Hybrid-data-driven SOH estimation for LFP batteries: An engineering-oriented paradigm based on interpretable feature

Data-driven state-of-health (SOH) estimation for lithium iron phosphate (LFP) batteries in practical energy storage systems remains challenging due to limited aging data and the lack of accurate SOH labels under operational conditions. This study proposes a hybrid-data paradigm that combines early-life operational data from an LFP battery pack with a labeled open-source full-life-cycle LFP cell dataset to establish an interpretable relationship between electrochemical features and capacity fade. An incremental capacity (IC)-based method is developed to identify the inflection point between the two voltage plateaus in charge-discharge profiles. The capacity accumulated beyond this point, corresponding to the high-voltage plateau associated with the LiFePO₄/FePO₄ cathode phase transition, is extracted as a physically interpretable feature. The proposed feature extraction algorithm shows stable inflection-point identification, with voltage variations below 30 mV in the full-life-cycle dataset. When trained using only 20% of the labeled full-life-cycle dataset, multiple machine learning models achieve low SOH prediction errors, with RMSE, MAE, and MAPE below 1%. These results indicate a strong feature-SOH correlation and demonstrate an efficient, interpretable framework for practical LFP battery SOH estimation which effectively bridge the gap between machine learning models and practical LFP battery SOH prediction.