删除或更新信息,请邮件至freekaoyan#163.com(#换成@)

结合序贯指示模拟的贝叶斯高斯混合线性反演

本站小编 Free考研考试/2022-01-03

印兴耀1,2,,
贺东阳1,2,
宗兆云1,,,
李坤1,2,
肖张波3,
张仟策4
1. 中国石油大学(华东)地球科学与技术学院, 山东青岛 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, 山东青岛 266071
3. 中海石油(中国)有限公司深圳分公司, 广州 510214
4. 中国石油集团测井有限公司辽河分公司, 辽宁盘锦 124010

基金项目: 国家自然科学基金(U1562215,U1762103),国家油气重大专项课题(2016ZX05024004,2017ZX05009001,2017ZX05036005,2017ZX05032003)和中国科学托举人才工程(2017QNRC001)联合资助


详细信息
作者简介: 印兴耀, 男, 1962年生, 博士, 教授, 博士生导师, 主要从事地球物理理论、方法与应用方面的研究.E-mail:xyyin@upc.edu.cn
通讯作者: 宗兆云, 男, 1987年生, 博士, 副教授, 博士生导师, 主要从事地球物理理论、方法与应用方面的研究.E-mail:zongzhaoyun@126.com
中图分类号: P631

收稿日期:2018-05-22
修回日期:2019-01-05
上线日期:2019-12-05



Bayesian Gaussian Mixture linear inversion combined with Sequential Indicator Simulation

YIN XingYao1,2,,
HE DongYang1,2,
ZONG ZhaoYun1,,,
LI Kun1,2,
XIAO ZhangBo3,
ZHANG QianCe4
1. School of Geosciences, China University of Petroleum, Qingdao Shandong 266580, China
2. Laboratory for Marine Mineral Resources, Qingdao National Laboratory for Marine Science and Technology, Qingdao Shandong 266071, China
3. CNOOC Ltd. Shenzhen Branch, Guangzhou 510240, China
4. CNPC Logging Liaohe Branch, Panjin Liaoning 124010, China


More Information
Corresponding author: ZONG ZhaoYun,E-mail:zongzhaoyun@126.com
MSC: P631

--> Received Date: 22 May 2018
Revised Date: 05 January 2019
Available Online: 05 December 2019


摘要
高斯混合模型(Gaussian Mixture Model,GMM)可以用来描述储层性质的多峰分布特性,多峰特性主要是由于它们在不同离散变量内的变化而引起的.在高斯混合模型中,高斯分量的权值代表离散变量的概率.然而,基于高斯混合模型的贝叶斯线性反演可能会对某些点的离散变量错误地分类,进而影响连续变量的反演结果,尤其存在强噪声的时候.在本文中,我们考虑了离散变量的空间变化性,并将高斯混合模型与序贯指示模拟(Sequential Indicator Simulation,SIS)相结合来确定离散变量的后验条件权值,形成了结合序贯指示模拟的贝叶斯高斯混合线性反演方法.该方法能够准确地对离散变量进行归类,且具有良好的抗噪性.通过模型试算,我们证明了这种方法的可行性,并在实际资料中取得了较好的结果.
高斯混合模型(GMM)/
离散变量/
线性反演/
连续变量/
序贯指示模拟(SIS)

In the Bayesian framework, the prior distribution of the model parameters is usually assumed to be Gaussian distribution. However, the model data does not obey the Gaussian distribution in practical applications. Model parameters tend to be multimodal due to different discrete variables, such as facies. The Gaussian Mixture Model can be used to describe the multimodal behavior of model parameters. In the Gaussian Mixture Model, the weights of the Gaussian components represent the probabilities of the discrete variable. However, Bayesian linear inversion based on Gaussian Mixture Model may misclassify discrete variable at some points, which may lead to a bad inversion result of continuous variable, especially when there is a strong noise. In this study, we consider the spatial variability of discrete variable and combine Gaussian Mixture Model with the Sequential Indicator Simulation to determine the posterior conditional weight of each discrete variable, and develop the method of Bayesian Gaussian Mixture linear inversion combined with Sequential Indicator Simulation. This method can accurately classify discrete variable and has good robustness. We certify the feasibility of this method on model data, and then apply this method to actual data with satisfactory results.
Gaussian Mixture Model(GMM)/
Discrete variable/
Linear inversion/
Continuous variable/
Sequential Indicator Simulation(SIS)



PDF全文下载地址:

http://www.geophy.cn/data/article/export-pdf?id=dqwlxb_15273
相关话题/博士 博士生导师 地球物理 海洋 实验室