Physics-informed machine learning estimation of the temperature of large-format lithium-ion batteries under various operating conditions

The thermal issues of lithium-ion batteries significantly affect their performance, safety, and lifespan, and are one of the main obstacles hindering the further development of electric vehicles. Accurate tracking and prediction of the temperature distribution of lithium-ion batteries provide feedback for the Battery Management System, which is crucial for ensuring the safe and efficient operation of the battery system. At present, temperature estimation methods for battery systems mostly rely on physical models with numerous parameters, which fail to reflect the overall temperature distribution of large-format batteries. On the other hand, data-driven models for battery temperature require large amounts of training data and are prone to overfitting and other problems. These models lead to problems such as low accuracy in temperature estimation for battery systems, a high demand for training data, and poor model interpretability. To address these issues, this paper proposes a PhysicsInformed Neural Network model based on the thermal physical model of large-format batteries. By embedding the thermal-physical model information of the battery into the neural network, the model can accurately predict the evolution of the lithium-ion batteries temperature under various operating conditions. Experimental results show that under different constant current and dynamic conditions, the maximum Root Mean Square Error and Mean Absolute Error of the Physics-Informed Neural Network model do not exceed 0.6 degrees C. Compared with traditional methods, this approach significantly enhances the precision and interpretability of the model with smaller datasets, providing a more accurate basis for temperature distribution in battery management system.