An Improved Approach Based on Transformer Network for Remaining Useful Life of Lithium-ion Battery

In the midst of a growing energy crisis, lithium batteries are attracting attention from various countries. With the massive use of lithium batteries in energy storage systems, accurately predicting the health status of lithium batteries has become a key issue. In battery management systems, predicting battery life is an important part, but it is difficult for current RNN-based neural networks to learn dependencies between distant positions, resulting in reduced accuracy. In this paper, we presented the EMD-Transformer method to achieve accurate prediction of battery life using only a small amount of online battery data. In the data pre-processing process, offline data were added to the dataset. Then EMD was used to extract lifetime degradation trends as well as capacity regeneration information. The Transformer model was used to get prediction results. Finally, a comprehensive SOH prediction result was generated by integrating all the separate results. Using a few battery lifetime data, we demonstrated that the RMSE of the prediction results is within 3%, which owes a higher accuracy than the existing Transformer and LSTM methods.