Multi-step Prediction of Battery Temperature and Thermal Fault Diagnosis Based on Informer

Lithium-ion batteries are prone to thermal failures under extreme conditions, leading to thermal runaway and safety risks such as fire or explosion. Therefore, effective temperature prediction and diagnosis are crucial. This paper proposes a thermal fault diagnosis method based on the Informer time series model. By extracting temperature-related features and conducting correlation analysis, a 9-dimensional input parameter matrix is constructed. Experimental results show that the model can maintain an absolute temperature prediction error within 0.5°C when predicting 10 seconds in advance, with higher accuracy than the LSTM model. Additionally, a three-level warning mechanism based on the forgetting coefficient further enhances diagnostic accuracy. Validation using test data and real vehicle data demonstrates that this method can efficiently diagnose and locate thermal faults in batteries, with low computational costs, making it suitable for online applications.