A transformer-based hybrid method with multi-feature for lithium battery remaining useful life prediction

During the cyclic aging process of lithium batteries, the capacity degradation exhibits significant non-stability and randomness due to external influences and their physical and chemical reactions. In this paper, a Transformer-based multi-feature decomposition prediction method is proposed for estimating the remaining useful life (RUL) of lithium batteries. For this study, the method utilizes the average discharge current, discharge voltage, battery temperature, discharge time, and discharge capacity from each charge–discharge cycle. Subsequently, the data is processed using Savitzky–Golay filtering (SGF) during the preprocessing stage. Some of the multi-feature data (current, voltage, battery temperature, and discharge time) are decomposed using the CEEMDAN method to obtain intrinsic mode function (IMF) components, and the theoretical capacity is calculated for prediction. Finally, the prediction is conducted using the Transformer network. The superiority of the proposed method is verified by comparing the evaluation metrics of the proposed method (SGF-CEEMDAN-Transformer) with other baseline prediction models and ablation experiments.