A Transformer Fault Diagnosis Method Based On Integrated Learning And Dissolved Gas Analysis In Oil

Power transformer is one of the most critical equipments in power transmission and transmission equipment,and its safe,stable and reliable operation is the key factor to ensure social stability and development.Using the on-line monitoring device of oil chromatography as a monitoring device,the typical gas data of transformers in operation can be obtained,and the method of dissolved gas in oil(DGA)widely used in domestic and foreign power industry can be used to detect latent faults as early as possible,reduce the loss of downtime and improve the level of operation and operation.Because Yunnan power grid is mainly based on threshold alarm and traditional methods of diagnosis,resulting in high false positive rate,increased maintenance costs.This paper fully investigates the transformer fault diagnosis technology,analyzes the application of traditional methods and intelligent methods in transformer fault diagnosis,and explores the new method of using machine learning algorithm for transformer fault diagnosis in combination with the practical application of yunnan power grid transformer oil chromatography on-line monitoring technology.The main research work of the thesis is as follows:(1)Build and deal with transformer fault characteristics based on DGA.The process of dissolving gas in transformer oil,the classification of transformer faults and the characteristics of different types of faults are studied,which provides a basis for judging transformer failures by machine learning methods.A series of new features are constructed and used for classification models by reference to the typical ratio and proportion relationship between the characteristic gases in different fault situations in the traditional method.In order to eliminate the information redundancy between features constructed by reference to traditional methods,reduce feature dimensionality and classification complexity,factor analysis methods are used to process the post-construction features,the processing results are used for the input of the fault diagnosis model,the differences before and after the factor analysis and processing are studied,and the diagnostic effect is compared with the pre-feature construction diagnostic effect.(2)The application of integrated learning algorithm in transformer fault diagnosis.Verify the performance differences of different integrated learning method models based on tree models and compare them with commonly used single classification models on data sets with different feature processing.(3)Transformer troubleshooting study of Bayesian Optimization(BO)integrated learning model.The main parameters of CatBoost algorithm are optimized by Bayes,the transformer fault diagnosis model based on BO-CatBoost is constructed,and the main parameters in SVM,RF and XGBoost are used by BO algorithm,and the simulation experiment is carried out on the data set after factor analysis.Comparing the classification results of different models,the experimental results show that BOCatBoost is an effective transformer fault diagnosis method.The diagnosis method based on integrated learning proposed in this paper can effectively diagnose transformer failure and have some practical value to the operation and maintenance of power transformer.