Non-linear Multi-trace Deconvolution Regularized By Spatial Reflection Structure And Its Software Development On GEOEAST Platform

Deconvolution technology,compresses seismic wavelet to make the interfered reflection events distinguishable,is widely applied for improving the vertical resolution of seismic data.There are two major deconvolution methods:least-square deconvolution(LSD)and sparse spike inversion(SSI).LSD,improves the resolution by turning up the amplitude of the high-frequency constituents within the effective band of original seismogram,has no ability to retrieve frequency components beyond the effective band.SSI,incorporates the sparse regularization term into deconvolution system,can partly retrieve high-frequency constituents beyond the effective band of observed data,but frequently provides instable,inaccuracy estimation,and causes the suppressing of the relatively weak reflection events.We develop a band-controlled sparse deconvolution(BCSD)algorithm under the single-trace deconvolution framework.BCSD inverts for band-controlled rather than traditional full-band estimation,so as to reduce the nonuniqueness in the inverse problem and improve the reliability of the output.Model and field data examples confirm that compared to LSD,BCSD algorithm can provide model parameter estimation with higher resolution and better stability.All of the single-trace deconvolution algorithms cannot consider the spatial dependence among model parameters at adjacent seismic traces,which further causes poor continuities of the structures.We develop a reflection structure characteristic(RSC),and propose a RSC-based multi-trace nonlinear deconvolution algorithm(RSC-MND).The RSC estimation,extracted from the observed stacked data,plays a key role in mapping the spatial association in data into the retrieved multi-trace estimation,so as to guarantee the lateral continuities of structures and protect weak reflection signals.We exploit the alternative direction method of multiple(ADMM)algorithm to solve the cost function of RSC-MND.Model and field datum examples demonstrate that compared to SSI,RSC-MND owns obvious merits on guaranteeing the spatially continuities of structures and retrieving relatively weak reflection events.Based on LINUX platform and C/C++language,we transplant RSC-MND algorithm into the GEOEAST system,the famous seismic data processing software belonging to BGP of PetroChina,as an independent processing module RSCmulti-traceNonlinearDec.The test results from the third party confirm that RSCmulti-traceNonlinearDec module is methodologically right and achieves the expected effect.Supported by the PetroChina's project in the 13th five-year plan,RSCmulti-traceNonlinearDec module processes many onshore and offshore field data from China.All processing results are relatively ideal and better than the alternative modules(i.e.,spike deconvolution;prediction deconvolution)in GEOEAST software.