Article: Reinforcement learning-based optimisation of intelligent battery thermal management system data Journal: International Journal of Energy Technology and Policy (IJETP) 2025 Vol.20 No.7 pp.48 - 68 Abstract: Effective battery thermal management (BTMS) is vital for safety, longevity, and performance, yet rule-based or PID schemes falter under rapid operational changes. We propose an RL-driven BTMS that learns control policies in a high-fidelity thermal simulator. The agent observes cell temperature, state of charge, and ambient conditions, and outputs continuous cooling/heating commands. We adopt deep deterministic policy gradient to cope with nonlinear dynamics and continuous actions. For safety and generalisation, the learned policy is fused with a rule-based controller via a confidence-aware hybrid scheme. Tests on real driving cycles show 3.2× faster response, 18.4% lower temperature-tracking MAE, and 18.0% less cooling energy than conventional BTMS, improving regulation efficiency and robustness. These results indicate deep RL with hybrid control is a scalable, adaptive, and safety-aware solution for intelligent BTMS. Inderscience Publishers - linking academia, business and industry through research

Title: Reinforcement learning-based optimisation of intelligent battery thermal management system data

Authors: Junda Ge

Addresses: Harbin Institute of Technology, No. 2, Wenhua West Road, Weihai, Shandong, 19213, China

Abstract: Effective battery thermal management (BTMS) is vital for safety, longevity, and performance, yet rule-based or PID schemes falter under rapid operational changes. We propose an RL-driven BTMS that learns control policies in a high-fidelity thermal simulator. The agent observes cell temperature, state of charge, and ambient conditions, and outputs continuous cooling/heating commands. We adopt deep deterministic policy gradient to cope with nonlinear dynamics and continuous actions. For safety and generalisation, the learned policy is fused with a rule-based controller via a confidence-aware hybrid scheme. Tests on real driving cycles show 3.2× faster response, 18.4% lower temperature-tracking MAE, and 18.0% less cooling energy than conventional BTMS, improving regulation efficiency and robustness. These results indicate deep RL with hybrid control is a scalable, adaptive, and safety-aware solution for intelligent BTMS.

Keywords: reinforcement learning; electric vehicles; intelligent control; thermal optimisation.

DOI: 10.1504/IJETP.2025.151790

International Journal of Energy Technology and Policy, 2025 Vol.20 No.7, pp.48 - 68

Received: 29 Apr 2025
Accepted: 25 Aug 2025
Published online: 19 Feb 2026 *

Title: Reinforcement learning-based optimisation of intelligent battery thermal management system data

Keep up-to-date