地震三维观测系统共中心点面元属性的并行计算方法  

Common mid-point bin attribute analysis of three-dimensional seismic geometry using parallel computation method

作　　者：李逢春[1,2] 王润秋[1] 赵薇薇 杜清波 郭武 王汉钧 

机构地区：[1]中国石油大学(北京)地球物理与信息工程学院,北京102249 [2]东方地球物理公司采集技术支持部,河北涿州072751

出　　处：《物探与化探》2015年第5期1074-1078,共5页Geophysical and Geochemical Exploration

基　　金：国家重点基础研究发展计划("973"计划)"深层油气藏地球物理探测的基础研究"(2013CB228602)

摘　　要：针对高密度地震勘探下的三维观测系统海量面元属性计算需求,基于Open CL并行编程模型,研究并实现了适用于异构环境下面元属性的并行计算方法。根据面元分析算法特点将整个网格分成多个子区域,子区域可由不同设备以粗粒度方式并行,而子区域内部以炮检对方式细粒度并行。采用HP Z820工作站,在CPU+GPU异构混合平台下的试算结果表明,异构多核处理速度是CPU(单核)的30倍以上,数据生成速率高于300 MB/s。For the purpose of common midpoint bin attribute analysis from high density exploration and threedimensional seismic geometry,this paper analyzes the calculation characteristics of bin analysis,and proposes an efficient parallel algorithm for bin attribute analysis based on Open Computing Language （ OpenCL） heterogeneous programming model. The algorithm divides bin grid into several subregions under hardware memory constraint.Tasks of subregions are executed with coarsegrained parallelization by assigning task to different devices,whereas pairs of sourcereceiver in subregion are executed with finegrained parallelization. With HP＇s Z820 desktop workstation,experiment was carried out with high density threedimensional geometry from seismic exploration.The experimental results show that the parallel algorithm takes full advantage of multiprocessor＇s parallel processing capability,and the processing speed is about 30 times faster than that on CPUbased implementation （ single core） under the condition that the precision is not changed.

关 键 词：地震观测系统 面元分析 并行计算 异构计算 OPENCL 

分 类 号：P631.4[天文地球—地质矿产勘探]