Method Study On Precise Identification And Processing Of Seismic Signal

Seismic survey is the most commonly used method for the oil and gas exploration.The key of oil and gas exploration lies on the precise imaging of the subsurface structure and the recognition of the reservoir feature,and these two aspects are highly depending on the quality of seismic data.Thus,seismic data processing,especially its early stages,plays an important role in the whole industrial exploration workflow.This paper focuses on the three early key proce-dures of the seismic data processing.First,analyze the problems existing in first-break picking,signal denoising and missing data reconstruction.Then,propose the corresponding improved algorithm or totally new framework.At last,conduct synthetic and field data experiments to validate the superiority of the proposed methods.First,seismic waveform identification(first-break picking)is a significant step in the seis-mic data early processing.For instance,precise picking result can provide reliable input for the next static correction step,and consequently output more accurate time-shift inversion values.However,in some highly noisy seismic data,signal(first break)waveforms are submerged by noise,resulting in significant errors of the picking results,which cannot be used for subsequent correction.To solve this problem,this paper proposes a new method based on the iterative Gaussian Mixture Model(IGMM)to enhance the stability and precision of arrival picking in the extremely noisy environment.This method firstly uses the local singular value decomposi-tion(LSVD)to roughly find the first-break waves and divide the seismic section into first-break and non-first-break parts,which can reduce the searching range effectively.In the first-break area,it utilizes the feature that first breaks have stronger inter-traces coherency and amplitude envelop compared to the noise to extract multi-channel attributes,then input them to IGMM to separate the first-break points from the non-first-break points.Synthetic data and three field examples all show that this method has stability in the strong noisy environment,and its picking precision is better than the conventional methods.Signal denoising is also a key step in the field of seismic data processing.Proper denois-ing procedures can effectively enhance the regularity and consistency of the seismic data,and reduce the errors of subsequent imaging and inversion.Traditional denoising methods usually rely on the assumption that distribution of noise is even over the whole section,and use one set of unchanged parameters to process it.When the situation of noisy data is relative simple,this strategy can achieve good result.However,most of the real seismic data are not stable in differ-ent time-space locations,thus this assumption is not available for the complicated noisy data.To solve the problem,this paper adopts the non-stationary processing strategy,improves the con-ventional structural filter and proposes the non-stationary structural filter.Adaptive smoothing radii are applied to the optimization operator during the estimation of slope field.More specifi-cally,in the noisy part,small radii are used,and vice verse.The test results of synthetic and field noisy data show that this method significantly improves the accuracy of slope field estimation and enhances effect of subsequent structural filtering.In addition to the signal denoising,reconstruction is another tool to improve the quality and quantity of seismic data.Due to the complicate surface conditions or acquisition cost,raw seismic data tend to contain blank area(or missing traces).Filling the missing data via proper reconstruction methods can make the seismic events more consistent and enrich the reflection information.Commonly used reconstruction methods suffer from low processing speed and be-ing incapable of dealing with the extremely sparse data.To alleviate these problems,this paper proposes two correspondingly novel methods,the fast dictionary learning based on sequential generalize K-means(SGK)and Sparse5 D fast reconstruction framework.The former one re-places dictionary updating based on K-SVD and the sparse coding procedures with the SGK operator,which obviously speeds up the reconstruction process without degrading the recon-structed results.The latter one is a totally new framework.By merging the adjacent common-midpoint gathers,an initial model with richer event information can be constructed.Then,a 2D filter is used to further enhance the regularity of signals and fill in the trace gap.This algorithm exhibits an excellent robustness in the extremely sparse 5D data,and its processing efficiency is one order of magnitude higher than the other methods.Overall,first-break picking,signal denoising and data reconstruction are three key pro-cedures in the seismic data processing,and have direct impact on the following imaging and inversion.Aiming at the problems in these three processes,this paper proposes correspond-ing improved algorithms.These novel methods enhance the precision of first-break picking,fidelity of the denoising,speed and effect of the reconstruction,providing the key technologies for developing more robust seismic data processing workflow.