Energy Reports (Nov 2022)
Data-driven coordinated control method for multiple systems in proton exchange membrane fuel cells using deep reinforcement learning
Abstract
To improve the stability and operating efficiency of a proton exchange membrane fuel cell (PEMFC) system, a distributed deep reinforcement learning-based data-driven coordinated control method is proposed for realizing the coordinated control of a PEMFC gas supply system and heat management system. In addition, a siphonophora multiagent double-delay deep deterministic policy gradient (SMA-4DPG) algorithm is proposed for this method. The design of the algorithm is based on bionics in that it imitates the feeding and survival strategies of a siphonophora, a jellyfish creature with a hydra-like structure. The algorithm utilizes different exploration principles for exploring the PEMFC environment to improve the robustness of the coordination strategy in a manner similar to the siphonophora with its different prey-seeking organs. With this algorithm, the gas supply system and the heat management system are treated as two agents. Through centralized training, agents with different objectives and time scales can coordinate with each other and improve the stability of the PEMFC output voltage and stack temperature. The effectiveness of the proposed algorithm is demonstrated in a series of experiments in which its performance is compared with that of a conventional controller.
