2023 9th International Conference on Control, Decision and Information Technologies

CoDIT 2023 | Rome, Italy / July 03-06, 2023 Technically co-sponsored by IEEE & IFAC

State of health prediction of lithium-ion battery

using machine learning algorithms
Jamila Hemdani1,2, Laid Degaa2, Nassim Rizoug2, and Abdelkader Chaari1
1 Department of Electrical Engineering, The National Higher Engineering School of Tunis (ENSIT),
2023 9th International Conference on Control, Decision and Information Technologies (CoDIT) | 979-8-3503-1140-2/23/$31.00 ©2023 IEEE | DOI: 10.1109/CODIT58514.2023.10284459

University of Tunis, Laboratoire d’Ingenierie des Systèmes Industriels et des Energies Renouvelables, BP 56-1008, Tunis, Tunisia.
2 Higher School of Aeronautical Techniques and Automobile Construction (ESTACA),
Laboratoire des Systèmes et Energies Embarqués pour les Transports, Parc universitaire Laval-Changé, Rue Georges Charpak, 53000 Laval.

Abstract—In the last years have seen an increasing usage of Machine learning methods, on the other hand, are based
Electrical Vehicle (EV). To guarantee safe and reliable operation, on the analysis of battery measurement data, such as voltage,
it’s necessary to possess the capability to monitor, in real time, the current, and temperature, to predict the health status of the
state of health (SOH) of the battery . This paper presents a deep
learning method which utilizes a Deep Neural network (DNN) for battery. These methods are often easier to implement than
cell-level capacity estimation based on the voltage, current, and physical models and can provide more accurate forecasts
State Of Charge. First, a multi-physical models of the battery is using algorithms such as neural networks, decision trees, and
done to extract input and output data for the different learning regression methods. Studies [35]- [39] have demonstrated that
and testing phases. Second, two machine learning algorithms, neural networks remain a very active research area, offer-
including DNN and Convolution Neural Network (CNN), are used
to predict SOH. Mean Absolute Error (MAE) and Mean Square ing great potential for estimating the state of health (SOH)
Error (MSE) are selected as the evaluation index. The results of batteries in complex aging conditions, especially when
show that the proposed algorithm DNN has the weakest error, data is abundant. Machine learning-based SOH estimation
which makes it possible to accurately predict the SOH and to approaches are therefore promising research areas that will
have a better stability. have a significant impact on the future of electrification of
Index Terms—Deep learning, Lithium-ion battery, DNN, CNN,
transportation. According to the literature, feed-forward neu-
state-of-health, Multi-physics battery model.
ral networks (FNNs), convolutional neural networks (CNNs),
and long short-term memory (LSTM) networks are the most
I. I NTRODUCTION performing. A comparative study conducted by Sungwoo Jo
et al. [40] shows that LSTMs have the best performance
Given the rising prices of petroleum and environmental compared to the other two types. However, their use can
pollution, car manufacturers have chosen to develop clean ve- be costly in terms of computation and memory due to the
hicles with very low CO2 emissions we use renewable energies complexity of their structure and memory cell. Indeed, the
[1]–[3]. Indeed, electric vehicles (EVs) have been considered embeddability constraints for the real-time deployment of an
as a favorable alternative to conventional fuel vehicles. By artificial intelligence approach for predicting the health status
2030, the global VEs market is expected to grow to 39,208,000 of a battery depend on several factors, such as the model
units, from an estimated in 2022, almost 8,151 thousand units, size, the complexity of the algorithms used, the processing and
at a compounded annual growth rate of 21.7% [4] [5]. To storage capacity of the embedded hardware, as well as latency
make the Electric market grow, a continued support from the and accuracy requirements. That is why we used CNN and
government will be needed along with the growing research DNN models to minimize computation costs and maximize
of EV batteries, charging systems and design to reduce the reliability.
initial cost of electric vehicles. Additionally, increasing vehicle This paper is constructed with the following sections: Sec-
range, reducing charging time and adding technology features tion I, we provide a brief description of the battery modeling
to the EV’s would act as other major factors for faster EV to extract a database for use it later in the different learning
market growth [6]–[8]. The largely retained technology is the and testing phases. In Section II, we present the proposed
lithium-ion accumulator that offers satisfactory characteristics, methodology for battery capacity prognosis and describe the
such as superior energy and power density, stability, low machine learning process and the different algorithms using
discharge rates when compared to other rechargeable batteries, the multi-physical model. In section III, we compare the
lower weight, low self-discharge rates, and excellent cycle capacity prediction results with the DNN and CNN algorithms.
performance [9] [10]. Indeed, the prognostics and health The conclusions are presented in section IV.
management (PHM) for lithium-ion battery are of great impor-
tance for properly scheduling maintenance and offering failure II. BATTERY DATASET
alarm. To improve performance and assuring the safe operation The goal of battery modelling is to predict the behaviour
of the battery energy storage system the accurate state-of- of the studied system [20]–[22]. The first part of the multi-
health (SOH) prediction of lithium-ion batteries is necessary physical model is the one that represents the electrical be-
[11]–[15]. haviour of the battery, and that allows to estimate losses

979-8-3503-1140-2/23/$31.00 ©2023 IEEE -2729-

Authorized licensed use limited to: ULAKBIM UASL - Okan Universitesi. Downloaded on January 09,2024 at 14:32:44 UTC from IEEE Xplore. Restrictions apply.
 CoDIT 2023 | Rome, Italy / July 03-06, 2023 Technical Co-Sponsors: IEEE CSS, IEEE SMC, IEEE RAS & IFAC.

within the battery. Then, the thermal model to to determine

the evolution of the temperature of the cells according to the
ambient temperature and the losses identified in the electrical
model. Finally, the aging model which consists of two parts:
a calendar aging model and a cycling aging model. The two
aging natures add up to form the global aging model. The
different models will finally be combined to represent the
dynamic behaviour of the battery. The developed model is used
to create a database for the learning and testing phase. The

Fig. 3. Current variation as a function of capacity variation and SOC

III. M ACHINE L EARNING P ROCESS

The battery is the main component in electric vehicles
so it is important to estimate their State Of Health (SOH)
to ensure their efficiency and safety. The performance of
lithium-ion batteries, as a source of energy, are affected by
the environment, for instance, the ambient temperature, age,
Fig. 1. Voltage variation as a function of capacity variation and SOC charge/discharge cycles and its chemical, resulting in complex
calculations of their SOH. As an inner state quantity that can-
not be directly measured, SOH prediction methods based on
main objective of creating a database is to predict the evolution
artificial intelligence (AI) have been widely used. Owing to its
of aging according to the history of use of the components.
easy application and independence on the detailed knowledge
Several tests are applied for different values of current, State
in terms of degradation mechanism, machine learning (ML)-
Of Charge (SOC) and initial capacity. At each test, we vary a
based methods are widely investigated for SOH estimation
single parameter to build a very rich database. Current varies
[23]–[27].
from 3C down to 0.5C, initial SOC from 80% up to 70% and
In this section, we present the proposed methodology and
initial capacity from 100%C up to 60%C. After extracting
data collected from multi-physics model for Li-Ion battery
the data from the multi-physics model, the final step is to
capacity prognosis and a comparative study is conducted of
accurately estimate the state of health of the battery in order
artificial learning algorithms. Figure4 shows the extraction
to predict its state of degradation. The figures 1,2 and 3 present
of data and the use of voltage, current, SOC and capacity
the results of inputs outputs obtained from the Multi-physics
values as input data in the learning phase to learn input/output
battery model.
and then accurately predict battery capacity. There are several
methods for predicting SOH in the literature, among which we
find the artificial neural network (ANN), convolutional neural
network (CNN) [28] [29], short-term memory (LSTM) [30]
[31] and deep neural network (DNN) [32] [33], describe the
behavior of a battery without prior knowledge of the internal
structure of the battery.

The following are the characteristics of each deep learning

algorithm. The Deep Neural Network (DNN) is generally a
stack of multiple hidden layers instead of only one hidden
layer in the standard ANN architecture. The DNN hidden
layers are the multiple feed-forward layers that are trained with
a back-propagation stochastic gradient descent. The hidden
layers consist of neurons nodes with tanh, rectifier (ReLU),
and maxout activation functions. The CNN is better handling
Fig. 2. Variation of SOC as a function of variation of capacity
multidimensional data, as it has adopted types of convolutional
techniques. Additionally, as aforementioned, the layers of

-2730-
Authorized licensed use limited to: ULAKBIM UASL - Okan Universitesi. Downloaded on January 09,2024 at 14:32:44 UTC from IEEE Xplore. Restrictions apply.
 CoDIT 2023 | Rome, Italy / July 03-06, 2023 Technical Co-Sponsors: IEEE CSS, IEEE SMC, IEEE RAS & IFAC.

B. Performance Evaluation
The loss functions to evaluate the error was expressed as
the mean absolute error and mean squared error as follows:
n
1 X
M AE = . |ybi − yi | (1)
n i=1
n
1X
M SE = (yi − yˆi )2 (2)
n i=1

Fig. 4. Database creation procedure

where yi is the true value of capacity, ŷi is the estimated
capacity, and n is the number of datasets.

IV. E STIMATION R ESULTS AND D ISCUSSION

CNN and RNN are far more complex than those of DNN. To validate the proposed method, the capacity estimation
Therefore, CNN and RNN take more time for training the results were compared in this Section. To compare and deduce
model, which is their major drawback. These reasons make which one has better performance between the two algorithms.
DNN more suitable for employment in real applications for Figures 6,7 shows capacity estimation results using the CNN
the most case. and DNN algorithms.

A. Model Training
As shown in Figure 5, the proposed process for battery’s
SOH estimation consisted of three stages: Creation of a
database, training model, and a testing model. The feature
data were split into a training dataset and test dataset. We
trained DNNs and CNN using the training data. In the testing
stage, the trained model was tested using the test dataset, and
its performance was evaluated using the Mean Absolute Error
(MAE) and Mean Squared Error (MSE).

Fig. 6. Evolution of real capacity and estimated by the CNN algorithm

Fig. 5. Flowchart of the proposed methodology

1) Convolutional Neural Network: CNN is a feature extrac-

tor that is widely used in in different fields. It usually consists
of convolutional layers,pooling layers, and fully connected lay-
ers. The con-volutional layer performs convolution processing
on the original data through the slip of the convolution, which
can enhance some of its features and reduce noise [34].
2) Deep Neural Network: The DNN, which is essential for
machine learning to form the basis of many architectures,
consists of multiple layers of perceptrons with a nonlinear
activation function. We employed the structure of the DNN,
which had a five-layer structure, consisting of an input, three
hidden, and output layers. The hiddens layers contained 32
hidden neurons, and the ReLU function was used as an Fig. 7. Evolution of real capacity and estimated by the DNN algorithm
activation [19].

-2731-
Authorized licensed use limited to: ULAKBIM UASL - Okan Universitesi. Downloaded on January 09,2024 at 14:32:44 UTC from IEEE Xplore. Restrictions apply.
 CoDIT 2023 | Rome, Italy / July 03-06, 2023 Technical Co-Sponsors: IEEE CSS, IEEE SMC, IEEE RAS & IFAC.

V. C ONCLUSION
This article presented a methodology for estimating and
prognosis of Li-ion battery capacity of electric vehicles. The
framework includes the use of DNN and CNN algorithms to
predict the future of battery behavior based on the use the
multi-physics model to extract the input and output of the
model. The performance of the proposed approach is evaluated
through a comparative study on data collected from the multi-
physics model of battery. The results obtained show a strong
resemblance between the actual values and those predicted by
the proposed methodology for the different cycles. Therefore,
the proposed method follows the evolution of the actual
battery behavior with satisfactory accuracy. However, we have
challenges to improve the method developed in this work for
a more powerful prediction of battery capacity taking into
Fig. 8. Evolution of the loss function by the CNN algorithm
consideration the effect of temperature.

R EFERENCES
[1] W.Yosafat, P.Ofita, S.Wiwik and T.Bambang, “Zero emission and clean
energy concept for campus area in hot-humid tropical climate,” ArchNet-
The SOH estimation results of the CNN showed a trend IJAR, 2022.
[2] R.Zahedi, M.Ghodusinejad, A.Aslani and C.Hachem-Vermette, “Mod-
elling community-scale renewable energy and electric vehicle manage-
ment for cold-climate regions using machine learning,” Energy Strategy
Reviews, 2022.
[3] Y.Ahoutou, A.Ilinca and M.Issa “Electrochemical Cells and Storage
Technologies to Increase Renewable Energy Share in Cold Climate
Conditions— A Critical Assessment,” Energies 2022.
[4] ”Electric Vehicle Market by Type, Vehicle Type, Vehicle Class, Top
Speed and Vehicle Drive Type: Global Opportunity Analysis and Indus-
try Forecast, 2021-2030,” ID: 5578613, 501 Pages, January 2022.
[5] ”Electric Vehicle Market by Component, Vehicle Type, Vehicle Class,
Propulsion (BEV, PHEV, FCEV), Vehicle Drive Type (FWD, RWD,
AWD), Vehicle Top Speed (125 mph), Charging Point Type, Vehicle
Connectivity, End Use, Region-Global Forecast 2030,” ID: 5578613,
326 Pages, April 2022.
[6] A.Chakrabarti, N.Marati, A.K.Bhoi, V.H.C.De Albuquerque and
A.Kalam, “AI Enabled IoT for Electrification and Connected Transporta-
tion,” Springer, 2022.
[7] F.Heinrich and M.Pruckner, “Virtual experiments for battery state of
health estimation based on neural networks and in-vehicle data,” Journal
of Energy Storage, April 2022.
[8] W.Cieslik, F.Szwajca, W.Golimowski and A.Berger, “Experimental
Fig. 9. Evolution of the loss function by the DNN algorithm
Analysis of Residential Photovoltaic (PV) and Electric Vehicle (EV)
Systems in Terms of Annual Energy Utilization,” Energies, 19 February
2021.
[9] M.Al-Saadi, J.Olmos, A.Saez-de-Ibarra, J.V.Mierlo and M.Berecibar,
similar to those of the DNN, as shown in Figures 6,7. To “Fast Charging Impact on the Lithium-Ion Batteries’ Lifetime and Cost-
make a more straightforward comparison, we calculated the Effective Battery Sizing in Heavy-Duty Electric Vehicles Applications,”
MSE and MAE of the SOH estimation, as shown in TableI. Energies, 2022.
[10] J.A.Sanguesa, V.Torres-Sanz, P.Garrido, F.J.Martinez and Johann
M.Marquez-Barja, “A Review on Electric Vehicles: Technologies and
TABLE I Challenges,” Smart Cities, 2021.
P ERFORMANCE INDICES OF THE DNN AND CNN ALGORITHMS
[11] Y.Zhang and Yan-Fu.Li, “Prognostics and health management of
Lithium-ion battery using deep learning methods: A review,” Renewable
- MAE MSE and Sustainable Energy Reviews, June 2022.
DNN 0.2253 0.0875 [12] J.Hemdani, M.Soltani, A.J.Telmoudi and A.Chaari, “Prediction of aging
CNN 0.7220 0.5860 of battery for electric vehicles based on a modified version of neural
networks,” 29th Mediterranean Conference on Control and Automation
(MED), IEE, 2021.
[13] I.Remadna, S.L.Terrissa, , M.Sayah, S.Ayad and N.Zerhouni, “Boosting
The DNN model had the lowest MAE of 0.2253 and MSE RUL Prediction Using a Hybrid Deep CNN-BLSTM Architecture,”
of 0.0875 compared with CNN model.Indeed, CNN take more Automatic Control and Computer Sciences, 2022.
time for training the model, which is their major drawback. [14] J.Hemdani, M.Soltani, A.J.Telmoudi and A.Chaari, “Prediction of aging
of battery for electric vehicles based on a modified version of neural
These reasons make DNN more suitable for employment in networks,” 29th Mediterranean Conference on Control and Automation
real applications for the most case. (MED), PUGLIA, Italy, 15 July 2021.

-2732-
Authorized licensed use limited to: ULAKBIM UASL - Okan Universitesi. Downloaded on January 09,2024 at 14:32:44 UTC from IEEE Xplore. Restrictions apply.
 CoDIT 2023 | Rome, Italy / July 03-06, 2023 Technical Co-Sponsors: IEEE CSS, IEEE SMC, IEEE RAS & IFAC.

[15] J.Hemdani, L.Degaa, M.Soltani, N.Rizoug, A.J.Telmoudi and A.Chaari, [38] J.Qu, F.Liu, Y.Ma and J.Fan. ”A Neural-Network-based Method for RUL
“Prediction of aging electric vehicle battery by multi-physics modeling Prediction and SOH Monitoring of Lithium-Ion Battery”. IEEE Access
and deep learning method,” 2022 8th International Conference on 2019.
Control, Decision and Information Technologies CoDIT’22 — Istanbul, [39] Y.Fan, H.Wu, W.Chen, Z.Jiang, X.Huang and S.Chen. ”A Data Aug-
Turkey, May 2022. mentation Method to Optimize Neural Networks for Predicting SOH of
[16] P.Khumprom and N.Yodo, “A Data-Driven Predictive Prognostic Model Lithium Batteries”. In Proceedings of the InternationalConference on
for Lithium-ion Batteries based on a Deep Learning Algorithm,” ener- Robotics Automation and IntelligentControl (ICRAIC 2021), November
gies, February 2019. 2021.
[17] S.Jo, S.Jung and T.Roh , “Battery State-of-Health Estimation Using [40] S.Jo, S.Jung and T.Roh . ”Battery State-of-Health Estimation Using
Machine Learning and Preprocessing with Relative State-of-Charge,” Machine Learning and Preprocessing with Relative State-of-Charge”.
energies, 2021. Energies 2021.
[18] S.Choi, H.Jang, H.Han, S.Park, M-I.Choi and S.Park, “Artificial
Intelligence-based Battery State-of-Health (SoH) Prediction through
battery data characteristics analysis,” Industrial and Commercial Power
Systems Technical Conference, 2022.
[19] Y.Che, Z.Deng, P.Li, X.Tang, K.Khosravinia, X.Lin and X.Hu, “State
of health prognostics for series battery packs: A universal deep learning
method,” Energy, 2022.
[20] H.Dai, X.Zhang, X.Wei, Z.Sun, J.Wang and F.Hu “Cell-BMS validation
with a hardware-in-the-loop simulation of lithium-ion battery cells for
electric vehicles,” International Journal of Electrical Power Energy
Systems, 2013.
[21] A.Hentunen, T.Lehmuspelto, J.Suomela and all, “Electrical battery
model for dynamic simulations of hybrid electric vehicles,” 2011 IEEE
Vehicle Power and Propulsion Conference, 2011.
[22] R.Xiong, H.He, H.Guoa and Y.Ding, “Modeling for Lithium-Ion Battery
used in Electric Vehicles,” Procedia Engineering, 2011.
[23] L.Yao, S.Xu, A.Tang, F.Zhou, J.Hou, Y.Xiao and Z.Fu, “A Review of
Lithium-Ion Battery State of Health Estimation and Prediction Methods,”
World Electric Vehicle Journal, 2021.
[24] Y.Zhi, H.Wang and L.Wang,“A state of health estimation method for
electric vehicle Li-ion batteries using GA-PSO-SVR,” Springer, 2022.
[25] G.Li, Q.Chu, Z.Wang and L.Li, “State of Health Prediction for Battery
Based on Ensemble Learning ,” International Conference on Electronics,
Circuits and Information Engineering (ECIE), 2021.
[26] H.Lin, L.Kang, D.Xie, J.Linghu and J.Li, “Online State-of-Health Es-
timation of Lithium-Ion Battery Based on Incremental Capacity Curve
and BP Neural Network,” Batteries, 2022.
[27] Y.Li, K.Li, X.Liu, Y.Wang and L.Zhang, “Lithium-ion battery capacity
estimation — A pruned convolutional neural network approach assisted
with transfer learning,” Applied Energy, 2021.
[28] C.Wang, R.Xiong, J.Tian, J.Lua and C.Zhang, “Rapid ultracapacitor
life prediction with a convolutional neural network,” Applied Energy,
January 2022.
[29] Y.Li, K.Li, X.Liu, Y.Wang and L.Zhang “Lithium-ion battery capacity
estimation — A pruned convolutional neural network approach assisted
with transfer learning,” Applied Energy, March 2021.
[30] P.Li, Z.Zhang, Q.Xiong, B.Ding, J.Hou and D.Luo, Y.Rong, S.Li “State-
of-health estimation and remaining useful life prediction for the lithium-
ion battery based on a variant long short term memory neural network,”
Journal of Power Sources, May 2020.
[31] S.Wang, S.Jinb, D.Bai, Y.Fan, H.Shi and C.Fernandez “A critical re-
view of improved deep learning methods for the remaining useful life
prediction of lithium-ion batteries,” Energy Reports, November 2021.
[32] J.Tian, R.Xiong, W.Shen, J.Lu and X.Yang “Deep neural network
battery charging curve prediction using 30 points collected in 10 min,”
Sciencedirect, June 2021.
[33] K.Phillip, C.Vidal, M.Naguib, and Michael Skells. “LG 18650HG2
Li-Ion Battery Data and Example Deep Neural Network XEV SOC
Estimator Script.” Mendeley, March 2020.
[34] M.Haris, G.Student, M.N.Hasan and S.Qin, “Degradation Curve Pre-
diction of Lithium-Ionn Batteries Based on Knee Point Detection
Algorithm and Convolutional Neural Network,” IEEE Transactions on
Instrumentation and Measurement, 2022.
[35] X.Sui, S.He, S.B.Vilsen, J.Meng, R.Teodorescu and D.Stroe. “A review
of non-probabilistic machine learning-based state of health estimation
techniques for Lithium-ion battery”. Appl. Energy 2021.
[36] Z.Bao, J.Jiang, C.Zhu and M.Gao. “New Hybrid Neural Network
Method for State-of-Health Estimation of Lithium-Ion Battery“.Energies
2022.
[37] J.Zhou, Z.He, M.Gao and Y.Liu. “Battery state of health estimation
using the generalized regression neural network”. In Proceedings of the
2015 8th International Congress on Image and Signal Processing (CISP),
October 2015.

-2733-
Authorized licensed use limited to: ULAKBIM UASL - Okan Universitesi. Downloaded on January 09,2024 at 14:32:44 UTC from IEEE Xplore. Restrictions apply.