Useful Life Prediction of Lithium-Ion Batteries Based on Mode Decomposition, Reconstruction, and Transfer Learning

To improve the accuracy of lithium-ion battery remaining useful life (RUL) prediction and ensure safe system operation, a prediction method integrating mode decomposition and transfer learning is proposed. This method first employs principal component analysis (PCA) on the CALCE dataset to fuse indirect health indicators (HIs). Then, the fused HI is processed using the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), and the dominant modal components are identified using permutation entropy to reconstruct the signal, completing the data preprocessing process. The reconstructed signal is then fed into a Transformer network for pretraining, followed by fine-tuning with the original capacity data based on a transfer learning strategy, thereby establishing the RUL prediction model. Finally, the model’s generalization capability is evaluated on the NASA dataset. Tests conducted on the CALCE and NASA datasets indicate that the proposed approach achieves higher prediction accuracy than mainstream models, while maintaining robust stability and superior generalization capability.