Abstract：In view of the traditional machine learning algorithm having such defects as low precision, easily erroneous judgment in the field of transformer fault diagnosis, a model for power transformer fault diagnosis based on convolutional neural network was proposed. On the basis of the analysis of dissolved gases in the oil, the characteristic gas values ( decimal number) were transformed into binary and were represented by two-dimensional data. The two-dimensional data were used as inputs of convolution neural network to train and optimize the model. The simulation results showed that the precision and real-time performance of the proposed model were superior to that of such models as deep belief network, support vector machine and artificial neural network. The model had a certain advantage in shortening the maintenance time of the transformer and in improving the reliability of the power system.

Keyword：transformer;convolutional neural network;characteristic gas;fault diagnosis;