Research on Fault Diagnosis Method of Power Transformer Based on Graph Neural Network

Artificial neural networks are often used on transformer fault diagnosis. It takes the independent characteristic gas content or gas ratio as the input parameter, lacks the consideration of the correlation between gases. To solve this problem, a transformer fault diagnosis method based on graph neural network(GNN) is proposed in this paper. Relationship between gases is described by a directed graph. For the advantages on dealing with unstructured data, A GNN is designed in this paper to extract correlations characteristic between gases. The relationship between the gas characteristics and the fault type is established by a multi-layer linear neural network, then, the fault type is finally determined. The experimental results show that, compared with the methods using convolutional neural network(CNN), multilayer perceptron(MLP), K-nearest neighbor(KNN) and support vector machine(SVM), the method proposed in this paper achieves deep extraction of gas characteristics and improves diagnosis accuracy for each fault type of power transformers.