Research on Regional Power Grid Scheduling Strategy With Flexible Resource Clusters Based on Multi-Agent Deep Reinforcement Learning

The increasing integration of distributed energy resources, controllable loads and energy storage systems is reshaping power systems by enhancing flexibility in supply–demand balancing. However, their large-scale deployment imposes significant communication and computational burdens on dispatch centres. Traditional model-driven scheduling methods often struggle to maintain efficiency and fairness among stakeholders, whereas existing deep reinforcement learning approaches lack mechanisms to address real-time response deviations within resource clusters leading to unstable policy performance. To tackle these challenges, this paper proposes a real-time scheduling strategy for partitioned power grids based on multi-agent deep reinforcement learning. A hierarchical distributed control framework is developed, where different agents manage regional grids and coordinate decision-making across flexible resource clusters. The framework adopts centralised training and distributed execution integrating real-time regulation performance as a regularisation term in agent rewards to improve learning stability and decision efficiency. Simulation results under varying renewable energy penetration levels demonstrate that the proposed method enhances scheduling performance and system robustness. This approach provides a promising solution for managing large-scale flexible resources and contributes to the intelligent operation of new type power systems.