A Dual-Graph Attention-Based Approach for Identifying Distribution Network Topology

The distribution network needs to obtain the correct topology in time during the operation to adjust the control strategy and ensure the safe operation of the distribution network. For the current problems of frequent changes of distribution network topology and difficulties in obtaining topology structure in real time, in this paper, we propose a distribution network topology identification method based on dual graph attention mechanism (TI-DGA), which firstly obtains the attention score by inputting different features and weight matrices in the graph convolution layer, and then uses a pooling mechanism to select top-k node features as the features of the whole graph to achieve the classification purpose. The simulation computation part adds noise to the training data to better fit the real environment, and the effectiveness of the proposed model is verified on IEEE33 and IEEE57 node distribution networks. Meanwhile, the distribution network topology recognition framework built based on the model supports offline training of the model and online recognition of distribution network topology, which considerably improves the operation efficiency of the relevant system.