基于流形学习的地震属性特征提取方法及应用  

Seismic Attribute Feature Extraction and Application Based on Manifold Learning

作　　者：刘杏芳[1] 郑晓东[1] 徐光成[1] 李劲松[1] 王玲[1] 

机构地区：[1]中国石油勘探开发研究院

出　　处：《岩性油气藏》2010年第F07期144-146,158,共4页Lithologic Reservoirs

基　　金：国家科技重大专项课题“海相碳酸盐岩储层地震预测、油气藏描述技术研究”（编号：2008ZX05004-006）资助.

摘　　要：地震属性与地质特征的关系通常是非线性的．传统的基于线性变换的地震属性降维优化方法不能充分反映这种非线性关系，降低了储层预测的精度。流形学习通过保持数据局部结构的方式将高维数据投影到低维空间，挖掘和发现隐藏在数据中的内在特征与规律性，是地震属性优化研究的新领域。文中提出了基于流形学习的地震属性特征提取方法，把基于线性降维的主成分法（PCA）和基于非线性降维的局部线性嵌入法（LLE）提取地震属性特征的方法进行了对比。理论模型分析和实际应用均表明：在处理非线性问题上．流形学习具有更好的聚类分析能力和特征提取性能．LLE提取的地震属性比PCA提取的属性更加准确地刻画了有利储层的展布特征，说明流形学习在地震属性优化方面具有较好的应用前景。Commonly, the relationship between seismic attributes and reservoir features is non-linear,therefore the traditional linear dimensionality reduction optimization method cannot well reflect this non-linear relationship and reduce the precision of reservoir prediction. Manifold learning is format projecting high dimensionality data to low dimensionality space while maintaining local structure of data and discovering the intrinsic features and rules plus is new area for seismic attribute optimization study. The new seismic attribute extraction based on manifold learning is introduced in this paper and the comparison between principle component analysis （PCA） and locally linear embedding （LLE） based on non-linear dimensionality reduction for extraction of seismic attribute is conducted. Both theoretical model and case studies show that when tackling non-linear problems, manifold learning possesses better cluster analysis and attribute extraction performance and that new seismic attribute extracted through LLE can describe the distribution characteristics of favorable reservoir more accurately than that of PCA, indicating that manifold learning poses fairly good prospect in terms of seismic attribute optimization

关 键 词：属性优化 降维映射 局部线性嵌入方(LLE) 流形学习 主成分分析(PCA) 

分 类 号：P631.42[天文地球—地质矿产勘探]