Research On Fault Diagnosis Technology Of Large Transformer Based On Deep Network

The fault diagnosis of large oil-filled transformers refers to the timely and accurate identification of the internal discharge or overheating,so as to ensure the smooth and safe operation of the power equipment.For a long time,the fault diagnosis of large oil-filled transformers is mainly based on analyzing the composition and content of the characteristic gas dissolved in the internal insulating oil of the transformer in different states to identify and classify t he transformer faults.However,the causes of power transformer faults are complex,and there is no obvious physical boundary between different faults,which brings great challenges to power transformer fault diagnosis.This thesis deeply analyzes the predecessor's methods of transformer fault diagnosis based on dissolved gas analysis technology.In view of the shortcomings of these methods,this thesis uses the deep neural network's good feature extraction and recognition capabilities to effectively diagnose power transformer faults.The specific work is as follows:1.The research background,significance and development status of transformer fault diagnosis at home and abroad are described;then the relevant theories of Restricted Boltzmann Machine and deep belief network are studied emphatically.2.Aiming at the problems of low accuracy rate of traditional transformer f ault diagnosis methods and the impact of redundant data on training efficiency,a transformer fault diagnosis method based on PCA-DBN is proposed.The method first uses principal component analysis to reduce the dimensionality of the fault data,remove the redundant information in the fault data,then trains the deep belief network,and finally uses the trained network to diagnose the faults in the power transformer.Experiments show that this method can not only accurately identify common transformer faults,but also improve the training efficiency of the network.3.Aiming at the problem that transformer faults have no obvious physical boundary and traditional fuzzy methods have low diagnostic accuracy,a transformer fault diagnosis method based on DBN-IFCM is proposed.This method uses DBN as the feature extractor of the fault data,which is used to obtain the deep features of the fault data.Then use the improved fuzzy clustering as the classifier of the entire network to classify the extracted fault features.The experimental results show that this method can better solve the problem of misdiagnosis caused by the unclear fault boundary,and further improve the power transformer fault recognition ability.At the same time,it provides an idea for the pattern recognition problem of scarce labels.