GIS Insulation Fault Diagnosis Based on GraphSAGE

Deep learning has been widely used in insulation fault diagnosis of GIS. Common deep learning methods include convolutional neural network (CNN) and graph convolutional neural network (GCN). CNN relies on a large amount of data as a training set. Although GCN solves the problem of data dependence, it is transductive learning and cannot be generalized to unseen nodes, which makes the performance of GCN unsatisfactory in the practical application of GIS insulation fault diagnosis. In order to solve this problem, this paper proposes a GIS insulation fault diagnosis method based on graph Sampling and Aggregation algorithms (graphSAGE). First, the GIS voltage signal is converted into a graph structure using the Adaptive K-nearest neighbors graph construction method (AKNN), and then the graph structure is input into the graphSAGE network. On the one hand, the graphSAGE network can determine the number and number of layers of sampled neighbors, making the model more expressive. On the other hand, graphSAGE is inductive learning, which enables it to learn a set of aggregation functions to process unseen nodes, making a big step forward in the practical application of graph neural network in GIS insulation fault diagnosis. Finally, the effectiveness of the proposed method is verified by experiments, and the results show that the accuracy of this method is significantly higher than that of traditional deep learning networks.