摘要
GIS局部放电模式识别的研究对于保证GIS的安全可靠运行有着十分重要的意义。为此,采用具有优良平移时不变特性的二元树复小波变换对超高频局部放电(UHF PD)信号进行分解,提出了一种采用二元树复小波变换的时频域特征提取方法,阐明了该算法的原理,给出了具体计算公式和步骤;通过伸缩和平移等运算功能对实验室获得的大量UHF PD信号进行多尺度细化分析,综合选用了UHF PD信号在各频带投影序列的能量、在各个尺度下的模极大值和统计参量,构造了完整的UHF PD信号特征空间;并将UHF PD信号特征空间分别输入到RBF组合神经网络分类器的成员分类器中,获得了优良的识别效果,总体识别率>86%。
In this paper,a new method of feature extraction of UHF PD signal which is on basis of dual-tree complex wavelet is proposed and the principle,formulas and algorithm of the method are presented in detail.UHF PD signals which are acquired in lab are analyzed by translation and scaling.The feature space is constructed on basis of the energy on all frequency bands and modulus maxima on all scales.And then feature space is inputted into a component classifier of combined neural network classifier.It is proved that this method are effective way of recognition of UHF PD signals and the recognition rate of UHF PD signals is about 86%.
出处
《高电压技术》
EI
CAS
CSCD
北大核心
2010年第3期553-558,共6页
High Voltage Engineering
基金
国家重点基础研究发展计划(973计划)(2009CB724506)
国家自然科学基金(50577070)~~