State-of-health estimation for EV battery packs via incremental capacity curves and S-transform

Accurate battery state-of-health (SOH) estimation in electric vehicles (EVs) plays a crucial role in mitigating user range anxiety. However, the suboptimal quality of cloud-based battery management system (BMS) data combined with the material heterogeneity of battery cathodes creates substantial barriers to developing universal SOH estimation methods for real-world EV applications. In this study, we propose a generalizable feature extraction framework based on the charging process. The method extracts time-domain features from incremental capacity (IC) curves and frequency-domain features using the S-transform, while also incorporating inter-cell inconsistency indicators. To assess the robustness of the extracted features, validation is conducted using laboratory data. Additionally, the influence of temperature on battery capacity and extracted features is analyzed through tests on batteries with varying capacities and cathode materials. Furthermore, real-world operational data from 37 EVs over a three-year period are employed to develop machine learning (ML) and deep learning (DL) models. Based on these results, a fusion model combining gated recurrent units (GRU) and LightGBM (LGB) is proposed, achieving material-independent battery SOH estimation with a mean absolute percentage error (MAPE) below 1.99 % and a maximum error (MAXE) under 6.57 %.