Core Temperature Estimation and Prediction for Batteries Using the Multitask Coordinated Framework With Limited Data

Abstract

The core temperature of lithium-ion batteries (LIBs) substantially impacts their safety and reliability in electric vehicles and energy storage systems. However, its direct measurement remains challenging, particularly under complex operating conditions, wide temperature variations, and limited data due to sparse temperature sensing. To address these challenges, we propose a method that integrates a temperature-sequence diffusion model (TSDM) for data augmentation with a Bayesian optimization (BO)-gated recurrent unit (GRU)-attention mechanism (AM) multitask coordinated framework. TSDM learns the thermal properties of batteries to realistically augment data, supplementing extreme-condition samples, while a multitask coordinated framework enables efficient multitask coordination through the joint optimization of network parameters and task weights. This integrated approach accurately estimates the core temperature over a wide temperature range (from <inline-formula> <tex-math notation="LaTeX">$- 10~^{\circ }$ </tex-math></inline-formula>C to <inline-formula> <tex-math notation="LaTeX">$55~^{\circ }$ </tex-math></inline-formula>C) and achieves precise prediction up to 120 s ahead, while maintaining strong performance for aging batteries. Experimental results show a maximum root mean square error (RMSE) of <inline-formula> <tex-math notation="LaTeX">$0.1720~^{\circ }$ </tex-math></inline-formula>C in core temperature estimation and <inline-formula> <tex-math notation="LaTeX">$0.8504~^{\circ }$ </tex-math></inline-formula>C in 120-s ahead prediction, substantially outperforming existing methods. The proposed scheme provides crucial technical support for the battery management system (BMS) in real-time thermal safety monitoring and long-term fault early warning.