Informative frequency band selection from EIS data for lithium-ion battery RUL prediction

Abstract Lithium-ion batteries inevitably experience aging and performance degradation during operation, which significantly impacts the lifespan and reliability of energy storage systems. As such, accurately predicting the remaining useful life (RUL) is of critical importance. Compared to time-domain features, electrochemical impedance spectroscopy (EIS) can provide more comprehensive information on battery states. However, acquiring full-spectrum EIS data is often hindered by lengthy measurement times and high noise levels. To tackle this problem, this study introduces a feature extraction method for EIS data based on frequency-band selection, integrated with machine learning algorithms for RUL prediction. Pearson correlation analysis is employed to identify frequency bands of the impedance spectrum that are highly correlated with battery aging. Specifically, a continuous five-point frequency band with high correlation coefficients is selected, and the imaginary components of impedance within this range are extracted as input features for the following model. A Bayesian-optimized support vector regression model is constructed for validation. Comparative analysis with other feature extraction methods demonstrates that the proposed approach maintains high prediction accuracy while achieving good generalization across different battery cells. The prediction results show a reduction in root mean square error and mean absolute error by 66.5% and 71.3%, respectively, compared to models using full-spectrum EIS features.