Rapid battery pack state of health estimation for electric vehicles considering polarization features in multi-stage charging

Accurate and reliable state of health (SOH) estimation is of great significance for lithium-ion batteries degradation monitoring, ensuring the safe operation of electric vehicles (EVs). Although numerous SOH estimation methods have been proposed and validated under laboratory conditions, the uncertainty of operating conditions and data quality issues present in field data limit the widespread application of these methods. In addition, most of the current methods ignore the polarization features of batteries caused by current switching, which is widely presented in multi-stage constant-current fast charging conditions of EVs. To fill the gap, this study introduces an SOH estimation method for battery packs based on polarization features extracted from multi-stage charging process. Firstly, the labeled capacity of the battery pack is calculated by the ampere-hour integration method. Secondly, a cross-stage approach is adopted to extract polarization features at charging current switching moments to map the battery aging process. Lastly, a Random Forest regression (RFR) model is utilized to efficiently establish the mapping relationship between polarization features and labeled SOH, with its hyperparameters optimized using Bayesian optimization method. The estimation results of fifteen machine learning algorithms are compared, and the results showed that the RFR-based algorithm utilizing the polarization features of the selected three charging current switching moments had the best estimation results with a mean absolute percentage error of 1.83 %, and it remains as low as 1.91 % when using just 7 s of measurements. The proposed approach reduces the model's dependence on data completeness and can be widely applied for large-scale, rapid SOH estimation in EVs.