摘要
本文基于GPU/CPU协同系统,将计算量最大的波场逆时外推通过GPU实现,并利用随机速度边界的思路提高波场外推算法的并行性,解决了大规模存储的I/O问题。通过优化拉普拉斯算子压制由互相关成像条件引入的低频噪声。数值试验表明,GPU/CPU协同系统的计算效率非常高,在实际应用中取得良好的成像效果和时效比。理论模型试算和实际盐丘数据的处理验证了算法的正确性。
Based on the GPU/CPU co-parallel system,we achieve reverse-time extrapolation of the wavefield for the largest amount of computation through the GPU,improve wavefield extrapolation method parallelism using the idea of random velocity boundary,and solve the I/O problems of large-scale storage.Through optimizing the Laplace operator,the low-frequency noise from cross-correlation imaging condition is removed.Numerical experiments show that GPU/CPU co-parallel system has very high computational efficiency.In practical applications,we obtain not only high computational efficiency but also good imaging results.The theoretical model and real salt dome data processing resufts show the correctness of the algorithm.
出处
《石油地球物理勘探》
EI
CSCD
北大核心
2012年第5期712-716,844+676,共5页
Oil Geophysical Prospecting
基金
"天然气复杂储层预测与烃类检测地球物理技术研究及应用"(2011ZX05007-006)资助
关键词
逆时偏移
波动方程
成像条件
GPU/CPU协同计算
随机速度边界
reverse time migration,wave equation,imaging condition,GPU/CPU co-operating parallel computation,random velocity boundary