Health Prognosis With Optimized Feature Selection for Lithium-Ion Battery in Electric Vehicle Applications

Abstract

The widespread use of lithium-ion batteries in electric vehicles has attracted widespread attention in both academia and industry. Among them, lithium-ion batteries' prognosis and health management are important research problems that need to be resolved urgently. This article proposes a novel computationally efficient data-driven state-of-health (SOH) estimation approach based on an optimized feature selection method. The difficulty of feature acquisition is defined based on voltage data distribution from more than 11 000 charging processes. The ridge regression is applied to model the battery aging process with features obtained from the charged capacity and incremental capacity data. Afterward, the feature set is downsized by solving a multiobjective optimization issue with the particle swarm optimization algorithm. The comparison experiments validate that our proposed optimized feature set performs better than the conventional feature set via manually selecting. Moreover, by collaborating with the selected features, the ridge regression can provide more reliable SOH estimation results using fewer computing resources than some nonlinear algorithms. Our approach is the first application of quantified feature acquisition difficulty in battery SOH estimation to the best of authors' knowledge.