Health feature extraction and capacity prediction of lithium-ion batteries based on group screening

This study introduces a group-screening-based approach for health feature extraction and usable capacity prediction of lithiumion batteries. The method integrates a novel feature selection strategy that combines sequential bifurcation selection with Sobol indices, together with an enhanced Gaussian process model, to enable efficient processing of electrochemical impedance spectroscopy (EIS) data and accurate capacity prediction. By embedding Sobol global sensitivity analysis within the sequential bifurcation algorithm, a quantitative criterion for ranking the importance of EIS frequencies is established, addressing the limitations of conventional methods in capturing feature interactions. Experimental validation on a public dataset demonstrates that the enhanced Gaussian process model substantially reduces the root mean square error (RMSE), further decreases prediction errors compared with existing approaches in the literature, and improves the R² value, thereby highlighting considerable application potential in the domain of battery health state prediction.