基于Spark大数据处理框架的逆时偏移成像技术研究

基于双程波动方程的逆时偏移成像技术是当前高精度地震偏移成像的主流手段,可实现地下复杂构造的精确成像。近年来,随着高密度地震采集技术的不断发展,日益增长的海量地震数据对偏移成像过程中数据的存储和高效计算提出了更高的要求。提出了一种基于Spark大数据处理框架的逆时偏移成像技术,该技术充分利用了Spark框架分布式内存计算及其分布式文件存储系统支持海量数据读写的优势。采用数据均衡服务实现了分布式存储数据的均衡分布并利用数据动态分块算法实现了计算资源的充分利用和并行粒度的合理分配,从而保证了并行作业的负载均衡。与MPI及MapReduce两种并行方式下的实际数据偏移测试对比分析结果表明,所提出的基于Spark大数据处理框架的逆时偏移成像技术能够在保证成像精度的前提下实现计算效率的提升并满足海量数据存储的要求。

Reverse-time migration based on the two-way wave equation is the main method of high-precision seismic migration imaging,which can be used to accurately identify complex underground structures.With the development of the high-density acquisition technology,the size of the seismic data has increased considerably,posing a challenge to data storage and to the efficient computation of the pre-stack depth migration.In this study,a reverse-time migration imaging technique based on the Spark big data processing framework was proposed.The technique takes full advantage of the distributed computing and storage capabilities of the framework,as well as of the system to support the input and output of massive data.A data balancing service was used to balance the distribution of the stored data,and a dynamic data assignment algorithm was implemented to achieve the full utilization of the computing resources and a reasonable granularity,thereby ensuring an optimal load balance among the parallel jobs.Tests on field data showed that the proposed method can improve the computing efficiency compared to methods based on parallel MPI and MapReduce.

国家重点研发计划(2017YFC0602804-02)和国家科技重大专项(2016ZX05014-001-002)共同资助。