Seismic Signal Processing Methods Based On Sparse Representation Theory

With the deepening development of oil and gas exploration, the exploration and exploitation targets gradually shift to the complex, subtle, deep and unconventional hydrocarbon reservoirs, the requirements for the quality of seismic data and the prediction accuracy of reservoirs is getting higher. Under the influence of complex acquisition conditions, the acquisited field seismic data is often contaminated by serious noise, even the effective signal is completely submerged; in addition, due to the existence of some waste shots and traces, as well as the obstacles or forbidden zone, the field seismic data is often incomplete or irregular. How to ensure the integrity and regularity and improve the signal-noise ratio, the resolution and fidelity of seismic data, in order to facilitate the subsequent interpretation and oil and gas exploration, is the goal of seismic data processing personnel.Sparsity is a very attractive feature of signal, which endows seismic signal processing with new vitality and provides convenience for solving many seismic data processing problems. Whether the sparsity of seismic reflection coefficient sequence itself, or the sparsity of seismic data in overcomplete dictionaries, can effectively promote the solving of inverse problems, such as noise suppression, deconvolution, seismic data reconstruction, etc. Compared with other traditional methods, the results obtained by solving the sparsity-regularized inverse problems have obvious advantages. In this thesis, the sparse representation and compressive sampling theory is introduced into the field of seismic signal processing, so as to provide technical support for high-precision seismic data processing.Sparsity promotes the solving of seismic noise suppression problems. Based on sparse representation theory, in order to obtain the ideal denoising results, the adopted transform must represent seismic data as sparsely as possible. Considering the real seismic data usually contains a variety of forms of content, if only a single dictionary is adopted to sparsely represent it, may not be able to achieve the optimally sparse representation of all the features, therefore, the seismic data is viewed as a combination of different morphological components, and the amalgamated dictionary is adopted for sparse representation of seismic data. In view of this, in order to deal with the random noise suppression problem, the seismic random noise suppression method based on morphological component analysis is developed; in order to solve the surface wave suppression problem, by rearranging a single shot into one trace, the ground-roll suppression method based on morphological component analysis is developed; in order to deal with reverse-time migration imaging noise suppression problem, the reverse time migration noise suppression method based on morphological component analysis is put forward; in order to solve the harmonic noise suppression problem, inspired by the feature of monofrequency interference in F-K domain, the monofrequency interference suppression method via sparse reconstruction strategy is proposed.Sparsity promotes the solving of seismic data reconstruction problems. Compressive sampling theory establishes a direct link between the sample and the sparsity, and provides theoretical support for the solving of reconstruction problems, in which case, the data is irregular or has large gap, or the data may be not satisfy the Shannon sampling theorem, once the best dictionary and the solving strategy of the sparsity-regularized inverse problems are found, the complete data can be recovered from the incomplete data. In the light of this, to solve the interpolation problems of2D seismic data, the seismic data reconstruction method based on morphological component analysis is developed; to solve the interpolation problems of3D seismic data, starting from the characteristics of seismic data, the3D seismic data interpolation method based on2D dual-tree complex wavelet transform (DTCWT2D) is proposed.Sparsity promotes the solving of seismic deconvolution and reflection coefficient inversion problems. Since the generalized Gaussian probability density function has the ability to approximate arbitrary probability density function, in view of the sparsity of seismic reflectivity series itself, the generalized Gaussian probability density function is adopted to fit the probability density function of the reflectivity series, combining Bussgang blind deconvolution algorithms, the objective function which uses Kullback-Leibler distance as sparsity measure is established, and finally, the seismic blind deconvolution method based on generalized Gauss distribution is implemented. The seismic reflection coefficient inversion problem can be viewed as the problem to solve the sparse representation coefficients of seismic data in sparse dictionary, by selecting the best dictionary, basis pursuit seismic reflection coefficient inversion method and basis pursuit poststack seismic blind inversion method are developed. In consideration of the problem that the effective reflection information from coal measure strata is submerged in the strong reflection information caused by the coal seam, the matching pursuit coal seam strong reflection separation method is put forward.Sparse representation plays an important role in seismic data processing field, the key lies in the optimally sparse representation of seismic data. In order to obtain the optimally sparse representation coefficients, the design of optimal dictionary and the solving of the sparsity-regularized inverse problems are both essential. In the processing of real data, the sparse dictionary can be chosen by means of data characteristics and sparsity analysis. By combining model data test and real data verification, the effectiveness of the proposed methods in this thesis is demonstrated, and provides new ideas for high signal-to-noise ratio, high resolution and fidelity seismic data processing.