Research on Remaining Useful Life Prediction of Batteries Based on Explainable Machine Learning: A Case Study of NASA Battery Data

This paper proposes a Remaining Useful Life (RUL) prediction method for lithium-ion batteries based on explainable machine learning. The study uses NASA battery data as its foundation and, through comprehensive analysis of multidimensional features such as charge-discharge cycles, current, voltage, and temperature, constructs a high-precision RUL prediction model. By comparing various machine learning algorithms, the Lasso regression model demonstrates optimal performance, achieving low mean absolute error (MAE $=14.828$) and mean squared error $(\mathrm{MSE}=253.601)$ on the test set. More importantly, this study introduces the Local Interpretable Model-agnostic Explanations (LIME) technique to interpret the prediction results, revealing the influence mechanisms of key factors—such as cycle count, discharge voltage, and charge temperature—on battery life prediction. Experimental results indicate that the proposed method not only accurately predicts battery RUL but also provides interpretable decision support, serving as an effective tool for lithium-ion battery health management and predictive maintenance.