引用本文:
【打印本页】   【HTML】   【下载PDF全文】   查看/发表评论  【EndNote】   【RefMan】   【BibTex】
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被:浏览次   下载 本文二维码信息
码上扫一扫!
分享到: 微信 更多
基于集合和神经网络架构搜索的自动历史拟合方法
张黎明1,陈昕晟1,李国欣2,马小鹏1,张凯1,谷建伟1,姚军1,王健3,孙海1
(1.中国石油大学(华东)石油工程学院,山东青岛 266580;2.中石油勘探与生产分公司,北京 100007;3.中国石油大学(华东)理学院,山东青岛 266580)
摘要:
由于人工经验选取的局限性,难以获取决定模型重构精度的最优网络参数是目前使用深度学习方法在自动历史拟合中对油藏地质模型进行降维时的难点之一。针对此问题,通过将深度自编码器与粒子群优化算法相互结合实现最佳网络架构的自动搜索,并以此构建一种基于集合数据同化和神经网络架构自动搜索的油藏自动历史拟合方法。分别对一个二维河流相油藏渗透率场分布模型以及SPE-10单层油藏数值模型应用该方法,并与单一的自动历史拟合方法进行对比验证。结果表明,经优化后自动搜索出最优神经网络构架的自动历史拟合方法要比优化前及单一的自动历史拟合方法能够更准确地提取出油藏数值模型的地质特征。
关键词:  自动历史拟合  深度学习  复杂地质特征  深度自编码  网络架构搜索  数据同化
DOI:10.3969/j.issn.1673-5005.2022.02.013
分类号::TE 33
文献标识码:A
基金项目:国家自然科学基金项目(51722406,52074340,51874335);山东省自然科学基金项目(JQ201808); 中央高校基本科研业务费(18CX02097A);中石油重大科技项目(ZD2019-183-008);山东省高等学校青创科技支持计划(2019KJH002);国家油气重大专项(2016ZX05025001-006);“111”计划(B08028)
An automatic history matching method based on ensemble and neural architecture search
ZHANG Liming1, CHEN Xinsheng1, LI Guoxin2, MA Xiaopeng1, ZHANG Kai1, GU Jianwei1, YAO Jun1, WANG Jian3, SUN Hai1
(1.School of Petroleum Engineering in China University of Petroleum(East China), Qingdao 266580, China;2.PetroChina Exploration and Production Company, Beijing 100007, China;3.College of Science in China University of Petroleum(East China), Qingdao 266580, China)
Abstract:
Due to the limitations of manual selection via human experience, it is difficult to obtain the optimal network parameters that determine the accuracy of model reconstruction, which is currently one of the difficulties when using deep learning methods to reduce the dimensionality of reservoir geological models in automatic history matching. In response to this, the automatic search of the best network architecture was realized by combining the deep auto-encoder and the particle swarm optimization algorithm, and an automatic reservoir history matching method was constructed, based on aggregate data assimilation and automatic search of the neural network architecture. A two-dimensional permeability distribution model of a fluvial reservoir and a SPE-10 single-layer reservoir numerical model were used to verify the proposed method, in comparison with a single automatic history matching method. The results show that the automatic history matching method can automatically searches the optimal neural network framework after optimization, which can extract the geological characteristics from the reservoir numerical model more accurately than the single automatic history matching method.
Key words:  automatic history matching  deep learning  complex geological features  deep auto-encoder  neural architecture search  ensemble smoother
版权所有 中国石油大学学报(自然科学版)编辑部 Copyright©2008 All Rights Reserved
主管单位:中华人民共和国教育部 主办单位:中国石油大学(华东)
地址: 青岛市黄岛区长江西路66号中国石油大学期刊社 邮编:266580 电话:0532-86983553 E-mail: journal@upc.edu.cn
本系统由:北京勤云科技发展有限公司设计