重庆理工大学学报2025，Vol.39Issue(21)：55-62,8.DOI:10.3969/j.issn.1674-8425(z).2025.11.007
氢燃料重卡燃料电池热管理强化学习控制研究
Reinforcement learning-based control for fuel cell thermal management in hydrogen-fueled heavy-duty trucks
摘要
Abstract
Hydrogen-fueled heavy-duty trucks have emerged as a key pathway for decarbonizing the long-haul transportation sector,thanks to their extended driving range and zero carbon emissions.The performance and longevity of their fuel cell system(FCS)depend heavily on maintaining a stable operating temperature.However,conventional thermal management systems(TMS)struggle to achieve precise temperature control while minimizing parasitic energy losses.Thus,the design and implementation of a sophisticated and highly responsive thermal management system(TMS)are fundamentally critical to ensuring the reliability of hydrogen-powered heavy-duty trucksand advancing their commercial viability.This necessitates the development of intelligent control strategies that adapt to the dynamic operating conditions inherent in real-world driving scenarios. Conventional approaches to fuel cell thermal management heavily relies on classical controllers,such as proportional-integral-derivative(PID)control,or model-based strategies like model predictive control(MPC).Despite their straightforward implementation,PID controllers often exhibit sub-optimal performance,including significant temperature overshoots and slow response times,when faced with the highly nonlinear and transient thermal dynamics of a heavy-duty truck's powertrain.Advanced model-based methods,though more effective,demand high computational resources and depend on the fidelity of a system model,pose big challenges for development and calibration.A more significant challenge,however,lies in the inherent trade-off between achieving precise temperature regulation and minimizing the parasitic energy consumption of the TMS components,namely the coolant pump and radiator fan.Current research has focused on single-objective optimization—either prioritizing temperature stability or energy efficiency,failing to address the complex,multi-objective control problem crucial for enhancing the overall vehicle economy and system durability under demanding,real-world operational cycles. To address this critical gap,this paper proposes a thermal management control strategy for hydrogen-fueled heavy-duty trucks based on deep reinforcement learning(DRL).The twin delayed deep deterministic policy gradient(TD3)algorithm is employed to train a joint control policy for the coolant pump and radiator fan.The core innovation of our approach is the design of a multi-objective composite weighted reward function that intelligently balances competing control goals.This function simultaneously penalizes temperature deviations for both the fuel cell coolant outlet and inlet,while also integrating the instantaneous energy consumption of the TMS actuators. To validate our strategy,a comprehensive co-simulation platform in Simulink is developed,which couples the TMS controller with a top-level adaptive equivalent consumption minimization strategy(A-ECMS)for vehicle energy management.It is evaluated under the dynamic China heavy-duty commercial vehicle test cycle(CHTC-TT).Compared to a conventional PID strategy,the proposed energy-optimized TD3-E strategy demonstrates vastly superior temperature regulation,reducing the cumulative time-integrated value of outlet temperatures exceeding the 80 ℃ threshold by 88.5％.Crucially,this is achieved with only a marginal 2.9％increase in TMS energy consumption.The results confirm DRL-based approach achieves an effective and robust balance between thermal control performance and economic efficiency,markedly enhancing the comprehensive operational performance of the fuel cell TMS.关键词
氢燃料重卡/燃料电池热管理/深度强化学习/多目标优化控制Key words
hydrogen-powered heavy-duty trucks/fuel cell thermal management/deep reinforcement learning/multi-objective optimal control分类
信息技术与安全科学引用本文复制引用
JU Fei,ZHUANG Weichao,XU Xiaomei,WANG Liangmo..氢燃料重卡燃料电池热管理强化学习控制研究[J].重庆理工大学学报,2025,39(21):55-62,8.基金项目
国家自然科学基金青年科学基金项目(52402476) （52402476）
江苏省自然科学基金项目(BK20240677) （BK20240677）
