人工智能在油气压裂增产中的研究现状与展望  

Research Status and Prospect of Artificial Intelligence in Reservoir Fracturing Stimulation

作　　者：盛茂[1] 李根生[1,2] 田守嶒 廖勤拙[1] 王天宇 宋先知[1] SHENG Mao;LI Gensheng;TIAN Shouceng;LIAO Qinzhuo;WANG Tianyu;SONG Xianzhi(State Key Laboratory of Petroleum Resources and Prospecting,China University of Petroleum(Beijing),Beijing 102249,China)

机构地区：[1]油气资源与探测国家重点实验室·中国石油大学(北京) [2]中国工程院

出　　处：《钻采工艺》2022年第4期1-8,共8页Drilling & Production Technology

基　　金：国家自然科学基金优秀青年科学基金项目“油气井流体力学与工程”(编号:52122401);中国石油天然气集团公司—中国石油大学(北京)战略合作项目“物探、测井、钻完井人工智能理论与应用场景关键技术研究”(编号:ZLZX2020-03)。

摘　　要：针对油气压裂增产技术发展需求,阐述了人工智能在油气压裂增产中的研究现状,分析了压裂人工智能发展所面临的关键理论问题,展望了压裂人工智能研究的主攻方向和应用场景设计。国内外现阶段在压裂设计优化、压裂工况诊断与风险预警、压裂返排优化控制等方面已取得一定研究进展,总体处于从学术型研究向工业级应用的过渡阶段,面临小样本少标签数据问题、数据驱动与机理模型深度融合问题、模型可解释性差等关键理论问题。文章围绕所存在的问题,展望未来压裂人工智能研究的主攻方向包括数据治理与特征工程、小样本学习场景下的压裂数据深度挖掘、基于知识嵌入和知识发现的可解释性压裂智能算法、基于强化学习的压裂参数动态优化与风险预警调控方法等。基于上述研究,建议构建压裂设计智能优化、压裂施工闭环调控、压裂返排智能控制等三类应用场景,最终实现高质量均衡造缝和安全压裂目标。In view of the development needs of oil and gas reservoir fracturing stimulation technology, the research status of artificial intelligence(AI) in fracturing stimulation is described, the key theoretical problems in the development of AI in fracturing are analyzed, and the future main research directions and application scenarios are prospected. At home and abroad, some progress has been made in fracturing design optimization, fracturing condition diagnosis and risk pre-warning, fracturing flowback optimization control. But it is still in the transition stage from academic research to industrial application, facing with the key theoretical problems including small sample size and lack of label data, poor model interpretability and the requirement of deep integration of data-driven and physical models. Based on the existing problems, the main research directions of artificial intelligence for fracturing in the future are prospected in this paper, such as data management and feature engineering, deep mining of fracturing data in small sample learning scenarios, interpretable fracturing intelligent algorithms based on knowledge embedding and knowledge discovery, and dynamic optimization of fracturing parameters and risk pre-warning and control methods based on reinforcement learning, etc. Three application scenarios are suggested, including intelligent optimization of fracturing design, closed-loop control of fracturing construction, and intelligent control of fracturing flowback, aiming for the high-quality balanced fracture formation and safe fracturing.

关 键 词：人工智能 机器学习 水力压裂 应用场景 压裂设计 风险预警 压裂返排 

分 类 号：TE357[石油与天然气工程—油气田开发工程] TP18[自动化与计算机技术—控制理论与控制工程]