Seismic Signal High-resolution Inversion Constrained By Stratigraphic Parameter Sparse Representation

Signal inversion predicts the original state or parameters of the target by analyzing the observed signals,enabling researchers to extract more intuitive information from limited and indirect observed data.This is crucial across many scientific and engineering fields.When artificial seismic waves propagate downward and encounter geological interfaces,they reflect back to the surface,carrying comprehensive information about the strata’s structure,lithology,and fluid.Seismic signal inversion can reveal various attribute parameters of subsurface structures by analyzing and transforming seismic reflection signal.It is not only the principal method for human to analyze deep subsurface structures and rock physical attributes,but also the most important means for the exploration of subsurface resources,especially oil and gas.Seismic signal inversion is an interdisciplinary subject of common concern in the fields of signal processing and exploration geophysics.Due to the earth’s absorption effects,most of the frequency band information of the seismic signal is lost.Along with noise in the observed data,the inversion of subsurface rock physical parameters from seismic signals often faces non-uniqueness issues,that is,it is impossible to obtain the unique solution with high resolution.To mitigate these non-uniqueness issues and improve the resolution and stability of the inversion,it is usually necessary to assume that the characteristics of the target parameters follow specific patterns according to the existing prior knowledge or experience,and to narrow the solution space by adding prior constraints to the algorithm.Nevertheless,the prior constraints used in most seismic signal inversion methods are model-driven.The agreement between the prior model and the actual data is low,so the accuracy and resolution of the inversion are greatly challenged.The data-driven prior model construction method represented by sparse representation can obtain a prior model that is relatively consistent with the actual data,and achieve good inversion accuracy and resolution.However,the previous methods still have problems such as insufficient ability to represent the strata with complex structural features,and cannot comprehensively solve the non-uniqueness issues of seismic signal inversion.Based on sparse representation,this dissertation developed an accurate representation method for stratigraphic structural information,and completed the data-driven prior model construction method based on sparse representation,which effectively narrows the solution space and improves the resolution of inversion results.The main work and novelties of this dissertation mainly include the following four aspects:(1)Initial model construction method constrained by sparse representation of stratigraphic sequence: Aiming at the problem that the initial model obtained by interpolating well logging data is not completely consistent with the sedimentary law of subsurface strata,this dissertation proposed a method for constructing initial models constrained by stratigraphic sequence representation.Based on a collaborative sparse representation algorithm,this method learns a collaborative dictionary of seismic data and stratigraphic sequence.Each atom in the dictionary contains the association relationship between seismic data and stratigraphic sequence.Then,the accurate representation of stratigraphic sequence in space is realized when the seismic data are represented by the collaborative dictionary.Finally,a reliable initial elastic parameter model is constructed under the constraint of the stratigraphic sequence.The research indicates that the reliability of the initial model constructed under the constraint of stratigraphic sequence is greatly improved,which provides a solid foundation for subsequent high-resolution seismic inversion.(2)Facies-controlled inversion constrained by dynamic representation of sedimentary structure: Aiming at the problem of inversion accuracy and resolution caused by using the same sparse dictionary to represent the strata of different facies in the existing facies-controlled inversion methods,this dissertation proposed a facies-controlled seismic inversion method constrained by dynamic representation of sedimentary structures.The method first divides seismic facies through seismic attribute analysis,then employs various sparse representation patterns to describe sedimentary structural features within corresponding facies.Meanwhile,an online dictionary learning algorithm is introduced to dynamically update the feature library in real time,thus dynamically representing reservoir structural features.The research indicates that the method proposed in this dissertation significantly improves the ability to represent the reservoir structure in complex sedimentary environments,while improving the accuracy and resolution of the inversion results.(3)Seismic signal intelligent inversion constrained by sparse representation:Aiming at the problem that the existing intelligent inversion methods often fail to fully exploit and integrate the potential information in geological data by constructing prior constraints,this dissertation developed an innovative data-driven inversion framework that combines inversion network with prior information neural network,allowing for flexible integration of various prior constraints.The method first generated rich training samples based on the statistical characteristics of the data,then extracted and utilized the key features in the data based on sparse representation technology.It was integrated into the prior information network through pre-training,and finally the inversion network was constrained based on back propagation.The research indicates that this data-driven inversion framework combined with sparse representation can more effectively exploit and integrate the constraint information in the data,which paves new technical pathways for accurate inversion of complex geological structures.(4)Stochastic inversion constrained by sparse representation of sedimentary structures: The existing stochastic inversion methods cannot use the sparse representation of stratigraphic sedimentary structure,which leads to the problem that the inversion results are not completely consistent with the actual stratigraphic sedimentary structural features.To solve the problem,this dissertation proposed a stochastic inversion method constrained by sparse representation of sedimentary structure,thoroughly extracting the sedimentary structural features of the reservoir,and using these features as key prior information for geological statistical inversion.The research results demonstrate that the proposed method not only strengthens the description of parameter space variability and improves the resolution of inversion results,but also achieves an assessment of the uncertainty in inversion results.This method is significantly important for addressing challenges in analyzing complex reservoir structures.In summary,this dissertation significantly improves the representation ability to subsurface stratum structure through the sparse representation,and completes the overall seismic inversion methodology centered around sparse representation technology.Meanwhile,it provides essential technical support for high-resolution seismic inversion in oil and gas exploration,thus has great theoretical and practical significance.