深度学习算法在白云凹陷深水区密度反演中的应用  

Application of depth learning algorithm to density inversion in deepwater area of Baiyun Sag

作　　者：陈兆明 李志晔 张卫卫 张振波 吕华星 陶禹 CHEN Zhaoming;LI Zhiye;ZHANG Weiwei;ZHANG Zhenbo;LYU Huaxing;TAO Yu(CNOOC China Limited,Shenzhen Branch,Shenzhen,Guangdong 518000,China;CNOOC Deepwater Development Limited,Shenzhen,Guangdong 518000,China)

机构地区：[1]中海石油(中国)有限公司深圳分公司,广东深圳518000 [2]中海石油深海开发有限公司,广东深圳518000

出　　处：《中国海上油气》2022年第3期55-61,共7页China Offshore Oil and Gas

基　　金：中海石油(中国)有限公司科技项目“南海大中型天然气田形成条件、勘探潜力与突破方向(编号:KJZH-2021-0003-00)”部分研究成果。

摘　　要：密度属性参数与储层孔隙度、含油气性和流体饱和度的相关性高,但相对于纵波阻抗和横波阻抗,叠前三参数反演中的密度参数反演依赖于大角度地震信息,往往存在较大的不确定性。本文从单井岩石物理建模开始,对南海北部白云凹陷深水区珠江组ZJ110砂层28 m厚度的砂岩进行厚度和含气饱和度替换,生成121口井的纵波速度、横波速度和密度曲线,进而正演出121口井的道集数据,得到近、中、远道3个地震数据体。进而对地震数据体开展属性分析,优选出9种地震属性作为特征向量,通过深度学习算法建立多个地震属性与密度的最佳非线性关系。深度学习算法反演的密度参数与实际井曲线吻合度高,反演误差远小于叠前三参数反演,也证实了该方法的实用价值。The density attribute parameter is highly correlated with reservoir porosity,oil-bearing property and fluid saturation,but compared with P-wave impedance and S-wave impedance,the inversion of density parameter in three-parameter pre-stack inversion relies on large-angle seismic data,and often has some large degree of uncertainties.In this paper,the thickness and gas saturation of ZJ110 sandstone layer with thickness of 28 m in the Zhujiang Formation in the deepwater area of Baiyun Sag,northern area of South China Sea are replaced in turn by single well petrophysical modeling,the P-wave velocity,S-wave velocity and density curves of 121 wells are produced,the channel data of 121 wells are obtained by forward modelling,and further 3 seismic data volumes of near,middle and far channels are acquired.Moreover,the attributes of seismic data volume are analyzed,9 seismic attributes are selected as feature vectors.Finally,the optimal nonlinear relationship between multiple seismic attributes and density is established by deep learning algorithm.The density parameters obtained from inversion by deep learning algorithm is in good agreement with the actual logging curve,and the inversion error is far less than the pre-stack threeparameter inversion,which also confirms the practical value of this method.

关 键 词：深度学习 密度反演 属性分析 白云深水区 

分 类 号：TE132.1[石油与天然气工程—油气勘探]