Data-driven on-line health assessment for lithium-ion battery with uncertainty presentation

Lithium-ion battery has been widely applied in various industrial systems including distributed grids, electric vehicles, and spacecraft. The reliability and health state of battery energy storage system (BESS) directly influence the safety of the whole system. Particular emphasis is placed on approaches that can be used to assess the health state of lithium-ion battery especially for on-line operation conditions. As the battery is working in real applications, there is thereby an urgent need but it is still a significant challenge to estimate the battery health state based on on-line available parameters. This paper offers a new routine for lithium-ion battery on-line health assessment just based on the voltage measurements from constant current charge and discharge phases. Two different degradation features extracted from on-line measurable parameters are mapped to the battery health state space directly. Relevance vector machine (RVM) is applied to train the nonlinear mapping model by virtue of the sparsity of the model considering the computing complexity in on-line applications. On the other hand, the RVM algorithm can not only obtain the accurate health assessment results but also can represent the uncertainty involved in the results, which makes the health assessment more informative for system mission planning, condition based operation optimization, and other strategies to increase the system reliability and decrease the maintenance cost. The experimental results indicate that the proposed approach could potentially achieve on-line battery health assessment and present the uncertainty at the same time.