Reverse-time Migration Imaging And GPU Parallel Implementation Of3D Multi-wave Seismic Data

Seismic migration imaging is the core process of seismic data processing, whichis also a time-consuming part, and usually turns out to be the actual productionbottleneck. As one of the best seismic migration imaging technologies currently,calculation speed and storage demand of reverse time migration (RTM) are theproblems that must be faced, therefore, this paper use a heterogeneous computersoftware platform that combined by GPU (graphics processing unit) and CPU(Central Processing Unit) to accelerate the process of RTM imaging of seismic data.This paper first discusses the operation mechanism of GPU briefly and thendescribes the related CUDA programming theory. Based on the basic principle ofelastic wave RTM, use high-order finite difference method in staggered grid to solvethe three-dimensional first-order velocity-stress elastic wave equation, then derived itshigh-order finite difference scheme, and use perfect match layer(PML) to suppress theboundary reflection. Extend the source wave field and receiver wave field ofthree-dimensional elastic wave equation respectively, during which P-waves andS-waves decoupling and three-component S-vector standardization are accomplished.Choose the proper imaging condition to complete RTM imaging of elastic wave. Asthe fact that imaging condition plays a vital role in RTM, this paper devotes to studythe mechanism of noise in RTM and suppress those noise by using the normalizedwave-field separation cross-correlation imaging condition and Laplace filter.Meanwhile, GPU high performance computing and random border strategy areintroduced to solve the mainly problems of RTM, namely, massive computation andhuge hard drive storage. At the premise of ensuring the high accuracy of migrationoperators, the parallel computing strategy has greatly enhanced the computationefficiency of RTM, and maximized the GPU’s parallel computing capabilities. Finally,this paper accomplishes the parallel algorithm on CPU and GPU platform of2D and3D elastic wave pre-stack reverse time migration, and proves the RTM imaging effects GPU acceleration effect through the different model computation. The modelcomputation shows that:(1)In three-dimensional isotropic medium, the completely P-waves and S-wavesdecoupling can be implement by solving the divergence and curl field of elastic wavefields, thus obtained pure P-waves and three-component S-wave vector. Since thecross-correlation imaging condition of3D elastic RTM requires a scalar, sothree-component S-wave vector needs to be scalarization into a scalar S-wave;(2)Improve the class cross-correlation imaging condition by introducingPoynting vector. The normalized wave-field separation cross-correlation imagingcondition and Laplace filter can suppress the interbed reflection and low-frequencynoise effectively and thus a more accurate imaging result can be acquired;(3)It is easy to tell that the migration algorithm has a good scalability and linearacceleration in GPU computing platform, At present, GPU calculation speed of elasticwave RTM has reached about20times of the CPU calculation speed; Based on thecore ideal of using calculation instead of storage, choose the random boundarystrategy to avoid the drawback of storing a large number of wave field data duringRTM. The comparison of several imaging results of various models shows that therandom boundary strategy turns out to be the best method for GPU computing inRTM because of its high computational efficiency.