油藏数值模拟智能化技术思考与展望  

Thoughts and Prospects on The Intelligent Transformation of Reservoir Simulation

作　　者：吴淑红[1] 毕剑飞 范天一 徐明源 杨义轩 王宝华[1] 李华[1] 李巧云[1] Wu Shuhong;Bi Jianfei;Fan Tianyi;Xu Mingyuan;Yang Yixuan;Wang Baohua;Li Hua;Li Qiaoyun(PetroChina Research Institute of Petroleum Exploration&Development,Beijing 100083,China)

机构地区：[1]中国石油勘探开发研究院

出　　处：《石油科技论坛》2024年第6期96-106,共11页PETROLEUM SCIENCE AND TECHNOLOGY FORUM

基　　金：中国石油天然气股份有限公司科技项目“油气勘探开发人工智能关键技术研究”(编号:2023DJ84)。

摘　　要：油藏模拟作为油气藏开发与管理的关键技术,对老油田提高采收率和非常规油气资源增储上产至关重要。近年来,深度学习技术的突破为提升油藏模拟的精度与效率提供了新的路径和思路,推动了油藏模拟的智能化转型。文章深入分析了涉及纯数据驱动模型、算子学习模型、物理约束模型等各类深度学习模型在油藏模拟领域的适用场景与潜在价值,人工智能在辅助油藏数值模拟、构建高效代理模型方面的研究进展,明确了深度学习在油藏模拟领域的应用前景和局限。研究提出油藏模拟智能化未来将聚焦于构建专业大模型、探索渗流偏微分方程(PDE)求解、开发智能体技术等方向发展。Reservoir simulation,as a key technology for oil and gas reservoir development and management,is crucial for enhancing recovery in mature fields and increasing reserves and production in unconventional resources.In recent years,breakthroughs in deep learning have provided new approaches and insights for improving the accuracy and efficiency of reservoir simulation,driving its intelligent transformation.This article delves into the applicable scenarios and latent value of diverse deep learning models,including pure data-driven models,operator learning models,and physics-informed models,within the domain of reservoir simulation.It also examines the research progress of artificial intelligence in facilitating reservoir numerical simulation and constructing efficient surrogate models,clarifying the application prospects and limitations of deep learning in reservoir simulation.The research proposes that the future development of intelligent reservoir simulation will concentrate on directions such as constructing specialized large models,exploring the solution of the partial differential equation(PDE)for fluid flow,and developing agent-based technologies.

关 键 词：油藏模拟 深度学习 智能化转型 辅助模拟 代理模型 

分 类 号：TE34[石油与天然气工程—油气田开发工程] TP319[自动化与计算机技术—计算机软件与理论]