Reinforcement Learning DP-Based Technique for Hydrogen Production and Storage in PEMFC

A reinforcement learning (RL) model is proposed to solve the cost optimization problem of a grid-connected proton exchange membrane fuel cell (PEMFC) power plant. In this model, in addition to power generation and heat recovery, the utilization and selling of hydrogen strategies are considered. Excess hydrogen can be synthesized during periods of low demand and subsequently stored in a reservoir for prospective applications. The hydrogen stored in the reservoir may be commercialized or utilized for electricity generation later, contingent on the system's economic considerations. The paper aims to minimize the overall operational cost of the plant under different operational constraints. Q-learning algorithm with epsilon greedy policy is adopted to find the schedule that gives the lowest system cost. Electrical and thermal load profiles of the PEMFC are used to test the RL-based approach. The result shows a significant reduction in the operational cost for both hydrogen utilization and selling strategies, which proves the viability of the used technique. In addition, the proposed technique is compared to the solution of the problem using the Evolutionary programming (EP). The proposed technique showed results that are 1.73% and 2.43% better than those of the EP for hydrogen storage and sales, respectively.