Research Of High Accuracy Full Waveform Inversion With Nonsmooth Constraint

Seismic velocity model rebuilding is always the key step in seismic data processing and interpretation.Velocity model describes the changes of speed of seismic wave propagation in the underground medium and affects the quality of migration imaging directly.It is one of the most fundamental and most important parameters in seismic data interpretation and geological interpretation.In recent years,with the rapid development of global economy and the increasing mature of geophysical prospecting technology,the goal of oil and gas exploration has turned from the land and the shallow ocean area to the deep ocean water area gradually.Although the oil and gas resources are abundant in deep ocean area,the oil and gas exploration is facing new challenges due to the deep target layer and complicated geological conditions.The traditional velocity model rebuilding methods such as stack velocity analysis,ray tomography,migration velocity analysis and so forth usually cannot provide a high quality velocity model in terms of accuracy and resolution and cannot meet the requirements of geophysical and geological interpretation of these complex areas.The full waveform inversion(FWI)which has become a popular velocity model rebuilding technology in recent years is a kind of high precision and high resolution inversion method of underground medium velocity.It can provide a high precision velocity model for pre-stack depth migration by taking advantage of the amplitude and phase information in seismic data and updates the medium velocity and minimizing the error between the simulated data and real data.Although the superiority of full waveform inversion technique is obvious in theory,there are still many problems in the practical application due to the limitations of the technology itself such as strong nonlinear,shallow depth of effective inversion,low computing efficiency,large memory and strong dependence of the initial model and low frequency information.In this paper,we perform a research on the application of frequency domain full waveform inversion in the ocean data and put forward the corresponding solutions according to the problems and difficulties in application.Based on the total variation constraint and depth weighting method,this paper proposed a joint full waveform inversion algorithm which can not only improve the effect of the deep part,but also reduce the nonlinear during inversion,meanwhile,reconstruct the discontinuity interfaces in the velocity model effectively.The offset is limited on account of the acquisition of ocean data is often towed streamer acquisition geometry,therefore the signals of diving wave and refraction from deep layers are weak or cannot be acquired which leads to the poorer inversion effect in the deep layer.In addition,the geological condition is complex and the change of velocity is discontinuous in practice as well as the high velocity region and speed abrupt change interfaces usually exist which enhance the nonlinear of the inversion greatly and affect the precision of inversion results.The medium velocity can be layered inversion by depth weighting method.The weighting of shallow signal is apparently higher than the weighting of deep signal in actual data.The weighting of deep signal can be increased gradually in the layered inversion and the deep inversion effect can be improved.The total variation regularization method as a kind of unsmooth constraint can deal with the discontinuity results effectively,reconstruct the discontinuous interfaces and keep the edge information of high velocity region.In this paper,we apply the depth weighting method and the total variation regularization to full waveform inversion in frequency domain.This new method,a combination of the two kinds of technology advantage,can not only increase the effective depth of the inversion,improve the inversion effect of deep part,but also improve the reconstruction effect of discontinuous interfaces,thus has improved the precision of the full waveform inversion.The dependence of full waveform inversion on the initial velocity model is strong.When the accuracy of initial model is very low,the cycle skipping phenomena will appear and make the inversion trapped in local minimum.So we can't get a correct inversion result.In this paper we propose a vertical total variation constraint full waveform inversion based on the hinge loss function.The L1 norm of the hinge loss function of the total variation in the vertical direction and the total variation constraint are used as the regularization constraint of the objective function at the same time.They can restrain the model updates in the vertical direction.We adopt a loop iteration inversion strategy that gradual increase the threshold value and decrease the weight of the total variation constraint item continuously during the inversion.The inversion result of a smaller threshold value is the initial model of the inversion with a larger threshold value.In the end the method can not only reconstruct the discontinuous interfaces and keep the edge information of high velocity region and greatly reduces the dependence on the initial velocity model.Forward modeling is the core part of full waveform inversion,so choosing a suitable forward modeling method determines the precision and efficiency of full waveform inversion.We need calculate the inverse of large coefficient impedance matrix in the frequency domain full waveform inversion.This needs huge computational resources for three dimensional case and even it can't be solved sometimes for the existing situation of computer hardware.Therefore,this paper proposes a three dimension multi-scale full waveform inversion algorithm in hybrid domain.We use the forward modeling method in time domain,and then extract the single frequency wave field from time domain wave field via discrete Fourier transform for the frequency domain inversion.This method is easy to apply,has little demand for computer memory and combines the advantages of both the time domain forward and the frequency domain inversion.Full waveform inversion need multi source forward modeling for each iteration optimization.Especially in dealing with practical problems,the huge calculation increases the computation cost and reduces the computational efficiency.In this paper,we apply the parallel strategies based on MPI and the source coding technology to the full waveform inversion algorithm and divide the model into a number of different areas.The wave field of different parts is calculated by different CPU,and different CPU needs to communicate with each other to exchange the boundary wave field data.Meanwhile,we use random phase encoding technology to perform multi source simulating at the same time.This method greatly improves the computational efficiency,and can suppress the effects of crosstalk noise effectively.As a kind of high precision velocity modeling method,full waveform inversion has a high demand for data quality and the initial model.In practical application,it cannot be used independently,and often needs to combine with other data processing technologies to achieve an ideal inversion effect.Based on the predecessors' research,this paper proposes a full waveform inversion velocity model rebuilding process which is suitable for towed streamer data.This process mainly includes the following several parts.Firstly,pre-process the raw data which includes band pass filter,geometric spreading compensation,multiple wave suppression,tow-dimensional spread transformation,Fourier transformation and data normalization.Secondly,estimate the source wavelet which is in order to eliminate the influence caused by the differences between inconsistent wavelets.Thirdly,build initial velocity model by using other methods,such as migration velocity analysis,travel time tomography and the Laplace domain full waveform inversion.Fourthly,determine the parameters of the full waveform inversion and set acquisition system according to the actual data.Fifthly,rebuild the velocity model using full waveform inversion.Finally,test the accuracy of the full waveform inversion results using Kirchhoff pre-stack depth migration.In this paper,this process is applied on a set of towed streamer data and a set of plow cable streamer data respectively.The tests show that the full waveform inversion technique is feasibility and effectiveness in practical application meanwhile the process we use in this paper is also feasible.