基于混合物理数据驱动的油藏地质体CO_(2)利用与封存代理模型研究  

Study on CO_(2)Utilization and Storage Proxy Model for Reservoir Geobodies Based on Hybrid Physics Driven-Data

作　　者：芮振华[1,2,3] 邓海洋 胡婷 RUI Zhenhua;DENG Haiyang;HU Ting(State 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 Carbon Neutrality Future Technology,China University of Petroleum(Beijing),Beijing 102249,China)

机构地区：[1]油气资源与工程全国重点实验室·中国石油大学(北京) [2]中国石油大学(北京)石油工程学院 [3]中国石油大学(北京)碳中和未来技术学院

出　　处：《钻采工艺》2025年第1期190-198,共9页Drilling & Production Technology

基　　金：国家重点研发计划项目“二氧化碳提高油藏采收率与地质封存一体化关键技术及应用示范”(编号:2022YFE0206700);国家自然科学基金项目“碳酸盐岩油藏二氧化碳在提高采收率与封存中的运移演化规律研究”(编号:42302272);国家资助博士后创新人才支持计划项目“碳酸盐岩油藏中二氧化碳提高采收率与封存的协同机制与模型研究”(编号:GZB20230862)。

摘　　要：在全球能源转型与能源需求持续增长的背景下,碳捕获、利用和封存(CCUS)已成为极具前景的研究方向。CO_(2)利用与封存协同优化通常依赖大量的组分正演模拟,但三维高分辨率模型计算成本高昂,限制其广泛应用。基于混合物理数据驱动的GPSNet模型以其高效的计算效率已成为一种理想的代理模型,然而现有的GPSNet模型难以准确捕获复杂的相行为和组分间的相互作用,为此,文章提出了一种新型专用于组分模拟的comp-GPSNet模型,通过标准失配最小化方法和基于伴随的梯度优化算法对comp-GPSNet模型进行训练,以拟合从高分辨率模拟中获取的井响应数据。将训练后的模型应用到PUNQ-S3油藏中,全面评估复杂条件下comp-GPSNet模型的预测能力,结果表明,comp-GPSNet模型在单井和区块范围内均表现出良好的预测精度,CO_(2)利用率和封存率的预测误差分别为0.16%和3.13%。该模型为CO_(2)利用与封存协同优化提供了一个稳健的代理框架,以推动油田数字化与智能化发展。In the context of the global energy transition and increasing energy demand,carbon capture,utilization and storage(CCUS)has become a highly promising research direction.The collaborative optimization of CO_(2) utilization and storage often depends on extensive compositional forward simulations.However,the high computational cost of three-dimensional high-resolution models limits their widespread application.The GPSNet model based on hybrid physics data-driven has become an ideal surrogate model due to its computational efficiency.However,existing GPSNet models struggle to accurately capture complex phase behaviors and interactions among components.Therefore,the paper proposes a novel comp-GPSNet model specifically designed for component simulation.The comp-GPSNet model is trained by standard misfit minimization methods and the adjoint-based gradient optimization algorithm to match the well response data obtained from high-resolution simulations.The trained model is applied to the PUNQ-S3 reservoir to comprehensively evaluate the predictive ability of the comp-GPSNet model under complex conditions.The results show that the comp-GPSNet model has good prediction accuracy in both single well and block ranges,and the prediction errors of CO_(2) utilization factor and storage factor are 0.16%and 3.13%,respectively.The model provides a robust proxy framework for the collaborative optimization of CO_(2) utilization and storage,and promotes the digital and intelligent development of oilfields.

关 键 词：CCUS comp-GPSNet 混合物理数据驱动 代理模型 组分模拟 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]