The Apply Research On Fuzzy Rule System For Power Transformer Fault Diagnosis

Large power transformer that causes most of the power system incidents is crucial equipment in the power system, and its operating state influences the system’s security level direetly, so the application of transformer oil dissolved gas analysis (DGA) data for its fault diagnosis is of great significance.The paper systematically analyzes the production mechanism about the gas in transformer oil, and the qualitative relationship with transformer fault and transformer oil dissolved gas. The contemporary research status and the existing problems of the of transformer fault diagnosis technologies are summarized. The thesis determines diagnosising transformer fault with DGA data based on the fuzzy inference system, and indepth studies the issues involved.The fault characteristic variables in transformer DGA data are more, and there is close correlation in the variables, therefore this will lead to fuzzy diagnosis rules too much and complex diagnostic reasoning process. The principal component analysis (PCA) is applied to the DGA feature variable dimensionality reduction processing, and the features input variables of the fuzzy diagnostic system are reduced without loss of fault characteristic information, and it is convenient for fuzzy diagnostic system establishment and application.According to that the composition of current large number dissolved gas analys is the real samPles, a fuzzy inference system studing method based on particle swarm optimization (PSO) combining with simulated annealing algorithm (SA) is put forward, and is applied to multi-class classification. The particle swarm optimization is used to optimize each membership function parameters of each variable, and an incorporative strategy for the membership function is adopted, and candidate rule set is formed. The process of obtaining fuzzy inference system from the candidate rule set can be viewed as portfolio allocation problem, the simulated annealing algorithm is utilized to choose excellent rules from candidate rule set and reconstruct concise fuzzy inference system.According to the characteristics of the transformer DGA data, the sample data pre-processing method is determined. Based on the dimension reduction of the transformer fault feature variables, using of fuzzy inference system studing method establish transformer running condition fuzzy recognition system, and experiments show that the system has behaves well in accuracy and generalization.