基于半监督高斯混合模型与梯度提升树的砂岩储层相控孔隙度预测  

Facies-controlled porosity prediction of sandstone reservoirs based on semi-supervised Gaussian mixture model and gradient boosting tree

作　　者：魏国华[1] 韩宏伟[1] 刘浩杰[1] 李明轩 袁三一[2] WEI Guohua;HAN Hongwei;LIU Haojie;LI Mingxuan;YUAN Sanyi(Shengli Geophysical Research Institute of Sinopec,Dongying,Shandong 257000,China;State Key Laboratory of Petroleum Resources and Prospecting,China University of Petroleum(Beijing),Beijing 102249,China)

机构地区：[1]中国石化胜利油田分公司物探研究院,山东东营257000 [2]中国石油大学(北京)油气资源与探测国家重点实验室,北京102249

出　　处：《石油地球物理勘探》2023年第1期46-55,共10页Oil Geophysical Prospecting

基　　金：国家自然科学基金项目“五维叠前地震信息驱动的深度学习致密砂岩储层表征机制及含气性预测”(41974140);中国石油天然气集团有限公司—中国石油大学(北京)战略合作科技专项“多源信息深度融合的储层预测和精细描述”(ZLZX2020-03)联合资助。

摘　　要：孔隙度是一种描述储层物性特征的重要参数。考虑砂岩与泥岩的孔隙度存在明显差异,提出了一种基于半监督高斯混合模型与梯度提升树的相控孔隙度预测方法,以实现砂岩储层孔隙度的精细描述。首先利用少量具岩相标签的测井数据确定高斯混合模型的初始聚类中心及对应的岩相类别;其次利用大量无标签测井数据优化高斯混合模型,实现砂岩与泥岩的准确划分;再次基于地质认识将泥岩孔隙度解释为固定的极小值,从而后续只开展砂岩孔隙度预测;然后将测井曲线拟合方法导出的孔隙度先验信息和测井敏感属性作为梯度提升树算法的多元输入信息,通过学习统计性岩石物理关系建立砂岩孔隙度的计算模型;最终根据岩相结果将砂岩段与泥岩段的孔隙度进行组合得到相控孔隙度。D油田的18口井数据测试结果表明:半监督高斯混合模型的岩相分类效果优于K均值、支持向量机、随机森林等机器学习算法,2口盲井的岩相分类准确率达到94.5%;所构建方法对2口盲井预测的相控孔隙度结果与真实孔隙度具有较高的一致性,平均相关系数达0.805。Porosity is an important parameter to describe the physical properties of reservoirs.Considering the obvious differences in the porosity of sandstone and mudstone,this paper proposes a new method for facies-controlled porosity prediction that combines a semi-supervised Gaussian mixture model and a gradient boosting tree to achieve the fine porosity description of sandstone reservoirs.First,a small amount of logging data with lithofacies labels is used to determine the initial cluster center of the Gaussian mixture model and the corresponding lithofacies types.Then,a large amount of unlabeled logging data is used to optimize the Gaussian mixture model so that sandstone and mudstone can be classified correctly.Depending on geological knowledge,the mudstone porosity is interpreted as a fixed minimum value,and only sandstone porosity is predicted subsequently.The porosity prior information and logging sensitive attributes derived from logging curve fitting are taken as the multiva-riate input information of the gradient boosting tree algorithm,and the calculation model of sandstone porosity is built by learning the statistical petrophysical relationship.Finally,according to the lithofacies results,the porosity of the sandstone section and the mudstone section is combined to obtain the facies-controlled porosity.The method is tested with the data of 18 wells in Oilfield D.The results show that the lithofacies classification effect of the semi-supervised Gaussian mixture model is better than those of K-means,support vector machine,random forest,and other machine learning algorithms,and the lithofacies classification accuracy of two blind wells reaches 94.5%.In addition,the facies-controlled porosity predicted by the proposed method in two blind wells is highly consistent with the true porosity with an average correlation coefficient of 0.805.

关 键 词：相控孔隙度预测 岩相划分 半监督高斯混合模型 梯度提升树 机器学习 

分 类 号：P631[天文地球—地质矿产勘探]