A multi-objective energy coordinative and management policy for solid oxide fuel cell using triune brain large-scale multi-agent deep deterministic policy gradient

In this paper, a data-driven multi-objective energy coordinative management policy is proposed in order to enhance the net output power and efficiency of a solid oxide fuel cell (SOFC) and prevent constraint violations. This study focuses on the optimization agent and controller design for a SOFC power system to maintain stable oxygen excess ratio (OER) and fuel utilization (FU) ratio as well as meet the load demand simultaneously. The optimization agent is responsible for output the reference OER and FU, aiming to achieve maximum net output power and operational efficiency as well as dynamic constraint satisfaction times in terms of OER and FU in real time. By applying reference OER and FU settings, the air and hydrogen flow within the SOFC can be effectively controlled by coordination of the air control agent and hydrogen control agent, respectively. In addition, a triune brain large-scale multi-agent deep deterministic policy gradient algorithm (TBL-MADDPG) is proposed. In order to improve the robustness of the proposed policy, the design of TBL-MADDPG entails curriculum learning, imitation learning and a large-scale multi-agent training framework. The performance of this proposed method is verified by the experiment.