Volume 64

Transfer Reinforcement Learning-Based Optimal Scheduling for Distribution Networks Considering Topological Changes Jiankai Ling, Yuguang Song, Zekuan Yu, Haiwang Zhong

Abstract

With the large-scale integration of a high proportion of renewable energy into distribution networks, its inherent volatility and randomness introduce challenges such as voltage limit violations and power flow limit violations to distribution network operation. To address these challenges, topology changes are required to optimize power flow distribution, improve voltage quality, and enhance renewable absorption capacity. However, traditional methods that rely on precise physical models struggle to meet real-time scheduling demands, while conventional data-driven approaches lack generalization capability. This paper, based on transfer reinforcement learning, conducts corresponding research on the optimal scheduling problem under distribution network topology changes. Experiments on a modified IEEE 123-bus system demonstrate that compared to traditional methods, the proposed algorithm significantly improves decision-making efficiency across two typical topological change scenarios: network reconfiguration and change in the number of power sources.

Keywords distribution network, deep transfer reinforcement learning, optimal scheduling, topology changes, renewable energy