Topology-Aware Reinforcement Learning-Based Multi-Agent Carbon Emission Flow Calculation for Active Distribution Networks

Active Distribution Networks (ADN) can integrate distributed energy resources such as solar energy, wind energy and energy storage devices, and therefore have illustrated great potential to improve the efficiency and flexibility of the power grid and reduce carbon emissions. However, topological uncertainty in the operation of ADN poses new challenges for fast carbon emission flow calculation and power dispatching. To achieve swift calculation of carbon emissions in the power system, this paper proposes an intelligent carbon emissions flow calculation method for ADN that can better adapt to changes of ADN topology. First, we employ multi-agent coordination to enhance the state space extension for carbon emission flow calculation. Second, a Q-learning-based reinforcement learning approach is introduced to accelerate the speed of carbon emission flow calculation. The proposed method efficiently computes carbon flows under topological changes, thereby enhancing calculation speed. The effectiveness of the proposed method under different topological changes is verified through simulation analysis.