State of Health Estimation of Lithium-Ion Batteries Based on Combination of Gaussian Distribution Data and Least Squares Support Vector Machines Regression

Abstract

Lithium-ion batteries play a critical role in the reliability and safety of a system. Battery health monitoring and remaining useful life (RUL) prediction are needed to prevent catastrophic failure of the battery. The aim of this research is to develop a data-driven method to monitor the batteries state of health and predict their RUL by using the battery capacity degradation data. This paper also investigated the effect of prediction starting point to the RUL prediction error. One of the data-driven method drawbacks is the need of a large amount of data to obtain accurate prediction. This paper proposed a method to generate a series of degradation data that follow the Gaussian distribution based on limited battery capacity degradation data. The prognostic model was constructed from the new data using least square support vector machine (LSSVM) regression. The remaining useful life prediction was carried out by extrapolating the model until reach the end of life threshold. The method was applied to three differences lithium-ion batteries capacity data. The results showed that the proposed method has good performance. The method can predict the lithium-ion batteries RUL with a small error, and the optimal RUL starting point was found at the point where the battery has experienced the highest capacity recovery due to the self-recharge phenomenon.

An open access tool for exploring machine learning model choice for battery life cycle prediction

“…These features can be based on electrochemical signals, such as capacity, current, or internal resistance; auxiliary signals such as temperature or impedance spectroscopy; or on metadata such as cell chemistry and form factor. While featurebased linear models achieve reasonable performance, improved performance has been shown for more complicated models, such as tree-based models, (Nuhic et al, 2013;Liu et al, 2015;Patil et al, 2015;Berecibar et al, 2016;Mansouri et al, 2017;Susilo et al, 2018;Fermín-Cueto et al, 2020;Paulson et al, 2022) Gaussian models, (He et al, 2011;Guo et al, 2015;Hu et al, 2015;Li and Xu, 2015;Richardson et al, 2017;Wang et al, 2017;Susilo et al, 2018;Aitio and Howey, 2021) or neural network models. (Liu et al, 2010;Berecibar et al, 2016;Wu et al, 2016;Mansouri et al, 2017;Zhang et al, 2017;Ren et al, 2018;Khumprom and Yodo, 2019;Venugopal and Vigneswaran, 2019;Hong et al, 2020;Ma et al, 2020;Shen et al, 2020;Li et al, 2021;Strange and dos Reis, 2021).…”

Data-driven charging strategies for grid-beneficial, customer-oriented and battery-preserving electric mobility

“…Data-driven methods, e.g. Bayesian networks and neural networks [31,32] hold feasible alternatives for SOH estimation [33]. However, the existing approaches barely use user-related data, for which reason the influence of charging behavior on battery aging is not represented properly.…”

Research on residual life prediction method of lithium ion battery for pure electric vehicle

“…Compared with other batteries, lithium-ion batteries have been widely used because of their large specific energy, long cycle life and good safety performance. However, the lithium-ion battery still has low recovery rate, resulting in short life time due to overcharge and over discharge, resulting in the performance degradation of the whole system (Susilo et al, 2018;Salah et al, 2019). The remaining service life (RUL) prediction is an effective method to provide lithium-ion battery life information for manufacturers and users.…”