Hybrid Transformer Model for Lithium-Ion Battery Lifetime Prediction With Multi-Scale Features

With the increasing demand for high-performance lithium-ion batteries in electric vehicles (EVs), accurate prediction of their remaining useful life (RUL) has become crucial for optimizing battery management and ensuring safety. This paper proposes a novel hybrid model combining Transformer Encoder and Feed-Forward Neural Network (FFNN) to predict battery lifetime. By integrating both local and global features, the model captures short-term fluctuations and long-term degradation trends in battery performance. The local features focus on cycle-level variations, while global features summarize long-term aging dynamics. Mutual information is employed to select the most informative features, reducing redundancy and improving model efficiency. Experimental results on TOYOTA Research Institute (TRI) lithium-ion battery dataset demonstrate that the proposed hybrid model outperforms existing methods in terms of predictive accuracy, achieving lower root mean square error (RMSE) and mean absolute percentage error (MAPE). Additionally, an ablation study confirms the effectiveness of the hybrid approach. This work provides a robust framework for battery lifetime prediction, offering significant improvements over traditional models.