Remaining Useful Life Prediction of Lithium-ion Batteries Based on Interpretable Machine Learning Methods

Predicting the remaining useful life (RUL) of lithium-ion batteries is a core issue in battery health management. However, the “black-box” nature of traditional machine learning models limits their reliability and application value. This paper proposes a machine learning framework that integrates interpretability analysis to achieve high-precision and transparent RUL prediction. Based on the NASA lithium-ion battery aging dataset, eight key indicators are selected through feature engineering and Bayesian optimization is employed to fine-tune the parameters of the machine learning models: XGBoost, Random Forest, Support Vector Regression, and Stacking. Experiments demonstrate that the Stacking model exhibits the best performance on the test set, which indicates a significantly superior predictive performance of the model. Furthermore, Shapley values are utilized to conduct an interpretability analysis of the model, revealing that the charging current area has the highest contribution ratio, suggesting it as the most critical feature influencing RUL. By integrating model optimization with feature importance quantification, this study enhances prediction accuracy, provides physical insights into battery aging mechanisms, and offers a reliable and transparent solution for battery maintenance decision-making in industrial scenarios.