Seismic Wavelet Blind Extraction And Non-Linear Inversion

With the deepening of the degree of prospecting, the exploration activity turns from structural hydrocarbon reservoir to lithologic reservoir. As the most effective exploitation method, seismic exploration is playing an more and more important role. Seismic data contain abundant lithology, physical property and fluid information, which can be acquired by various kinds of seismic inversion methods, and therefore offer more abundant materials for reservoir prediction and characterization, espicially in complicated reservoir. This dissertation study in depth several key technologies in seismic inversion such as seismic wavelet extraction, poststack inversion and prestack inversion.Most seismic inversion problems are ill-posed and non-linear, they are neither sheer underdetermined problem nor over-determined problem. Its'characteristic is that the observed data are more than the unknown parameters, but the observed data are not linear independence, and it's this kind of ill-posed problem that makes ambiguity and unstability in inversion. There are three methods to solve ill-posed problems, i.e. (1) Regularization methods; (2) Probabilistic methods; (3) Information theory methods.Regularization methods divide ill-posed problems into underdetermined problem and over-determined problem. Probabilistic methods include probability distribution matching, maximum likelihood and Bayesian methods, among them the first two kinds of methods have only considered the uncertain question that exist in the data themselves, and not considered the priori information of the unknown parameter. In contrast, Bayesian methods consider both of them, So possess greater superiority. There are relations among least square, maximum likelihood and Bayesian solution. In the case of linear model and likelihood function obeys Gaussian distribution, it is easy to show that the maximum likelihood estimate becomes equivalent to least square one. In the case of prior information obeys uniform distribution, the Bayesian estimate becomes equivalent to maximum likelihood one.In contrary to maximum likelihood approach, the data and model are assumed to be exact and only the uncertainty of unknown parameter is modeled in Information methods. These methods are mainly used in blind signal processing, and in blind deconvolution while seismic exploration. Seismic wavelets extraction is a key problem in seismic exploration data processing. However, wavelets can neither be acquired without using well-logging information nor by using traditional auto-correlation statistical method. Therefore, it is important to extract seismic wavelet correctly by using seismic data only, i.e. wavelet blind extraction. Under blind signal processing theory, Subspace method seismic wavelets extraction based on second statistic, and cumulant matrix equation method, phase estimation method, cumulant matching based on high-order statistic are deeply studied. receive reliable and accurate wavelets extraction results finally.Well logging information are always needed while poststack seismic inversion, they are primarily used to calibrate seismic data and combined with seismic data to extract seismic wavelet. However,there are often no logging or few logging information can be used in some area while performing seismic inversion, in addition, some logging information can't reflect the actual stratum information for they are easily to be affected by various kinds of subjective factors and objective factors(e.g. wall cavitation, instrument difference). These situations may exist on the different layers in same well, and the same layers of different wells. Therefore, it has important actual meaning about how to extract seismic wavelet correctly and perform acoustic impedance inversion directly by using seismic data only under the situation that there are no logging information or the information are not reliable(i.e. PSBI), and offering reliable stratum Lithology and physical property parameters for elaborate reservoir characterization. A kind of PSBI method has been put forward, first of all, extracting mixed-phase wavelet by using genetic algorithm from seismic data based on high-order statistics theory; then performing seismic inversion based on Bayesian theory; Finally, acoustic impedance are acquired under the situation that there are no well logging information.Prestack inversion possess better fidelity and much information than poststack inversion, it's not only suitable for reservoir physical property inversion but also suitable for oil and gas property inversion. This dissertation study in depth several prestack inversion methods. Points constraint sparse spike prestack inversion based on linearized approximation of the Zoeppritz equation and Bayesian parameter estimation theory, Covariance matrix is used to describe the degree of correlation between the parameters, P-wave impedance, S-wave impedance, density information are used to constrain the inversion results, thus it is reliable to make the result more accurate. Non-linear quadratic programming prestack inversion is transformed into non-linear quadratic programming problem, and inversion results are acquired under several constraints, tests on synthetic data and practical application show that all inverted parameters were almost perfectly retrieved for further analysis even the SNR is low. Prestack inversion based on SVM is a kind of novel method and circumvents many of the limitations associated with conventional inversion methods without sacrificing performance in parameter estimation, it exhibits the ability to directly estimate independent elastic parameters using seismic data. This method eliminates the need for any approximations of the Zoeppritz's equations or assumptions about the independent elastic parameters contrast, initial model and constraint of well logging are not needed also, and therefore is suitable for further popularization and application.