Study On Nonlinear/Non-gaussian Seismic Inversion Problems And Its Applications

Seismic inversion is one of the most important technologies that used in geophysics and reservoir detection. Common inversion approaches are based on Gaussian distribution, the defects and limits of Gaussian distribution are recognized by researchers. The main task of this dissertation is to answer the following two questions: firstly, what kind of non-Gaussian distribution does the real seismic data follow? Secondly, what kind of seismic inversion method is available if the seismic data follows non-Gaussian distribution? Besides, the dissertation have done some researches on seismic noise suppression, seismic wavelet extraction and model building, and proposed improvements, respectively. These methods are all applied to real seismic inversion and reservoir detection applications.To answer the first question, the statistical characteristics of a group of stacked seismic data and a group of prestacked seismic data, chosen from two different fields, studied in the dissertation. The statistical characteristics, including mean, variance, skewness and kurtosis, compared with Gaussian and non-Gaussian α stable distribution statistical characteristic. The result illustrates that the real seismic data has similar statistical characteristics with non-Gaussian α stable distribution. And the α stable distribution parameters, estimated from resl seismic data with different methods, are within the correct ranges. It certifies that the real seismic data follows non-Gaussian distribution.To answer the second question, the dissertation proposes two inversion methods, the least p norm inversion method and covariation inversion method. Both of the inversion methods are based on fractional lower order moment theory and nonlinear. In this dissertation, two iteraton algorithms are designed for the above inversion method, respectively. Theory model and real data seismic inversion experiments are conducted. Both of the application obtaine results are superior to ones of the Gaussian inversion methods.Several seismic inversion related techniques, seismc noise suppression, seismic wavelet extraction, model building, are studied in this dissertation and improved methods are propoded.On seismc noise suppression, the traditional Gaussian distribution noise elimination methods are unable on impulse noise. The dissertation takes the assumption that the impulse noise follows non-Gaussian distribution. The Myriad filter, designed for impulse noise eliminating, is improved. The improved Myriad method has excellent abilities of both noise eliminating and fast computating.On seismic wavelet estimation, this dissertation follows the assumption of non-Gaussian distribution and proposes covariation seismic wavelet estimating method. The ability of proposed wavelet estimating method is evaluated by booth theory data and real data.On modeling building, the dissertation proposes a new modeling building method that added high frequency component into model, and the model has high accuracy to constraint the inversion result.On the problem of the elastic impedance is unable to identify reservoir if the angle range is small, the dissertation proposes a novel angle gradient elastic impedance(AGEI) reservoir predicating method. Since the AGEI is more sensitivity angle changes, the reservoir predicting accuracy is improved.