神经网络诊断变压器故障的新方法

New Approach to Transformer Fault Diagnosis Based on ANN

介绍了一种用于信号分类识别的小波神经网络，其网络权值由小波函数集充当，并在学习过程中应用了共轭梯度法。将此小波神经网络用于变压器油色谱诊断，经仿真计算，证实它具有比ＢＰ网络更好的逼近性能，可以达到较高的准确率。

This paper describes a wavelet artifi- cial neural network(ANN)for signal classification, and applies it for transformer fault detection with dissolved gas analysis (DGA).The weights of the network are replaced by wavelet hations and are corrected by conjugate gradient method in the training iteration. Preliminary simulation results show the wavelet ANN for DGA can get a good correct diagnosis rate superior to BP ANN.