Remaining Useful Life Prediction of Lithium-Ion Batteries Based on Conditional Variational Autoencoders-Particle Filter
Ruihua Jiao, Kaixiang Peng, Jie Dong
Abstract
Accurate prediction of remaining useful life (RUL) is of great significance to the safety and reliability of lithium-ion batteries, which is able to provide useful reference information for maintenance. Particle filter (PF)-based prognostic methods have been widely used in the RUL prediction of batteries. However, due to the degeneracy of particles, the prediction accuracy of the traditional PF is not high. In this article, a novel PF framework based on conditional variational autoencoder (CVAE) and a reweighting strategy is proposed to predict the RUL of batteries. First, the CVAE algorithm is described in detail and embedded into the PF framework to substitute the traditional prior distribution so as to alleviate particle degradation. Furthermore, a reweighting strategy is introduced during particle resampling to prevent the loss of particle diversity. Afterward, the state-space model of battery capacity is established on the basis of data analysis. In the end, the proposed CVAE-PF is employed to predict the degradation of the battery capacity, and the RUL can be obtained when the capacity drops to a predefined failure threshold. From the experimental results it can be concluded that the new method is able to achieve better prediction performance compared with some traditional methods.