A Lithium-Ion Battery RUL Prediction Method Based on ConvTrans and tAPE Under Capacity Regeneration Noise and Low-Dimensional Time Series Data

Abstract

Remaining useful life (RUL) prediction of lithium-ion batteries (LIBs) is essential for ensuring their optimum performance and longevity in energy storage solutions. However, it is difficult to predict RUL accurately under the capacity regeneration noise, especially with low-dimensional long time series data. Therefore, a novel RUL prediction method of LIB is proposed based on convolutional Transformer (ConvTrans) combined with time absolute position encoding (tAPE). First, complete ensemble empirical mode decomposition with adaptive noise is introduced to decompose the capacity degradation sequence into several components. Then, to accurately assess the importance of each component in reconstructing the original signal, random forest (RF) regression and Gini index are applied to obtain the weights of each component, which measures its ability to interpret the original sequence. Next, a ConvTrans is proposed to capture both short-term and long-term dependencies in battery data, which can depict the overall degradation trend of the battery without missing important local details. Moreover, combined with tAPE, which fits for processing low-dimensional long time series data, the improved ConvTrans can accurately model the battery degradation process and evaluate the RUL. Finally, comprehensive cases have been studied, and the results validate the effectiveness and superiority of the proposed method.