An Automatic Weak Learner Formulation for Lithium-Ion Battery State of Health Estimation

Current pulses are convenient to be actively implemented by a battery management system. However, the short-term features (STF) from current originate various sensors with uneven qualities, which hinder one powerful and strong learner STF for state of health (SOH) estimation. This article, thus, proposes an optimized weak formulation procedure lithium-ion SOH estimation, further enables automatic initialization integration learners into efficient estimation framework. A Pareto front-based selection strategy is designed select representative solutions nondominated fed knee point driven evolutionary algorithm, guarantees both diversity accuracy learners. Afterward, learners, whose coefficients obtained self-adaptive differential evolution, integrated weight-based structure. The proposed method utilizes boost overall performance validation proved LiFePO ${_4}$ /C batteries under accelerated cycling ageing test including mission profile providing primary frequency regulation service grid constant profile.