基于地震多属性利用ES-DBN网络估算裂缝孔隙度  

Estimation of fracture porosity using ES-DBN network based on seismic multiple attributes

作　　者：张圣泽 张广智[1,2] 周游 刘俊州 韩磊[3] ZHANG ShengZe;ZHANG GuangZhi;ZHOU You;LIU JunZhou;HAN Lei(School of Geosciences,China University of Petroleum(East China),Qingdao 266580,China;Marine Mineral Resources Evaluation and Detection Technology Functional Laboratory of Marine National Laboratory,Qingdao 266071,China;Sinopec Petroleum Exploration and Production Research Institute,Beijing 100083,China)

机构地区：[1]中国石油大学(华东)地球科学与技术学院,青岛266580 [2]海洋国家实验室海洋矿产资源评价与探测技术功能实验室,青岛266071 [3]中国石油化工股份有限公司石油勘探开发研究院,北京100083

出　　处：《地球物理学进展》2022年第6期2492-2497,共6页Progress in Geophysics

基　　金：国家自然科学基金项目“基于深度学习的深层裂缝储层参数地震反演方法”(42074136);国家自然科学基金企业创新发展联合基金项目“渤海潜山裂缝性储层地震响应机理及精确成像方法”(U19B2008);国家科技重大专项“中西部地区碎屑岩领域勘探关键技术”(2016ZX05002-005);中国石油大学研究生创新工程项目“基于集成学习的变质岩潜山油气藏裂缝特征参数预测方法研究”(YCX2020014)等联合资助。

摘　　要：地震数据中蕴含丰富的地层信息,裂缝孔隙度作为表征致密碎屑岩储层品质及产能评价的关键参数,目前无法利用常规线性反演方法从地震数据中挖掘出裂缝孔隙度的有效信息.为了有效获取致密碎屑岩地震信号中储层与非储层数据特征之间的差异,本文借助深度置信网络(DBN)强大的特征提取能力,利用回声状态网络(ESN)中的回归层代替DBN网络中的误差反向传播算法组合搭建出ES-DBN网络.ES-DBN网络能够较好的捕捉地震数据中的动态时序信息,每次训练都独立于前次的DBN学习过程,且只考虑回归层最终的输出权值矩阵,通过模型测试对比发现,该网络的学习效率和预测准确率均高于传统的DBN网络.以井中裂缝孔隙度为预测目标,基于提取的多种井旁敏感的叠前叠后裂缝属性,利用构建出的网络模型对研究工区地下地层裂缝孔隙度进行预测,预测结果表明:该网络能够较好的实现致密碎屑岩储层的裂缝孔隙度的三维表征,取得了良好的应用效果.There are abundant stratigraphic information in seismic data.As a key parameter to evaluate the quality and productivity of tight clastic reservoirs,the effective information of fracture porosity can not be extracted from seismic data by conventional linear inversion methods In order to effectively obtain the difference between reservoir and non-reservoir data features in seismic signals of tight clastic rocks,this paper uses the powerful feature extraction ability of deep confidence network(DBN),and uses the regression layer in ESN to replace the combination of error back propagation algorithm in DBN network to build an ESN-DBN network ESN-DBN network can better capture the dynamic time series information in seismic data.Each training is independent of the previous DBN learning process and only considers the final output weight matrix of the regression layer.Through model test comparison,it is found that the learning efficiency and prediction accuracy of this network are higher than that of the traditional DBN network In fracture porosity prediction target in the well,based on the extraction of a variety of prestack post-stack cracks of sensitive attributes,using network model to construct the ESN-DBN research work area underground strata fracture porosity prediction,prediction results show that the network can achieve good dense clastic rock reservoir fracture porosity of the three dimensional characterization,has obtained the good application effect.

关 键 词：地震属性 神经网络 裂缝预测 非线性反演 

分 类 号：P631[天文地球—地质矿产勘探]