Optimal Iteration Algorithms And Their Application To The Attenuation Of Noise In Marine Seismic Data

With the deepening of marine oil and gas exploration,the target gradually extends towards the deep water.Due to the complex geological conditions,such as multiple ultralow velocity layers,abnormal high pressure,and vertical fissure development,etc.,the signal-to-noise ratio(SNR)and resolution of seismic data acquisition in the middle and deep water are relatively low.In recent years,the broadband,wide-azimuth and simultaneous source acquisition technology is developed to solve the marine deep detection.With the development of marine acquisition technology,corresponding seismic data processing techniques are also need to be developed.Among them,broadband variable depth streamer(VDS)acquisition technology needs to suppress receiver ghost caused by sea surface.This paper,considering the data features of interference wave for marine variable depth streamer,optimization algorithm is presented and applied in marine variable depth streamer deghosting and the supresstion of random noise.The superior results show that the proposed methods can effectively improve the SNR and resolution of marine seismic data,which can provide the high quality seismic data for marine oil and gas exploration.In this paper,the following optimization algorithms are developed: alternativeoptimization based on non-linear objective function,least-squares inversion for noisy data,Hankel low-rank approximation algorithm and optimal rank selection strategy.The research includes: adaptive optimization of iteration invertion algorithm to deghost for marine variable-depth streamer data in F-K domain and frequency-slowness domain,the hankel and low-rank structure analysis of the useful signal in seismic data,the application of hankel low-rank approximation and optimal rank selection strategy to seismic random noise suppression.Experiments on both synthetic and field seismic data demonstrate the superior and reliable performance of the proposed methods for seismic interference wave suppression.The main results and recognition of this paper are as follows:(1)The notch effect caused by receiver ghost is varying with different towed streamer depth.With the streamer towed deeper,the low frequency becomes stronger and high frequency is weaker.On the contrary,with the streamer towed shallow,the high frequency is rich the low frequency becomes weaker.Variable depth streamer acquisition can take advantage of both the deep towing depth and shallow towing depth,which is characterized by various notches.It is beneficial to suppress ghost and achieve broadband data.(2)Different variable depth streamer shapes can result in different boradband character.Undulate cable has better broadband than plow cable,and plow cable has better broadband character than slant cable.The slant cable has better broadband character than constant cable.The longer the slant portion of the plow cable is,the richer the high frequency components are.Otherwise,the low frequency components are relatively abundant.The shorter the changing wavelength of the undulate streamer is,the larger the fluctuation amplitude will be,and the low frequency and high frequency will be taken into account.(3)This paper derived alternative-optimization iteration algorithm based on nonlinear objective function,which can solve the objective function minimization problem with very high nonconvex inversion.Secondly,the least squares inversion algorithm for noisy data is derived.The objective function of the inversion is to minimize the residual noise under a least squares criterion,which improves the SNR of the inversion results.The Hankel low-rank approximation method(HLR)for noisy data is proposed,and the Hankel structure and low-rank property of noise-free data are considered.The alternativeoptimization algorithm is used to solve the HLR approximation problem,which provides a mathematical basis for interference wave suppression of marine seismic data.(4)In this paper,the least square inversion algorithm and the alternativeminimization iteration algorithm are applied to suppress the ghost in the frequency-wavenumber domain and frequency-slowness domain.The inversion stability can be improved by combining the mirror shot gather data and the original shot gather.In terms of the error of the ghost-delay time with respect to primary reflection and the amplitude difference coefficient between ghose and primary reflection variers with offset,the adaptive optimization iteration inversion algorithm is proposed to optimally calculate the amplitude difference coefficient and ghost-delay time between ghost and primary reflection.At the same time,the up-going wave can be extracted.The deghosting results can be improved in rugged submarine and long offset acquisition.(5)This paper proposed hankel low-rank approximation(HLR)method and provide rigorously convergence analysis of the proposed algorithm.Experiments on both synthetic and field seismic data demonstrate the proposed HLR method can better suppress seismic data than the traditional low-rank approximation methods.Meanwhile,the paper proposed an optimal rank selection strategy which is data driven and is independent of the noise parameters such as the variance.The proposed optimal rank selection can be incorporated into the classical LR approximation method and many other recently developed methods such as those by shrinking the singular values.Experiments on both synthetic and field seismic data demonstrate the superior performance of the proposed rank selection strategy for seismic data denoising.The innovation of this paper is mainly reflected in the following two aspects:(1)The paper proposed the least square inversion algorithm and the alternativeminimization iteration algorithm and applied them to suppress the ghost in the frequencywave-number domain and frequency-slowness domain.This method can optimally calculate the amplitude difference coefficient and ghost-delay time between ghost and primary reflection.At the same time,the up-going wave can be extracted and the deghosting results can be improved.(2)The paper proposed hankel low-rank approximation method,which consider both the hankel structure and the low-rank structure of useful signals.The proposed HLR method can better suppress seismic data than the traditional low-rank approximation methods.The paper also proposed an optimal rank selection strategy and applied to the traditional low-rank approximation methods,which can adaptively select the rank.