The increasing use of distributed generation (DG) in power systems can result in frequent online voltage problems. In scenarios in which substantial DG prediction errors occur because of high DG accommodation levels, ...
详细信息
The increasing use of distributed generation (DG) in power systems can result in frequent online voltage problems. In scenarios in which substantial DG prediction errors occur because of high DG accommodation levels, traditional technical solutions cannot meet the online voltage regulation requirements. Hence, new resources for online voltage regulation are needed. Here, flexible network reconfiguration is proposed to coordinate with the existing resources for severe online voltage deviations. For online topology-based voltage regulation (OTVR), the authors develop a deep reinforcement learning (DRL) algorithm based on the following specially designed modelling to enhance the computationalperformance. The mechanism of action incorporates the concepts of local research, branch exchange, and action separation, and it effectively simplifies the action dimension and action space. In addition, for the graph data in OTVR, a graph convolution network (GCN) is applied to obtain better feature extraction. Case studies performed on IEEE 14-bus, 33-bus, 141-bus systems and a practical system verify that our proposed algorithm can obtain close to optimal solutions in 2 s which can meet the needs of online voltage regulation. Moreover, we verify that the developed OTVR effectively increases DG penetration and decreases the need for investment in additional regulating devices.
暂无评论