Research On And Application Of The Seismic Signal Processing In The Sparse Domain

The shortage of oil and gas resources is an important issue facing China at present and in the future.With the increasing exploration and exploitation of hydrocarbon resources,there are far fewer newly discovered structural reservoirs.The exploration prospects and targets are gradually shifting to complex oil and gas reservoirs,such as stratigraphic and lithologic reservoirs.Reservoir prediction from surface seismic data is more dependent on high efficiency,high fidelity and high resolution signal processing methods,and a new set of theories and methods is urgently needed to be applied to the entire life cycle of seismic exploration.Seismic signals are typically characterized by high information redundancy,that is,the information carried in massive volumes of seismic data is limited,can be modeled in terms of sparsity.The sparse theory with sparse representation as its core has got rid of the shortcomings of traditional information theory and revealed the information contained by a small number of coefficients.The convolution model of seismic signal and sparse representation share the same mathematical formulation,and the sparse theory can be used to process seismic data efficiently and easily.The thesis focuses on the sparse seismic signal processing and its application in prestack gather conditioning and reservoir characterization.The main research results and key conclusions include:1.According to the structure and noise characteristics of seismic signals,three improved schedules were proposed,that include:?1?Outliers may be generated during the data acquisition and processing,and the L₂-norm and L_F-norm fidelity are incapable of handling outliers scenario.L₁-norm and L_2,1-norm fidelities were introduced into the objective function.?2?Wavelet dictionary constructed by seismic wavelet has higher coherence and does not meets the requirements of sparse representation.L_1-2-norm regularization was introduced into seismic data processing.?3?L₁-norm and L_2,1-norm regularizations have the disadvantage of imbalanced penalization of source vector,and iteratively reweighted is used to hybrid with the prototype.Based on the above,three iteratively reweighted sparse representation algorithms,i.e.,L₁-L₁-norm,L_1-2-L₁-norm,and L_2,1-L_2,1-norm were proposed.2.The traditional multi-channel algorithms are typically not suitable for the common shot gather and CMP gather.The direct and refracted waves in these two types of gathers are characterize by inclined events,and the reflected waves are characterize by hyperbolic events.The inclination angles of the events are different,that cannot benefit from the advantages of multi-channel algorithms.The common offset gather is characterized by horizontal events,and the reflected waves of each trace share the same two-way travel times and reflection structures,which satisfies the common sparsity condition.By taking the advantages of both joint sparse representation and common offset gather,we apply the joint sparse representation algorithm to the common offset gather,that can utilize the spatial coherence of the signals and improve the performance of the denoising algorithms.3.Commonly used seismic data processing methods are based on homogeneous and isotropic layered media.The residual moveout still exists in the prestack gathers after dynamic and static correction,and we cannot obtain much flatter gathers.However,the residual moveout correction method based on dynamic time warping may causes waveform distortion.A method for the removal of residual moveout by combining sparse representation and dynamic time warping was proposed.The reflection coefficient sequence can be regarded as the unit impulse responses of subsurface,which is a pulse sequence.After constructing a dictionary from seismic wavelets and using a sparse representation to invert the reflection coefficient sequences,the influence of the waveform can be eliminated.Correcting the residual moveout of the reflection coefficient gather can avoid waveform distortion.4.The application of seismic inversion to thin reservoir prospecting faces three dilemmas:?1?Thin layer and insufficient resolution of seismic data,due to the tuning of adjacent layers;?2?Poor continuity of seismic events,resulting in poor horizontal continuity of single trace inversion results;?3?Weak effective signal and strong noise.To solve these problems,a multi-trace inversion based on joint sparse representation is proposed.In order to improve the resolution of inversion results,a wedge wavelet dictionary is constructed by dipole decomposition.The pre-stack seismic data after NMO and residual moveout correction is characterized by horizontal events,which meets the assumption of common sparsity.Using the joint sparse representation to invert the pre-stack seismic gathers can improve the robustness and lateral continuity of the inversion results.The local stratigraphy is approximately horizontal,which satisfies the common sparsity hypothesis.Implementing multi-trace post-stack inversion using joint sparse representation is feasible.And multi-trace participation in the calculation is equivalent to adding constraints,which can reduce the solution space and the multiplicity.5.Seismic signals are non-stationary signals,but truncated signals can be regarded as piecewise stationary,and the time-frequency spectrum is approximately sparse.The signal truncation used in time-frequency analysis can be regarded as the signal sampling.Compressed sensing is introduced into time-frequency analysis,and the high-resolution time-frequency spectrum can be recovered by using sparse representation.At the same time,changes in the sedimentary facies,physical and fluid properties of the subsurface cause changes in the seismic waveform and time-frequency spectrum.Changes in the time-frequency spectrum are more pronounced than waveform changes.However,time-frequency analysis causes data to be highly redundant,which increases interpretation workload.Combining time-frequency analysis and clustering,a time-frequency domain waveform classification is proposed.Using logging to calibrate clustering results,it can be used as a direct hydrocarbon detection method to aid reservoir prediction.At the same time,the uncertainty of interpretation methods such as low-frequency shading is reduced.6.Validation of the proposed methodologies using an actual field data.Random noise suppression and gather flattening are achieved through pre-stack gather optimization processing,which provides a high SNR and flat gathers for reservoir characterization.The test results of the main producing interval?P5 sand?verify the applicability and accuracy of reservoir prediction method based on sparse theory.