摘要
叠加能量寻优反射波剩余静校正方法是解决剩余静校正问题的有效方法之一,该方法精度高,但计算量大且需要进行多域数据切换。随着原始数据量的不断增长,其算法实现的高效运行成为主要的应用瓶颈问题。分析了叠加能量寻优反射波剩余静校正方法计算密集、通讯密集的特征,针对算法难以实现并行计算的难点,提出了基于Spark分布式内存计算模型的技术解决方案,实现了海量地震数据弹性分布式数据集的高效流转和多域数据的灵活切换,完成了叠加能量寻优反射波剩余静校正方法的多节点分布式并行计算,提高了大数据情形下方法的适应性和计算效率,提升了其在地震数据处理中的实用化程度。实际生产数据的应用结果表明,基于Spark的叠加能量寻优反射波剩余静校正的软件模块在复杂近地表地震数据的处理中取得了能满足实际生产要求的应用效果,兼具适应性强和计算效率高的特点。
Reflection residual static correction optimized using stacked energy is an effective means to solve the problem of residual static correction with high accuracy.However,it suffers from a large amount of calculation and multi-domain data switching.Due to the growing volume of raw data,efficient computation has become a bottleneck in practical applications.In this paper,we analyze the computation-intensive and communication-intensive characteristics of reflection residual static correction optimized using stacked energy,and put forward a technical solution based on the Spark distributed in-memory computing model to address the issue of parallel computation.We realize high-efficiency transfer of elastically distributed mass seismic data and flexible switching of multi-domain data,and accomplish multi-node distributed parallel computing of reflection residual static correction optimized using stacked energy,with significantly improved computational efficiency and feasibility.Practical applications show that the software implementation of our method,featuring strong adaptability and high computational efficiency,is good enough for seismic data processing in the areas with complicated near-surface conditions,e.g.mountainous and desert areas.
作者
袁联生
YUAN Liansheng(SINOPEC Geophysical Research Institute Co.,Ltd.,Nanjing 211103,China)
出处
《石油物探》
CSCD
北大核心
2024年第4期807-816,共10页
Geophysical Prospecting For Petroleum
基金
中国石化科技部项目“π地震处理系统新技术集成(P23025)”
“π处理系统特色技术集成与软件工程化管理(P22185)”
“系列特色应用模块的集成(P20053-1)”
“π应用系统关键应用功能完善与实用化(P21066-2)”共同资助。
关键词
反射波剩余静校正
Spark框架
工程化实现
分布式并行计算
reflection residual static correction
Spark computing framework
engineering realization
distributed parallel computing
