期刊
JOURNAL OF PETROLEUM SCIENCE AND ENGINEERING
卷 195, 期 -, 页码 -出版社
ELSEVIER
DOI: 10.1016/j.petrol.2020.107837
关键词
Reservoir characterization; Lithofacies classification; Permeability modeling; Statistical analysis; Machine learning; Clastic reservoirs
An integrated multidisciplinary workflow of machine learning and data analytics was conducted for the multivariate geostatistical characterization of clastic reservoirs. This workflow was adopted for the estimation of lithofacies and permeability distributions in non-core intervals and wells on data from the upper shale member/Zubair formation in Luhais oil field, southern Iraq. Specifically, the workflow includes field and reservoir description, review of univariate, bivariate, and multivariate statistics for facies and petrophysical property distribution. In univariate and bivariate statistical analysis, the correction of core and log scale was adopted in order to have core permeability corrected to log scale. The discrete lithofacies distribution has been obtained for a few wells to be later incorporated in the permeability modeling and prediction. At first, the measured discrete lithofacies distributions were incorporated into classification algorithms to model the facies given the well logging data (shale volume, neutron porosity and water saturation) in order to predict the discrete facies distribution at other wells of missing records. The Multinomial Logistic Regression (Multinom), Logistic Boosting Regression (LogitBoost), and Extreme Gradient Boosting (XGBoost) were all comparatively adopted for lithofacies classification. After that, the resulting most accurate discrete facies distribution by LogitBoost were included along with the well logging interpretations into the multivariate permeability modeling through advanced machine learning approaches to model and predict the corrected core permeability given well logging interpretations for all wells in the reservoir. More specifically, the Multivariate Adaptive Regression Splines (MARS) and Smooth Generalized Additive Models (SGAM) were comparatively adopted to model and predict the core permeability at all wells. From this workflow, it was indicated that LogitBoost and MARS were the best machine learning algorithms to produce the most realistic lithofacies classification and permeability modeling, respectively.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据