Coordinated Optimization of Topology Reconfiguration and Distributed Resource Scheduling for Overload Mitigation in Distribution Systems

The development of new electricity system presents significant challenges for distribution networks, which are characterized by dual peaks with the unregulated integration of new loads, such as ultra-fast charging stations for electric vehicles, resulting in short-term peak loads that exacerbate local overloading. To address these challenges, this study proposes a cyber-physical fusion co-optimization method that combines Graph Reinforcement Learning(GRL), Deep Q-Network (DQN), and Second-Order Cone Programming (SOCP) to achieve real-time coordinated optimization of distribution network reconfiguration (DNR) and distributed resource scheduling. The method leverages the strength of GRL in capturing network structure, effectively mitigating local overloads and enhancing voltage stability by dynamically adjusting the network topology and optimizing dispatch strategies. A safety layer is incorporated to ensure reliability, and the Gurobi solver is employed for efficient optimization. The results demonstrate that the proposed method effectively reduces voltage deviations under overload conditions. These findings highlight the potential of the proposed control strategy as an effective tool for enhancing both the economic performance and safety of distribution networks.