基于支持向量机的高含CO_(2)凝析气藏油环体积预测方法  

Prediction of oil ring volume in condensate gas reservoirs with high CO_(2) content based on a support vector machine learning method

作　　者：陈浩[1,2] 蒋东梁 邢建鹏 王红平 左名圣 王朝锋 杨柳 刘希良 于海增 袁志文 CHEN Hao;JIANG Dongliang;XING Jianpeng;WANG Hongping;ZUO Mingsheng;WANG Chaofeng;YANG Liu;LIU Xiliang;YU Haizeng;YUAN Zhiwen(College of Safety and Ocean Engineering,China University of Petroleum(Beijing),Beijing 102249,China;State Key Laboratory of Petroleum Resources and Prospecting,Beijing 102249,China;PetroChina Hangzhou Research Institute of Geology,Hangzhou 310023,China)

机构地区：[1]中国石油大学(北京)安全与海洋工程学院,北京102249 [2]油气资源与探测国家重点实验室,北京102249 [3]中国石油杭州地质研究院,浙江杭州310023

出　　处：《中国石油大学学报（自然科学版）》2023年第2期90-98,共9页Journal of China University of Petroleum(Edition of Natural Science)

基　　金：国家自然科学基金青年基金项目(51704303);北京市自然科学基金项目(3173044)。

摘　　要：准确、高效地预测高含CO_(2)凝析气藏油环体积对于开发方案的制定至关重要,但海上深水凝析气藏难以通过大规模钻探来探明油环体积,且高浓度CO_(2)的萃取作用使油环体积变化更加复杂。首先通过CO_(2)充注实验还原高含CO_(2)凝析气藏成藏过程,以数值模拟结果为基础数据开展数据预处理,建立样本数据库,并通过关联分析优选其主控因素,明确不同地层条件和气顶组成下油环体积的变化规律,最后基于支持向量机开展油环体积预测训练,搭建油环体积的预测模型,实现输入主控因素以精确、快速预测油环体积的目的。预测结果表明,采用三次核函数的机器学习模型与数值模拟、物理模拟、矿场实际的油环体积误差分别为3.43%、5.10%和7.21%。An accurate and efficient prediction of the oil ring volume in condensate gas reservoirs with high CO_(2) content is very important for the development of the reservoir.However,it is difficult to detect the oil ring volume for offshore deep-water condensate gas reservoirs through large-scale drilling,and the extraction effect of CO_(2) can make the oil ring volume more complex.In this study,in order to establish a machine learning model to predict the oil ring volume,the reservoir forming process of high CO_(2) condensate gas reservoirs was firstly simulated through CO_(2) filling experiments,and data fusion processing was conducted based on the numerical simulation results to establish a sample database,and optimize its main controlling factors through correlation analysis.The variation of oil ring volume under different formation conditions and gas cap composition was then clarified.Finally,the oil ring volume prediction training was carried out based on a support vector machine learning method,and the prediction model of the oil ring volume was established with formation conditions and main gas cap compositions as the main input and control parameters.The prediction results show that the errors between the machine learning model using a cubic kernel function with numerical simulation,experiment and actual oil ring volume are 3.43%,5.10%and 7.21%,respectively.

关 键 词：高含CO_(2) 凝析气藏 油环体积 支持向量机 机器学习 

分 类 号：TE372[石油与天然气工程—油气田开发工程]