Data-driven research on battery pack temperature prediction

The operating temperature of lithium-ion battery systems is crucial for thermal management and safety in electric vehicles. However, physical modeling is challenging to implement at the vehicle or cloud level due to high computational resources and time requirements. Meanwhile, purely data-driven models require vast training data and may lack accuracy. This study proposes a data-driven training model based on the principle of battery thermal equilibrium. By analyzing model accuracy, the validity of input feature values, and stability under different predictive operating conditions, the study evaluates the training and predictive performance of various neural network models. The results indicate that the Backpropagation Neural Network (BP), using the proposed feature set, performs best in training and prediction, with a maximum error of approximately 0.1 °C and a final temperature prediction accuracy exceeding 99 %. The study also discusses the prediction error of about 0.6 °C under various operating conditions using this model. This model not only provides precise predictions for the operating temperature of lithium-ion batteries in electric vehicles but also offers valuable insights for optimizing thermal management strategies, which are crucial for improving the safety, performance, and lifespan of battery systems in real-world applications. • Enhanced data-driven model via battery thermal equilibrium principles. • Achieves >99 % temperature prediction accuracy with <0.1 °C error margin. • Hybrid model integrates physical properties for untrained location accuracy. • Computationally efficient: 70–90 % faster than conventional methods.