Applying self-attention mechanism to realize battery life prediction based on early cycle features

This work is based on the prior paper, “Data-driven prediction of battery cycle life before capacity degradation”, by Kristen A. Severson et al. Using the same dataset, the effect of the Variance group in the prior paper is regenerated first, and then three types of input data and three models based on self-attention modules are proposed. One of the models is selected to conduct research comparisons on different input types of data. The advantages and disadvantages of different data types based on the self-attention module are analyzed, and finally the advantages and disadvantages of the self-attention mechanism compared with the Elastic Net used in the original research and the possible future improvement directions of battery life prediction based on the self-attention mechanism are discussed.