A Low-carbon Optimal Operation Method for Microgrids Based on Deep Reinforcement Learning

In this paper we present a low-carbon optimization strategy for microgrids with the long-term goal of developing a new type of power system. This method, rooted in deep reinforcement learning, integrates the concepts of electricity-to-gas conversion and carbon capture to attain environmentally conscious operations. Primarily, by conducting an assessment of renewable energy generation and storage apparatus, as well as incorporating electricity-to-gas conversion and carbon capture systems, the integration of a carbon-capturing electricity-to-gas turbine within the power generation framework establishes a diverse and multi-source energy microgrid system. Subsequently, to achieve optimal operation of the proposed microgrid system in an environment of source-load uncertainty, the multi-period optimization problem is transformed into a Markov decision problem based on deep reinforcement learning theory. Additionally, a knowledge fusion deep reinforcement learning framework is proposed. The utilization of the distributed proximal policy optimization algorithm facilitates the realization of optimal performance across diverse power sources within the microgrid system. The empirical findings from the simulations provide substantial evidence for both the efficacy of the envisaged deep reinforcement learning framework and algorithm, as well as the economic feasibility of the proposed operational strategy．