深度学习网络在非常规油气开发中的应用研究  

Research on application of deep learning network in unconventional oil and gas development

作　　者：李道伦[1] 查文舒[1] 刘旭亮 李祥[3] 沈路航 周霞 郝玉祥 汪欢 LI Daolun;ZHA Wenshu;LIU Xuliang;LI Xiang;SHEN Luhang;ZHOU Xia;HAO Yuxiang;WANG Huan(School of Mathematics,Hefei University of Technology,Hefei 230601,China;Hefei University,Hefei 230000,China;School of Engineering Science,University of Science and Technology of China,Hefei 230051,China)

机构地区：[1]合肥工业大学数学学院,合肥230601 [2]合肥大学,合肥230000 [3]中国科学技术大学工程技术学院,合肥230051

出　　处：《非常规油气》2024年第6期1-7,共7页Unconventional Oil & Gas

基　　金：国家自然科学基金面上项目“基于深度学习的渗流方程求解方法研究”(12172115);国家自然科学基金面上项目“基于流动方程的致密油气藏深度学习数值模拟方法”(12372244)。

摘　　要：以深度学习为代表的人工智能已被公认为是石油勘探开发技术转型升级的关键核心技术。借助深度学习强大的学习能力,试井解释正在向模型自动识别及试井自动解释方向快速演化;早期的产量预测多基于已知的特征,从实测数据中自动提取特征的研究引发关注,仅基于井口压力的产量预测已取得积极进展;以多点地质统计为基础的数字岩心重建方法正被生成对抗网络所取代,但如何满足裂缝、孔隙度及渗透率等约束仍是难题;偏微分方程求解正经历着颠覆性变化,基于非线性方程的传统求解方法正迎来智能求解时代。基于物理本质,单一的神经网络结构正在向复杂网络结构演化。随着人工智能技术的不断发展,大型模型将成未来趋势。因此,相关研究多头并进,加强数据收集,构建大型模型,是我国未来智慧油气开发的核心任务。Artificial intelligence represented by deep learning has been recognized as the core technology for the transformation and upgrading of oil exploration and development technology.Well test interpretation is rapidly evolving towards automatic model recognition and artificial interpretation with the powerful learning ability of deep learning.Early production prediction was mainly based on known features,and the research on automatically extracting features from measured data was attracted attention.In addition,positive progress has been made in production prediction based solely on wellhead pressures.Generative adversarial networks are replacing the digital core reconstruction method based on the multi-point geological statistic,but how to meet the constraints such as fractures,porosity,and permeability remains challenging.Partial differential equation solving is undergoing subversive changes.Traditional solving methods based on nonlinear equations are ushering in an era of intelligent solving.Based on the physical nature,the single neural network structure evolves into a complex network structure.Large-scale models will be the future trend with the development of artificial intelligence technology.Therefore,the relevant research goes hand in hand,and strengthening data collection and building large-scale models will be one of the important tasks for future oilfield development in China.

关 键 词：深度学习 油气开发 试井解释 产量预测 人工智能 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] O241.82[自动化与计算机技术—控制科学与工程]