Research on Lightweighting of Electric Vehicle Battery Models Based on Physics-Informed Causal Convolutional Neural Networks

The accurate execution of numerous complex tasks in electric vehicles relies on precise battery models, yet high-precision models often entail prohibitive computational costs. This study proposes a battery model lightweighting technique based on a Physics-Informed Causal Convolutional Neural Network (PI-CCN). Specifically, three twin models are constructed to address the challenging solid- and liquid-phase diffusion processes in the Single Particle Model (SPM). These models are trained with physical constraints derived from governing equations and leverage the parallel computing advantages of convolutional neural networks. While maintaining solution accuracy, the proposed framework achieves as-fold acceleration in single-process solving speeds and a 2.27-fold acceleration in full-model computational efficiency compared to conventional methods, thereby facilitating the broader adoption of high-precision battery models.