Research On Fault Diagnosing Methods For Power Transformers

Power transformer is a key equipment in the power system, and research on its fault diagnosis methods is of great significance for the early detection of transformer potential failure and the improvement of the power system security. In view of the complex fault mechanism of transformer, the paper selects vibration signal, partial discharge signal and dissolved gases in the oil as research object, and focus on transformer fault diagnosis methods combining with status feature extraction and machine learning theory.Local wave method for analyzing transformer body vibration signal is proposed. In view of transformer body vibration signal can effectively reflect the status of winding and iron core of transformer, local wave method is used for vibration signal pattern recognition of transformer body vibration. It can better understand the characteristics included in the transformer fault vibration signal, and thereby can discriminate whether the transformer is fault.A novel transformer fault diagnosis method based on factor analysis (FA) and gene expression programming algorithm (GEP) is proposed. First, the principal components are extracted by FA from original DGA data, and then GEP based transformer fault diagnosis model is builded by intelligent training. It can effectively reduce DGA attribute dimension, overcome correlation between attributes and improve the diagnostic accuracy.A transformer fault diagnosis method based on matter-element theory combining with cloud model is proposed. Cloud model is used to reform the structure of matter-element, which can resolve the information’s uncertainties of transformer fault diagnosis. And the model can be builded of no data sample, so it is suitable to slove the transformer fault diagnosis problem of fewer samples or fewer fault samples.A novel combination model for transformer fault diagnosis is proposed, which uses multiple fault diagnosis models to initial diagnosis and then use support vector machine to secondary combination diagnosis. Based on intelligent complementary idea, the combination model overcomes weakness of single diagnosis model and improves the diagnostic accuracy and the application scope.Windowed segments denosing algorithm based on EEMD for partial discharge signal is proposed, which can effectively reduce EEMD computation. And also, by selecting and computing the eight kinds of feature of two-dimensional spectra of discharge signal and adopting Fisher method, the experimental samples are classified. The results show that the selected characteristic quantities and the proposed method can well discriminate the partial discharge type.