State of Health Estimation of Lithium-Ion Batteries Across Temperatures Using EIS-Based Features

Accurate evaluation of the state of health (SOH) of lithium-ion batteries (LIBs) is essential for ensuring their reliability and safety during operation. This study proposes an interpretable data-driven framework that integrates electrochemical impedance spectroscopy (EIS) measurements with features derived from an equivalent circuit model (ECM) and distribution of relaxation times (DRT). To mitigate temperature-induced spectral scaling while preserving degradation-related patterns, an in-line normalization strategy is applied to the impedance data. An Extreme Gradient Boosting (XGBoost) regression model is then trained using the fused feature set. On the cross-temperature test cells, the proposed method achieves an average coefficient of determination (R 2 ) of 0.936, with a root mean squared error (RMSE) of 1.08% and a mean absolute error (MAE) of 0.77%, which significantly outperforms the unnormalized baseline model (R 2 = 0.874, RMSE = 1.63%, MAE = 1.26%). Shapley additive explanations (SHAP) indicate that the model relies primarily on low-to-mid frequency interfacial features associated with SEI evolution and charge-transfer processes, supporting a physically meaningful interpretation of the prediction mechanism. These results suggest that the proposed approach enables temperature-robust and interpretable SOH estimation for LIBs across temperatures.