论文部分内容阅读
从大量的地震属性中提取最能反映地质特征的综合属性是储层预测技术的关键,通常选用降维方法来优选属性。目前应用最为广泛的线性降维方法。但是,由于地震属性与地质特征的关系通常是非线性的,基于线性变换的地震属性降维优化方法不能充分地反映这种非线性关系,降低了储层预测的精度。流形学习是一种新的非线性学习方法,它是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,开拓了地震属性降维优化研究的新领域。本文首次实现了3D地震数据的层问属性特征提取,讨论了LLE方法及其关键技术,并以奥陶系礁滩相储层实例说明LLE和PCA两种方法降维及聚类的不同效果。理论模型分析和实例应用表明:LLE较好地保持了数据本身的原始结构;提取的综合属性和聚类相图较好地刻画了沉积相带、储层和流体的特征。这说明流形学习具有更好的特征提取性能。
It is the key of reservoir prediction technology to extract the most comprehensive attribute that reflects the geological features from a large number of seismic attributes. Usually, the dimension reduction method is used to optimize the attributes. Currently the most widely used linear dimension reduction method. However, since the relationship between seismic attributes and geological features is usually nonlinear, the dimensionality-reduction optimization method based on linear transformation can not fully reflect this nonlinear relationship and reduce the accuracy of reservoir prediction. Manifold learning is a new non-linear learning method, which is to project high-dimensional data into low-dimensional space by maintaining the local structure of data, to excavate and discover the intrinsic features and regularity hidden in the data, and to open up the seismic attributes The new field of dimensionality optimization research. This paper firstly realizes the layer-by-layer attribute feature extraction of 3D seismic data, discusses the LLE method and its key technologies, and illustrates the different effects of LLE and PCA methods on dimension reduction and clustering based on the Ordovician reef-shoal facies reservoir examples. The theoretical model analysis and practical application show that LLE preserves the original structure of the data well. The extracted comprehensive attributes and the clustering phase diagrams characterize the sedimentary facies belts, reservoirs and fluids well. This shows that manifold learning has better feature extraction performance.