低频振荡主导模式的滑窗谱分析方法  

Low frequency oscillation sliding window spectrum analysis method for dominant modes of low frequency oscillation

作　　者：竺炜[1,2] 蒋頔[1] 马建伟[3] 曾喆昭[1] 

机构地区：[1]长沙理工大学电气与信息工程学院,湖南长沙410004 [2]中国电力科学研究院,北京100192 [3]贵阳供电局,贵州贵阳550001

出　　处：《电力科学与技术学报》2013年第1期48-55,共8页Journal of Electric Power Science And Technology

基　　金：国家自然科学基金(61040049);湖南省自然科学基金(11JJ6032);湖南省科技计划项目(2010FJ4095)

摘　　要：实际电力系统低频振荡复杂,具有多模式且模式时变的特点,但在秒级时间窗内,仍可采用非时变特征根来描述机电振荡模式.采用滑窗后谱分量比较的办法,解决阻尼识别和模式变化判别问题;针对振荡带宽较窄的特点,采用最小二乘递推的傅里叶基神经网络谱分析方法提高抗干扰能力,并从窗口权值分析得到主导模式的频率;通过滑窗训练,识别各模式的阻尼和幅值以及模式的变化.开窗和滑窗分析符合实测数据在线分析的实际过程;对频谱分析方法的改进,即保留了原有工程经验,又解决了实际问题.仿真表明:该方法在干扰和多模式的情况下抗干扰性强,模式识别准确.Power system low frequency oscillation are complex, and often have the characteristics of multi-mode and mode changing, whereas, it can still use the non-variable characteristics roots to describe the electromechanical oscillation modes in second time window. Through the compari- son of sliding window spectral components, the damping and the changing of mode can be identi- fied. Because of LFO narrow bandwidth, in order to improve the noise immunity, spectrum anal- ysis method based on Fourier Basis Functions neural network are used. The dominant mode fre- quency can be calculated by the weights. Through the sliding window training, the damping, magnitude and mode changes can be identified. Sliding-window analysis are suitable for the meas- ured data on-line analysis process. The improved spectral analysis method not only retains theoriginal engineering experience, but also solves practical problems. In the case of interference and multi-mode, simulation results show that the method has good noise resistance characteristics, and identifies more accuracy.

关 键 词：低频振荡 主导模式 滑窗 谱分析 神经网络 

分 类 号：TM711[电气工程—电力系统及自动化]