The Research On Fourier And Neural Network Analysis Methods Of Low Frequency Oscillation Mode

This article focuses on analysing the dominant mode of power system lowfrequency oscillation, discusses the advantages and disadvantages of the currentanalytical methods.Then three new kinds of pattern recognition methods areproposed ,which do not need to consider the mechanism of the low-frequencyoscillations, and directly use field field measurement data to analyse, achieve goodresults.These methods have better noise immunity, and have a positive significance inpower system low frequency oscillation.Base on analysing the characteristics of the power system low frequencyoscillation signal,the spectral distribution of the windowed low frequency oscillationsignal is analysed .Then a new method based on sliding window FFT algorithm isproposed. Through sliding a fixed window, analysing the amplitude change of thecorresponding spectral components, we can identify the damping characteristics; forthe barrier effect of the FFT algorithm, the pattern recognition methods and steps arerespectively proposed. The simulation results show that this method can effectivelyidentify low frequency oscillation characteristics, is robust to random white noise.Through the analysis of side-lobe characteristics of several windowfunctions ,FFT combined neural network pattern method was proposed. FFT algorithmbased on Blackman window is used to analysing the frequency , which can reduceenergy leakage influence between the frequency.Then energy weight can also beanalysed,so the dominant oscillation mode can be extracted rapidly, and thenamplitude, phase and the damping can be solved by neural network.The simulationresults show that the method can be reliable, accurate identification of low-frequencyoscillation dominant mode, has better noise immunity compared with pronyalgorithm.Because of narrow band width and fewer low-frequency oscillation dominantmodes, a neural network algorithm was proposed. The damping can be obtained bysegmental Fourier coefficients. Neural network model with finite neurons was used toovercome the difficult of solving the Fourier coefficients directly.In addition, it isanalyzed and discussed that the dominant mode of frequency leakage can bediscriminated by the weight of energy and the fitting error .The simulation resultsshow this algorithm can identify the dominant mode reliably and accurately,which hasbetter noise immunity compared with Prony algorithm.