Methods And Application Of Full Waveform Inversion For Abyssal Seismic Data

Velocity is an important physical parameter describing subsurface structures andreservoir characteristics. Rebuilding an exact velocity model from seismic data playsan important role during both seismic data processing and seismic interpretation.Traditional methods to rebuild velocity model during seismic data processing includestack velocity analysis, tomography and migration velocity analysis. In recent years, asthe rapid development of offshore oil and gas exploration in China and techniques ofoffshore seismic acquisition and data processing, exploration targets tend to theabysmal area or complex stratum. In this condition, traditional velocity rebuildingmethods have become increasingly unable to satisfy the requirement of quality andresolution on velocity model by seismic data processing and interpretation. FWI (Fullwaveform inversion) is a high-resolution and high-quality velocity rebuilding method.It is realized by matching the theoretical data according to initial model to the realobserved data, and gradually updating the velocity model until obtain a depth-domainvelocity model which can correctly describe subsurface velocity distribution. However,FWI haven’t formed a well-developed system during the industrial production, becauseof some of its inherent limitations and some problems during its application.This paper performed a deep research on the application of frequency-domain FWIin abysmal seismic data. For several problems and difficulties of FWI when we useFWI to rebuild the velocity model from abysmal seismic data, this paper proposed somesolutions.This paper proposed a joint high-efficiency frequency-domain FWI algorithm togreatly improve its computational efficiency. The computational time cost andmemory requirement of FWI are huge because of that the model updating needsnumerous iterations and for each iteration, we need to do the wave-equation forwardmodeling for numerous sources in order to match the observed seismic data. This paperintroduced three methods, source-encoding technique, principal component analysismethod, and variable grid method, into theory of frequency-domain FWI to improve itscomputational efficiency. Source-encoding technique combined several different sources into a supershot to reduce the total number of sources within the forward andinversion problem. However, crosstalk noise introduced by the encoding will reducethe quality of final result. This paper proposed a frequency-group encoding strategy tosuppress crosstalk noise by using orthogonal encoding matrix; Principal componentanalysis method can reduce the computational costs of FWI by singular valuedecomposition of residual matrix, accumulated energy analysis and projection oforiginal observed data matrix and source matrix into a lower dimensional data space;Variable grid method is realized by calculation of suitable grid spacing correspondingto different frequency, hence FWI can obtain reliable result using smaller modeldimension, especially for the lower frequencies. For above three methods, this papercompared their merits and drawbacks for different frequency scales. Based on this,these methods are combined within the framework of frequency FWI theory, takingadvantage of these methods at different frequency scales. This joint method greatlyreduced the computational cost of FWI in the promise of inversion quality.This paper proposed and improved a seismic data low-frequency componentrebuilding method based on sparse constraint to improve the robustness of FWI.Generally in the application, FWI is realized by changing to be a local optimizationproblem based on Born-approximation and solved with some local optimizationalgorithms, which makes FWI a high nonlinear problem. In addition, model updatingcan easily meets the local minimum problem because of the complexity of observedseismic data during the wavefield matching. Therefore, FWI usually needs a highquality initial velocity model and high signal-to-noise ration low-frequencycomponents to assure validity and stability of inversion. In this paper, low-frequencyband energy of data are recovered in frequency-domain by using compressive sensingtheory and sparse inversion method based on the sparsity of seismic reflectioncoefficients in time domain. In this way, we can not only satisfy the requirement of low-frequency information by FWI, but also reduce its dependency on the initial velocitymodel.This paper proposed a multi-scale frequency-domain FWI algorithm to improvethe inversion quality for the deep part of velocity model. During the iterative modelupdating using FWI, the information of diving wave and refraction wave are used torebuild low-wavenumber components of velocity model. However, these signals can beonly observed on the middle/large offset locations. For the conventional offshore towed streamer acquisition geometry, the length of streamer is usually between6~8km. So thesignals of diving wave and refraction wave corresponding to the deeper layers are hardto acquire, and then rebuilding model can’t satisfy the requirement of abyssal seismicdata processing. For this problem, this paper proposed an inversion-constraining basedfrequency-domain layer stripping method. This method first slices the model accordingto the change rate of updated model in space during the iterations, and then calculatesa group of depth related weighting factors based on previous layering. By using thismethod, deep part of velocity model are better recovered during the inversion.During the seismic data processing, FWI is hard to be used as an individualvelocity rebuilding technique. In general, it needs some pre-processing steps and qualitycontrol evaluation methods as assistant tools. Some necessary pre-processing can notonly provide appropriate seismic data and initial velocity model for subsequent FWI,but also realize the optimistic choice of parameters and robustness analysis for FWI;Migration and stack are not only necessary steps of seismic data processing, but alsodesirable tools to verify the result of FWI. This paper proposed a velocity modelrebuilding procedure facing abysmal conventional towed streamer data based on FWI.This procedure mainly includes following steps: Firstly, some pre-processing methodslike high-pass filtering, geometric spreading compensation, multiple attenuation and2D transmission are applied to field data; secondly, migration velocity analysis andtomography are used to build initial velocity model; thirdly, rational and effectivestrategies and corresponding parameters are best selected according to the datacharacteristics and inversion targets; fourthly, FWI is used to rebuild the depth velocitymodel; finally, pre-stack depth migration and stack are used to test the inversion quality.This paper apply this procedure to a set of towed-streamer data and a set of variabledepth streamer data, respectively. The application of FWI to these two types of datashows the validity of the proposed method...