Transformer Fault Diagnosis Based On Selective Ensemble Least Squares Twin Support Vector Machine

As the core equipment of power system,power transformer is mainly responsible for the transformation,distribution and transmission of power in the power grid,and it’s running state determine the safety and reliability of power supply directly.Once the transformer failure,it might cause the interruption maintenance of transformer and local blackout,or it may even cause the transformer explosion,which would lead to major accidents and economic losses.Therefore,it is of high importance to diagnose the latent fault of transformer in time and accurately.A variety of intelligent methods have been developed for the fault diagnosis of transformer,but these intelligent methods have some defects,also it is difficult to establish an accurate fault diagnosis model with just such single method.In order to improve the accuracy of transformer fault diagnosis further,this paper introduces the idea of selective ensemble learning and LS-TSVM,and studys the selective ensemble fault diagnosis method of transformer based on DGA.Firstly,the different fault types of transformer and their causes is analyzed,and the generation mechanism of dissolved gases is studied.The gas production mechanism of insulating materials and course of dissolution of fault gas are discussed in detail.On this basis,The relationship between transformer fault type and gas component in oil are studied.Secondly,the classification method of LS-TSVM is studied.The basic principle of LS-TSVM is studied deeply,and according to the existing problems of multi-class classification methods,a multi class classification model based on Huffman tree is proposed,the detailed construction process of which is given.In order to solve the parameter selection problem of LS-TSVM,use CSO that has a preferable search performance to optimize parameters,which would make the performance of the multi-class classification model based on LS-TSVM reach optimum state.Thirdly,On the basis of studying the theory of selective ensemble learning,the selective ensemble method of LS-TSVM is proposed.The method use LS-TSVM as base classifier,and constructing multiple base classifiers based on Bagging algorithm.After using CSO optimize the weight of base classifier,the base classifier is selected to participate in the integration according to the preset weight threshold.By introducing selective ensemble learning,the classification performance of LS-TSVM can be improved significantlyFinally,the method of selective ensemble LS-TSVM is introduced into the fault diagnosis of transformer,and a selective integrated fault diagnosis model is built.The model is emulated by the DGA datas of the field transformer,and the results show that,the proposed method can obtain a high fault diagnosis accuracy by just selecting part of the base classifiers to integrate,and compared with single multi-class classification method of LS-STVM and SVM,it’s accuracy is higher,which demonstrates the effectiveness and practicability of the proposed method in transformer fault diagnosis.