Key Technique And Application Of Structure Oriented Filter For Seismic Wave Field

With the improvement of oil-gas exploration, complex area and complexreservoir have become the main geophysical target. For seismic data acquired fromcomplex areas, data processing directly affects the precision of interpretation ofseismic data. In seismic data processing, complexity of surface and geologicalstructure results in the difficulty on characterizing and analyzing of seismic wave field,which rises difficulties for establishment of the velocity field and migration imaging,etc. Furthermore, since the massive interferences are involved in seismic acquisition,the signal to noise ratio and resolution of seismic data is reduced. And then duringprocessing seismic data generated from complex surface and structure, how tomeasure the random noise suppression and structural information (such as folds andfaults, etc.) protection has become the key point.Due to the special nature of seismic exploration technology implementation, thecorresponding data processing has its unique property, different types of seismicwaves with time-distance curve are all described as seismic events in seismicexploration. Except to filtering based on time and spatial attribution, seismic dataprocessing can also be filtered based on the different tracks. Obtaining a nonstationarychange of amplitude values along the event (post-stack data structure) and applyingdifferent filtering methods constitute the structure-oriented filtering. This methodprotects construction information as well as suppresses random noise.Structure-oriented filtering consists of two core elements, which arecharacterizing of structure direction and nonstationary filtering. There are two structure-oriented strategies proposed in this study: one is to use characteristics oflocal dips of seismic event to build predictive data volume, and use predictive datavolume as a structure direction; the other is to determine local signal orientationdirectly from local dip. According to the different morphological characteristics ofpre-stack and post-stack seismic waves, three different methods are developed fordetermining the local dip of seismic events. First, for pre-stack CMP (common middlepoint) data, once relatively accurate velocities are acquired, seismic local dip ofpre-stack CMP data can be calculated by the corresponding formula derived frombasic time-distance curve relationship. For the post-stack seismic data, formulamethod is studied and concluded, respectively. Second, because partial derivativedirectly calculated from seismic data in the formula method will strengthen randomnoise energy with high frequency, this study proposes a local dip formula of seismicwave based on Hilbert transformation. Approximation in frequency domain (FiniteImpulse Response, FIR) is derived by Hilbert transform and the derivative operator, itis able to obtaine a stable non-iterative method for computing local dip of seismicevent. Finally, we review and compare the method of calculating local dip based onthe plane-wave destruction(PWD) are summarized and compared.In this dissertation, the content analyzes noise immunity, computationalefficiency, and spatial anti-aliasing capability of the above three dip determiningmethods. First, synthetic models with different signal to noise ratio are designed, theresults show an order for three noise immunity descending order: PWD, Hilberttransform, basic formula method. Second, with the same computer hardware condition,and a similar dip accuracy, computation cost of Hilbert transform and PWD areobtained. Computational efficiency of Hilbert transform is much better than the PWD.Finally, if seismic data show spatial aliasing, time-distance curve, Hilbert transform,PWD can appear the anti-aliasing capabilities.After one characterizes structure of seismic data, the study focus on selection ofthe appropriate filter. Commonly used filtering methods are various, many naturalsignals, including seismic data, are nonstationary, their property is more complicatedthan stable data. Nonstationary signals is also often referred to as time-varying signals, the reason is that some statistic of non-stationary signals (such as mean andcovariance function) is a time-varying function, and can not be simply interpreted as aconstant value or whether the waveform signal with time. Therefore, in the seismicdata processing, the filter method should be able to adapt to the characteristics ofnon-stationary signals. Non-stationary polynomial fitting is chosen in this dissertation.Compared with median filter, the research gives the analysis of its merits and demerits.According to the research, non-stationary polynomial fitting is more suitable fordealing with seismic event.Combination of different methods for determining local dip and different typesof filtering methods are applied to test several synthetic models, which includepre-stack models: CMP synthetic data, CMP synthetic data with AVO and post-stackmodels:"sigmoid" model,3D "qdome" model. Merits and demerits of the effects ofdifferent filters are summarized by evaluating local similarity coefficients. Theproposed methods in this dissertation are also applied to the field data, the results offield data show the effectiveness of two strategies. The dissertation also gives thescope of application according to different methods.