| With the rapid development of electric power industry in our country, the scale of power system expands continuously and the interconnected grid is formed. Although the operation efficiency and reliability of generation and transmission system can be improved in the interconnected grid, the low frequency oscillation would be caused after multiple regional grid interconnections. The process of low frequency oscillation has strong randomness and nonlinearity. Studying the dynamic characteristics of low frequency oscillation can not only describe the system's nonlinearity and time-varying features, but also reveal the oscillation modes and propagation characteristics. Effective tracking power system's oscillation mode can forecast the unstable phenomena accurately and provide important references for system analysis and emergency control. However, oscillation modes are diverse and oscillation process is complicated with various external noises mixed, which make the existing analysis methods for low frequency oscillation dynamics ineffective. Therefore, studying the novel methods for analyzing low frequency oscillation dynamic characteristics has important theoretical and practical application values. In this dissertation, the novel methods using modern signal processing techniques were proposed, which could be applied for modal parameters identification, dominant modes detection and modal nonlinearity interaction study. The main contributions of this dissertation were summarized as follows.First of all, the research statuses of low frequency oscillation were introduced including the phenomena, mechanisms and analysis methods. The applications of wide area measurement system (WAMS) in the low frequency oscillation analysis are parameters identification, safe warning and disturbance source location.The basic concepts of atom decomposition (AD) theory were introduced and two major factors affecting the AD algorithm were focused on. The factors were selections of atom library and matching pursuit methods. The novel modal parameter identification method based on AD was proposed and the transition process from Gabor atom to oscillation atom was also introduced. In order to improve the efficiency of pattern recognition, modal atom method was proposed for revealing the oscillation modes'time-varying performance. This method applied particle swarm optimization (PSO) to the atom decomposition process and the best modal atoms were selected by comparing the particles'best values. Thus, the process for selecting modal atoms was turned into solving the function optimization problems by PSO. This method can adaptively track the oscillation modes' time-varying process and has the abilities of anti-noise and robustness.The basic concepts and principles of Hilbert-Huang transform (HHT) were stated, including the empirical mode decomposition (EMD)'s basic principle, intrinsic mode function (IMF)'s definition and Hilbert tansform's theory. The conventional HHT's shortages such as the problems of sampling rate settings, loop termination conditions, the envelope fitting, boundary effects and modal aliasing were analyzed and the corresponding improvements were also introduced. A hybrid improved-HHT (HI-HHT) was presented, which adopted pre-morphology filter unit to remove high frequency noises and applies the modal intensity index to indicate the modal aliasing. In order to avoid the bad effects of improper heterodyne frequency selection in frequency heterodyne, the inverse damping factor was added for improving the modal frequency resolution and the effectiveness of EMD was also enhanced. The robust mirror extension method was applied for handling boundary effects and Hermite interpolation was used for the curve fitting. The adaptive correlation coefficient method was used for EMD termination. The decomposition process would terminate if the requirements for threshold could not be met and the false component would be eliminated. The HI-HHT method can overcome the shortcomings of traditional EMD by which the low frequency oscillation modal parameters can be identified. Case study indicates that the HI-HHT is valid and reliable.Typical detection methods for low frequency oscillation dominant mode were summarized. The basic concepts and principles of eigenvalue method, normal form method, Prony energy method and EMD signal energy method were introduced in detail. A novel detection method named atom decomposition energy entropy (ADEE) for dominant mode was put forward. According to oscillation signal characteristics, damped sinusoid model was selected to represent atomic library. Using the power angle trajectory, the modal parameters were identified from the atomic library by AD method and ADEE was calculated during the identification process. By comparing the energy entropies, the dominant inertial modes were identified. This method, unrestricted from the system order, has a strong ability of data processing and can reveal the complex dynamic characteristics and nonlinear effects among the modes. So the method can be applied to low frequency oscillation on-line analysis. The accuracy and effectiveness of the proposed method were validated by simulation cases. Higher order spectrum theory in statistical signal processing was presented. According to this theory, the intrinsic characteristics of signal including signal amplitude and non-minimum phase information could be preserved and it could be used to analyze the system's nonlinear process. The basic theory and properties of higher order spectrum were focused on. There were two kinds of methods, nonparametric method and parametric method respectively in higher order spectrum estimation. A novel algorithm for analyzing low frequency oscillation modes nonlinear interactions was proposed to avoid the disadvantages of the eigenvalue analysis method and the traditional nonlinear numerical solutions. The eigenvalue analysis method could not reveal the compound modes in oscillatory processes following large perturbations, and the traditional nonlinear numerical solutions had truncation errors and low computational efficiency. The non-parametric direct method of higher order spectrum theory was applied to analyze low frequency oscillation modes and the quadratic phase coupling information among modes were identified by using bispectrum estimation and bicoherence analysis. The method can acquire the nonlinear coupling degree between modes quantitatively which provides a new way to study the stability and the dynamic behavior of power systems with strong nonlinearity. The feasibility and effectiveness of the proposed approach were verified by the numerical simulation results.Finally, the innovative results of this dissertation were summarized and a perspective on the future research work was proposed. |
PDF Full Text Request