Distribution Network Topology Identification with Graph Transformer Neural Network

The safety and stable operation of distribution networks is of great significance to ensure reliable power supply. Obtaining the dynamic topology of the distribution network quickly and accurately is important to carry out system state assessment and ensure system operation quality. However, it is difficult to directly collect the real-time status of all switches and lines in a distribution system, due to the limitation of the deployment capability and data noise or mistake of the measurement device. In this paper, a topology identification model based on graph transformer neural network is proposed to calculate and infer the current topological distribution based on incomplete measurement information, aiming to overcome the insufficiency and inaccuracy of traditional methods. First, the data of distribution network is converted into a graph model and effective features of system operation are selected. Then, the graph-transformer-network-based topology identification model of power distribution system is jointly trained by node features, node labels and node connections, which can improve computational efficiency while increasing the amount of input information. Finally, the effectiveness and generalization of the proposed method are verified on IEEE 33 case and PG&E 69 case.