State of Health Estimation for Lithium-Ion Battery via Recursive Feature Elimination on Partial Charging Curves

Abstract

Knowing the batteries’ health state in electric vehicles (EVs) accurately and effectively would directly enhance the system’s reliability and safety. Accordingly, this article proposes a state of health (SOH) estimation method based on the collaboration of feature selection and machine learning methods. Specifically, actual EV data from more than 1200 charging processes are analyzed to designate the constrained voltage range for data processing. Features are then derived, extracted, and selected from the capacity-based curves to depict the battery degradation process and estimate the SOH. Afterward, to find the relevant SOH estimation features, a recursive approach is utilized for pruning the unimportant features cooperated with a linear regression (LR) model. SOH estimation is therefore realized based on the above-obtained features and a low computation cost linear regressor. Six lithium-ion batteries are used to verify the proposed method. The maximum estimation error can be quantitatively limited in the range of −1%–1% with the synergy of the model and features. Furthermore, comparative experiments demonstrate that the LR with optimized features could achieve a similar SOH estimation accuracy as the complicated nonlinear model but using less operating time.