Optimal Feature Extraction and State of Health Estimation for Incremental Capacity Curves Based on Bayesian Optimization

The aging process of lithium-ion batteries is a complex nonlinear process involving multiple electrochemical reactions. During different charge-discharge cycles, the battery exhibits different degradation characteristics. Accurate State of Health (SoH) estimation is a fundamental requirement for battery prediction and health management. Existing SoH estimation methods often rely on manually selected features, which introduce subjective bias, limiting their generalizability and robustness. This paper proposes a feature extraction framework based on a Bayesian optimization algorithm. First, partial incremental capacity (IC) analysis is performed within a specific voltage range, and this framework searches for effective Cycle-Voltage-IC features in the multidimensional feature space based on the self-fluctuation of the features and their correlation with battery capacity. Bayesian optimization is used to search for feature indices and intervals to extract features. Then a stacking ensemble model is developed that combines Bayesian ridge regression and randomized consistent sampling regressor to enhance the robustness of SoH estimation. This method provides interpretable feature extraction and improves the accuracy of SoH estimation. Compared with the feature extraction method based on manual experience, the feature extraction framework and SoH estimation model proposed in this paper can better fit the capacity degradation process of the four battery models, and the absolute errors of their SoH estimation are reduced by 21.88%, 66.67%, 8.92%, and 32.59%, respectively.