Research On Transformer Fault Diagnosis Method Based On Rough Set And Probabilistic Neural Network Theory

The transformer is one of the most important equipments in Power System,the technology of fault diagnosis for transformer is always taken into account by savants all over the world. Rough Set theory is a new intelligent information process technology. It can analysis and deduce all kinds of incomplete data,it can find the relationship between the data, it can pick up the useful characters,reduce the information process,it can research the imprecise, incertitude knowledge's expression, learning and induction. A new method based on probabilistic neural networks (PNN) to transformer fault diagnosis is presented,the major features of the probabilistic neural stems from its modular architecture design and can be easily extended to adapt to a changing environment by incremental learning.The genetic algorithm is introduced to train the smoothing factor of PNN in order to increase the accuracy of diagnosis.This thesis combines Rough Set theory with Probability Neural Networks, applying it in the transformer fault diagnosis. It can fully develop the two methods' advantages and learn from other's strong points to ofset one's weakness.The paper includes: Basic concept and the developing situation in application field of the transformer fault diagnosis have been summarized;relative theory of RS has been introduced,and the applications in determination and fault diagnosis has been discussed.the contrast analysis was done between the probability neural network and the BP network,and then strong point of the probability neural network was elaborated,based on probability neural network fault diagnosis mode has been established,In training process,we uses the genetic algorithm to optimize the probability neural network smooth factor,Thus the classification result of the probability neural network is better than the ordinary probability neural network; Using RS theory,the original data reduction is performed first to form a simple rule collection, based on which the structure of probabilistic neural network is then completely determined,which has a better topological structure with greatly reduce network scale and greatly improved learning speed,while keeps the good classification ability of network.Finally, setting up a simple system of fault diagnosis about Transformer by the Visual VB and Matlab7.0,The results of simulation show that this method has high fault-tolerance performance and it is efective.