Enhanced Deep Reinforcement Learning-based Thermal Management Strategy for PEMFCs

Researchers from Wenzhou University have developed a novel thermal management strategy for proton exchange membrane fuel cells (PEMFCs) that leverages advanced machine learning techniques to optimize fuel cell performance and efficiency. The study, led by Dongji Xuan, utilized the twin delay deep deterministic policy gradient (ALEDE-TD3) algorithm to develop an adaptive learning and exploration mechanism that minimizes parasitic cooling power while maintaining system stability. The proposed strategy demonstrates significant improvements in key performance indicators, including a 63.1% reduction in temperature overshoot and a 46.7% reduction in average settling time.

Key Takeaways:

• The proposed ALEDE-TD3 strategy outperforms traditional control methods, including TD3, DDPG, and PID algorithms, in terms of temperature overshoot and average settling time.
• The research highlights the potential of reinforcement learning in fuel cell control, demonstrating its immense potential in advancing renewable energy goals.
• The study utilizes adaptive learning and exploration mechanisms to address nonlinear dynamic characteristics and multivariable coupling issues in fuel cell systems.
• The proposed strategy minimizes parasitic cooling power while maintaining system stability, making it a promising solution for thermal management in PEMFC applications.
• The research emphasizes the importance of high-efficiency and stability in fuel cells for driving the transition to clean energy.
• The study's findings have significant implications for the development of advanced thermal management systems for fuel cells, with potential applications in the transportation and stationary power sectors.

Statistics:

• The ALEDE-TD3 strategy reduces temperature overshoot by 63.1%, 87.4%, and 88.9% compared to TD3, DDPG, and PID algorithms, respectively.
• The proposed strategy reduces average settling time by 46.7%, 56.7%, and 59.1% compared to traditional control methods.
• The study utilizes the twin delay deep deterministic policy gradient (ALEDE-TD3) algorithm, which demonstrates significant improvements in system response speed and control stability.

Sources:

• Enhanced Deep Reinforcement Learning-based Thermal Management Strategy for Pemfc Considering Coolant System Parasitic Power. International Journal of Hydrogen Energy, 2025;146.
• Pergamon-elsevier Science Ltd, The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, England (Elsevier - www.elsevier.com; International Journal of Hydrogen Energy - www.journals.elsevier.com/international-journal-of-hydrogen-energy/)
• Dongji Xuan, Wenzhou University, College of Mechanical and Electrical Engineering, Wenzhou, People's Republic of China.