Distributed deep reinforcement learning-based multi-objective integrated heat management method for water-cooling proton exchange membrane fuel cell

Abstract

In order to improve the operational efficiency and stability of proton exchange membrane fuel cell (PEMFC), a distributed deep reinforcement learning (DDRL)-based integrated control strategy is proposed to solve the coordinated control problem of water pump and radiator in stack heat management system. This strategy substitutes the independent controllers of the water pump and radiator in the traditional control framework, and employs multi-input multi-output (MIMO) agents which simultaneously control the cooling water velocity of the water pump and the air velocity of the radiator, whilst monitoring the optimal global stack temperature control performance. To this end, an efficient curriculum exploration distributed double-delay deep determinate policy gradient (ECE-5DPG) algorithm is proposed for the strategy, the design of which is based on the concepts of curriculum learning, imitation learning, and distributed exploration, thus improving the robustness of the proposed strategy. The experimental results show that the proposed integrated control strategy can effectively control the cooling water velocity and air velocity simultaneously, thereby improving the operating efficiency of the PEMFC.