| Power system low frequency oscillation(LFO) is inherent problems of interconnection of power system, which seriously restricts power transmission capability and threatens the safety and stability of power greid. The growing installation of phase measurement units(PMU) in the power systems makes it possible to analyze low frequency oscillation.PMU data reflects the current operating status of the system and does not require mathematical modeling of complex systems, which efficiently complement traditional methods based on detailed system model. Most detecting low frequency oscillation methods based on wide-area measurements PMU measured data often simply consider gaussian white noise as background noise, ignoring the influence of gaussian colored noise.Based on previous expert, first the paper studied the gaussian white noise sensitivity of modal identification algorithms. Second, we focuses on the research on modal analysis of power system low frequency oscillation considering gaussian colored noise. Details are as follows:First, the identification efficiency of Prony, Hankel Total Least Squares, Matrix Pencil and SVD-Prony were studied at different levels of white Gaussian noise. In view of the good noise immunity of HTLS and MP, SVD filter of the sampling data directly established hankel matrix combining with Prony, SVD-Prony algorithm was proposed with better noise immunity. Prony, HTLS, MP and SVD-Prony algorithm were verified by deal test signal, four machine two areas simulation system and a set of real PMU measurement obtained from Hz grid.Second, modal analysis methods of LFO based on were focused on High-order Mixed Cumulants(HOMC) considering gaussian colored noise. Fourth-order Mixed Cumulants in infinite data and single recording limited data were derived in detail. Withthe advantages of blindness to gaussian noise of HOS, FOMC-Prony?FOMC-HTLS and FOMC-MP were proposed, which can suppresse Gaussian noise(including colored gaussian noise) and obtain the original signal modal information. The analysis of deal test signal and two set sof real PMU measurements obtained from Hz grid indicates that proposed methods can suppress Gaussian colored noise and promote the accuracy of detection results.what is more they gain modal information of original signal in FOMC algorithm, realizing quantitative evaluation of proposed algorithms.Besides, modal analysis methods of LFO based on were focused on Cross Correlation Function(CCF) considering gaussian colored noise. CCF in infinite data and single recording limited data were derived in detail at background of gaussian noise(including colored Gaussian noise). With the property that original signal and its CCF hold same oscillation Frequency and damp, and CCF can restrain Gaussian colored, CCF-Prony,CCF-HTLS and CCF-MP were proposed, which can suppresse Gaussian noiseand obtain the original signal modal information. The analysis of deal test signal and two sets of real PMU measurements obtained from HZ grid indicates that proposed methods can suppress Gaussian colored noise and promote the accuracy of results.what is more they gain modal information of original signal in FOMC algorithm, realizing quantitative evaluation of proposed algorithms.Finally, modal parameters is not easy to obtain since the modals of FOMC and CCF sequence are different from original signa. Besides, gaining modal information utiliting total least squres has some shortcomins that the precision is inadequate, and morbid issues occur at the process of calculation. In order to overcome these shortcomings, obtaining modal methods based on Adaline neural network were put up. The paper gives the parameters detailed steps of calculating original signal with the help of Adaline neural network in the background of Gaussian colored noise, when frequency attenuation factor are known. Four algorithms including FOMC-HTLS/MP-Adaline and CCF-HTLS/MP-Adaline were proposed. Similarly, the algorithms were verified by deal test signal and two sets of real PMU measurements obtained from HZ grid. Besides, the paper explores the relationship between the value of learning rate ? and regulation accuracy and convergence rate. lastly, all proposed methods were for comparison from the view of operational efficiency and fitting accuracy. Results indicates that FOMC-HTLS,CCF-HTLS, FOMC-HTLS-Adaline and CCF-HTLS- Adaline have a higher operational efficiency than others, and Adaline Neural network algorithms related algorithms, such as FOMC-HTLS-Adaline?FOMC-MP-Adaline?CCF-HTLS- Adaline and CCF-MP-Adaline,own higher accuracy than others. |
PDF Full Text Request