Transformer Fault Diagnosis Based On Improved Bp Neural Network Research

Power transformer is an important equipment in the modern industrial system which always breaks down, its reliability directly affects the security and stability of power system. Therefore, predicting early, dectecting and exhaling internally potential failure of transformer is a subject which still has great significance. Transformer fault types are various, it will produce features gas before fault, therefore, based on the dissolved gas analysis of transformer fault diagnosis is very effective. In this paper, based on the principle of dissolved gas analysis in oil, AdaBoost algorithm as a data mining algorithm and BP neural network which has been improved by the grey relation analysis theory have been applied in transformer fault diagnosis, to a certain, the fault recognition rate has been improved.First of all, the traditional BP (Back Propagation) neural network is not unstable and always falls into local minimum when is used in transformer fault diagnosis, this article will combine AdaBoost.M2as the expansion of AdaBoost algorithm and BP neural network, and build a transformer fault diagnosis model based on the AdaBoost.M2and BP neural network. In this model, AdaBoost.M2use BP neural network as a weak classifier, change weights of the given transformer fault diagnosis samples in each iteration so that it can get subset samples, and then generate a series of weak classifiers by using BP neural network to train those subset samples, and finally those weak classifiers will be combined into a strong classifier by weighted voting, this strong classifier will produce the final result of classification which also called transformer fault type. This model can overcome BP neural network's instability and lack of falling into minimum, and so as to effectively improve the accuracy rate of transformer fault diagnosis.Secondly, be aimed at the problem that the hidden layer neuron number of traditional BP neural network is selected by experience value, the grey relation analysis theory will be introduced to determination of network hidden layer neurons in this paper. By analysing the relevance between hidden layer neurons and output layer neurons of BP neural network, we can determine the number of neurons in hidden layer according to the size of the correlation, so that achieve the purpose of optimizing neural network structure. In this paper,the optimal BP neural network will be used in transformer fault diagnosis, the simulation results show that the optimal BP neural network have faster convergence speed and better recongnition rate, and optimizing hidden layer neurons of BP neural network by the grey relation analysis is more practical and scientific.Finally, we summarize the results of the study in this paper, and analysis the problem in the study, and prospect the further research direction based on those.