Research On Transformer Fault Diagnosis Based On DGA Technology

Power transformer is an important large-scale equipment of the power system,and it is an important factor to ensure the reliable operation of the power system.Accurate diagnosis of transformer faults can reduce the loss caused by transformer faults.Therefore,transformer fault diagnosis has always been a research hotspot in the field of electrical engineering.At present,the diagnostic technology based on Dissolved Gas Analysis(DGA)is the most widely used in the field of power transformer fault diagnosis.The three-ratio method and Rogers method mainly used in actual engineering are all diagnostic methods developed based on this technology.The fault boundary of these methods is too absolute,and the data classification effect near the boundary is not good,which affects the fault diagnosis accuracy of the transformer.The reason is that the dissolved gas data in transformer oil has strong nonlinear characteristics,the lack of a clear functional mapping relationship between the fault and the gas content in the oil,and the distribution characteristics are difficult to accurately measure,so how to accurately extract sample data The characteristic of the fault is the key to improve the accuracy of diagnosis.Based on transformer fault diagnosis,this paper analyzes a small sample data set of dissolved gas in oil,and uses feature extraction,machine learning and other methods to accurately diagnose transformer faults.The research content of this article mainly includes the following aspects:1.Research the source of gas in transformer oil and the dissolution process of gas,analyze the mechanism of gas production in transformer oil,study the mapping relationship between fault types and characteristic gases,and analyze the current mainstream methods of fault diagnosis.Transformer fault diagnosis is essentially a classification problem,and the amount of reliable data that can be used in experiments is limited.Support vector machines(SVM)have good performance when dealing with small sample data classification problems.Therefore,this paper uses SVM as the main means of transformer fault diagnosis.2.The classification accuracy of SVM depends on its penalty factor C and kernel function parameter g.The diagnostic accuracy of a single SVM model is limited,and the parameters of SVM can be optimized through intelligent algorithms.Based on this,a transformer fault diagnosis model of WPA-SVM is proposed.The wolf pack algorithm(WPA)is used to optimize the two important parameters of the SVM,and the method is verified.Results It shows that the accuracy of the model reaches 86.67%,which is 5% higher than the accuracy of using SVM alone.This shows that the WPA-SVM model can improve the classification performance of SVM by optimizing parameters in the fault diagnosis of transformers.3.There is no clear functional mapping relationship between transformer faults and gas content in oil.The distribution characteristics of gas are difficult to accurately measure.Difficulty in extracting fault features from sample data is an important factor that affects the accuracy of diagnosis.This paper introduces Kernel Principal Component Analysis(KPCA)to extract features of transformer fault data,and then input the extracted features into the WPASVM transformer fault diagnosis model.The accuracy of fault diagnosis based on the KPCAWPA-SVM model is increased to 93.33%.4.In actual engineering,some fault data have abnormal values,missing data,and data duplication.In order to reduce the impact of the above reasons on the accuracy of fault diagnosis,the isolation forest algorithm(iForest)is introduced to clean the DGA data set,and the abnormal The data is eliminated,and then the processed data is input to the KPCA-WPA-SVM model for diagnosis,and the accuracy rate rises to 97.32%.The results show that removing abnormal data can improve the accuracy of the transformer fault diagnosis model.