Health State Estimation Method of Lithium Ion Battery Based on NASA Experimental Data Set

Abstract

Based on the experimental data set of NASA lithium-ion battery, this paper proposes two novel methods for estimating the health status of lithium-ion battery. Firstly, the definition of battery health status is introduced. Secondly, based on the data preprocessing and visualization analysis, four features related to actual capacity degradation are extracted from the data. Thirdly, Two machine learning models, regression tree and random forest, are compared in this work. Both models are used Bootstrap methods for performance evaluation. Finally, The experimental results show that both have high estimation accuracy. The regression tree final model predicts a mean square error of 0.0006, while the random forest final model predicts a mean square error of 0.0002, indicating that the random forest is a better model.