Abstract
Electrochemical impedance spectroscopy (EIS) offers abundant dynamic information related to battery degradation. However, the distribution of relaxation time (DRT) methods often produces ambiguous timescale features due to overlapping peaks and the subjective selection of regularization parameters. We propose a DRT method based on RQ elements. This approach significantly reduces parameter sensitivity and enhances the stability of extracted features. The most informative peaks and their combinations are optimized using a feature selection algorithm and are then used as inputs to a multilayer perceptron (MLP) for SOH estimation. The proposed method is validated on two commercial LIB chemistries under different states of charge (SOC) as well as on an open-access dataset. The results demonstrate that the health indicators extracted via the RQ-DRT method and feature combinations effectively reduce redundancy and improve SOH estimation performance. By bridging impedance-based timescale analysis and machine learning, this work provides a practical pathway for SOH monitoring in battery management systems.