Fault Diagnosis Of Transformer In Distribution Network Based On Data Mining

As the prime hub equipment in the power grid,power transformers play apivotal role in it,and their operating status directly affects the safe operation of the power grid.Timely and accurate judgment of transformer fault types has always been a hot topic for scholars.The traditional three-ratio method is the most widely used method on fault diagnosis in the past few decades,but the correct rate of fault diagnosis is not high caused by defects such as boundary blurring.With the explosive growth of the test data of power equipment,traditional diagnostic methods are more and more exhaustive,and data mining technology has unique advantages in dealing with massive data and mining hidden information.Therefore,this paper combines data mining technology to propose a transformer fault diagnosis method based on data reduction,improved gravity search and fuzzy kernel clustering.In this paper,the traditional power transformer fault diagnosis method basedon dissolved gas in oil is theoretically studied.The fault types of oil-immersed power transformers are summarized and summarized.The principle of gas generation in transformer oil and its relationship with corresponding faults are analyzed.The defects of the three-ratio method are pointed out and used as the entry point for the next fault diagnosis.Outlineing the basic process of data mining and several commonly used intelligent algorithms.Focusing on fuzzy kernel clustering method,the principle characteristics and implementation steps of the algorithm are studied in detail.H₂?CH₄?C₂H₂?C₂H₄?C₂H₆?CH₄/H₂?C₂H₂/C₂H₄?C₂H₄/C₂H₆ were selected as the original feature quantities of transformer fault diagnosis,and the main model of fault diagnosis was constructed.The simulation experiment found that the diagnostic accuracy of the basic fuzzy kernel clustering method was excessively dependent in the initial cluster center vector.Considering the different condition attributes in the knowledge system contribute differently to the decision-making,the redundant condition attribute may make the decision-maker unable to correctly identify the useful information.Moreover,the complexity of the fault diagnosis process is positively correlated with the data dimension.Therefore,rough set and kernel principal component analysis are introduced.The results show that the linearly inseparable raw data turn to linearly separable after the kernel principal component analysis.After rough set processing,it not only reduces the data dimension but also maintains the category information of the original data.Both of the above reduction methods are helpful to data mining.Integrating gravitational search algorithm into the fuzzy kernel clustering.The diagnostic model uses kernel function parameters and cluster centers as search particles to find the optimal solution in the initial population space.Chaotic initialization,dynamic approximation strategy and vertical crossover strategy are used to enhance the diversity of algorithm search,improve the distribution of the initial population,and optimize the efficiency of the algorithm.This paper builds a transformer fault diagnosis model based on rough set-universal gravitational kernel clustering and kernel principal component analysis-universal gravitational kernel clustering.The simulation results show that the diagnosis result of kernel principal component analysis-universal gravitational kernel clustering is slightly better than rough set-universal gravitational kernel clustering,and both better than fuzzy kernel clustering,which indicates that the proposed method has certain applicability in transformer fault diagnosis.