Transformer Fault Diagnosis Based On Normal Cloud Model And Improved Bayesian Model

As the main electrical equipment in power system transmission and distribution network, the power transformer is of great significance to the safety and stability of the power system. The transformer is a complex system, which needs to consider many kinds of uncertainty factors in fault diagnosis. Due to the development of the current monitoring technology, the field of the transformer related data access is growing and the need for effective use of massive data is required. Establishing transformer fault diagnosis model based on data mining has the important theoretical significance and practical value, by looking for the law of the transformer fault and state information from a large number of testing and monitoring data.In the data pre processing stage, this paper finds that most applications based on DGA data of transformer fault diagnosis data mining methods have the problem that numerical region is divided without considering random and fuzzy of boundary element. In order to solve the problem, the normal cloud model is used to analysis and discrete the continuous quantitative DGA data in this paper. It not only makes numerical regional division to be more objective, but also improves efficiency of association rule mining.In the stage of establishing the model of data mining, the model of fault diagnosis based on Naive Bayesian(NB)classifier do not conform to the actual situation with assuming that each attribute is independence. In order to solve the problem, the association rules of forest representation method and properties of joint probability algorithm is induced to improve NB classifier in this paper, and the transformer fault diagnosis model which is based on positive normal cloud model & Improved Bayesian classifier is established. Meanwhile, there is a problem that diagnostic correct rate of the new model is low when the fault examples are not adequate. Therefore, the Support vector regression(SVR)technology is induced to combine a set of transformer fault diagnosis classifier, including the cloud reasoning classifier, the NB classifier, the TAN classifier and the improved Bayesian classifier. The new model is proved to have higher accuracy by examples.