3D TTI Reverse Time Migration And Optimization Of Multi-GPUs

With the rapid development of oil and gas prospecting and increasing complexity of exploration target, one-way acoustic wave equation migration based on the tradi-tional assumption of isotropic is insufficient to achieve practical exploration require-ments,particularly wide azimuth seismic data. Anisotropic reverse time migration method has attracted more and more attention . Accurate anisotropic imaging is rec-ognized as a breakthrough in areas with complex anisotropic structures. Tilted trans-versely isotropic (TTI) media are typical earth anisotropy media from practical ob-servational studies. TTI reverse time migration (RTM) is an important method for these areas. In this paper, We implement a fast algorithm for 3D TTI RTM based on multi-GPUs. The research mainly contains the following aspects:(1). Using an tenth-order spatial and second-order temporal accuracy stencil, this paper implements the TTI wave propagation simulation on a regular grid. Besides, the stable condition of numerical simulation is derived to implement fast algorithm of 3D TTI RTM. And adopting BP2007 TTI and 3D salt TTI model to validate the correct-ness and effectiveness of our multi-GPUs-based TTI RTM method,which achieves the great performance and imaging results.(2). Adopting matching of the anisotropy parameters to suppress the SV artifacts,which are coupled with P wave in the pseudo-acoustic wave equation.(3).Updating acoustic random velocity boundary and extending it to TTI wave equation aiming at the problem of large wave-field storage in RTM, which greatly reduce storage and 10.(4). Discussing the optimization method of GPU parallel computing, from three aspects: code parallelism,kernel and CPU / GPU-GPU interaction.(5). Extending the single GPU code to multi-GPUs and presenting a correspond-ing high concurrent strategy with multiple asynchronous streams. At the same time,the optimization method is extended to forward of the conventional acoustic equation.Both of them have achieved the ideal acceleration ratio 2: 1, which greatly improves the computational efficiency.