Transformer Fault Diagnosis Technology

Transformer is key equipment in the power system. Its operation affects the security and reliability and stability of the system, so we must prevent and reduce the transformer fault's occurrence. But in the long operation of transformer, because of the influence of the aging insulation and the force, the faults can't be avoided completely. Transformer's fault mechanism and reasons are complex, which brings a huge challenge to the transformer fault diagnosis. Transformer fault diagnosis is helpful to find fault early and to be maintained timely, reduce the loss rate of the accident's capacity, improve the power efficiency and obtain the huge economic benefits. Therefore, transformer fault diagnosis technology's research has important actual significance.Transformer fault diagnosis is a nonlinear mapping process from fault information to fault type. We can't use an accurate mathematical model to describe it and we get plenty of fault sample is difficult when he transformer fault happened. So the traditional fault diagnosis method is limited to use. But the neural network has excellent non-linear approximation ability and support vector machine is suitable for small sample's classification. Due to the neural network and support vector machine's advantages, this paper studies the neural network, the artificial fish-swam neural network and support vector machine's application in the transformer fault diagnosis. This paper mainly includes:Based on the neural network's non-linear approximation ability and L-M algorithm's quick convergence, this paper puts forward the transformer fault diagnosis method based on the BP neural network. The performance of the method was established by Neural Network Tools Box available in MATLAB. And this paper analyses and discusses the training sample's selection, input layer and output layer and hidden layer's design, the learning function's selection. Through the test in examples, this model can well realize the transformer fault diagnosis.Based on artificial fish-swarm algorithm having good ability in overcoming local extreme and obtaining global extreme, this paper puts forward the transformer fault diagnosis method based on the artificial fish-swarm neural network. This paper uses artificial fish-swarm algorithm to train neural network, and the influence of network performance parameters are discussed. Last, test the method. Results show that the method's diagnosis accuracy is higher than the BP neural network.Considering the shortcomings of binary tree support vector machine (SVM), this paper puts forward the transformer fault diagnosis method, which combines the separation measures among types and binary tree support vector machine. First, the type most easily separated is calculated by the separation measures among types and the transformer fault diagnosis model is established. Then support vector machines at all levels of the model are trained. Last, the performance test's results show that the model can classify fault samples well.