Reinforcement learning-based active thermal management for sustainable electric-vehicle battery packs

• A GNN-enhanced DRL framework is proposed for EV energy and thermal management. • The model integrates SpikGRU2+ for temporal–thermal feature extraction. • The EMS is formulated as a Semi-Markov Decision Process for adaptive control. • Temperature-dependent reward ensures balance between efficiency and safety. • Achieves improved energy efficiency and thermal stability under all drive cycles. The rapid emergence of electric vehicles (EVs) emphasizes the importance of the adequate thermal control of lithium-ion battery systems to guarantee their safety, prolong their service life, and improve the energy efficiency. The proposed investigation suggests a smart control framework of sustainable battery heat management, relying on the reinforcement learning (RL), with a Spiking Gated Recurrent Unit (SpikGRU2+), and a Semi-Markov Decision Process (SMDP) model. The suggested methodology uses real-time cell-level data to continuously adjust the rate of coolant flow and the speed of the fans to achieve uniformity of temperatures in battery modules. The framework is verified with four conventional driving cycles: NEDC, FTP-75, HWFET, and US06 and compared with the well-known control methods, such as PID, fuzzy logic, DDPG, and FRL-ITEM. The outcomes of the simulation indicate that the temperature stability and the overall energy savings made a considerable improvement proving the practicability of the SpikGRU2+-SMDP-based controller to be used in the dynamic low-energy thermal management of a battery. The results not only facilitate the development of the further advanced and energy saving control solutions in the next generation electric mobility systems, but also facilitate the path to the real-life applications of the same.