Research On Acoustic Reverse-time Migration And Least-square Reverse-time Migration

Migration is the key technology of seismic exploration.With the development of exploration in oil industry,there is an increasing requirement for a sharp and amplitude-preserving imaging result.Reverse time migration(RTM)offers a clear image of the underground for larger dip angles and complex velocity contrasts.However,several difficulties arise for the RTM image,such as low-frequency noise and weak energy in the deep layers.Additionally,these problems lead to an image with low resolution.Involving RTM into the inversion framework,Least-square RTM and produces image with high resolution.Because of the inaccuracies on gradient calculated by cross-correlation imaging condition and the large computational cost in the iterative method,LSRTM suffers from inaccurate image result and heavy computational cost.The main aim of our research is to obtain higher resolution result and decrease the computational cost.The research foucuses on reverse time migration and least-square reverse time migration method.The focuses are specifically on the finite-difference forward modeling,absorbing boundary conditions,imaging conditions,amplitude preservation,storage strategies,and parallel computing strategies.The results obtained are as follows:(1)The wavefiled decomposition method,including the upgoing and downgoing wavefield decomposition method,leftgoing and rightgoing wavefield decomposition method,upgoing downgoing leftgoing and rightgoing wavefield decomposition method is implemented.A multi-direction wavefield decomposition method is developed to elimate the low-frequency noise.Comparing with the traditional RTM and other wavefield decomposition method,this method shows high resultion and overcome the disadvantages in other wavefield decomposition method.(2)The traditional finite-difference method,time-space finite-difference method and optimal time-space finite-difference method are compared by analysing the dispertion relationship and numerical simulation accurancy.The optimal time-space finite-difference method in this research demonstrates higher resolution that other methods.And the optimal finite-difference method is introduced into LSRTM in order to increase the quality of LSRTM image.At the same time,a time accumulation method is used to correct the distortion after the Laplacian filtering.Finally,a comparison is made with the LSRTM results calculated with optimal time-space finite-difference method and time-space finite-difference method.The results show that optimal method can produce image with higher resolution and better iteration speed.(3)Adaptive variable-length spatial operators strategy,an effective saving boundary and improved conjugate gradient method is introduced into LSRTM.At the same time,a hybrid accelerating method combining with MPI and CUDA is introduced.A comparison is performed between single node and multi-nodes GPU cluster.The results show that the hybrid accelerating method improves the calculation efficiency in the computation of LSRTM.(4)Research on the implement of RTM and LSRTM with model data and real data.The model test shows that these methods are stable and effective.In additionally,the multi-direction wavefield decomposition RTM with real data is tested.The result shows that this method produce images with higher resolution.