Research on multi-time-scale flexible operation method of distribution network considering deep reinforcement learning

• A multi-time scale flexible operation method for distribution network considering deep reinforcement learning is proposed. • In the intra-day optimization stage, using the traditional rolling optimization method limits the energy storage output power. • By using the proposed method, the capacity utilization rate of energy storage is improved, and the operation effect is better. • In the real-time stage, the deep reinforcement learning method is used to make decisions on the power of the soft open point. • In the real-time stage, scenario clustering is first carried out, and the application effect of real-time optimization is improved. To fully exploit continuous and discrete devices in the distribution network, this paper studies a multi-time-scale optimal operation method for the distribution network that considers deep reinforcement learning. Firstly, this method comprehensively considers the response times and action characteristics of different regulating devices in the distribution network and then matches them to different time scales for optimization. Then, during the intra-day stage, in order to address the issue of insufficient ESS capacity release, the global time-window construction method based on the sliding time domain for ESS is adopted. In the real-time stage, typical scenario clustering of loads is completed, and it is combined with DRL for decision-making. Finally, a case study is conducted in the modified IEEE 33-node distribution network, and an extensibility analysis is conducted in the modified IEEE 118-node distribution network. The results show that the proposed method can effectively improve the voltage deviation of the distribution network, reduce the operating cost, and at the same time, it has good real-time performance and extensibility.