Accurate state estimation is essential for the safe and efficient operation of lithium-ion batteries, with temperature being a key variable affecting performance and degradation. electrochemical impedance spectroscopy-based temperature estimation has been widely explored due to its advantages over surface-mounted sensors. However, systematic feature selection methods capable of quantifying the effect of all variance sources on impedance, including their interactions, have not yet been reported. Here, we propose a feature selection approach based on analysis of variance (ANOVA). The method calculates the effect size (ES) of all variance sources and their interactions for each impedance feature. Since ES quantifies relative contributions, it enables objective comparison and ranking of features. Although demonstrated here for temperature estimation, the method can be applied to any state variable. Validation with synthetic data from a Doyle-Fuller-Newman-based model confirmed the ability to identify features primarily driven by temperature. The method was then applied to experimental data across varying temperatures and state of charge (SoC). Finally, three features of different ranks were tested under dynamic conditions, where the highest-ranked feature yielded the lowest root-mean-square error of 1.25 °C without SoC information. This systematic approach advances impedance-based diagnostics by improving state estimation accuracy and providing a standardized feature selection framework. • ANOVA-based feature selection for impedance-based state estimation is presented. • The approach is validated using a physicochemical impedance model. • ANOVA-based feature selection is applied to commercial cell data. • Accurate temperature estimation is achieved without SoC information. • The method provides a systematic framework transferable to other state variables.
Gomez et al. (Wed,) studied this question.