State of Health Estimation for Lithium-Ion Batteries Based on Improved Gaussian Process Regression

Precise state of health (SOH) estimation for lithium-ion batteries (LIBs) can provide guidance for the rational use, reduce the failure rate of batteries, and extend the battery life. Due to the complex aging mechanism, and uncertainty of aging paths of LIB system, the model-based method needs to establish a complex mechanism model, which requires a large amount of calculation. In this paper, the Gaussian process regression (GPR) method is introduced to analyze battery experimental data and estimate battery SOH. GPR is a data-driven method with the ability to describe uncertainty. Firstly, the aging experiment of LIB was carried out. In the experiment, three LIBs of the same type were operated under three different charging and discharging current rates until the end of life. Secondly, based on the charging curve of LIB, four features reflecting battery aging were extracted. In addition, the grey relational analysis method was introduced to analyze the correlation between the extracted features and SOH. Thirdly, a SOH estimation framework based on improved GPR is proposed, and the GPR model is improved by using an explicit mean function and a combined covariance function. Finally, based on the experimental data, the proposed framework is used to estimate the SOH, and the results are analyzed.