The Research Of Forward Modeling Based On CPU/GPU Cooperative Acceleration In VTI Media

The forward modeling of seismic waves has important theoretical and practical significance for people to understand the propagation law of seismic waves correctly,verify the correctness of the Earth model,and perform geological interpretation and reservoir prediction of actual seismic data,as well as the development of Earth resources.In traditional seismology and seismic exploration,isotropic seismic data processing methods are used to process seismic data of anisotropic media,which leads to reduced seismic resolution and imaging errors.Therefore,the study of seismic anisotropy forward modeling is a necessary part of seismologists to study underground media.In the forward modeling of anisotropic media,the computational efficiency is reduced due to the complexity of the wavefield component parameters of the elastic wave.At the same time,considering the existence of the converted wave field at the reflection interface of the elastic wave forward modeling,the pseudo-acoustic wave with the acoustic approximation can directly reflect the underground medium structure.However,there is still a huge gap between theoretical maturity and the practical use of production in forward modeling,which is the computational efficiency.In actual production,the blocks that need to be simulated are often very large,and the number of meshing points defined thereby is also extremely large.However,the finite difference method has high parallelism.When performing time step updating,the calculation of each grid point is independent,and only depends on the values of the previous grid points that have been calculated.Therefore,the calculation of each grid point can be performed in parallel.Therefore,this paper uses the CPU/GPU cooperative acceleration technology.Based on CUDA,the wave field extension part of the high-order differential forward modeling is transferred from the CPU to the GPU to further improve the calculation speed,thus achieving better economic benefits.Based on the basic theory of anisotropic media and the basic theory of finite-difference forward modeling in staggered grids,the paper derives the elastic wave equations in isotropic media and VTI media.Based on the idea of acoustic approximation,the pseudo-acoustic wave fluctuations in VTI media are derived.The equation is discretized by the finite difference method under staggered grid,and the difference scheme of its absorbing boundary condition is derived.Then,the CPU/GPU co-acceleration theory is analyzed.Combined with the computational characteristics of large computation and high parallelization in finite difference forward modeling,the CUDAC language is used to accelerate the parallelization of the wavefield extension.Finally,based on the principle of CPU/GPU cooperative acceleration,the elastic wave forward modeling of isotropic media and VTI media is realized.The results show that the elastic wave forward modeling produces complex converted wavefield at the interface.On this basis,the pseudo-acoustic forward modeling of 2D and 3D velocity models in VTI media is carried out,and the idea of 3D data volume calculation using 2D thread blocks is used to improve the tolerance of model size in forward modeling.The correctness of the CPU/GPU cooperative acceleration algorithm is verified by model test,and the effect of parallel acceleration and its production application are analyzed.