基于随机森林算法的二氧化碳驱油与封存主控因素研究  

Research on the main control factors of carbon dioxide flooding and storage based on random forest algorithm

作　　者：任俊帆 薛亮[1,2] 聂捷 肖镭 廖广志[1,3] REN Junfan;XUE Liang;NIE Jie;XIAO Lei;LIAO Guangzhi(National Key Laboratory of Petroleum Resources and Engineering,China University of Petroleum(Beijing),Beijing 102249,China;College of Petroleum Engineering,China University of Petroleum(Beijing),Beijing 102249,China;College of Geophysics,China University of Petroleum(Beijing),Beijing 102249,China;Chuanqing Drilling Engineering Company Limited,China National Petroleum Corporation,Chengdu 610051,China;Alibaba Cloud Computing Company Limited,Hangzhou 310030,China)

机构地区：[1]中国石油大学(北京)油气资源与工程全国重点实验室,北京102249 [2]中国石油大学(北京)石油工程学院,北京102249 [3]中国石油大学(北京)地球物理学院,北京102249 [4]中国石油集团川庆钻探工程有限公司,成都610051 [5]阿里云计算有限公司,杭州310030

出　　处：《地质科技通报》2024年第3期147-156,共10页Bulletin of Geological Science and Technology

基　　金：国家自然科学基金项目(52274048);北京市自然科学基金项目(3222037)。

摘　　要：在碳达峰、碳中和目标背景下,二氧化碳驱油与封存是经济可行的碳减排的主要技术手段。明确影响二氧化碳驱油与封存效果的主控因素,是实现二氧化碳高效驱油与封存的基础。在行业标准算例PUNQ-S3模型的基础上,综合考虑二氧化碳与原油混相作用和二氧化碳构造、残余、溶解、矿化封存机理,构建了二氧化碳提高原油采收率与地质封存一体化数值模拟模型,结合随机森林智能算法,开展了影响二氧化碳驱产油量和封存量的储层和生产参数特征重要性分析,考虑驱油与封存时间尺度的差异,建立了参数时序特征重要性分析方法,实现了在不同二氧化碳驱油与封存阶段的主控因素分析。结果表明,二氧化碳驱油与封存时序随机森林模型准确性高,在二氧化碳驱油与封存前期,二氧化碳构造封存量受储层含水饱和度控制,溶解封存量受地层水矿化度控制;在二氧化碳驱油与封存中、后期,二氧化碳构造封存量则受储层渗透率控制,溶解封存量则受储层渗透率与地层水矿化度控制;残余封存量在二氧化碳驱油与封存前期较小,导致其主控因素不明显,在二氧化碳驱油与封存中后期受储层渗透率与含水饱和度控制;二氧化碳矿化封存量在整个二氧化碳驱油与封存阶段受地层水pH值与矿化度控制;二氧化碳驱油量在整个二氧化碳驱油与封存阶段受储层渗透率及含水饱和度控制。时序随机森林算法可以明确不同二氧化碳驱油与封存阶段的主控因素,为二氧化碳提高原油采收率和地质封存的高效实施提供了技术支撑。[Objective]To achieve carbon peak and carbon neutrality goals,carbon dioxide flooding and storage are the main technical means for carbon emission reduction.It is crucial to clarify the main controlling factors of carbon dioxide flooding and storage under reservoir conditions,which provides the basis for realizing the efficient development of carbon dioxide flooding and storage.[Methods]In this study,with the widely used PUNQ-S3 case study as the basis,an integrated numerical simulation model of carbon dioxide flooding and geological storage is constructed.It considers the miscible interaction between carbon dioxide and crude oil as well as storage mechanisms,including structural,residual,dissolved,and mineral trapping.By employing the random forest intelligent algorithm,a feature importance analysis of reservoir and production parameters during the carbon dioxide flooding and storage process is carried out.The differences between carbon dioxide flooding and storage at different time scales are considered.A time series-based feature importance analysis method is established,and the main controlling factors in the different carbon dioxide flooding and storage stages are analysed.Through the fluctuation of the evaluation index,the influence of reservoir and production parameters on different stages of carbon dioxide flooding and storage is inferred.[Results]The results show that the time series-based random forest model for carbon dioxide flooding and storage has high accuracy.In the early stage of carbon dioxide flooding and storage,the amount of carbon dioxide structural storage is controlled by the reservoir water saturation,and the amount of dissolved storage is controlled by the salinity of the formation brine;In the middle and later stages of carbon dioxide flooding and storage,the amount of carbon dioxide structural storage is controlled by reservoir permeability,while the amount of dissolved storage is controlled by reservoir permeability and formation water salinity;The residual storage capacity is small i

关 键 词：二氧化碳驱油与封存 随机森林算法 特征重要性分析 原油采收率 数值模拟 地层水矿化度 储层 

分 类 号：TE357.45[石油与天然气工程—油气田开发工程]