【作者】 王小宇；

【导师】 贾嵘；

【作者基本信息】 西安理工大学 ， 电力系统及其自动化， 2007， 硕士

【摘要】 电力系统的高可靠性要求诊断系统能及时发现水轮发电机组的故障，并给出调整建议；同时要求诊断系统能根据机组状况给出优化运行方案。传统的基于傅立叶变换的故障诊断方法只能从频率的角度发现机组的故障，不能准确定位机组发生故障的时间及故障发展趋势。本文以水轮发电机组状态监测与故障诊断方法为研究对象，深入研究了Hilbert-Huang变换理论基础的上，将其引入水轮发电机组故障诊断系统中。该方法首先对机组振动信号进行经验模态分解(EMD)获得固有模态函数(IMF)分量，再经过Hilbert变换获得信号的Hilbert谱，从信号的Hilbert谱中即可发现信号中的异常频率及其发生的时间，从而确定机组的故障和发生时间。Hilbert-Huang变换的端点效应限制了其应用，本文对端点效应的原理做了详细的分析后提出了两种解决端点效应的方法：可变长极值镜像拓延法避免了镜像闭合拓延法造成数据量庞大的缺点，按照实际情况选取拓延极值长度，取得较好效果；基于最小二乘支持向量机(LS-SVM)的端点拓延方法结合了水轮发电机组暂态数据短的特点和最小二乘支持向量机在小样本下可获得最优解的优点，较好的抑制了端点效应。信号特征提取方法是故障诊断的难点，本文针对水轮发电机组振动信号的特点，提出两种信号特征提取方法：基于EMD和AR模型的方法从波形入手提取信号特征，该方法将每个固有模态函数(IMF)分量提取为4个参数作为智能识别系统的输入；基于能量的特征提取方法是将每个IMF分量的能量作为智能识别系统的输入。这两种方法可以将信号的特征提取为数值特征，为故障的智能识别奠定了基础。本文最后将Hilbert-Huang变换应用于实际中，对贵州索风营电站1号机组进行了诊断，分析其性能和振动原因，并尝试将最小二乘支持向量机分类方法应用于水轮发电机组故障智能识别系统，得出相应的结论。结果表明，基于Hilbert-Huang变换和支持向量机的水轮发电机组状态监测与故障诊断方法能对机组性能做出较好的评价，准确定位机组的故障，值得推广应用。更多还原

【Abstract】 High reliability of electrical power system demands that a fault diagnosis system should find out failures of a water-wheel generator set in time, make a suggestion to correct them and give a optimistic scheme for operating as the states of the generator set. However, the traditional fault diagnosis method based on Fourier transform(FT) can only find failures from frequencies of vibration signals and can not know what time and how the failures will happen.This thesis investigates methodology of the state monitoring and fault diagnosis of a water-wheel generator set. Firstly the Hilbert-Huang transform is deeply studied. After that, it is applied into a fault diagnosis system of a water-wheel generator set. By this method, signals are decomposed by empirical mode decomposition (EMD) firstly and the intrinsic mode functions (IMFs) are obtained. Then Hilbert spectrum is obtained by Hilbert transform. Abnormal frequencies and their occurring time can be discovered from the Hilbert spectrum. So do the failures of the generator set.Applications of the Hilbert-Huang transform are restricted because of its point effect. Armed with analyzing the principle of point effect in detail, two methods of dealing with it are presented. The one is an alterable length extremism mirroring extension algorithm. The algorithm does not like the close mirroring extension algonithm which can produce an amount of data. Moreover, it has good effect by taking extension extremism length in the actual condition. The other is a point extension algorithm based on the least square support vector machine. This algorithm makes full use of the character of short transient vibration signal data of the water-wheel generator set and optimal advantage of the least square support vector machine with limited samples. So it can solve the point effect more effectively.It is difficult to distill signal characters for the fault diagnosis. With the character of vibration signals of a water-wheel generator set, the thesis gives two algorithms to distill signalcharaters. The first algorithm is based on EMD and AR mode to distill signal characters from the wave form. This algorithm takes four parameters as inputs of an intelligent recognizing system from every IMF; while the second algorithm is based on energy and it is to make energy of IMF as inputs of the intelligent recognizing system. Both the algorithms can transform signal characters into digital characters and they are the basis of the fault diagnosis.At the end of this thesis, Hilbert-Huang transform is used in fault diagnosis system of No.1 water-wheel generator set of SUO FENG YING Its performance and vibration reasons are analyzed. Least square classification vector methane is applied to the fault diagnosis intelligent recognizing system of the water-wheel generator set and results are analyzed. They indicate that state monitoring and fault diagnosis system based on Hilbert-Huang transform and support vector machine can give a good estimate for the performance of water-wheel generator set and locate the failures of the generator set. Thus, it is worth of spreading and application.更多还原