A Reinforcement Learning-Based Dynamic Network Reconfiguration Strategy Considering the Coordinated Optimization of SOPs and Traditional Switches

With the growing integration of renewable sources on a large scale into modern power systems, the operation of distribution networks faces significant challenges under fluctuating renewable energy outputs. Therefore, achieving multi-objective optimization over multiple time periods, including minimizing energy losses and maximizing renewable energy utilization, has become a pressing issue. This paper proposes a Collaborative Intelligent Optimization Reconfiguration Strategy (CIORS) based on a dual-agent framework to achieve a global collaborative optimization of distribution networks in a multi-time period environment. CIORS addresses goal conflicts in multi-objective optimization by designing a collaborative reward mechanism. The discrete agent and continuous agent are responsible for optimizing the switch states within the distribution grid while coordinating the control of both active and reactive power flows through Soft Open Points (SOPs), respectively. To respond to the dynamic fluctuations of loads and renewable energy outputs, CIORS incorporates a dynamic weighting mechanism into the comprehensive reward function, allowing the flexible adjustment of the priority of each optimization objective. Furthermore, CIORS introduces a prioritized experience replay (PER) mechanism, which improves sample utilization efficiency and accelerates model convergence. Simulation results based on an actual distribution network in a specific area demonstrate that CIORS is effective under high-penetration clean energy scenarios.