An Online Prediction Method for Distributed Temperatures of Lithium Battery Packs Based on Physics-Informed Spatiotemporal Graph Neural Network

Abstract

In response to the online prediction problem of temperature field in lithium batteries under limited sparse observation conditions, a temperature field modeling strategy is developed on basis of physical information spatiotemporal graph neural network (PI-SGNN) by integrating the thermal conduction process with the message transmission mechanism of GNN. First, the thermal diffusion dynamics is embedded into the graph neural network model regarded as a soft constraint, and a multi-head graph attention strategy is introduced to adaptively learn the timevarying thermal coupling strength between battery cells. Then, the gate recurrent unit (GRU) is used to efficiently extract the longrange temporal dependent features of thermal dynamics, and a gray box prediction model that combines physical consistency and data-driven adaptability is constructed, effectively improving the modeling and prediction accuracy of the battery pack temperature field.