An Innovative Transformer-Based Approach for State of Health Trajectory Prediction and Remaining Useful Life Estimation in Lithium-Ion Batteries

In this article, we introduce an innovative approach based on the Transformer neural network architecture for future state of health (SoH) trajectory prediction and remaining useful life (RUL) estimation in lithium-ion batteries. In this methodology, the same model handles predictions for both early and late cycles, thanks to intelligent management of variable-length input and output sequences. Additionally, the future SoH curve is obtained through a one-shot multistep (OSMS) prediction, which enables faster predictions and fewer accumulated errors compared to the standard autoregressive approach. We tested this model on the PoliMi-TUB battery dataset, achieving better results than previous models based on Long Short-Term Memory (LSTM) networks.