Application Of Adaptive Classification Algorithm Based On Stack AutoEncoder Neural Network In Transformer Fault Diagnosis

With the increase of the scale of power network equipment,the traditional power failure detection work has been gradually replaced by the state maintenance work based on live detection technology.Oil-immersed power transformers are the focus equipment of substations,and live detection of them is the focus of daily work.However,in the process of using the traditional three-ratio method to analyze the detected data,there will be some problems,such as code defect and insufficient criterion of critical value.Auto Encoder(AE)is an unsupervised learning algorithm with powerful feature extraction ability,which can effectively extract high-level features of detection data.Ada Boost,as an ensemble learning algorithm with better classification performance than other base classifiers,can solve the classification problem when the critical value criterion is insufficient.so this thesis studies the application of AE and Ada Boost classifier in oil immersed power transformer fault diagnosis.1.This thesis studies the basic principle and detection method of dissolved gas detection in oil,discusses the corresponding relationship between various faults of transformer and the type and content of gas produced in oil,and compare and analyze the advantages and disadvantages of the traditional diagnostic three-ratio method and the improved three-ratio method.Combined with the actual operating oil-immersed power transformer,the detection data is obtained.2.Stack Auto Encoder(SAE)is the depth model of AE.This thesis combines the SAE and Ada Boost to construct the SAE-Ada Boost fault diagnosis model.Use SAE to extract the advanced features of the input data and use them as the supplementary features of the three ratios.The two data are fused,and the Ada Boost classifier is used to identify the fault type.Finally,the structural parameters of the SAE-Ada Boost fault diagnosis model are determined through experiments,and the diagnostic ability of the model is tested.3.Analyze the data of dissolved gas in oil for oil-immersed power transformers in high altitude areas to determine their operating status and fault types.The research results show that the accuracy of the results obtained by using the SAE-Ada Boost fault diagnosis model to judge the fault of the oil-immersed power transformer is above 90%.Through infrared imaging detection and partial discharge detection of power transformers,it is verified that the use of autoencoder and Ada Boost classifier can improve the reliability and accuracy of transformer fault diagnosis results to a certain extent.