Research On Neural Network Based Power Transformer Fault Forecast Model

Power transformer is one of the most important equipment in power system. If the power transformer fails, it will cause power failure in local area, even in the whole power system. With the increase of power system's capacity and structure, the power transformer is developing towards large capacity and high voltage level. New requirement is proposed to meet the requirement of safety and reliability of power supply. Thus, traditional power transformer is unable to meet the modern smart power grids'needs of reliability, accuracy and real-time performance.Improved BP neural network has strong functions of multi-nonlinear data processing and function approximation. Moreover, it can obtain high precision predictive diagnosis model by using original sample data's self-learning. If the robust predictive diagnosis function of the modified BP neural network is introduced into the transformer fault processing, and the intelligent prediction model which reflect accurately the characteristics of transformer fault is built, the requirements of the transformer fault on-line detection can be realized and the level of integrated automation of substation can be improved.In this thesis, the features of on-line monitoring of transformer fault prediction diagnosis is combined, the combination method of improved three ratio method and artificial neural network are selected as the basic mathematics algorithm of transformer fault prediction. We choose the three improved BP neural networks as the research model, establish the transformer fault prediction of diagnosis model of three improved BP neural network by using the neural network toolbox of MATLAB.In the experiment,100 sets of original transformer fault sample data are selected to train the three layers modified BP neural network model, and 25 sets of transformer on-line monitoring data are used to simulate. The error convergence curves show that the transformer fault diagnosis model based on improved BP neural network compared with traditional methods have higher diagnosis rate. Furthermore, the treatment of the transformer fault diagnosis model based on improved BP neural network is reasonable and effective.