Combined Time-frequency Spectral Decomposition And Its Application On Seismic Data

Seismic spectral decomposition based on time-frequency analysis methods, is widely used in reservoir interpretation in recent years. To optimize the accuracy of seismic data interpretation, it is necessary to innovate a variety of existing methods including the traditional time-frequency spectral analysis methods, to improve the accuracy ofseismic data interpretation.In this paper, two new time-frequency analysis methods are introduced and implemented after discussing the conventional methods. To improve the poor resolution of short-time Fourier transform (STFT) due to fixed windows and solve the problem of different frequency component in the constraint least-squares spectral analysis using the same parameter, a new method is proposed. The content of this dissertation is arranged as follows:(1) Conventional time-frequency analysis methods. It gives detail instruction toSTFT, wavelet transform, Wigner-Ville distribution and its improved method, Stransform and generalized S transform, analyzes the advantages and disadvantages ofthese methods.(2) Constrained least-squares spectral analysis methods. Based on the theories ofFourier analysis, construct sine basis function coefficients equation. Thenhigh-resolution spectral are obtained by solving the equations. The method caneliminate the frequency smearing likes STFT, and it has higher resolution in lowfrequencies than the windowed time-frequency analysis method such as wavelettransform and S transform. Synthetic models prove that this method has higherresolution than conventional frequency analysis method in thin layer recognition.(3) Empirical Mode Decomposition (EMD) and its improved algorithm，completely empirical mode decomposition (EEMD) and complete set of empiricalmode decomposition (CEEMD), for seismic time-frequency analysis. EEMDovercomes the mode mixing problem, but big errors exist in reconstruction. CEEMDnot only overcomes the mode mixing problem of EMD, but also almost losslessreconstructs the original signal. Based on CEEMD and Hilbert transformtime-frequency analysis methods，high resolution spectral can be accessed. (4) Combined time-frequency analysis method. This paper proposed thecombination methods of STFT, least squares constraint time frequency analysis, EMD,EEMD and CEEMD. Different components(IMFs) choosing different length of timewindow(when STFT is used) or using different number of iterations and windowlength (when constrained least-squares spectral analysis is used),make eachcomponent reaching best time-frequency localization and, finally, obtaining highresolution spectrum. It provides a new way of thought to obtain high resolutionspectrum.