| The development of economy has promoted the increase of energy demand,which has made the oil exploration in the deep-water area develop vigorously.At present,in the deep-water area of marine,the underground structure is complicated,small structure development,oil and gas buried deeply,overlying layer thickness is large and reflection energy in deep profile is weak,conventional seismic processing and imaging methods can not meet the later interpretation requirements.The multi component seismic data acquisition by the ocean bottom cable contains rich seabed information with full azimuth angle and large offset,full waveform inversion can make these effective informations to build a velocity model,it is a high precision imaging method.However,due to the limitations of the method itself,such as the large amount of memory in the calculation process,time-consuming calculation and other issues,it is still in the theoretical stage and difficult to put into practical production.In this paper,we focus on the main problems existed in the time domain full waveform inversion,and combined with the characteristics of OBC multi component data.A complete elastic time-domain full waveform inversion of multi-component OBC seismic data is completed.Detailed studies of the forward efficiency,full waveform inversion theory,gradient calculation and gradient pre-processing are carried out,the main results are as follows:(1)The time-domain parallel forward modeling of time-domain FWI of any even order based on MPI is realized.The technology can advance the development of FWI in time domain.The parallel design based on region decomposition,by region decomposition,forward modeling required large memory is distributed into different computing nodes,and the calculation is divided into several processes and compute simultaneously,greatly improve computational efficiency.The parallel forward computation can be divided from X direction and Z direction,and the number of parallel processes can be chosen freely.(2)The CFS-NPML boundary condition is implemented in forward modeling.In the process,a large number of convolution calculations of conventional CPML are avoided by introducing auxiliary variables,and combined with the characteristics of staggered grid,the time of the auxiliary variable and the discrete wave equation expression of auxiliary variables are given.CFS-NPML boundary conditions only needto store auxiliary variables in the boundary,no need to split the wave field,the storage space is small and the mothed realized simple.CFS-NPML boundary condition can not only avoid the evanescent wave generated by the PML boundary condition at large angle,but also suppress the boundary reflection in the case of a few boundaries.(3)The technology of wave field reconstruction based on CFS-NPML boundary storage and random boundary is realized.Calculate instead of storage,the problem of a large amount of memory required by storing forward wave filed when the full waveform inversion gradient is obtained are solved.From the point of view of memory required,wave field boundary storage need to be stored the wave field values at all moments at the boundary and the wave field values at the last time of the whole model,Where CFS-NPML boundary condition requires the least variables at the boundary and well suppressed the boundary reflection with fewer grid points,so the needed memory is smaller.The random boundary wave field reconstruction method only needs to store the wave filed of the whole model at the last time.so the required memory is minimal.From the calculation point of view,based on the boundary storage method is equivalent to one more time of forward modeling,but only need to consider the internal calculation area,PML boundaries do not need to be calculated.The results of the inversion based on the boundary storage method are consistent with the conventional full wave field store method,and the results based on the random boundary method have a great influence on the gradient due to the random diffraction energy generated by the random boundary,which leads to poor inversion.(4)The energy preprocessing of seismic wave can avoid the complex solution of Hession matrix and its inverse,so improved the inversion accuracy in depth.Where the use of residual energy to calculate of detection point lighting,can directly obtained in the detection point residual back propagation process,without extra computation,but there will be some abnormal disturbance illusion.While the use of synthetic records will not lead to abnormal disturbance of illusion,but need to be an additional back-propagation.The approximate Green’s function method with minimum amount of computation,is used to assume that all the points on the surface of the earth have receivers,and the ground is a uniform model.The inversion result of the inverse synthetic record method is the best,the next is the approximate Green response method,and the last is the inverse of the residuals.(5)A multi-scale full waveform inversion based on Butterworth autocorrelationlow-pass filter is proposed.Butterworth filter is relatively flat in the passband and monotonically decreasing out the passband,it is a non-leaky low-pass filter.Butterworth filter autocorrelation is zero phase filter system,which can ensure that the seismic wavelet and seismic signal filtered waveform unchanged.On this basis,the corresponding frequency selection strategy is given.By the strategy,the cutoff frequency can be directly changed to realize the inversion of the next higher cutoff band,and the time domain multi-scale full-waveform inversion can be realized directly and improve the convergence of inversion.Three convolution interpolation techniques take into account the influence of the adjacent 16 points,the calculation accuracy is high,which can keep the fine structure well of the image,and the error with the true value is small.The interpolation technique can achieve the interpolation of the coarse grid size and the fine grid size.(6)For P-wave,the Vz component of the near-offset is larger than the Vx component,and the Vx component of the far offset is larger than the Vz component.So for the P-wave velocity inversion,the Vz component of the near offset and the Vx component of the far offset should be used.For S-wave,the Vx component of the near offset is larger than the Vz component,and the Vz component of the far offset is larger than the Vx component.So in the inversion of the S-wave velocity,the Vx component of the near offset and the Vz component of the far offset should be used.The results of the hydrophone component and the Vz component inversion are not very different.The P-wave velocity is better in the deep.While S-wave velocity in the deep better in Vx component inversion.In the inversion of joint Vx component and Vz component,the P-wave velocity and S-wave velocity in the deep both are better.Simultaneous inversion of P-wave and S-wave velocity and density,because of the coupling of multi-parameters,the density inversion error is very large.At last,the inversion strategy of OBC multi-component is proposed.Firstly,joint the Vx and Vz components to invert P-wave velocity and S-wave velocity,then taken the P-wave and S-wave velocity and initial density obtained from the inversion as a new initial model,to invert P-wave velocity,S-wave velocity and density. |
