Research On Condition-based Maintenance Of Transformer Based On Data Mining

Transformer is one of the major power supply equipments in the power system, whose reliability is directly related to operation of power system. As an important method of detecting the internal faults of transformer, the technology of dissolved gas chromatography is also improved in recent 30 years, and the relevant guidelines issued by the state also experienced several revisions. With the development of power system and heightening of electric voltage rank as well as the application of new technologies, the transformers'fault and the corresponding gas symptoms exhibit new features inevitably, and thus the conventional methods are insufficient. With the commitment and renewal of power system monitoring equipments, the data in the related power system database has an explosive growth. As data mining technique has unique advantage in processing massive data, finding hidden information and association rules, it is important to apply data mining technique to power system analysis.This thesis introduces the relationship between the generation mechanism of dissolved gas of the oil and the types of glitch, as well as the widely used technologies of gas monitoring and chromatography. At the same time, it discusses the present status of data mining application, the decision-tree generation algorithms, pruning techniques and back propagation neural network algorithm. Then the thesis introduces the decision- tree for fault data rule mining to form the rule base, and for the first time, it uses back propagation neural networks algorithm to predict the gas content in oil, and then establishes the transformer condition-based maintenance model for diagnose of transformer current and future condition. In addition, the fault data of a certain transformer is taken as the example for validating the model proposed.