Study On Power Transformer Fault Diagnosis Based On Rough Set Theory And Support Vector Machines

This dissertation taking the dissolved gas in transformer’s oil as thecarrier,through analysing the transformer’s fault mechanism,using artificial intelligencealgorithm to diagnose and research transformer’s fault. According to the transformer’sinternal fault mechanism and fault types, establish layered diagnosis model.Accordingto the importance of the various oil dissolved gas for diagnosis, reduct the sample databased on genetic algorithm for reduction of knowledge method,get more featuresubset.This dissertation puts forward a reduction set selection algorithm, the multiplefeature subset screening,get the optimal feature subset,and reduce the workload of therate of each feature subset for training and testing.Use the binary tree principle toestablish multistage stratified transformer fault diagnosis model,make the complextransformer fault diagnosis problem detailed step by step,avoid the problem of regionalimpartibility.By adjusting the inertia weight to improve the particle swarmoptimization(PSO) algorithm, not only accelerate optimization process of thesupport vector machine (SVM), spenalty factor and nuclear parameter,but alsoimprove the diagnosis accuracy.By using the algorithm to diagnose two transformer’saccident fault of one power generation company’s, diagnostic results shows that thisalgorithm applied to engineering practice better.To overhaul the transformer,the first to carry on the condition monitoring.Thecondition monitoring is divided into off-line monitoring and on-line monitoring,tomonitor the transformer online is the inevitable trend of intelligentized grid.Throughparticipating in the chromatogram intelligent monitoring system of transformer’sdesign,research the system of oil and gas separation unit,gas detection unit,themicrocomputer control and diagnosis unit ’s structure and working principle,and put itinto operation on the two transformer,the operation of the system status andconventional test results show that the system has the short test cycle,stableperformance, etc.