Time-Frequency Feature Fusion for Enhanced Forecasting of Lithium-Ion Battery Health and Capacity

Abstract Accurately diagnosing the state of health (SOH) and predicting future capacity of lithium-ion batteries (LIBs) are essential for enhancing battery safety, performance, and longevity. This study proposes a novel, hybrid feature-driven approach that integrates electrochemical impedance spectroscopy (EIS) and incremental capacity analysis (ICA) data to estimate SOH and predict future capacity across varying cycles. Using EIS, two mid-frequency components are identified as key frequency-domain features, while the peak value, peak area, and corresponding voltage from the ICA curve are extracted as time-domain features. These hybrid features, strongly correlated with battery SOH, serve as inputs for a genetic algorithm-optimized LightGBM (GA-LightGBM) model. The model achieves high prediction accuracy, with an R2 value close to 1 for both short and long-term predictions. Shapley value analysis further elucidates feature contributions, enhancing model interpretability across different temperatures and batteries. These findings underscore the potential of GA-LightGBM in real-time battery management, with implications for capacity forecasting, optimal charge protocols, and the extension of battery lifespan.