伪深度域交错网格逆时偏移成像方法及并行优化

基于声波方程的逆时偏移作为现阶段最主要成像手段之一,主要受限于计算机运算速度及内存占用。为减少模型网格数以提高计算效率和减少计算存储占用量,将曲坐标变换引入到正演及逆时偏移中,推导了伪深度域一阶声波方程及其离散形式,并给出了对应带PML吸收边界的离散方程。进一步,通过C++AMP并行架构实现了相应正演及逆时偏移算法优化,在减少内存占用的基础上还极大地提升了计算效率。最后,通过模型试算验证了算法及并行架构的正确性及高效性。

Parallel optimization of reverse-time migration on staggered grids in pseudo-depth domain

Reverse time migration based on the acoustic wave equation is one of the most important imaging methods at present,but it is constrained by computer speed and memory use. To reduce model grids for improving calculating efficiency and reduce memory use,this paper introduces curvilinear coordinate transform into forward modeling and reverse time migration,derives the first-order acoustic wave equation and its discrete form in the pseudo-depth domain,and gets the discrete equation with PML absorption boundaries,and then optimizes the algorithm of forward modeling and reverse time migration on the C++ AMP parallel architecture. The algorithm can greatly improve calculating efficiency while reducing memory occupation. Model test has proved the algorithm and parallel architecture correct and effective.