Deep reinforcement learning-enabled intelligent thermal management controls for hybrid fuel cell-battery powertrains under cold conditions

Energy and AI, vol. 24, pp. 100751

Abstract

This paper proposes an intelligent thermal management strategy for hybrid fuel cell-battery powertrains operating under subzero conditions by integrating a PCM-based waste heat recovery (WHR) system with a deep reinforcement learning (DRL)-enhanced controller. The objective is to mitigate the high auxiliary power demand and thermal instability encountered by conventional proportional-integral (PI) controllers during cold starts, particularly in the presence of nonlinear PCM phase transitions and tightly coupled multi-loop thermal interactions. A high-fidelity 1-D dynamic system model is developed in GT-Suite and co-simulated with MATLAB/Simulink to evaluate system behavior across −10 °C to −25 °C cold-start scenarios. A prioritized experience replay twin-delayed DDPG (PER-TD3) controller is implemented and benchmarked against PI, model predictive control (MPC), and neural network-assisted MPC-NN strategies. Results demonstrate that PER-TD3 maintains stable coolant and stack temperature regulation within the target operating range while suppressing oscillations observed in MPC-based controllers during PCM charge/discharge transitions. Compared to PI control, PER-TD3 reduces auxiliary pump/fan energy consumption by 10–25% across all ambient temperatures, while achieving comparable warm-up times. The soft-constraint PER-TD3 variant further improves actuation smoothness and reduces peak auxiliary power without sacrificing thermal performance. Seasonal-duty analysis reveals cumulative auxiliary energy reductions of up to ∼25%, highlighting the effectiveness of DRL-enabled thermal management in enhancing cold-start efficiency and operational robustness for hybrid fuel cell powertrains operating in cold climates.

Authors 4

1. Aezid-Ul-Hassan Najmi corresponding Aachen Faculty of Aerospace and Automotive Engineering

   RMIT University · FH Aachen

   Affiliation as printed

   Faculty of Aerospace and Automotive Engineering, FH Aachen University of Applied Sciences, Aachen, 52064, Germany

   School of Engineering, RMIT University, Bundoora, VIC 3083, Australia

2. Ankur Jain Aachen Faculty of Aerospace and Automotive Engineering

   FH Aachen

   Affiliation as printed

   Faculty of Aerospace and Automotive Engineering, FH Aachen University of Applied Sciences, Aachen, 52064, Germany

3. Thomas Esch Aachen Faculty of Aerospace and Automotive Engineering

   RMIT University · FH Aachen

   Affiliation as printed

   Faculty of Aerospace and Automotive Engineering, FH Aachen University of Applied Sciences, Aachen, 52064, Germany

   School of Engineering, RMIT University, Bundoora, VIC 3083, Australia

4. Bahman Shabani corresponding

   RMIT University

   Affiliation as printed

   School of Engineering, RMIT University, Bundoora, VIC 3083, Australia