Multi-agent Scheduling with Multi-stage Training for Mutual Assistance between Distributed Sustainable Energy and Flexible Loads

The integration of large-scale distributed renewable energy sources has brought tremendous operational pressure on the power grid. Coordinating these distributed renewable energy sources with adjustable resources in distribution grids will become an important means to enhance grid resilience. Addressing the challenges of modeling massive flexible resources and dealing with large decision spaces, this paper proposes a multi-agent scheduling method with multi-stage training algorithm. Firstly, a mathematical model for coordinated regulation of distributed sustainable energy and flexible loads is established. Secondly, a multi-objective phased learning framework is designed. Finally, through training and simulation on a power grid, it is verified that the proposed method can achieve power mutual assistance among distributed resources and improve the level of intelligent collaborative autonomy in distribution power grids.