天然气工业2026，Vol.46Issue(5)：25-36,12.DOI:10.3787/j.issn.1000-0976.2026.05.003
基于核磁共振和机器学习连续定量评价孔隙结构的新方法
A new method for continuous quantitative evaluation of pore structures based on NMR and machine learning
摘要
Abstract
Deep tight sandstone reservoirs are characterized by complex pore structures and strong heterogeneity,which makes traditional core experiments fail to support the vertical continuous quantitative evaluation of pore structure parameters.Therefore,it is in urgent need to establish a new method applicable to the continuous quantitative evaluation of pore structures in deep tight sandstone reservoirs.Taking the deep tight sandstone reservoir of Jurassic Sangonghe Formation in the Taibei Sag of the Tuha Basin as an example,this paper establishes a new method for continuous quantitative evaluation of pore structures by making comprehensive use of cores thin section,scanning electron microscopy(SEM),high-pressure mercury injection(HPMI),nuclear magnetic resonance(NMR)and conventional logging data,in combination with CatBoost machine learning algorithm and Optuna hyperparameter optimization technology,which reconstructs capillary force curves based on NMR and performs prediction through machine learning.The following results are obtained.First,based on the HPMI experimental results of 16 samples from the study area,the pore structures in the Sangonghe Formation reservoirs are classified into three types by analyzing capillary force curve shape,pore throat radius distribution and pore throat characteristic parameters,which lays a classification basis for the establishment of the method for continuous quantitative evaluation of pore structures.Second,based on the data of NMR logging and HPMI,a method for constructing capillary force curve based on NMR data is established by calibrating NMR logging with capillary force curve,so as to realize the quantitative continuous evaluation of pore structures in the Sangonghe Formation deep tight sandstone reservoirs.Third,the expulsion pressure prediction model based on CatBoost has a test set determination coefficient of 0.859,and can provide reliable expulsion pressure prediction for the real hole sections without NMR logging.Fourth,the individual-well application results indicate that this model can effectively support the quantitative evaluation of pore structures in the hole sections without NMR logging.In conclusion,the newly proposed method for continuous quantitative evaluation of pore structures based on NMR and machine learning can realize the continuous reconstruction of capillary force curve and the intelligent prediction of expulsion pressure,and provide reference and guidance for the fine characterization of pore structures in deep tight sandstone reservoirs and the identification of favorable reservoirs.关键词
吐哈盆地/三工河组/核磁共振/高压压汞/机器学习/孔隙结构/深层致密砂岩Key words
Tuha Basin/Sangonghe Fm/Nuclear magnetic resonance(NMR)/High-pressure mercury injection(HPMI)/Machine learning/Pore structure/Deep tight sandstone分类
能源科技引用本文复制引用
王贵文,田银宏,李红斌,何志斌,邵临波,赖锦..基于核磁共振和机器学习连续定量评价孔隙结构的新方法[J].天然气工业,2026,46(5):25-36,12.基金项目
中国石油天然气股份有限公司科技项目"吐哈盆地深层-超深层致密砂岩气富集机理与关键评价技术研究"(编号:2022DJ2017). （编号:2022DJ2017）
