng, Z. (2019). A (2018). Revisiting the dual extended Kalman
brief review on key technologies in the battery filter for battery state-of-charge and state-of-
management system of electric vehicles. health estimation: A use-case life cycle
Frontiers of Mechanical Engineering, 14(1), analysis. Journal of Energy Storage, 73-87.
47-64. Yu, J., Mo, B., Tang, D., Yang, J., Wan, J., & Liu, J.
Kempera, P., Li, S. E., & Kum, D. (2015). Simplification (2017). Indirect State-of-Health Estimation for
of pseudo two dimensional battery model using Lithium-Ion Batteries under Randomized Use.
dynamic profile of lithium concentration. Energies.
Journal of Power Sources, 286, 510-525.