Capacities prediction and correlation analysis for lithium-ion battery-based energy storage system

Abstract

Lithium-ion battery-based energy storage system plays a pivotal role in many low-carbon applications such as transportation electrification and smart grid. The performance of battery significantly depends on its capacities under different operational current cases, which would be affected and determined by its component parameters interacting with one another. Due to the complex interdependency of electrical, chemical, and mechanical dynamics within a battery, it is a key but challenging issue to predict battery capacities under various current cases and analyze correlations of key parameters within a battery. This paper proposes an XGBoost-based interpretable machine learning framework, which fills the gap of predicting and analyzing how battery capacities under different current rates with respect to battery component parameters of interest. The parameter importance ranking is obtained by using the Gini index within the XGBoost model, while the correlations of all parameter pairs are quantified by using the predictive measure of association. The proposed framework is tested in two popular lithium-ion battery types with three various current levels. Illustrative results show that the proposed XGBoost-based framework is able to not only produce satisfactory capacity prediction performance but also provide reliable importance as well as correlation quantifications of involved battery component parameters. These could promote the prediction and analysis of battery capacities under different current rates, further benefitting the monitoring and optimization of battery management for wider low-carbon applications.