Dynamic Reconstruction Method of Distribution Network Considering Human-machine Collaborative Decision-making

With the rapid development of modern industry and commerce and the rapid expansion of urban scale, distribution network as the "last kilometer" of power supply, its stability and reliability is more and more important. Especially in the event of distribution network failure, the decision of load transfer is directly related to the efficiency and quality of power supply restoration. Traditional load transfer methods often rely on manual decision making, which has strong subjectivity and risk of error. In this paper, we propose a load switching control strategy for distribution networks that combines human-machine collaboration and Deep Reinforcement Learning (DRL) methods. The machine decision-making interacts with the real-time data of the distribution network through Deep Q-Network (DQN), which is able to adapt to the uncertainty of the topology change of the distribution network and quickly give a high-quality transfer scheme. While human decision-making is limited by its irreproducibility and subjectivity, this paper adopts the Soft Actor-Critic (SAC) algorithm to complement it. The method can effectively solve the contradiction between speed and solution quality of traditional methods, and provides a reliable computational solution for load switching in distribution networks.