Lithium-ion battery remaining useful life estimation with an optimized Relevance Vector Machine algorithm with incremental learning

• Battery RUL estimation with prognostic uncertainty is realized.
• An incremental learning strategy is utilized in sparse RVM algorithm.
• An adaptive on-line data-driven prognostic method is proposed.

Lithium-ion battery plays a key role in most industrial systems, which is critical to the system availability. It is important to evaluate the performance degradation and estimate the remaining useful life (RUL) for those batteries. With the capability of uncertainty representation and management, Relevance Vector Machine (RVM) becomes an effective approach for lithium-ion battery RUL estimation. However, small sample size and low precision of multi-step prediction limits its application in battery RUL estimation for sparse RVM algorithm. Due to the continuous on-line update of monitoring data, to improve the prediction performance of battery RUL model, dynamic training and on-line learning capability is desirable. Another challenge in on-line and real-time processing is the operating efficiency and computational complexity. To address these issues, this paper implements a flexible and effective on-line training strategy in RVM algorithm to enhance the prediction ability, and presents an incremental optimized RVM algorithm to the model via efficient on-line training. The proposed on-line training strategy achieves a better prediction precision as well as improves the operating efficiency for battery RUL estimation. Experiments based on NASA battery data set show that the proposed method yields a satisfied performance in RUL estimation of lithium-ion battery.