广东电力2018，Vol.31Issue(6)：127-132,6.DOI:10.3969/j.issn.1007-290X.2018.006.019
基于卷积神经网络的变压器振动信号分析
Analysis on Transformer Vibration Signal Based on Convolutional Neural Networks
摘要
Abstract
In order to study relationship between vibration and operating state of the transformer,the paper makes analysis on transformer vibration signal on the basis of wavelet analysis method combining with the convolutional neural networks. It firstly makes simulation on two failure states including winding looseness and iron core looseness of one oil-immersed trans-former and respectively measures vibration signals. Then it makes wavelet transform of measured vibration signals and cre-ates wavelet gray scale maps,and conducts the convolutional neural networks training analysis. The training results indicate accuracy of this method is 84.03%,which explains the analysis method combining the convolutional neural networks and wavelet gray scale map can effectively recognize fault information in vibration signals. After comparing distribution of error results of verification samples,it discovers error results are greatly affected by locations of measuring points. It points out in the premise of improving measuring points and increasing training data,accuracy rate can be promoted.关键词
卷积神经网络/小波分析/灰度图/变压器/振动信号Key words
convolutional neural networks/wavelet analysis/gray scale map/transformer/vibration signal分类
信息技术与安全科学引用本文复制引用
苏世玮,郭盛,高伟,杨涛,赵家毅..基于卷积神经网络的变压器振动信号分析[J].广东电力,2018,31(6):127-132,6.基金项目
国网湖北省电力公司科技项目(52191614004V) （52191614004V）
