Remaining useful life prediction of lithium batteries based on multiscale feature extraction and multiscale feature aggregation

The estimation of the Remaining Useful Life (RUL) of lithium batteries is a highly effective method of monitoring their health status. The present paper proposes a novel approach for predicting the remaining useful life (RUL) of lithium batteries. The proposed Multiscale Feature Extraction Module (MFEM) and Learnable Multiscale Aggregation Module (LMAM) to forecast RUL based on the capacity degradation trend of lithium batteries. Unlike the single-scale modelling of previous approaches, the model primarily utilizes the feature extraction module’s ability to extract RUL-specific information at different scales. This information is then coded to integrate long-term trend data while detecting subtle fluctuations in the short-term. The efficacy of the proposed model is validated through experimental comparisons based on two publicly available datasets, NASA and CALCE. The results demonstrate the effectiveness of the prediction model presented in this paper. RE, MAE and RMSE, respectively. Moreover, a comparative analysis was conducted between the newly proposed model and previous classical prediction models. This analysis revealed consistent improvements across all comparison metrics. • A lithium battery RUL prediction model based on multi-scale feature extraction with a learnable multi-scale aggregation module is presented. • Learning aggregation of already extracted multi-scale features using CrossAttention to obtain a more comprehensive feature representation. • The feasibility of incorporating multi-scale information into the prediction of the remaining life of lithium batteries.