基于波形特征向量的凝聚层次聚类地震相分析  

Agglomerative hierarchical clustering seismic facies analysis based on waveform eigenvector

作　　者：刘仕友 宋炜[2] 应明雄 孙万元 汪锐 LIU Shi-You;SONG Wei;YING Ming-Xiong;SUN Wan-Yuan;WANG Rui(Zhanjiang Branch,CNOOC Ltd.,Zhanjiang 524057,China;College of Geophysics,University of Petroleum of China,Beijing 102249,China)

机构地区：[1]中海石油(中国)有限公司湛江分公司,广东湛江524057 [2]中国石油大学(北京)地球物理学院,北京102249

出　　处：《物探与化探》2020年第2期339-349,共11页Geophysical and Geochemical Exploration

基　　金：国家科技重大专项课题“琼东南盆地深水区大中型气田形成条件与勘探关键技术”(2016ZX05026-002)。

摘　　要：常规的基于地震沉积学原理的地震相分析,主要利用地震切片技术沿目标层提取均方根振幅属性,在地震信号信噪比较低,目标层厚度薄时,容易影响地震相分析的精度和可靠性。本文从地震沉积学原理出发,沿地层切片提取地震波形特征向量,然后引入地震波形特征向量凝聚层次聚类方法(agglomerative hierarchical clustering,AHC)开展地震相划分。波形凝聚层次聚类是一种无监督的机器学习算法,与传统的地层切片地震相分析方法相比较,基于波形聚类的分析方法,通过波形特征的变化,综合考虑了地震信号的振幅、相位和频率属性特征,具有更好的抗噪能力和更高的横向分辨率。物理模型数据测试和实际资料应用都证明了该方法的稳定性和适用性,验证了本方法具有较好的沉积相特征划分能力,是一类新的岩性分析工具,具有良好的应用前景。Conventional seismic facies analysis based on seismic sedimentology principle mainly uses seismic slicing technology to extract RMS amplitude attributes along the target layer.When the signal-to-noise ratio of seismic signals is low and the target layer is thin,the accuracy and reliability of seismic facies analysis will be easily affected.In this study,on the basis of the principle of seismic sedimentology,the feature vectors of seismic waveforms were extracted along stratigraphic slices,and then the Agglomerative Hierarchical Clustering(AHC) method was introduced to classify seismic facies.Waveform AHC is an unsupervised machine learning algorithm.Compared with the traditional method of seismic facies analysis for stratum slices,the method based on waveform clustering considers the amplitude, phase and frequency attributes of seismic signals synthetically through the change of waveform characteristics.It has better anti-noise capability and higher horizontal resolution.The stability and applicability of this method have been proved by physical model data testing and practical data application.It has been proved that this method has a good capability of distinguishing sedimentary facies characteristics,and hence it is a new kind of reservoir facies analysis tool and has a good application prospect.

关 键 词：机器学习 凝聚层次聚类 波形聚类 地震相 地震沉积学 

分 类 号：P631.4[天文地球—地质矿产勘探]