基于二元树复小波变换与流形学习的GIS局放信号识别研究.pdfVIP

基于二元树复小波变换和流形学习的GIS 局 放信号识别研究 1 1 2 周倩 ，游步新 ，谢颜斌 1. 重庆市电力公司江北供电局，401147 ；2. 重庆大学输配电装备及系统安全与新技术 国家重点实验室，400030 CLASSIFICATION FOR UHF PARTIAL DISCHARGE SIGNALS IN GIS BASED ON DUAL-TREE COMPLEX WAVELET TRANSFORM AND MANIFOLD LEARNING 1 1 2 Zhou Qian , You Buxin , Xie Yanbin 1. Chongqing Jiangbei Power Supply Bureau, Chongqing Power Corporation, 401147, 2. State Key Laboratory of Power Transmission Equipment System Security and New Technology, Chongqing University, 400044 ABSTRACT: In this paper, a new method of KEY WORD: Dual-tree complex wavelet feature extraction of UHF PD signal which is transform, Partial discharge, Manifold learning on basis of dual-tree complex wavelet is 摘要：本文采用具有优良平移时不变特性的二元树 proposed and the principle, formulas and 复小波变换对 UHF PD 信号进行分解，提出了一种 algorithm of the method are presented in detail. 基于二元树复小波变换的时频域特征提取方法，给 UHF PD signals which are acquired in lab are 出了具体计算公式和步骤；通过伸缩和平移等运算 analyzed by translation and scaling. The 功能对实验室获得的 UHF PD 信号进行多尺度细化 feature space is constructed on basis of the 分析，综合选用了 UHF PD 信号在各频带投影序列 energy on all frequency bands and modulus 的能量、在各个尺度下的模极大值和统计参量，构 maxima on all scales. 造了完整的 UHF PD 信号特征空间；根据流形学习 The energy of frequency band of each 的基本原理，提出了基于核主分量分析的 UHF PD projection sequence, modulus maximum and 信号特征空间降维方法，在保证原有信号的特性最 statistical parameters in each scale compose 大限度的保持不变的情况下，有效地降低了维数空 the whole feature space of UHF PD signals. 间，将降维处理后特征向量作为径向基神经网络分 On the basis of the principle of manifold 类器的输入向量，获得了优良的识别效果。 learning, a method of dimension reduction of 关键词：二元树复小波变换，局