Multi-Kernel Relevance Vector Machine With Parameter Optimization for Cycling Aging Prediction of Lithium-Ion Batteries

Aging prediction plays a vital role in battery prognostics and health management, which contributes to preventing unexpected failure and evaluating battery systems' residual value. In this study, a reliable cycling aging prediction based on a data-driven model is proposed to address the urgent issue of adaptive and early prediction of lithium-ion battery remaining useful life (RUL). First, to enhance the learning and generalization abilities of standard relevance vector machines (RVMs), a multi-kernel RVM model containing two kernel functions with different characteristics is constructed, followed by the particle swarm optimization (PSO) algorithm for determining the kernel and weight parameters. Then, a similarity criterion of the battery capacity curves is proposed to screen battery offline data for model training to achieve early life prediction. Battery cycling aging data from two types of batteries under different aging conditions are used for model training and verification. Quantitative experimental results demonstrate that the proposed multi-kernel RVM model can realize the accurate prediction of the failure cycle and capacity attenuation trajectory of different types of batteries. Moreover, the proposed method has also been proved to be able to learn the general fading characteristics from other types of batteries.