State of health estimation approach for Li-ion batteries based on mechanism feature empowerment

As lithium-ion batteries are increasingly being used in electric vehicles and renewable energy applications, real-time accurate assessment of battery health is critical to ensure battery performance and safety. This study aims to propose a mechanistic empowerment-based approach to estimate the health status of Li-ion batteries, which analyzes and understands the cell performance degradation mechanism from the perspective of internal cell processes. First, based on the operational data of the cell discharge process, combined with the data-driven principle, the characteristic curve of mechanism empowerment is reconstructed. To remove feature redundancy, interval voltage segmentation and Max-Relevance and Min-Redundancy (MRMR) algorithms are used to streamline the feature inputs that are fed into the four typical machine learning algorithms, and the estimated Root Mean Square Error (RMSE) of state of health (SOH) is kept within 3 % for all of them. It is found that achieving accurate capture of the electrochemical information implicit in the aging process is a crucial technique for stability prediction, even though these features can be hidden more deeply during high magnification discharge. Finally, tested with three available public datasets, these mechanistically empowered features have higher accuracy and stability than traditional methods. •The mechanism curves of IC, DV, QV and ICD are analyzed simultaneously.•Aging information was mainly concentrated in the early and late stages of discharge.•The mechanism empowerment features have better complementarity.•Multiple algorithms to cross-validate the stability of mechanism empowerment features•The elimination of feature redundancy can streamline the model and enhance its application potential.