基于波形特征向量的谱聚类地震相分析  

Seismic Facies Analysis Based on Spectral Clustering with Waveform Characteristic Vector

作　　者：秦德文 张岩 于杰 QIN Dewen;ZHANG Yan;YU Jie(Shanghai Branch of CNOOC(China)Ltd.,Shanghai 200335,China)

机构地区：[1]中海石油(中国)有限公司上海分公司,上海200335

出　　处：《CT理论与应用研究（中英文）》2024年第1期13-23,共11页Computerized Tomography Theory and Applications

基　　金：“十四五”科技重大项目(中国近海新区新领域勘探技术(KJGG2022-0304))。

摘　　要：本文提出一种基于地震沉积学原理沿层提取地震波形特征向量,并以谱聚类(spectral clustering)分析进行地震相划分的方法。谱聚类能够处理非线性的数据结构和高维数据的聚类问题,但其相似度矩阵的构建和谱分解的计算较为复杂,需要较高的计算资源和时间成本。为提高谱聚类算法的效率和可扩展性,本文提出将Mini-batch K-means算法与谱聚类算法结合起来的MKSC算法,在提高谱聚类算法精度的同时大大降低谱聚类空间的复杂度。经过对数值模拟、地球物理模型数据和实际地震资料的处理分析,证明该方法在沉积相划分、沉积相特征识别方面的效果明显,是一种具有良好应用前景的新型沉积特征分析工具。Based on the principle of seismic sedimentology,the feature vectors of seismic waveforms are extracted along stratum slices,and spectral clustering analysis is introduced to classify seismic facies.Spectral clustering is an unsupervised machine learning algorithm.Its essence is to simplify the expression of high-dimensional seismic data in the form of feature vectors,which belongs to the process of dimensionality reduction.Considering the traces with specific time windows in the seismic work area as nodes of the graph and the similarity between traces as the weight of the edges,a graph model can be constructed.Spectral clustering must determine the best segmentation method to complete the segmentation of the graph,so that different types of sedimentary characteristics can be distinguished.Physical model and actual data processing and analysis demonstrate that this method is capable of dividing sedimentary facies characteristics and is a new kind of facies analysis tool for reservoir classification,which has good application prospects.

关 键 词：机器学习 谱聚类 波形聚类 地震相 地震沉积学 

分 类 号：O242[理学—计算数学] P631.4[理学—数学] P315[天文地球—地质矿产勘探]