A support vector regression-based prognostic method for li-ion batteries working in variable operating states

Abstract

Prognostics of failures is very important for health management of Li-ion batteries and has received increasing attention from both researchers and practitioners in recent years. In practice, a Li-ion battery system often works under variable operating states, which is usually caused by the evolving environment or the different operational conditions. Thus, for remaining useful cycles (RUC) prognostics in this situation, it is important to estimate the current operating state of the system. This paper proposes a support vector regression (SVR) based data-driven approach using the possibilistic clustering classification and particle filtering to estimate the system state and select SVR parameters according to the system state. Experiments data provided by NASA Ames Prognostics Center of Excellence are introduced to testify the superiority of the proposed method.