基于人工智能的残余绕射多次波衰减技术研究及应用  

Diffracted multiples attenuation based on artificial intelligence

作　　者：刘金朋 钟明睿 杜皓 LIU Jinpeng;ZHONG Mingrui;DU Hao(Data Processing Center,Institute of Geophysical Exploration,Geophysical China Oilfield Services Limited(Zhanjiang),Zhanjiang 524057,China)

机构地区：[1]中海油田服务股份有限公司物探事业部物探研究院特普数据中心(湛江),广东湛江524057

出　　处：《石油物探》2024年第6期1177-1185,共9页Geophysical Prospecting For Petroleum

摘　　要：深水崎岖海底或含有复杂构造海域的地震资料存在严重的绕射多次波。传统的多次波衰减技术依赖于多次波模型与原始数据的相似度,当它们之间的差异较大时,则绕射多次波的衰减效果不佳。提出了利用人工智能深度学习中的目标检测网络检测绕射多次波区域;然后利用字典特征学习的方法,将多次波衰减问题转化为多次波字典特征学习与再重构问题,从而解决了绕射多次波模型与实际数据之间存在较大差异的问题。正演数据及实际数据应用表明,基于人工智能的残余绕射多次波衰减技术能够有效确定绕射多次波的位置并衰减残余绕射多次波,解决了绕射多次波预测不准、衰减效果不佳的问题。Seismic data acquired in a sea area with deep rugged seabed or complex structures suffer from serious diffracted multiples,which cannot be effectively attenuated using traditional techniques that depend on the similarity between modeled multiples and observed multiples.When there are great difference between the model and the real data,the attenuation effect is poor.This paper proposes a deep learning method using a target detection network to detect the regions with diffracted multiples.The method of dictionary learning is then used to transform multiples attenuation into the dictionary learning and reconstruction of multiples,so as to alleviate model-observation dissimilarity.The application of forward modeling data and actual data shows that this method can effectively locate residual diffracted multiples and thus greatly improving the efficiency and accuracy of attenuation,which solves the problem of inaccurate prediction and poor attenuation of diffracted multiples and greatly improves the quality of deep-water seismic data.

关 键 词：绕射多次波衰减 人工智能 目标检测 字典特征学习 

分 类 号：P631[天文地球—地质矿产勘探]