变压器绕组缺陷诊断系统仿真与实例验证

Simulation and Example Verification of Transformer Winding Fault Diagnosis System

准确、灵敏的变压器绕组缺陷诊断有利于电力系统安全运行。采用PSPICE软件建立变压器频率响应等效电路,利用小波变换法找到频率响应曲线的谐振点和幅值的变化情况,从而提取出特征向量,构造神经网络的训练样本;依据变压器绕组缺陷识别诊断的特点分别对BP神经网络、RBF神经网络和PNN神经网络的结构进行了设计,并对训练好的3种神经网络进行验证;采用已建立的3种神经网络诊断系统对实际变压器常见的2种变形方式(轴向变形和径向变形)进行检测和识别。试验结果表明3种方法都能够识别诊断变压器绕组变形缺陷,但是BP神经网络诊断系统具有更高的准确性和实用性。

The accurate and sensitive fault diagnosis of transformer winding is beneficial to the safe operation of power system. The transformer equivalent circuit of frequency response is established by PSPICE software. The changes in the resonant point and amplitude of frequency response curve are found out by using wavelet transform method, in order to extract feature vectors and establish the training samples of neural network. The structures of BP neural network, RBF neural network and PNN neural network are designed respectively based on the characteristics of transformer winding fault identification diagnosis, and the verification for three kinds of trained neural networks is made. Finally, the three kinds of neural network diagnosis systems are used to detect two kinds of deformations in the actual transformer, that is axial deformation and radial deformation. The results show that all the three methods can identify the winding deformation, but the BP neural network diagnosis system has higher accuracy and practicability.