Fuzzy Cluster Analysis And Its Application In Analyzing Chromatographic Data Of Transformer Oil

Power transformer is one of the most important electric apparatus in electric power system, and its operating state affect directly the safety and stability of power system. As the dissolved gas analysis (DGA) has the advantages of dispense with offline testing, convenient for online monitoring and impervious to external electric field and magnetic field, it is regarded as one of the best methods for monitoring and diagnosis the earlyhidder faults of oil-immersed transformer.At present, there are many methods for transformer fault diagnosis based on DGA. In this paper, fuzzy cluster analysis is introduced in analyzing the chromatographic data of transformer oil, and existing problems of this method are discussed.At first, the fuzzy ISODATA algorithm is used to analyze the chromatographic data of transformer oil. This method has some disadvantages in actual application, such as, the division of the pattern space is lack of basis; the sensitivities of different gases for fault have not been considered and so on. Some aspects of revision is proposed, i.e. a weight vector of index is introduced to describe the sensitivities of different gases for fault; the cluster centers are split and merged after each iterative calculation. 100 groups of transformer fault data are analyzed using the revised algorithm, and the results are compared with those of three-ratio method, revised three-ratio method and actual fault. The revised method has an exactness amounted to 92%, compared with former 88%, the revised method has higher accuracy.Fuzzy ISODATA algorithm and revised algorithm are very time consuming when realizing cluster analysis in large database. Although the revised method has greatly reduced the time of fault diagnosis, before the fault diagnosis the known fault data should be analyzed, and the process of analysis is still very complex. For this reason, the fuzzy logic cluster neural network is introduced to analyze the transformer fault data. As this method uses logic operators to complete the calculation, and the neural network has the ability of parallel processing, the computing time is greatly decreased and the exactness amounted to 91%.Fuzzy ISODATA algorithm has two major shortcomings: one is too much time consumed when applying in large database and the other is liability of falling into regional `extreme values or saddle point. The fuzzy logic cluster neural network can solve the first problem, but it has the liability of falling into regional extreme values or saddle point. For this reason, evolutionary strategy method is introduced to analyze the transformer fault data. As this method can search for global optimization, more reasonable results can be gained. The results show that, under the same conditions, this method can get smaller cluster target function than fuzzy ISODATA method, and the exactness is also increased.