Remaining Useful Life Prediction of Lithium Batteries Based on Transfer Learning and Particle Filter Fusion

Accurately predicting the remaining useful life (RUL) of lithium-ion batteries is critical for anticipating battery failure and enabling effective health management. However, existing RUL prediction methods often suffer from several limitations, including the need for large volumes of training data, significant differences across datasets, and insufficient accuracy in long-term forecasting, which hinder their applicability in real world scenarios. To address these challenges, this paper proposes a hybrid model that integrates transfer learning (TL) and particle filtering (PF) with the Mogrifier LSTM (MLSTM) network. Specifically, the model first employs a transfer learning-based Mogrifier LSTM (TL-MLSTM) to perform long-term prediction of battery capacity, thereby enhancing the modelś generalization ability to accommodate RUL prediction under varying operating conditions. Subsequently, the capacity predictions generated by TL-MLSTM are used as observations in the PF algorithm, which iteratively updates the battery state parameters and refines the capacity predictions, thereby further improving accuracy. The proposed model is validated using publicly available datasets comprising multiple types of batteries under various operational conditions. Experimental results demonstrate that the model achieves an average RMSE of 0.0199, MAPE of 0.5803%, MAE of 0.0167 and APE of 11 cycles across multiple test groups. Compared with standalone models or purely data-driven approaches, the proposed method exhibits significant advantages in robustness and accuracy for long-term capacity degradation prediction.