Optimal control of a hybrid microgrid for hydrogen-based heat supply using deep reinforcement learning

Abstract Green hydrogen is considered one of the key technologies of the energy transition, as it can be used to store surpluses from renewable energies in times of high solar radiation or wind speed for use in dark lulls. This paper examines the decarbonization potential of hydrogen for the heating industry. Worldwide, 99% of hydrogen is produced from fossil fuels, because hydrogen derived from renewable energy sources remains prohibitively expensive compared with its conventional counterpart. However, due to the expansion of renewable energy sources and the current energy crisis of conventional energy sources, hydrogen from renewable energy sources is becoming more and more economical. To optimize the efficiency of green hydrogen production and make it more price-competitive, the author simulates a hydrogen production plant consisting of a photovoltaic plant, a power grid, hydrogen storage, an electrolyser, a natural gas purchase option, a district heating plant and households. Using the deep deterministic policy gradient algorithm from deep reinforcement learning, the plant is designed to optimize itself by simulating different production scenarios and deriving strategies. The connected district heating plant is used to map how hydrogen can be optimally used for heat supply. A demonstrable outcome of this paper is that the utilization of deep deterministic policy gradient, over the course of a full year, can result in a competitive production of hydrogen derived from renewable or stored energy sources for the heating industry as a natural gas substitute.