Remaining useful life prediction of li-ion batteries based on an improved transformer model

Abstract

Precise Remaining Useful Life (RUL) prediction of Li-ion battery is crucial for health management and state estimation. With the rapid growth of new energy vehicles, it is a pressing need to enhance RUL prediction techniques. Under the development of artificial intelligence technology, physical methods no longer meet the needs of industrial Li-ion battery RUL prediction, so the data-driven model based on deep learning has become a hot spot. This paper proposed an improved Transformer model, where Discrete Wavelet Transform (DWT) is firstly employed to deal with the inherent noise during charge/discharge cycles. The serial Convolutional Neural Network (CNN) structure is utilized to mine local health factors, and position information based on residual connections encoded into the Transformer network and the trend fusion module is added to improve the network integration capability. Evaluations using both public Center for Advanced Life Cycle Engineering (CALCE) and experimental lifetime battery datasets B_X demonstrate the superiority and effectiveness of the DWT-CNN-Transformer model. It showcases faster convergence speed and higher optimization accuracy compared with other baseline approaches, significantly bolstering the precision and robustness of RUL predictions.

Keywords

• Li-ion battery
• remaining useful life
• feature reconstruction
• Transformer
• convolutional neural network