Low-carbon Dispatch of Microgrid Based on Soft Actor-Critic Algorithm

This paper proposes a microgrid low-carbon scheduling algorithm based on a deep reinforcement learning framework to solve the limitations of existing scheduling methods in complex target optimization, such as low convergence efficiency, poor dynamic adaptability, and insufficient exploration ability. Existing methods such as reinforcement learning based on deterministic strategies (such as DDPG) are too conservative due to insufficient exploration efficiency, and it is difficult to achieve global optimal scheduling in the scenario of wind and solar power output fluctuations and time-of-use electricity prices. To address the above problems, this paper first constructs a microgrid system model that includes photovoltaic, wind power generation, energy storage batteries and grid interaction, designs a collaborative optimization problem with the goal of minimizing operating costs and carbon emissions, and then uses the Soft Actor-Critic (SAC) algorithm to solve it. Simulation results show that compared with the traditional DDPG algorithm, the SAC algorithm reduces the total operating cost by ${1 4. 5 \%}$ and carbon emissions by ${8. 5 \%}$ within a 24 -hour scheduling cycle, while increasing the utilization rate of renewable energy by $1.06 \%$, verifying the effectiveness of the method proposed in this paper.