基于荧光光纤检测GIS局部放电的多重分形谱识别  

Multifractal Spectrum Identification of Partial Discharge in GIS Based on Fluorescence Optical Fiber Detection

作　　者：唐炬[1] 曾福平[1] 范庆涛[1] 刘永刚[1] 张晓星[1] 

机构地区：[1]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,重庆400030

出　　处：《高电压技术》2014年第2期465-473,共9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(5177181)~~

摘　　要：为了获取SF6气体绝缘电器(GIS)局部放电(PD)光信息特征,完善了反映GIS内部4种典型绝缘缺陷的物理模型,并利用研制的PD荧光光纤传感系统,获取了用绝缘缺陷模型产生的PD光信号,构建出了相应的φ-u-n三维图谱及其灰度图像。同时,针对简单分形特征不足以描述PD灰度图像而导致模式识别率较低的问题,采用具有分析灰度图像丰富特征信息能力的多重分形谱技术,提出了一种计算PD光信号灰度图像的多重分形谱概率算法。然后通过对多重分形谱中各特征信息的物理意义分析,提取了用于PD模式识别的特征量,并选用改进共轭梯度算法的反向传播(BP)神经网络作为分类器。研究结果表明:多重分形谱可有效描述PD灰度图像不均匀程度及其不同层次的几何结构特征,用于分类识别4种典型绝缘缺陷产生的PD时正确率均>87%,优于用盒维数与信息维数作为特征量的识别。In order to obtain the optical information characteristics of partial discharge （PD） in gas insulated switchgears （GISs）, we improved the physical models of four kinds of artificial insulation defects in GIS, obtained optical signals of PD using homemade fluorescence optical fiber sensor system, and built φ-u-natlas and grayscale images for the optical signals. To solve the problem that simple fractal characteristics are unable to describe grayscale images of PD, which leads to low pattern recognition rates of PD, we proposed a probability calculation method for calculating multi-fractal spectrogram of PD grayscale image to analyze the rich characteristic information of the grayscale images. After analyzing the physical significance of the features of obtained multi-fractal spectrum, we extracted some main features and use them as PD pattern recognition feature, and adopted the BP neural network with improved conjugate gradient algorithm as the classifier for identifying different PDs. The rates of successful recognitions are all more than 87% for all four kinds of PD, which is apparently improved comparing with using box-counting dimension or information dimension as characteristic feature in PD loattem recognition.

关 键 词：气体绝缘开关设备 光测法 局部放电 灰度图像 多重分形谱 模式识别 

分 类 号：TM595[电气工程—电器] TP391.41[自动化与计算机技术—计算机应用技术]