Research On High-resolution Seismic Inversion Method Based On Multi-band Information And Sparse Representation

Seismic inversion is an important technical means to transform seismic data into underground medium lithology and petroleum information.Due to the band-limit characteristics of seismic data and the influence of noise interference,seismic inversion has problems of low resolution and ambiguity.It is mainly reflected in the lack of high-frequency and low-frequency parts of vertical seismic traces and the discontinuity of horizontal multi-channel seismic data.The most direct way to solve the inversion resolution problem is to add high-frequency and low-frequency information to constrain the inversion through regularization,such as broadening the frequency band of seismic data by assuming that the high-frequency information satisfies certain mathematical assumptions.Studies have shown that conventional specific mathematical assumptions do not meet the actual stratigraphic conditions.Therefore,in recent years,some scholars have studied the method of acquiring priori information of stratigraphy through dictionary learning and using sparse representation regularization to introduce priori information into the inversion.This method provides an effective way to adaptively acquire different frequency band information and use it for seismic data inversion.Although the above-mentioned seismic inversion method based on dictionary learning and sparse representation introduces high-frequency information from logging and effectively improves the vertical resolution and accuracy of the inversion results,there are still many problems,mainly as follows: 1)Relying on the KSVD dictionary learning method cannot fully mine the high-frequency prior information of the logging data,resulting in insufficient resolution of the inversion results,2)The use of sparsely represented high-frequency logging prior information into the inversion lacks effective constraints,leading to poor lateral continuity of the inversion results,3)The existing dictionary inversion methods are usually based on redundant dictionaries.Although redundant dictionaries can provide effective sparse representation capabilities,they also have high computational costs and low dictionary learning and inversion efficiency.To solve the above-mentioned problems,this thesis introduces associative dictionary learning,orthogonal dictionary learning and joint sparse representation methods to carry out research on seismic impedance inversion with high resolution.The main research work of the thesis includes:(1)The introduction of logging prior information for the existing dictionary learning has the problem of insufficient feature information pickup in the inversion.This article draws on the idea of image super-resolution reconstruction,and introduces the associated dictionary learning and joint sparse representation method to reconstruct accurate formation high-frequency information.And used in seismic constrained inversion.This method can effectively expand the frequency band information of the inversion result,so as to achieve the purpose of improving the resolution of the inversion result.(2)In order to solve the problem of the lack of effective lateral constraints in the above-mentioned high-frequency information introduction process,which leads to poor lateral continuity of inversion results.This paper introduces a two-dimensional dictionary learning method to obtain the lateral distribution characteristics of the strata from the seismic images,and effectively introduces the low-frequency and high-frequency information of the strata under the constraints of this lateral information,so as to achieve the purpose of improving the lateral continuity of the inversion results.(3)In order to solve the existing dictionary inversion methods usually use redundant dictionary learning,this type of algorithm has the problems of high computational complexity and slow computational efficiency in the inversion process.In this paper,orthogonal dictionary learning is introduced to improve the above algorithm.Through the sparse reconstruction test of complex data,it is verified that the orthogonal dictionary provides more sparse representation capabilities than the redundant KSVD dictionary,and its dictionary learning and inversion efficiency is higher.Finally,the model and actual data test verify the effectiveness of the method.This article aims at providing effective solutions to improve the resolution,lateral continuity and calculation efficiency of seismic inversion so as to promote seismic inversion to a higher resolution,higher precision and higher efficiency.