Filter Methods Comparation for Incremental Capacity Analysis in Lithium-Ion Batteries Health Prediction

Accurate estimation of battery state of health (SOH) is crucial to ensure efficient, reliable, and safe operation of power battery systems in electric vehicles (EV) applications. The incremental capacity analysis (ICA) method is widely employed to evaluate battery SOH thanks to its non-invasive speciality. However, the inevitable error and noise in battery operation data hinder the acquirement of smooth incremental capacity (IC) curves and recognizable IC features, which is critical in the ICA method application. This paper systematically compares various filtering methods through a comprehensive scheme to choose an eligible filtering method for conducting the ICA method. Specifically, nine different filtering methods are carefully reviewed here, and the hyper-parameter selection process of compared filtering methods is analyzed in detail. Afterwards, a comprehensive comparison scheme is proposed among three aspects, correlation analysis, predictive accuracy and robustness, to examine the practicability and accuracy of IC curve filtering methods. After verifying through two public battery datasets, the robust Gaussian filtering (RGSF) shows superior performance than others in SOH estimation precision and robustness. Based on the Oxford and CALCE datasets, the predictive model for battery SOH, with the assistance of the RGSF method, can reduce the root-mean-square error by 10.37% and 5.29%, respectively. Finally, the RGSF is further utilized in the operation data for real-world EVs to assist in generating a smooth IC curve for investigating the capability of the RGSF in real-world applications.