地震叠前逆时偏移算法的CPU/GPU实施对策

Computational strategy of seismic pre-stack reverse time migration on CPU/GPU

相较于单程波偏移算法而言,逆时偏移成像方法以其物理基础为依托优势,几十年来一直备受国内外地球物理学家的青睐.目前的逆时偏移(RTM)若直接采用双程波动方程进行延拓,尽管可以回避上下行波的分离处理,然就已有算法而言,其计算量和I/O(输入/输出)量却是最大的.针对此问题,本文在分析现行逆时偏移的多种算法基础上,提出利用CPU/GPU(中央处理器/图形处理器)作为数值计算核心,建立随机边界模型,从而克服存储I/O难题和提高计算效率.在实际的数据测试中,本文的方法可以大幅度的提高计算效率和减少存储单元,从而促使其高效地应用于生产实际.

Comparing with one-way wave migration algorithm, reverse time migration （RTM） is more attractive because of the theory advantages. Two-way wave equation has been used to extrapolate wave field in RTM, instead of separating the up going wave and down going wave. However, due to the large amount of computation and I/O, RTM is most time-consuming in industrial applications. In this article, we analyze several computational strategies and propose our method, which uses CPU/GPU as computational core and builds random velocity boundary, for solving I/O problem and computational efficiency problem. In the actual test, it has been proved that this method can largely decrease storage memory units and improve computational efficiency.