State of health estimation for lithium-ion batteries through feature enhanced and genetic algorithm optimized adaptive neuro-fuzzy inference system

Accurate estimation of the state of health (SOH) of lithium-ion batteries is crucial for the safe operation of electric vehicles. Incremental capacity analysis (ICA) and adaptive neuro-fuzzy inference system (ANFIS) demonstrate significant potential for SOH estimation. Technically, features extracted from ICA are selected by correlation analysis and then input into ANFIS to yield plausible estimation results. However, such straightforward integration of ICA and ANFIS encounters two issues: (1) valuable information, such as Pearson correlation coefficient (PCC), is utilized solely for ICA feature selection and is discarded when fed into the estimation model, and (2) the presence of local minima due to improper ANFIS initialization. Accordingly, a novel PCC-based feature-wise normalization (PFN) is developed to explicitly leverage the rich ICA information within ANFIS. Moreover, to address the issue of local minima, a genetic algorithm (GA) module is introduced to provide reliable initial parameters for ANFIS, thereby enhancing its global optimization capability. Ablation and comparative experiments indicate that the proposed PFN-GA-ANFIS algorithm significantly outperforms other methods, achieving average estimation errors within 1%. This highlights the substantial potential of the proposed method for precise SOH estimation and practical application.