基于变分自编码器利用元素录井数据确定火成岩矿物含量的方法  

Method for Determining Igneous Rock Mineral Content Using Element Logging Data Based on Variational AutoEncoder

作　　者：贾瑞龙 潘保芝[1] 王清辉 李岩 管耀 王欣茹 JIA Ruilong;PAN Baozhi;WANG Qinghui;LI Yan;GUAN Yao;WANG Xinru(College of Geoexploration Science and Technology Jilin University,Changchun,Jilin 130026,China;Shenzhen Branch of CNOOC(China)LTD.,Shenzhen,Guangdong 518054,China)

机构地区：[1]吉林大学地球探测科学与技术学院,吉林长春130026 [2]中海石油(中国)有限公司深圳分公司,广东深圳518054

出　　处：《测井技术》2024年第4期407-415,共9页Well Logging Technology

基　　金：国家自然科学基金项目“基于T2-I模型的蚀变火山岩导电机理和饱和度模型研究”(42204122)。

摘　　要：火成岩由于岩浆类型及冷凝环境的不同导致矿物含量差异大,不同岩性的骨架参数明显不同。确定岩石骨架的矿物含量是评价储层的一项重要工作,在地层岩性划分、骨架参数计算以及沉积环境的研究等方面有着重要的意义。提出了一种火成岩矿物含量预测模型,该模型使用了元素录井得到的17种元素含量数据,基于变分自编码器(Variational AutoEncoder,VAE)方法预测矿物含量并重构元素含量。模型验证结果显示,该模型在数据集中预测平均绝对误差及均方误差小于BP神经网络(反向传播神经网络,Back Propagation Neural Network)、岭回归和支持向量机这3种典型方法。将该模型应用于南海某地区古潜山火成岩井段,应用结果表明,该模型跟典型算法相比具有优越性,同时具有良好的可应用性。Igneous rocks exhibit significant variations in mineral content due to differences in magma types and the environment in which they solidify,and the skeleton parameters of different lithology are obviously different.The determination of mineral content of the rock matrix is an important task in evaluating reservoirs,which is of great significance in stratigraphic lithology division,calculation of matrix parameters and study of depositional environments.In this study,a predictive model for mineral content in igneous rocks is proposed.The model utilizes data from 17 elements obtained through element logging.It employs a VAE(Variational AutoEncoder)approach to predict mineral content and reconstruct the elemental weight content.The model validation reveals that the proposed model has a smaller mean absolute error and mean square error compared to three typical methods:BP(Back Propagation)neural networks,ridge regression and support vector machines.Furthermore,the model is applied to a section of buried hill igneous rock well in the South China Sea.The results demonstrate the superiority of the proposed model over the typical algorithms while maintaining good applicability.

关 键 词：火成岩 矿物含量 变分自编码器 元素录井 

分 类 号：P631.84[天文地球—地质矿产勘探]