Research On Power Transformer Fault Diagnosis Based On Machine Learning

As an important hub equipment in power system,the stable operation of power transformer is the premise and foundation to ensure the stability,security and reliability of power system.Therefore,real-time monitoring of power transformer operation status to achieve early and accurate judgment of latent fault has become one of the main research contents of domestic and foreign scholars.This paper focuses on the application of machine learning in the field of power transformer fault diagnosis,combines the advantages of artificial neural networks and deep learning algorithms,and proposes three corresponding transformer fault diagnosis models for different practical application scenarios.The specific work is as follows:1)A transformer fault diagnosis model based on BP neural network is constructed.The SSA algorithm,GWO algorithm and CS algorithm are used to optimize the BP model respectively,and the simulation experiment analysis is carried out to verify the effectiveness of the BP model.The fault diagnosis accuracy of this type of model is about 70%,which is suitable for the practical application scenarios with low reliability and sufficient training data.2)A transformer fault diagnosis model based on SCGWO algorithm optimized Elman neural network is constructed.Elman neural network is used as the model architecture of transformer fault diagnosis.The gray wolf space multi-dimensional chaos optimization algorithm SCGWO is built to optimize Elman neural network,and simulation experiments are conducted to verify the advantages of SCGWO Elman model.The fault diagnosis accuracy of this model can reach 91.80%,which is suitable for practical application scenarios with high reliability and sufficient training data.3)A transformer fault diagnosis model based on SCGWO algorithm optimized DBN network is constructed.The DBN network with unsupervised learning mechanism is used as the model architecture of transformer fault diagnosis.The SCGWO optimization algorithm is used to optimize the network parameters of the DBN network.The Dropout mechanism is introduced to simplify the hidden layer neurons of the DBN network.Simulation experiments are carried out to verify the advantages of the SCGWO-DBN model.The fault diagnosis accuracy of the model can reach 95.08%,which is suitable for practical application scenarios with high reliability and lack of training data.