Collaborative optimization strategy of hydrogen fuel cell train energy and thermal management system based on deep reinforcement learning
Kangrui Jiang, Zhongbei Tian, T. Wen, Kejian Song, Stuart Hillmansen, Washington Y. Ochieng
Abstract
Railway decarbonization has become the main direction of future development of the rail transit industry. Hydrogen fuel cell (HFC) trains have become a competitive potential solution due to their zero carbon emissions and low transformation costs. The high cost of hydrogen, driven by the challenges in storage, transportation, and utilization, remains a major constraint on the commercialization of HFC trains. Temperature has a great impact on the energy conversion efficiency and life of HFC, and its thermal management requirements are more stringent than those of internal combustion engines. Existing HFC train energy management systems (EMS) generally overlook the impact of HFC temperature changes on energy conversion efficiency, and it is difficult to achieve real-time balance control of energy and thermal management according to environmental dynamic conditions. To address this issue, this paper proposes a collaborative optimization energy and thermal management strategy (ETMS) based on deep reinforcement learning (DRL) to minimize hydrogen consumption and control the temperature of the energy supply system near the optimal temperature, while ensuring the dynamic balance of battery charging and discharging. First, a complete physical model of the HFC train is established. Then, the ETMS is modeled as a Markov decision process (MDP), and the agent is trained through an advanced double deep Q-learning algorithm to interact with the real passenger line operation environment to make decisions on the output power of the HFC. Finally, a simulation test was conducted on the Worcester to Hereford line in the West Midlands region of the UK. The results show that within the UK's annual temperature range, the proposed method saves more than 5 % and 2 % of energy compared to the rule-based and GA-based methods, respectively. Additionally, it provides better temperature control and SOC maintenance for the energy supply system.