Physics-Informed iTransformer for Large-Capacity Lithium-ion Batteries Temperature Prediction Under Alternating Current Preheating With Few Initial Cycles

Alternating current preheating (ACP) serves as an effective solution for enhancing the performance of lithium-ion batteries (LIBs) during low-temperature operation. Accurately forecasting the temperature variation during ACP can facilitate the heating process and provide sufficient safety margins for battery thermal management. This paper proposes a physics informed iTransformer model for predicting LIB temperature under ACP, which is capable of forecasting the temperature variation of any unknown ACP process using only six initial cycles. To achieve high accuracy within few cycles, the physical mechanisms is integrated to the iTransformer during pre-training, while six initial cycles are utilized for transfer learning. Furthermore, a special designed data reconstruction method is proposed to enable temperature prediction for any future cycle with a single fixed input. Experimental validation is conducted using two groups of square batteries under ACP test at -17°C and 23°C. Results demonstrate that the proposed approach achieves mean absolute errors (MAEs) of 0.1939 and 0.0785 at -17°C and 23°C, respectively.