Advances in machine learning applications for battery thermal management systems: modeling methods, optimization strategies, and future trends

With the rapid development of renewable energy and energy storage technologies, battery systems are evolving toward high energy density and high integration. Thermal runaway has gradually become a key factor limiting system safety and reliability. The design of traditional battery thermal management systems has mainly relied on experimental testing and numerical simulation. However, high computational cost and long development cycles are often encountered in coupled analysis under multiple operating conditions and in the optimization of complex structures. In recent years, machine learning has been increasingly applied to battery thermal management, providing new methods for temperature prediction, performance evaluation, and structural optimization. This paper systematically reviews the research progress of machine learning in battery thermal management systems. Purely data-driven models, semi-physical models, physics-constrained models, and domain-specific large language models are summarized, together with common optimization strategies such as genetic algorithms and particle swarm optimization. Representative applications of machine learning are then reviewed for key tasks, including non-uniform heat generation and multiphysics thermal behavior, thermal safety-oriented battery material design, transient thermal runaway prediction, cross-scenario state estimation, and real-time control. Current limitations in data quality, physical consistency, cross-scenario generalization, and real-time deployment are further discussed. Future directions are also proposed, including multi-source data fusion, physics-informed modeling, lightweight computing, and intelligent closed-loop control. This paper can provide a reference for the modeling, optimization, and engineering application of machine learning-driven battery thermal management systems.