Study On Identification Of Power System Low Frequency Oscillation

In recent years, with the continuous expansion of power system, interconnection increases between power grids. At the same time, more and more fast excitation systems are being installed in the power system. All the above facts make the problem of low frequency oscillation more serious. Weak damping, weak link between power grids, fast excitation system applications and heavy loads have become the important reasons for the low frequency oscillation. Problem of low frequency oscillation in power system has been studied in deep, and usage of the WAMS/PMU can provide us the real-time data of generators, lines, transformers, load angle, power flow, and et al. It can help us to derive new analysis technique for low frequency oscillation.Prony algorithm considers any oscillation curves are composed by the linear combination of a series of exponential function with special amplitude, phase, frequency and constringent factors. It can directly extract the vibration signal characteristics, providing the basis for oscillation mode and damping analysis. So it has been widely used in the low frequency oscillation identification, load modelling, damping controller design, et al. However, the traditional Prony algorithm is very sensitive to the noise in the analysis signals, this limits its applications in power system.This thesis aims at finding new low frequency oscillation analysis method through combining the higher-order cumulant(HOC) method with Prony algorithm. HOC method is an efficient and important mathematical tool for modern signal processing. Since all the higher order cumulant of Gaussian serials equal to zero, HOC can suppress Gaussian noise completely in theory. It has been found that it has better noise performance than the traditional power spectrum or correlation function. Since Prony algorithm's poor antinoise performance and the higher-order cumulant's characteristics of suppressing noise, a new analysis method based on HOC method and Prony algorithm is proposed in this thesis. Typical signal serials and New England 39 nodes system are used to validate the effectiveness of the presented method in the low frequency oscillation analysis.