基于人工智能的测井地层划分方法研究现状与展望  

Research status and outlook of logging stratigraphic division methods based on artificial intelligence

作　　者：孙龙祥 韩宏伟[3] 冯德永[3] 刘海宁[3] 李泽瑞 康宇[4] 吕文君 SUN Longxiang;HAN Hongwei;FENG Deyong;LIU Haining;LI Zerui;KANG Yu;Lü Wenjun(AHU-IAI AI Joint Laboratory,Anhui University,Hefei City,Anhui Province,230601,China;Institute of Arificial Itelligence,Hefei Comprehensive National Science Center,Hefei City,Anhui Province,230088,China;Geophysical Research Institute,Shengli Oifeld Company,SINOPEC,Dongying City,Shandong Province,257000,China;Department of Automation,University of Science and Technology of China,Hefei City,Anhui Province,230026,China)

机构地区：[1]安徽大学安徽大学与合肥综合性国家科学中心人工智能研究院联合实验室,安徽合肥230601 [2]合肥综合性国家科学中心人工智能研究院,安徽合肥230088 [3]中国石化胜利油田分公司物探研究院,山东东营257000 [4]中国科学技术大学自动化系,安徽合肥230026

出　　处：《油气地质与采收率》2023年第3期49-58,共10页Petroleum Geology and Recovery Efficiency

基　　金：国家自然科学基金面上项目“地球物理测井资料智能处理与解释方法研究”(62273319);中国石化科技攻关项目“人工智能技术在井位部署中应用探索研究”(PE19008-8);胜利油田基础前瞻研究项目“面向‘十五五'胜利油田勘探开发技术发展方向及战略研究—地球物理专题”(YJQ2205);胜利油田重点科技项目“地震储层预测样本标注智能化方法研究及应用”(YKJ2201)。

摘　　要：基于地球物理测井地层划分相关概念及分类,将测井曲线自动分层方法分为传统方法和人工智能方法,从有监督学习方法和无监督学习方法2个方面分析人工智能方法的应用情况,并综合比较各类地层自动划分方法的优缺点。通过探索相关领域的发展情况,从不同角度思考测井地层划分方法进一步发展所存在的挑战及其解决方法。一是引入半监督学习方法,解决人工标签稀缺问题;二是从分割模型的角度,打破对测井数据的固有认识;三是采用测井曲线重构等方法,解决井段失真或缺失所导致的数据异构问题;四是通过样本加权,解决人工标签错误导致的数据偏差问题;五是采用迁移学习方法,解决不同地区数据分布差异问题。人工智能方法是解决地层划分、岩性识别、储层识别、生产运行中现有难题以及推进测井相关任务数字化转型的重要支撑。Given the related concepts and clssifcation of geophysical lgging stratigraphic division,this paper divided au-tomatic stratification methods of logging curves into traditional methods and artifcial intelligence methods and analyzed the application of artificial intelligence technology in logging stratigraphic division from the aspects of supervised and un-supervised learming.Then,it comprehensively compared the advantages and disadvantages of various automatic strati-graphic division methods.Finally,by exploring the development of related fields,this study considered the challenges and solutions in the future development of logging stratigraphic division from different perspectives.The specific solutions are as follows:①Semi supervised learning can be introduced to solve the problem of scarce manual labels.②A new under-standing of logging data can be obtained from the perspective of the segmentation model.③Methods such as logging curve reconstruction can be employed to solve the problem of data heterogeneity caused by the distortion or missing of well sec-tions.④The problem of data deviation caused by manual label errors can be resolved through sample weighting.⑤Trans-fer learning can be used to solve the problem of data distribution differences in different regions.Artificial itelligence tech-nology can provide vital support for solving existing problems in stratigraphic division,lithology identification,reservoir identification,as well as operation and production,and promoting the digital transformation of tasks related with logging.

关 键 词：测井曲线 地层划分 人工智能 半监督学习 迁移学习 

分 类 号：TE319[石油与天然气工程—油气田开发工程]