Transformer fault diagnosis method based on IFCM-DNN adjudication network

At present, the transformer fault diagnosis methods based on clustering and neural network generally ignore the influence of unbalanced data sets, which could lead the model fall into local optimum. In order to improve the accuracy of transformer fault diagnosis with unbalanced data as samples, a decision network based on improved fuzzy clustering and deep neural network is proposed. A distance correction term is introduced to modify the fuzzy membership function, improve the membership relationship of the boundary data to majority class cluster, and reduce the influence of the wrong division of the boundary data on the cluster center of a minority one, so that the cluster center tends to be stable in the ideal position, and the imbalanced data set is effectively divided, the resolution of data set is improved. In addition, referring to the process of human learning and voting activities, a DNN adjudication network is established after the IFCM. Through learning the partitioned data sets to different degrees, the common network and expert network are used for joint voting, so as to obtain more reliable prediction results. The experimental results show that this method can effectively improve the accuracy of transformer fault diagnosis under unbalanced data sets.