Temperature prediction of lithium-ion battery based on adaptive GRU transfer learning framework considering thermal effects decomposition characteristics

Lithium-ion batteries (LiBs) are widely used due to their high energy density and long cycle life, and the temperature has a significant impact on their performance and safety. Therefore, it is necessary to accurately and timely predict the temperature of LiBs in order to ensure the safe and stable operation effectively. This paper proposes a novel temperature prediction method of LiBs based on adaptive gated recurrent unit (Ada-GRU) and transfer learning framework considering thermal decomposition characteristics. The transfer learning framework constructed by the Ada-GRU model enables temperature prediction in different scenarios. Among them, the Ada-GRU characterizes the temporal distribution and splits the training data into diverse segments containing the larger distribution difference information. Then, the temporal distribution matching between network layers is performed during the training process to reduce the distribution divergence. Meanwhile, the thermal effects characteristics related to the reversible and irreversible heat are extracted using the mathematical decomposition algorithm. To verify the effectiveness, generalization, and stability of the proposed model, the validations were conducted on public and experimental datasets under different ambient temperatures and dynamic driving profiles. The results indicate that the proposed model has excellent transfer prediction performance under different temperatures, dynamic driving profiles, and batteries, respectively, and the model accuracy significantly outperforms the existing data-driven models and transfer learning models.