克希霍夫偏移成像分块策略对文件I/O的影响
详细信息 本馆镜像全文    |  推荐本文 | | 获取馆网全文
摘要
现在三维地震勘探数据通常都在TB量级,地震勘探处理中文件I/O的时间花费越来越引起了人们的关注。常规的偏移成像算法分块通常按照先inline再xline的顺序来分块,这样分块看上去比较简单直接,而且算法实现上比较容易。但这样分块会导致整个偏移过程中总的文件I/O量与原输入数据相比特别大。常规分块算法对小规模地震数据文件处理时,增加的文件I/O所花费的时间与偏移的时间相比不是很明显,因此因分块导致的文件I/O量的增加常常被人们所忽视。当处理的地震数据量大时,如现在的TB量级的三维地震数据,常规的分块算法导致I/O量的增加就突出了,通常达到了原数据的700~800倍,相当于读入700~800TB的数据,由此所消耗的读取时间也是惊人的,若换算到由单结点去读取,则需要花费4个月的时间。为此,我们提出二维分块算法,其基本思想来源于偏移成像的有效地震道分布依赖于定义的偏移孔径,将成像分块的方式从常规的inline优先改为inline和xline方向同等,形成一个个接近正方形的成像区域。在结点能装载的成像区域同等情况下,二维分块算法完成成像所需要读入的地震道数量比常规分块算法要明显的少。
Seismic measurement data processing is an important and indispensible part of seismic exploration. The I/O takes a large portion of time in processing huge amount of data. In 3D prestack time migration, the input data file is mostly in TB size. The traditional block division algorithms divide the image area in the order inline first then xline. The block division is relative simply and directly, and easy to implement. A core problem coming in tradition block division algorithms is the huge disk I/O access. When processing a small seismic data, the time proportion extra file I/O to migration is inconspicuous, and it is always neglected that the extra file I/O according to the migration division. As now we need process the 3D seismic data, the data size mostly in TB size level, the amount of file I/O in the migration processing of tradition division algorithms is remarkable, and in most cases it will come to 700-800 times of the original data size, which is equivalent to access 700-800 TB data. Only the time of disk I/O at a single node will take over four months. In this paper we propose a 2D division algorithm, which aims to decrease the amount access of disk I/O in seismic data processing. We divide the migration image area in both directions of inline and xline and the image area is proximately a square. For the same image area condition, it is obviously fewer to read the traces for the new 2D division than for the tradition division.
引文
[1]陈志德,刘振宽,李成斌.高精度克希霍夫三维叠前深度偏移及并行实现[J].大庆石油地质与开发,2001(3).
    [2]林明玉,匡斌.叠前深度偏移中的旅行时并行计算技术[J].断块油气田,2003,10(04):12-14.
    [3]陈辉,孙雷鸣,李录明,等.基于MPI+Open MP的多层次并行偏移算法研究[J].成都理工大学学报:自然科学版,2010,37(5):528-534.
    [4]曹国滨,张旭,张加海.叠前成像方法联合应用与配套处理技术分析[J].勘探地球物理进展,2005(05).
    [5]樊卫花,杨长春,孙传文,刘文卿.三维地震资料叠前时间偏移应用研究[J].地球物理学进展,2007(03).
    [6]王棣,王华忠,马在田,王立歆.叠前时间偏移方法综述[J].勘探地球物理进展,2004(05).
    [7]刘芳,邓笋根,聂宁明,王彦棡,王珏.克希霍夫时间偏移在GPU集群上的MPI/CUDA混合编程实现[J].科研信息化技术与应用,2012,3(5):34-41.
    [8]邓笋根,迟学斌,刘芳,聂宁明,王彦棡.数据流软件Flobus在地震勘探资料处理中的应用[J].科研信息化技术与应用,2011,2(6):83-91.
    [9]李振春.地震偏移成像技术研究现状与发展趋势[J].石油地球物理勘探,2014,49(1):21.
    [10]张锦涛,赵惊涛,王真理.FPGA与GPU并行计算分析——以Kirchhoff叠前时间偏移为例[J].地球物理学进展,2013,28(3):1464-1471.doi:10.6038/pg20130341.

版权所有:© 2023 中国地质图书馆 中国地质调查局地学文献中心