摘要
选用东海F气田的砂质辫状河三角洲的自然伽马数据作为训练数据构建深度卷积神经网络,并首次用于测井相识别。选用四种自然伽马曲线形态作为特征,将数值转变为图像形式,首先对图像做标准化、添加噪声、旋转和转灰度等处理,再对数据增强与扩充,建立训练和测试数据集;然后,训练卷积神经网络建立测井相识别模型,并在训练过程中加入了Dropout、局部响应归一化和L2正则化等策略限制了模型的复杂程度,提高了模型泛化能力;针对测井信息中不同级次沉积单元响应叠加带来的自动识别难题,使用不同尺度的小波基函数及极值分割处理和切分测井数据,最终有效划分了不同尺度沉积单元。通过与其他分类算法对比,验证了所提方法具有较好的测井相识别效果。
Natural gamma data of sandy braided river delta sedimentary environments in the Gasfield F,the East China Sea is selected as the training data to construct a deep convolutional neural network,which is used for log facies identification for the first time.Four kinds of gamma ray(GR)curve shapes are selected as characteristics,and their values are converted into the image form.Several processing steps are carried out for these images such as normalization,noise addition,rotation,and grayscale turning,and then this image data set is enhanced and expanded.In this way,training and test data sets are established.After that,a convolutional neural network is trained and used to establish the log facies identification model.During the training process,dropout,local response normalization,and L2 regularization are added to limit the complexity of the model and improve the generalization ability of the model.To automatically identify superposed deposition units of different grades in logging information,different scales of wavelet basis function and extreme value segmentation are used to classify different-scale deposition units of logging data.The comparison with other algorithms demonstrates that the proposed method achieves better log facies identification.
作者
何旭
李忠伟
刘昕
张涛
HE Xu;LI Zhongwei;LIU Xin;ZHANG Tao(School of Computer and Communication Engineering,China University of Petroleum(East China),Qingdao,Shandong 266580,China;Geological College,Shandong University of Science and Technology,Qingdao,Shandong 266590,China)
出处
《石油地球物理勘探》
EI
CSCD
北大核心
2019年第5期1159-1165,I0013,I0014,共9页
Oil Geophysical Prospecting
基金
国家自然科学基金项目“基于岩石物理实验的凝灰质砂岩岩相—成岩相测井精准识别方法研究”(41602135)
中国石油科技创新基金研究项目“基于大数据技术的超深低渗储层测井相分析”(2016D-5007-0305)联合资助
关键词
卷积神经网络
测井相
多尺度
小波基
convolutional neural network
log facies identification
multi-scale
wavelet base
作者简介
通讯作者:何旭,山东省青岛市黄岛区长江西路66号中国石油大学(华东)计算机与通信工程学院,266580。硕士研究生,1992年生,2015年获中国石油大学(华东)软件工程专业工学学士学位,目前在中国石油大学(华东)攻读计算机科学与技术专业硕士学位,研究方向为机器学习与计算智能。E-mail:hexu1114@163.com。
