Graph Convolution Network Considering Edge Node Feature Aggregation

Abstract

As the level of power grid intelligence continues to enhance, the identification of power grid topology becomes more and more important. Generating embedded vectors of power grid topology is an important method to identify power grid topology. At present, in the field of embedded expression of graph nodes and semi-supervised learning, graph convolutional networks (GCN) have been widely used. But the typical GCN model performs poorly in the classification performance of graph edge nodes. In order to improve the classification performance of graph edge nodes, this paper proposes a graph convolutional network considering edge nodes feature aggregation (Enfa-GCN). The Enfa-GCN model realizes feature aggregation of graph edge nodes by marking the missing key features of graph edge nodes. This paper conducts a large number of experiments on three benchmark data sets to evaluate the Enfa-GCN model. Compared with the most advanced models, Enfa-GCN has advantages in graph edge node classification and overall graph node classification performance. Among them, Enfa-GCN is 1.42%, 0.44%, and 0.87% higher than GCN on the Cora, Citeseer, and Pubmed data sets, respectively. It shows that the Enfa-GCN model has a significant effect in improving the classification performance of graph edge nodes and the classification performance of whole graph nodes.