High-imensional Seismic Inversion Via Sparse Representation

In the process of seismic wave propagation from ground to deep subsurface layers,massive high-dimensional seismic reflection signals with different incident angles will be generated,which comprehensively reflect the lithology,structure,and hydrocarbon information of subsurface geological media.The seismic inversion technique can infer such information from the high-dimensional signals,being a dominant method for oil and gas exploration and a frontier interdisciplinary subject of common concern in the signal processing and resource exploration fields.The difficulties of high-dimensional seismic inversion are mainly in the following four aspects:(1)the seismic inversion is inherently ill-posed,leading to instability in the inversion process and non-uniqueness of solutions?(2)different types of reservoir param-eters have different high-dimensional seismic responses,leading to low quality of the so-lutions of some parameters when inverting multiple reservoir parameters simultaneously?(3)the frequency band of seismic signals is narrow,leading to low resolution and accuracy of solutions?(4)the horizontal consistency of seismic signals is poor,results in low sta-bility of conventional trace-by-trace inversion methods,which in turn leads to insufficient lateral continuity of solutions.To reduce the impact of the above-mentioned problems,most existing methods add prior constraints on reservoir parameters in a model-driven manner.However,they no longer meet the needs of practical applications with increasing the high requirements of the exploration accuracy of complex reservoirs.In this disserta-tion,we propose a series of new methods for the seismic inversion problem by properly integrating model- and data-driven strategies.By using the sparse representation technol-ogy,we obtain a significant improvement in practical applications.The main work and novelties of the dissertation mainly include the following four aspects:(1)To overcome the ill-posedness of the inverse problem,traditional seismic inver-sion methods assume that subsurface reservoir parameters have specific structural fea-tures,which limits their application range,namely,low adaptability.When the actual situation is much more complicated than the assumptions,the accuracy of their solu-tions cannot reach a satisfactory level.To solve the problem,this dissertation proposes a novel seismic inversion method via dictionary learning and sparse representation.The method introduces an over-complete dictionary learning algorithm to capture the sedimen-tary structure features of reservoir parameters,then utilizes the inherent physical mecha-nism of the seismic inverse problem as a model-driven factor,and the sparse representation constraint of reservoir parameters as a data-driven factor,hence to implement a model- and data-driven procedure for seismic inversion problem.Experimental results demonstrate that,the proposed method can extract the prior information of reservoir parameters adap-tively,can improve the accuracy of solutions significantly,and is suitable for complex inversion tasks.(2)Traditional seismic inversion methods do not consider the spatial variation prob-lem of the correlation among different parameters while inverting multi-parameters si-multaneously,results in poor performance of solutions.Aiming at the above problem,this dissertation presents a novel approach for simultaneous seismic inversion of multi-parameters via collaborative sparse representation.The method learns a joint dictionary of multiple reservoir parameters,which concomitantly depicts the sedimentary structure features of each parameter and the correlation features among multi-parameters.By es-tablishing the objective function in a similar model- and data-driven manner,the method realizes a simultaneous inversion framework allowing taking into account the correlation information.Studies have shown that the proposed method can further improve the accu-racy of each reservoir parameter,especially for the density parameter that is insensitive to the amplitude information.(3)Limited by the frequency band of seismic signals,the traditional methods cannot make their resolution insufficient to describe complex reservoirs.To solve this problem,this dissertation proposes a high-resolution seismic inversion method via collaborative sparse representation and prediction of high-frequency components.This method draws on the idea of image super-resolution,utilizes the full-frequency information of well-log data,and executes a joint dictionary learning algorithm to capture the correlation features between low-frequency and high-frequency components of reservoir parameters.In do-ing this,the method can predict the high-frequency components of all parameters,which broadens the frequency band of inversion results.Studies have shown that the method can improve the ability to identify thin layers,and thus meet the high requirements of the inversion tasks of complex reservoirs.(4)The aforementioned inversion methods ignore the spatial structure features of subsurface parameters and have low lateral continuity in their solutions.The existing 3D inversion methods are either too simple,ignoring the real spatial structure features,or too complicated,suffering from low computational efficiency.Aiming at the above problems,this dissertation presents a structure-guided,collaborative-sparse-representation-based,and 3D seismic inversion method.The method first uses the structure tensor technol-ogy to extract the structure tensor field from observed seismic data,which represents the spatial structure features of subsurface parameters.Then,in the horizontal direction,the method uses the extracted features to enhance the spatial structure of solutions,and in the vertical direction,the method inherits the sparse-representation-based constraints of the sedimentary structure and correlation features,the constraints in two directions construct a 3D high-resolution seismic inversion framework.Experimental results demonstrate that the method can not only retain the advantages of the aforementioned research(i.e.,high accuracy,strong adaptability,and high resolution)but also further improve the lateral continuity of solutions.In summary,this dissertation provides a new idea for the high-dimensional seismic inversion of complex reservoirs and offers a reference for solving the inverse problems in related fields,thus has great theoretical and practical significance.