摘要
岩石孔渗特征是影响储层流体储集及渗流能力的主要因素。目前数字岩心孔渗参数计算通常采用孔隙尺度建模并进行数值模拟,具有建模复杂、耗时长的缺点。为此,本文根据天然岩心CT扫描结果,运用OpenFOAM生成654组训练样本,并通过机器学习算法建立数字岩心孔渗快速预测模型,并对模型超参数进行敏感性分析。当学习率为0.003时,模型具有较强的泛化能力,孔渗预测结果误差小于10%的占比90%以上,且能够在1 s内完成。研究结果实现了数字岩心孔渗高效率、高精度预测,能够有效降低生产成本,提高工作效率。
Rock porosity and permeability are the main factors affecting fluid storage and flow capacity in reservoirs. At present, pore scale modeling and numerical simulation are usually adopted in the estimation of properties of digital cores where modeling is complex and time-consuming. Therefore, based on the CT scanning results of natural cores, 654 sets of training samples were generated using OpenFOAM, and a fast prediction model was established by a machine learning algorithm. Sensitivity analysis was further conducted for model hyperparameters. When the learning rate is 0.003, the model displays strong generalization ability and prediction accuracy is above 90%. The time of prediction is reduced from more than one hour to less than one second. We propose a high efficiency and high-precision pore permeability prediction method of 3 D digital cores based on machine learning, which can effectively reduce cost and improve work efficiency.
作者
王依诚
姜汉桥
于馥玮
成宝洋
徐飞
李俊键
WANG Yicheng;JIANG Hanqiao;YU Fuwei;CHENG Baoyang;XU Fei;LI Junjian(State Key Laboratory of Petroleum Resources and Engineering,China University of Petroleum-Beijing,Beijing 102249,China)
出处
《石油科学通报》
2019年第4期354-363,共10页
Petroleum Science Bulletin
基金
国家重大专项课题(2017ZX05009-005)
中国石油大学(北京)优秀青年学者基金(2462019QNXZ04)资助
关键词
机器学习
数字岩心
渗透率预测
CT扫描
machine learning
digital cores
permeability prediction
CT scanning
