Battery state of health estimation with interpretable distance feature and dynamic weight model across-chemistry and working conditions

Accurate and interpretable estimation of battery state of health (SOH) is critical for ensuring the reliability, safety, and longevity of energy storage systems. Existing SOH estimation methods often struggle to generalize across-chemistries and operating conditions, while maintaining interpretability. To address these challenges, this paper proposes a novel framework for SOH estimation that integrates Shapelet-based distance features (DFs) with a dynamic weight model (DWM) and transfer learning. The voltage Shapelets are firstly extracted by sliding windows from CC charging voltages of different cycles. Then the interpretable DFs is derived from the distances between voltages from different cycles, which capture the capacity degradation tendency. Two different DFs features are proposed to adapt different battery chemistries and conditions. Furthermore, the real time error-based DWM adaptively adjusts the model selection to estimate SOH and transfer the models from source domain to target domain, ensuring robust and generalization performance across diverse environments. Finally, the proposed framework is validated on five battery datasets encompassing. Results show that the minimum MAE is 0.409 % and the minimum RMSE is 0.699 % by the minED method, while by the VMED method, the minimum MAE is 0.439 % and the minimum RMSE is 0.673 %, which proof the effectiveness of the proposed method.