Optimal hydrogen market participation and energy dispatch in multi-microgrids using dual-delayed gradient learning

Hydrogen has emerged as a promising option to complement sustainable generation units, balance demand, and ameliorate system flexibility. This paper concentrates on the energy management of a multi-microgrid combined into a 33-bus network in the context of a multi-carrier energy system (MCES), where the hydrogen is traded between microgrids and supplied by a centralized market. In the proposed configuration, sustainable units (wind turbine and photovoltaic), hydrogen storage units, fuel-cell combined heat and power (FC-CHP), and electrolyzers are adopted to meet various levels of energy demands (like electrical, thermal, and cooling). To address the complexities and dynamic specification of the MCES along with the intermittent of the sustainable units, dual-delayed actor-critic gradient (DDACG) learning is developed to obtain optimal energy management policies and ameliorate the system efficiency. To do this, a reward function is defined according to the economic and operational specifications of the MCES system including economic efficiency, hydrogen utilization, system reliability, and sustainable costs. The deep neural networks of DDACG are trained to maximize the reward function to obtain the defined objectives of MCES by interacting with its agent with the environment. Comprehensive simulation tests reveal the effectiveness of the DDACG framework in reducing operational costs, enhancing hydrogen utilization, and guaranteeing system reliability under various scenarios.