Methods And Applications Of Pattern Characterization For Nonstationary Seismic Data

Seismic exploration is an important tool for exploration of oil,gas,solid minerals and other resources,and currently oil and gas exploration is gradually expanding into“two deep” and “two new” areas,“deep water and deep layer area of the sea”,“new areas such as new polar regions and unconventional oil and gas”.At the same time,the exploration direction is gradually changing from conventional to unconventional oil and gas resources-tight oil and gas reservoirs,coal-bed methane,shale gas,and other unconventional oil and gas.These unconventional reservoirs are often accompanied by“three complexities(complex surface,complex structure,and complex lithology)”.Meanwhile,“wide azimuth,broadband,and high-densiy” integrated acquisition method is widely used.All of these pose higher requirements and new challenges to the traditional seismic data processing technology.On one hand,in the field data acquisition process,the complex near-surface geological conditions can severely limit the deployment of the observation system,and the field data exhibit irregular or undersampled distribution characteristics in space,while many subsequent processing steps are based on regularly distributed data as a prerequisite,and seismic data interpolation techniques provide an important way to solve this problem.On the other hand,noise interference is also unavoidable in seismic exploration.Random noise is caused by a variety of factors.In addition to environmental noise such as wind disturbance,the coherent noise generated by conventional processing of residual surface waves and efficient simultaneous-source data,etc.also show strong random noise characteristics in the special sorting gather;multiple is mainly generated by the reciprocal propagation of seismic waves between strongly reflective interfaces;ground-roll noise in onshore exploration is mainly caused by the propagation of seismic waves near the surface.It is difficult to obtain accurate effective wave information from seismic data with serious noise pollution,so noise suppression has been the core issue in high-precision seismic exploration.With the development of “wide-azimuth,broadband,and high-density” data acquisition technology,the amount of acquired field data has increased significantly,and conventional inefficient processing methods tend to seriously slow down the progress of data processing,and the computational efficiency of seismic data processing methods requires attention.Seismic data are affected by many non-deterministic factors in the acquisition process,such as the transformation effect of complex media on the source wavelet,non-repeatable stimulation and reception conditions at different times and various noise interferences,etc.Due to the limited observation conditions,the seismic data cannot yet be represented by a deterministic time(or space)function,and exhibit uncertain timevarying,space-varying and frequency-varying characteristics in physical form.Although sometimes the seismic data exhibit some deterministic characteristics locally,they are essentially regarded as nonstationary stochastic processes and can be defined as nonstationary seismic data,which have some unique patterns.Ignoring the corresponding characteristics of seismic data and rigidly applying methods from other fields to seismic data often do not yield the desired results.Therefore,it is necessary to deeply analyze the specific patterns of nonstationary seismic data and study the corresponding characterization methods in order to obtain more reasonable seismic data processing results.This dissertation focuses on characterization methods of the powerspectrum prediction and amplitude sparsity for nonstationary seismic data.The representation methods for power-spectrum prediction include adaptive predictionerror filtering method,streaming prediction filtering method,and streaming local polynomial fitting method,meanwhile,the characterization methods for amplitude sparsity include linear Radon transform method and stationary seislet transform.This dissertation firstly investigates the theory of adaptive prediction-error filtering in the time domain,analyzes the differences in the characteristics of adaptive prediction-error filtering operators corresponding to effective seismic signal,random noise,and ground-roll noise,and also gives the construction method of the nonstationary-pattern characterization operator.The dissertation establishes the leastsquares equations of the pattern characterization corresponding to the signal-noise separation problem,combines the pattern-based signal-noise separation framework with the adaptive prediction-error filter,and proposes a new two-step signal-noise separation technique to improve the separation accuracy between the nonstationary seismic signal and the two types of noise.The adaptive prediction filtering theory in the frequency domain is studied,the streaming prediction filtering method is extended to the frequency domain,the extended Sherman-Morrison algorithm for the solution of complex inverse matrices is derived.The classical "amplitude oscillation" problem in frequency-domain predictive filtering is solved by means of multi-directional smoothness constraints,and the highdimensional structure of the filter is designed.Development of a “snake” processing path to match streaming computation,which solving the multiple initialization problem of streaming prediction filtering in high-dimensional data processing,effectively suppresses the boundary effect.The proposed high-dimensional streaming prediction filter in frequency-space domain balances the amplitude fidelity of nonstationary seismic signals and the computational efficiency of large data volumes in the seismic random noise suppression and missing data interpolation problems.The theory of the adaptive polynomial fitting is also studied,and the local smoothing constraint is introduced into the solution of the adaptive polynomial fitting coefficients,furthermore,the smoothing constraint parameters are designed to vary with the data.A new mathematical inverse problem equation corresponding to the adaptive polynomial fitting is proposed,and a non-iterative analytical algorithm is given through the streaming computation framework.Finally,a streaming local polynomial fitting method is proposed to better characterize the drastic changes in data amplitudes to protect the complex structural details.The basic principle of linear Radon transform is investigated,and three implementation methods of linear Radon transform with amplitude preservation based on radial trace transform,slope decomposition,and seislet transform are proposed.Through theoretical derivation,the detailed calculation processes of different algorithms for linear Radon transform with amplitude preservation are given.Meanwhile,the accuracy and computational time of the forward and inverse transforms for these three methods are tested and compared with the mainstream algorithms.The linear Radon transform domain under different schemes provides a more effective and flexible implementation scheme for removing ground-roll noise in the shot gather.The basic principle of seislet transform is studied,meanwhile,a new stationary seislet transform method is proposed based on the theoretical framework of seislet transform.The relationship between the data at all scale levels of stationary seislet transform is analyzed,and the translation invariance of the transform method is tested.The stationary seislet transform predicts and updates the data based on the continuity of events or reflection layer,thus determining the structure information,such as faults and unconformities in seismic data,by the decomposed high frequency information.The stationary seislet transform can decompose the data into different scales,and the differential processing for different scales of data can better suppress the noise.Combining stationary seislet transform method with projection onto convex set method provides a new technical solution to the problem of missing data reconstruction.In this dissertation,several new methods to effectively characterize nonstationary seismic data patterns are proposed through the study of power-spectrum prediction and amplitude sparsity characterization methods for seismic data.It provides the theoretical basis and technical support to solve the problems of nonstationary data reconstruction and noise suppression of seismic data.