Online Estimation Of Power System Low Frequency Oscillation Mode Based On Hybrid Method Of Time-Frequency

With the continuous development of interconnected power system, low-frequencyoscillation(LFO) surfaces from time to time and becomes one of the critical factors thatlimit power transfer and jeopardize the safety and stability of grid operation. Effectivesupervising and controlling to LFO is an impotant component part of self-healing abilityin smart power grids. the time-domain algorithm based on Prony, one of theconventional LFO analysis mostly used for offline condition, lacks the ability oftracking LFO mode under normal operating condition. Thus, in this thesis the problemsof how to realize the method based on frequency-domain decomposition(FDD) andProny for LFO mode identification in power system is analyzed following the order thatrealization of FDD、nonlinear detrend method for dynamic data and detection algorithmof disturbance.①The dynamic data from power system are analyzed for removing nonlinear trend.To overcome the shortcomings of low speed calculation and big relative error in currentdetrend algorithm, a new detrend algorithm based on smoothness prior approach (SPA)is proposed in this thesis, which applies SPA principle and selects appropriateregularization parameter. This proposed method can successfully remove nonlineartrend from the dynamic data and improve the speed of computation, as well as theprecision of mode identification, which is suitable for online identification especially.②To deal with the disturbance detection of active power data.This thesis proposesa new strategy based on mathematical morphology for detection of dynamic data inpower system, taking morphological gradient in first-order derivation method and softthreshold function value in the soft threshold processing as indicator function. It utilizesamplitude changes of these indicator function to determine the start and end data point,which improves the level of dynamic data detection effectively.③The improved FDD algorithm for LFO mode identification based on the relationexpression, between maximum singular value of PSD matrix and system eigenvalue,resolves the solution of modal parameters. Based on the FDD principle, this thesisproposes an improvement of adopting modal amplitude coherence(MAC) to determinemode area size in the neighborhood of peak point on maximum singular valuedecomposition curve. It calculates exact modal frequency and damp ratio with leastsquare method simply and effectively, which avoiding the loss of precision caused by inverse Fourier transforms in conventional frequency-domain method.④The proposed methods are implemented based on Matlab program and tested bythe data from simulation model and PMU in some grid, and all the results demonstrateeffectiveness and feasibility of these methods. Finally, according to the research above,it accomplishs the application of LFO online identification based on hybrid method withimproved FDD and Prony.