The Applied Research Of Integrated Fault Diagnosis Methods In Power Transformer

Normal operation of power transformer is related to the security, reliability, quality of power system. Transformer in the event of failure will reduce power generation and bring inconvenience to the life of people in the Light effect. In the serious effect it will endanger people’s lives and property security, and greatly hinder the development process of the entire national economy. In order to accurately diagnose the fault type, this paper puts forward the integrated fault diagnosis strategy of power transformer, which researches the improvement of intelligent algorithm, the fusion method and so on.First of all, the paper introduces the research background and significance, and analyzes the research status of transformer fault diagnosis technology at home and abroad. Based on the above background, the thesis has carried on the basic research of power transformer fault diagnosis method, which is three ratio method. Coding three ratio method is adopted to establish the transformer fault diagnosis model. And the paper summarizes the advantages and disadvantages from the above process, and lays the foundation for the next study.Secondly, the paper studies the current commonly used method of artificial neural network. The paper uses Back Propagation(BP) neural network as an example, trains and simulates the BP neural network, and analyses the application of neural network in the transformer fault diagnosis strategy.Thirdly, in view of the traditional fault diagnosis methods by using single parameters for the diagnosis of complex systems with incomplete and uncertain information, the paper puts forward multi-source information fusion fault diagnosis methods, which is based on Principal Component Analysis(PCA) and Dempster/Shafer(D-S) evidence theory. The method of information fusion based on PCA carries out on the multidimensional data dimension reduction processing, and realizes the correct reasoning of accurate information by using the evidence theory. In conclusion, it can get more accurate diagnosis.Finally, the paper combines rough set(RS) theory and probability neural network(PNN), using the rough set attribute reduction advantages in knowledge representation space, as a probabilistic neural network input layer node information pretreatment. The method can simplify diagnosis network scale with the pattern classification ability of probabilistic neural network, so that the accuracy and rapidity of fault diagnosis system is improved. Furthermore it can improve the disadvantages of the diagnostic method, such as too much redundant information, too complex process, etc. And let Diagnostic strategy have the stronger engineering practical value.