电力电缆局部放电监测与识别

发布时间：2018-04-14 22:40

本文选题：电缆 + 局部放电　；参考：《中国矿业大学》2016年硕士论文

【摘要】：随着电力电缆在电网中的应用越来越广泛,电缆绝缘状况成为研究人员研究的热点,局部放电是绝缘状况劣化的一个重要的外部表现形式,同时,局部放电会导致绝缘状况进一步劣化。为了及时发现电缆的绝缘缺陷,保障电网正常运行,有必要对电缆局部放电进行在线监测。国内外学者对电缆局部放电在线监测做了相关的探索研究,本文在此基础上,对其中的几个关键环节进行了研究探索。本文首先阐述了电缆局部放电机理,接着说明了几种常用的检测方法,分析检测现场存在的干扰源和常见干扰的性质,依据局部放电的特点,选取4种放电数学模型,为信号仿真提供依据。并进一步分析电缆绝缘缺陷的形成原因,并对4种典型绝缘缺陷进行模拟实验,获取大量的4种绝缘缺陷下的放电信号,为电缆局部放电模式识别提供样本。电缆局部放电在线监测技术的第一个关键环节就是检测信号的预处理,也就是对检测放电信号中混入的噪声进行抑制,本文重点对背景噪声和周期性窄带干扰噪声的消除进行了研究,根据电缆局部放电信号和噪声干扰信号的特点,选取小波包分析法对白噪声进行抑制,小波包阈值法是去噪的常用手段,基小波的选取和阈值处理是小波包去噪的关键,本文从阈值函数入手,通过对传统阈值函数处理方法的分析,提出了一种改进阈值函数的小波包去噪算法,仿真和实际应用表明,改进算法去噪效果优于传统算法。考虑到周期性窄带干扰频率固定,适合在频域对其进行分析,本文采用FFT技术对窄带干扰进行抑制,仿真结果表明,FFT技术在去除周期性窄带干扰方面具有优越性。电缆局部放电在线检测的第二个关键环节就是对局部放电的识别,检测到的电缆局部放电信号可能是由4种不同绝缘缺陷造成的,正确识别放电类型对于迅速做出判断,采取相应措施至关重要。模式识别的输入对象往往不是放电信号本身,而需要我们从局部放电信号中提取特征向量,作为模式识别的输入,所以说特征量能否表征不同局部放电信号的特征非常重要。本文根据电缆局部放电信号的特点,对放电信号分别进行经验模态分解和小波包分解,然后分别计算固有模态函数的能量系数和小波包分解系数的频带能量作为放电信号的特征向量,分别采用BP神经网络算法和模糊C-均值聚类算法进行识别。结果表明:固有模态分量能量更能表征放电信号特征;对同一特征向量,BP神经网络算法识别率整体高于模糊C-均值聚类算法。
[Abstract]:With the increasing application of power cable in power network, cable insulation has become a hot research topic. Partial discharge (PD) is an important external manifestation of insulation deterioration, and at the same time,Partial discharge leads to further deterioration of insulation.In order to find out the insulation defect of cable in time and ensure the normal operation of power grid, it is necessary to monitor the partial discharge of cable on line.Scholars at home and abroad have made relevant research on on-line monitoring of partial discharge of cables. On the basis of this, several key links have been studied and explored in this paper.In this paper, the mechanism of partial discharge of cable is first expounded, then several commonly used detection methods are explained, and the interference sources and properties of common interference are analyzed. According to the characteristics of partial discharge, four kinds of mathematical models of discharge are selected.It provides the basis for signal simulation.Furthermore, the causes of cable insulation defects are analyzed, and four typical insulation defects are simulated. A large number of discharge signals under four insulation defects are obtained, which provide samples for partial discharge pattern recognition of cables.The first key step in the on-line monitoring of cable partial discharge is the preprocessing of the detection signal, that is, the suppression of the mixed noise in the detection discharge signal.This paper focuses on the elimination of background noise and periodic narrowband interference noise. According to the characteristics of cable partial discharge signal and noise interference signal, wavelet packet analysis method is selected to suppress white noise.Wavelet packet threshold method is a common method for denoising, and the selection of wavelet base wavelet and threshold processing are the key of wavelet packet denoising. This paper starts with threshold function and analyzes the traditional threshold function processing method.An improved threshold function wavelet packet denoising algorithm is proposed. Simulation and practical application show that the improved algorithm is better than the traditional algorithm.Considering the fixed frequency of periodic narrowband interference, it is suitable to analyze it in frequency domain. In this paper, FFT technique is used to suppress narrowband interference. The simulation results show that FFT technique is superior in eliminating periodic narrowband interference.The second key step in the on-line detection of cable partial discharge is the recognition of partial discharge. The detected partial discharge signal may be caused by four different insulation defects.It is essential to take appropriate measures.The input object of pattern recognition is usually not the discharge signal itself, but we need to extract the feature vector from the partial discharge signal as the input of pattern recognition, so it is very important to say whether the characteristic quantity can represent the characteristics of different partial discharge signal.According to the characteristic of cable partial discharge signal, the discharge signal is decomposed by empirical mode decomposition and wavelet packet decomposition, respectively.Then the energy coefficients of the inherent mode function and the band energy of the wavelet packet decomposition coefficient are calculated as the eigenvectors of the discharge signal respectively. The BP neural network algorithm and the fuzzy C-means clustering algorithm are used to identify the discharge signals.The results show that the inherent modal component energy can better characterize the characteristics of the discharge signal and the recognition rate of the BP neural network algorithm for the same eigenvector is higher than that of the fuzzy C-means clustering algorithm.
【学位授予单位】：中国矿业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TM855

【相似文献】