A physics-guided method for predicting the core temperature of lithium-ion batteries

With the rapid expansion of electric vehicles and renewable energy storage, accurate lithium-ion battery core temperature prediction has become critical for safety and thermal management. Current methods face challenges including high computational complexity, insufficient accuracy, and poor generalization under diverse operating conditions. To address these limitations, this paper proposes a physics-guided method for predicting the core temperature of lithium-ion batteries. First, a multi-layer electrothermal coupling model (METCM) is developed based on a second-order equivalent circuit model and internal stacked structure, encompassing electrochemical heat generation, interlayer heat conduction, and surface heat dissipation while ensuring computational efficiency and capturing key thermal processes. Second, to effectively reflect the dynamic influence of the environment on heat dissipation, an online adaptive convective heat transfer coefficient estimation method using the Extended Kalman Filter(EKF) algorithm is proposed. Finally, a Physics-Guided Long Short-Term Memory (PG-LSTM) method is proposed, which utilizes multi-path thermal power and state of charge (SOC) from the METCM as input features, enabling the data-driven model to focus on compensating for nonlinear coupling factors that METCM cannot capture. Experimental validation shows that PG-LSTM achieves an average RMSE of 0.3343 °C across three temperature conditions on the Dynamic Stress Test (DST) test dataset, representing a 32.9 % improvement compared to the METCM with adaptive convective heat transfer coefficient. This method provides a reliable and computationally efficient solution for battery thermal management systems in electric vehicles and stationary energy storage applications.