地震属性是地下介质的综合反映,与地质目标往往不具备一一对应的关系,这就导致单一属性在解释时不可避免的存在多解性,为解决这一难题,地震属性融合技术应用而生.传统的基于线性变换的主成分分析(Principal Component Analysis,PCA)方法是一种较为有效的地震属性融合技术,但它在面对复杂非线性的地震属性融合问题时,不能有效提取其中的非线性特征.因此,本文提出了基于栈式稀疏自编码器(Stack Sparse Auto Encoders,SSAE)的非线性地震属性融合技术.SSAE是一种深度学习网络,能够充分地挖掘数据的非线性特征,通过不断学习,自适应地融合各种属性中蕴含的有效信息.本文首先介绍了地震属性的优选、标准化处理方法,然后阐述了基于PCA、SSAE的属性融合方法的基本原理;最后通过两种方法在两个经典模型、一个正演模型及三个应用实例的对比分析,表明SSAE对于非线性数据拥有更好的融合效果,且对于多尺度、多属性、宽方位等不同类型的属性数据也具备普适性.SSAE融合属性集合了多种属性的特征信息,有效降低了解释的多解性,提升了储层预测的精度,可为同类型工区提供借鉴.
Seismic attribute fusion technology based on SSAE
Seismic attribute is a comprehensive reflection of the underground medium and don't simply coincide with geological targets.Which leads to the inevitable existence of multiple solutions in interpretation when using a single attribute.Principal Component Analysis(PCA)is the most widely used linear seismic attribute fusion technique,but can't effectively extract nonlinear features in the face of complex nonlinear seismic attribute fusion problem.According to this,a non-linear seismic attribute fusion technique based on Stack Sparse Auto Encoders(SSAE)is proposed.SSAE is a kind of deep learning network,which can fully mine the deep characteristics of the seismic attribute data.Through continuous learning,it adaptively merges the information contained in various attributes.In this paper,firstly,introduces the optimization and standardization of seismic attributes.Secondly,the principle of attribute fusion method based on PCA and SSAE is clarified.Finally,through the effect analysis of two methods in the two classical models,one synthetic seismogram and three application examples,it shows that SSAE has a better fusion performance for nonlinear attribute data,and it is also effective and applicable for different types of attribute data,such as multi-scale,multi-attribute,wide azimuth and so on.The SSAE fusion attribute has the characteristics of multiple attributes,which can effectively discover the reservoir information hidden in the attribute data and reduce the multi-solution of reservoir interpretation and improve the accuracy of reservoir prediction.This paper can also provide reference for the similar project.
Multiple solutionsNonlinearSeismic attributes fusionPrincipal Component Analysis(PCA)Stack Sparse Auto Encoders(SSAE)
NETL
NSTL
万方数据
