SOH evaluation and RUL estimation of lithium-ion batteries based on MC-CNN-TimesNet model

Abstract

Due to the increasing interest in the security of the battery system, precise and rapid work on the state of health (SOH) and remaining useful life (RUL) evaluation of lithium batteries (LIBs) is necessary in practice. In this article, a MC-CNN-TimesNet model is proposed to predict the SOH and RUL of lithium batteries. This model captures the deep-state characteristics of lithium battery aging by capturing the dependencies within and between different time scales. In addition, a Tree-structured Parzen Estimation (TPE) algorithm is used in the optimization of model parameters. In this study, we also conducted correlation analysis by Pearson Correlation Coefficient (PCC) on the input voltage, current, temperature, time, and capacity data to select the features with higher correlation with SOH and RUL. Based on the Principal Correlation Analysis (PCA), the result of the PCC is reconstructed to remove the redundant characteristic information. Then, the min–max character scaling algorithm is used to regularize all characters to speed up the training process. Finally, a comparative validation of different models was performed on the NASA dataset, CALCE dataset, and MIT dataset. The results indicate that SOH and RUL can be predicted with an average root mean square error (RMSE) within 1.5%.