Research Of Power Transformer Fault Diagnosis System

Power transformer is one of the most important equipment in the power system. Its operational state is related to security and stability of the entire power grid, so it is necessary to monitor its state and fault diagnosis. The ratio method which based on the dissolved gas analysis (DGA) is widely used in transformer diagnosis. It is easy to use, but also has same defects such as incomplete fault codes and fault divided interval is too absolute. In order to address the defects of traditional methods, in this paper the radial basis function (RBF) neural network is used, which is optimized by particle swarm optimization (PSO) algorithm.RBF neural network is one of the supervised networks, and its diagnosis performance is a large extent dependent on training samples and the network parameters. As to select of training samples, the usually method is random selection which selects the training samples from all samples randomly. This method is simple, but when the total number of all samples is less, the training samples selected is very uneven. This will seriously affect the knowledge acquisition of RBF neural network and deteriorate the performance of neural network. Therefore, in the paper the clustering method is used for training samples selection. First gets the distribution of samples using clustering method. Then select the training samples from all samples according to the principle suggested in this paper. As to compute of RBF neural network parameters, which the critical problem is to determine the hidden layer radial basis function centers, clustering method is usually used. It puts the number of clusters as the number of radial basis functions, and uses centers of clusters as the centers of radial basis functions. One of shortcoming of this method is that it can't get the optimal parameters sometimes and easy to converge to local optimum values. So, in the paper the particle swarm optimization (PSO) algorithm is employed to determine the optimum center parameters. Three tests were done to demonstrate the proposed method is effective and reasonable. At the end of paper, transformer fault diagnosis system based on PSO-RBFNN is designed by using MATLAB/GUI programming. The designed diagnosis system has three features. First, it has a clarity user interface. Second, it not only gives out the type of transformer faults that occurred, but also gives out the reasons which may be cause to fault. Third, it can constantly update, by putting the samples which the previous diagnosis system can't diagnosis into neural network training samples database. So that when the neural network is retrained, the performance of diagnosis system is improved.