Seismic Inversion Regularized By Local Prior Information

Seismic inversion uses seismic records and borehole data to predict the distribution of subsurface elastic parameters.Accurate and high-resolution inversion results will be beneficial to further fine reservoir characterization and subsequent oil and gas reservoir prediction.During seismic inversion,the following three problems should be considered.First,nonlinearity.Since seismic inversion is conducted based on the seismic forward modeling mechanism,the current commonly used wave equation,the imaging and convolution forward modeling is nonlinear.Second,nonuniqueness.Due to the limited observation data,the inversion has multiple solutions to match well with observed data.Third,instability.The smaller noise of observation data will cause a large disturbance of inversion model.To overcome the above problems,this paper introduces the basic theory and method of post-stack inversion and prestack AVO inversion based on convolution model.Through Taylor expansion approximation methods,the original non-linear problem is transformed into linear problem.Through the model constraint and the data-driven extraction method for structure characteristics,the inversion results regularized by local prior information can be stable and reliable.Seismic inversion methods based on convolution model can provide subsurface elastic parameter models for quantitative interpretation by using poststack seismic data,prestack seismic data,and borehole data.Poststack seismic inversion can provide underground acoustic impedance and the prestack seismic inversion can provide P-wave velocity,S-wave velocity and density,etc.Since the traditional inversion methods with L1 norm constraint cannot provide accurate sparse inversion results,this paper proposes an approximately unbiased sparsity,namely the non-convex L1-2 norm constraint,and uses the norm to reconstruct the reflection coefficient.The inversion results of elastic parameters are further obtained by the generalized linear inversion and Gaussian constraint inversion methods.Accurate reflection coefficient is more beneficial to obtain high-resolution inversion results.Considering the underground medium holds the continuous spatial distribution characteristics,this paper,taking advantage of the high lateral resolution of poststack seismic records,extracts the permutation matrix to record subsurface spatial distribution for multichannel poststack impedance inversion.The patchordering regularized scheme can simultaneously improve lateral continuity and the resolution of inversion results,which is beneficial for further interpretation.Due to band-limited seismic record and discrete characteristic of fixed temporal sampling interval,it is restricted for traditional seismic inversion method to get high-resolution inversion results.This paper comprehensively use conventional inversion results and high resolution borehole data to extract low-and high-resolution compenents by dictionary learning.Then the sparse coefficients of high-and low-resolution compenents are utilized to build the prediction model.The high-frequency components can be predicted by conventional inversion results to improve the accuracy and resolution of inversion results.Based on the above results,this paper uses the Gaussian mixture model of patch-based data to obtain the characteristics of underground local structures,which constrains the inversion of underground acoustic impedance and provides reliable inversion results.