Study Of Transformer Fault Diagnosis And Location Based On Machine Learning

In power systems,Power transformer is the key equipment to undertake voltage conversion and transmission,therefore,adopting appropriate condition monitoring methods and fault diagnosis techniques for transformers,timely discovering hidden fault information,and formulating maintenance strategies are important for stable operation of the power grid.Based on the dissolved gas analysis(DGA)method commonly used for fault diagnosis of oil-filled equipment,targeting transformer DGA Sample Features and limitations of related diagnostic algorithms.In this paper,the feature quantities are first reduced based on the neighborhood rough set,mining the relationship between DGA feature information and faults,secondly,a more practical and effective transformer multilayer fault diagnosis and location model based on algorithm fusion is established,the main research work is as follows:A fault diagnosis model based on neighborhood rough set and DGA ratio was constructed,this model uses a rough set tool to make decision reductions on alternative fault feature samples composed of multiple gas ratios,combine support vector machine(SVM)for fault identification,improved the problem of complicated DGA features and unknown fault information.Contrast proof,this method can improve the feature quantity,reduce the redundancy of elemental information to 50% of the original information,and get better fault diagnosis effect.A fault diagnosis model based on particle swarm optimization(PSO)SVM is constructed.After obtaining the preferred ratio sample in this model,use PSO to optimize the settings of the kernel function parameters and penalty parameters of theprevious classifier.In order to realize the function of nonlinear multi-classification in practical fault diagnosis applications,the influence of selecting relevant parameters on the performance of support vector machines is overcome.After case analysis,finds that PSOSVM is 4 times faster than CVSVM,the classification accuracy of the model is increased by about 8%.Based on the PSOSVM model,in order to quickly obtain more detailed fault information in the actual project,it is convenient to formulate subsequent maintenance plans,proposed primary fault nature classification and secondary fault location system.Analysis of DGA characteristics of transformer fault-prone locations and fault points,considering fault location and fault nature,different DGA feature data are used,fusion of multi-layer fault diagnosis ideas,the obtained NRS-PSOSVMs multi-layer model realizes the function of simplifying the feature input and classification of each classifier,refined fault categories and improved the overall efficiency and accuracy of fault diagnosis.The engineering example verifies that this model accuracy of fault diagnosis first layer reached 93.3%,and the accuracy rate of fault location level diagnosis is 83.5%.