Research On Transformer Fault Diagnosis Based On The Deep Neural Networks

Power transformer is an important primary side device,the core converts the energy grid and transmission of energy,but also an important support equipment safe operation of power system.Any accident in the operation of the transformer will bring economic losses and even cause serious social impact.At present,artificial intelligence algorithms are often used for fault diagnosis technology of transformers.Although the introduction of artificial intelligence algorithms has greatly improved the shortcomings of traditional diagnostic methods,the accuracy of fault diagnosis has been greatly improved,but there is still a slow convergence rate,poor stability,limited learning ability,and not suitable for a large number of sample training.Therefore,it is of great significance to study the fast and accurate transformer fault diagnosis technology and method,and to eliminate faults in a timely and effective manner.In order to solve the problems existing in transformer fault diagnosis,the gas production principle,gas production process and dissolution principle of Dissolved Gas Analysis(DGA)in transformer oil are analyzed firstly,and the different components and transformers of dissolved gas of transformer are analyzed.The corresponding relationship and method of faults are described.The common methods of transformer fault diagnosis based on DGA technology are expounded,and the fault diagnosis analysis of transformers is determined by split method.With the continuous development of deep learning,deep learning can discover the characteristics of complex data compared with traditional machine learning algorithms.Deep Belief Network(DBN)is a kind of deep learning and has better feature extraction and the ability to classify.Then analyze the application of DBN in transformer fault diagnosis,and build a Deep Belief Network Classification(DBNC),which can automatically extract the gas features from a large number of samples and make better use of it.The unlabeled sample training model improves the accuracy of power transformer fault diagnosis and reliably identifies fault types.DBN is tested and tuned using a large number of engineering example sample data,and compared with two methods of fault propagation methods:Back Propagation Neural Network(BPNN)and Support Vector Machine(SVM)..The results show that DBN has better fault diagnosis performance and more scalability,which can fully meet the actual engineering needs.