基于机器学习的CO_(2)封存致密砂岩储层孔隙度预测  

Machine learning-based prediction of porosity in tight sandstone reservoirs for CO_(2) geological sequestration

作　　者：路萍 郭京哲 高春云 赵军辉 张亚芹 南右泽 谭富荣 杨桂林 钟高润 李阳阳 巨浩波 焦尊生[11] LU Ping;GUO JingZhe;GAO ChunYun;ZHAO JunHui;ZHANG YaQin;NAN YouZe;TAN FuRong;YANG GuiLin;ZHONG GaoRun;LI YangYang;JU HaoBo;JIAO ZunSheng(Shaanxi Institute of Energy Resource and Chemical Engineering,Xi'an 710127,China;Huairou Laboratory,Beijing 101499,China;Northwest University,Xi'an 710127,China;Shaanxi Yulin Energy Chemical Industry Research Institute Co.,Ltd.,Yulin 719000,China;Changqing Oilfield Eighth Oil Extraction Plant,Xi'an 710018,China;Xi'an Abbott Environmental Analysis and Testing Technology Co.,Ltd.,Xi'an 710018,China;Shanghai Youye Information Technology Co.,Ltd.,Shanghai 201799,China;Shaanxi Mineral Geological Survey Center,Xi'an 710069,China;Yulin University,Yulin 719053,China;Yan'an University,Yan'an 716000,China;University of Wyoming,Laramie 82071,America)

机构地区：[1]陕西省能源化工研究院,西安710127 [2]怀柔实验室,北京101499 [3]西北大学,西安710127 [4]陕西榆能集团能源化工研究院有限公司,榆林719000 [5]长庆油田第八采油厂,西安710018 [6]西安阿伯塔资环分析测试技术有限公司,西安710018 [7]上海优也信息科技有限公司,上海201799 [8]陕西省矿产地质调查中心,西安710069 [9]榆林大学,榆林719053 [10]延安大学,延安716000 [11]怀俄明大学,拉勒米82071

出　　处：《地球物理学进展》2024年第3期1129-1140,共12页Progress in Geophysics

基　　金：陕西省自然科学基础研究计划项目(2022JQ-231);中国博士后科学基金-面上项目(2023M741321);北京市博士后科学基金(2023-ZZ-64)联合资助。

摘　　要：CO_(2)地质封存技术被认为是降低大气CO_(2)浓度的有效措施,是我国实现“双碳”目标的关键技术之一.孔隙度是评价储层储集性能的关键参数,对其高精度预测是CO_(2)封存潜力评估的一项重要内容.对于CO_(2)封存致密砂岩储层而言,由于孔隙类型多样,非均质性较强,在以往储层物性评价工作中,利用已有模型和测井解释方法,预测结果与实际测试结果往往偏差很大,实际应用效果不佳.如何利用前沿的数学算法充分挖掘测井数据中隐含的物性信息,建立预测精度较高泛化性能较强的孔隙度模型是致密砂岩储层精细勘探高效开发的关键,更是二氧化碳地质封存潜力评估的关键.本文基于机器学习主成分回归(PCR)、高斯过程回归法(GPR)、随机森林(RF)、支持向量机(SVM)、BP神经网络(BP-ANN)以及极致提升算法(XGBoost)构建致密砂岩储层孔隙度预测模型,经综合对比发现,基于XGBoost的孔隙度预测模型预测精度最高、泛化性能最强,该方法为CO_(2)地质封存致密砂岩储层孔隙度预测模型的构建提供新的思路.CO_(2) geological storage technology is a crucial measure in reducing atmospheric CO_(2) concentration and is considered as one of the key technologies for achieving China's “dual carbon” goals.Porosity,as a key parameter for evaluating the reservoir storage performance,plays a vital role in CO_(2) storage potential evaluation.However,in the case of CO_(2) storage in tight sandstone reservoirs,due to the diverse pore types and strong heterogeneity,the predicted results using existing models and logging interpretation methods often deviate greatly from the actual test results,resulting in poor actual application effects.Therefore,the establishment of a porosity model with high prediction accuracy and strong generalization performance by using cutting-edge mathematical algorithms to fully explore the porosity information hidden in logging data is crucial for the fine exploration and efficient development of tight sandstone reservoirs,as well as the evaluation of CO_(2) geological storage potential.This paper presents a porosity prediction model for tight sandstone reservoirs based on machine learning techniques,including Principal Component Regression(PCR),Gaussian Process Regression(GPR),Random Forest(RF),Support Vector Machine(SVM),BP Neural Network(BP-ANN),and Extreme Gradient Boosting Algorithm(XGBoost).Through comprehensive comparison,the XGBoost-based porosity prediction model is found to have the highest prediction accuracy and strongest generalization performance,providing new ideas for the construction of porosity prediction models for tight sandstone reservoirs.

关 键 词：CO_(2)地质封存 致密砂岩储层 孔隙度模型 机器学习 XGBoost 

分 类 号：P631[天文地球—地质矿产勘探]