【作者】 段大鹏；

【导师】 孙才新； 江秀臣；

【作者基本信息】 上海交通大学 ， 高电压与绝缘技术， 2009， 博士

【摘要】 气体绝缘金属封闭开关设备(Gas Insulated Switchgear-GIS)也称作气体绝缘变电站(Gas Insulated Substation-GIS),以其结构紧凑、运行可靠、易于维护等优点而在电力系统中得到广泛应用。但由于GIS结构复杂、制造与检修工艺精细,使部分常规的预防性试验方法难以有效实施,因此,GIS一旦发生故障后果相当严重。为了及时掌握GIS的绝缘状态,保证GIS的安全可靠运行,国内外对GIS局部放电检测技术开展了大量的研究,其中超高频(Ultra High Frequency-UHF)局部放电(Partial Discharge-PD)检测方法具有优越的抗干扰性能与较高的检测灵敏度,成为近年来的研究热点。然而,到目前为止,GIS中的UHF PD检测还是一个尚待深入研究的前沿课题,其难点主要包括:GIS中PD机理及其电磁波传播特性、稳定可靠的检测系统研制、UHF PD信号特征空间的构造与降维以及合理的缺陷类型模式识别方法等。论文从GIS局部放电机理、PD实际电流激发的电磁波以及GIS实际结构条件下电磁波传播特性出发,通过GIS PD检测试验,应用自主开发的UHF PD检测系统获得了大量试验数据,并在此基础上提出了UHF PD包络信号的特征空间构造与降维方法,应用仿生模式识别理论对GIS中的典型绝缘缺陷类型进行了成功识别。论文首先对GIS中UHF PD检测的国内外研究与应用现状进行了综述,概括阐述了GIS内部电磁波传播特性、UHF PD信号分析模式、缺陷类型模式识别以及UHF PD检测系统的研究与应用等方面的研究进展及存在的问题。然后详细分析了GIS中典型绝缘缺陷引发的PD的放电机理,并应用时域有限差分算法,以PD实际电流脉冲作为激励源,研究了GIS实际结构对电磁波传播的影响。仿真结果表明,PD实际电流脉冲与理想高斯脉冲激励的电磁波在上升时间、振荡衰减以及频率成分方面都具有明显差异,而且GIS中的盆式绝缘子、三相共筒以及断路器等实际结构均会对电磁波产生幅值、频率等方面的影响。然后,论文共设计制作了5种典型绝缘缺陷模型,并分别在正泰公司生产的126kV GIS与SIEMENS公司生产的252kV GIS上进行了局部放电UHF PD检测试验,应用自主研制的GIS UHF PD检测系统录取了大量试验数据。进而,论文以UHF PD时域包络信号为分析对象,提出了改进的自适应小波去噪算法及特征空间构造方法,将Weibul分布参数及Mel复倒谱系数引入特征空间,在时域、频域与分布域共提取37个特征参数。在此基础上,提出了改进的核主成分分析法与基于相关系数矩阵的降维算法,同时引入基于粗糙集理论的属性约简方法,用于特征空间的降维。并指出了三种降维算法的适用场合。最后,论文将仿生模式识别理论首次引入局部放电模式识别领域,用于GIS典型绝缘缺陷类型的模式识别。该方法以“认识”及“同源连续性”为基点,完全打破了以往基于“划分”的模式识别思想,具有崭新的特色。本文详细论述了仿生模式识别的基本思想、理论基础与实现方法,在前人研究的基础上提出了新颖的理论解释与计算方法,并与传统的BP神经网络及支持向量机分类方法进行了比较,显示了仿生模式识别方法的优越性。论文结尾,对全文进行了总结,并对下一步的研究工作作了展望。本文的研究成果对于GIS中局部放电机理、UHF PD检测与缺陷类型模式识别及其对GIS运行与检修的指导都具有重要理论价值与现实意义。更多还原

【Abstract】 GIS (Gas Insulated Switchgear or Gas Insulated Substation) has been widely used in electrical power system because of its advantages such as compact structure, reliable operation and easy maintenance. At the same time, there are several disadvantages of GIS including complex structure, precious manufacture and maintenance technologies and the integrated way to install of various equipments. So that the conventional preventive test methods are hard to effectivly implement. Therefore, in case of GIS failure event, it would be very serious. In order to ensure the safe and reliable operation of GIS, a substantial amount of research works have been carried out on the GIS PD (partial discharge) detection techniques at home and abroad. One UHF (Ultra-High Frequency) PD detection method, which has superior anti-disturbance performance and high sensitivity, has become the research focus in recent years.So far, UHF PD detection of GIS is still a advancing front issue to be researched more thoroughly. The difficulties mainly include the PD mechanism and the propagation of EMW (ElectroMagnetic Wave) in GIS, development of stable and reliable detection system, construction and dimension reduction of feature space based on UHF PD signals, reasonable methods of pattern recognition for defects type in GIS, and long-term accumulation of data from experiments and local sites. In this paper, based on the PD mechanism and the propagation of EMW excited by PD current pulse in real condition of GIS, a large number of experimental data were acquired by the developed UHF PD detection system in 126kV and 252kV GIS respectively. Afterwards, using the evelope of UHF PD signals, the feature space construction and dimension reduction methods were proposed. At last, the biomimetic pattern recognition was used to classify the typical insulation defects.Firstly, the current research and application situation of UHF PD detection in GIS was reviewed. The research status and problems were summarized including characteristics of EMW propagation within GIS, UHF PD signal analysis model, defects pattern recognition, as well as detection system research and application. Afterwards, the PD mechanism excited by typical insulation defects in GIS was analyzed, and the FDTD (Finite-difference time-domain) was used to simulated the propagation of EMW excited by actual PD current pulse in the actual GIS constructure. The simulation results show that differences of rise time, oscillation attenuation, as well as frequency component existed between PD current pulse exciting and ideal Gauss pulse exciting. In addition, the existence of basin spacer, shields and three conductors in actual GIS should effect the EWM propagation such as model transfer. Subsequently, five defect models were designed and set in 126kV GIS and 252kV GIS respectively, through PD experiments, a large number of UHF PD data was acquired by the developed detection system. After that, an improved adaptive wavelet threshold was proposed to filter the white noise in UHF PD signals. At the same time, a novel feature space were presented with 37 feature parameters. To reduce the dimension of the feature space, the improved KPCA (Kernel Principal Component Analysis) and the proposed CCMDR (Correlation Coefficient Matrix Dimension Reduction), as well as the RST (Rough Set Theory) were used to depress the sapace dimension. The three algorithms have their own advantages and achieved outstanding effects.Finally, the BPR (Biomimetic Pattern Recognition) was introduced for the first time into the fields of PD pattern recognition even into electrical engineering. Based on the“understanding”and“homologous continuity”, BPR completely broke the tranditional pattern recognition methods based on“division”with new features. After the discussion of basic idea, theory and implementation of BPR, a novel algorithm was improved to realize BPR and was used to classify the typical defects. The comparison with BP neural network and support vector machine demonstrates the superiority of BPR. Further more, conclusion and prospect were presented. The research results in this paper have important theoretical value and practical significance for PD mechanism, UHF PD detection, defects recognition research as well as guidance of GIS operation and maintenance.更多还原