基于增强域数据微调Yolo模型的储气库断层智能识别方法及应用  

Enhanced Domain Tuned Yolo-Driven Intelligent Fault Identification Method:Application in Selection and Construction of Gas Storage

作　　者：白雪峰 张峰源 邹环宇 黄发木 李俊伦 赵世杰 张莉[5] 汤继周 BAI Xuefeng;ZHANG Fengyuan;ZOU Huanyu;HUANG Famu;LI Junun;ZHAO Shijie;ZHANG Li;TANG Jizhou(Oil&Gas Pipeline Network Corporation,Shanghai 200011,China;School of Ocean and Earth Science,Tongji University,Shanghai 200092,China;State Key Laboratory of Marine Geology,Tongji University,Shanghai 200092,China;School of Earth and Space Sciences,University of Science and Technology of China,Hefei,Anhui 230026,China;College of Earth Sciences,Yangtze University,Wuhan,Hubei 430199,China)

机构地区：[1]国家管网集团储能技术公司,上海200011 [2]同济大学海洋与地球科学学院,上海200092 [3]同济大学海洋地质全国重点实验室,上海200092 [4]中国科学技术大学地球和空间科学学院,安徽合肥230026 [5]长江大学地球科学学院,湖北武汉430199

出　　处：《测井技术》2025年第1期47-56,87,共11页Well Logging Technology

基　　金：国家自然科学基金项目“长宁页岩气开发区地震活动性实时监测与机理研究”(U2139204);上海市“科技创新行动计划”启明星项目“超深层油气资源开采与二氧化碳协同封存多场多相耦合机制研究”(24QA2709700);中国石油科技创新基金项目“多模式压裂裂缝竞争扩展机理与多相态耦合效应研究”(2021DQ02-0501)。

摘　　要：地质断层作为储层油气聚集和运移的重要通道,是评价储层特征和圈闭性的重要指标,也是储气库构造样式选择的先决条件。然而,从地震图像资料中识别断层存在依赖专家知识、时效性差和多解性强等问题。近年来,以深度学习和大模型技术为代表的人工智能方法,凭借其高效的非线性数据分析能力极大地改变了传统工业任务范式。鉴于此,提出一种基于增强域数据微调Yolo(You Only Look Once)模型的断层智能识别方法。首先,针对现场数据稀疏问题,使用基于强化学习的图像自增强算法,通过下游任务需求定向训练优化算法,实现地震体图像最优增强组合方案;然后,根据地质领域专家知识,在三维地震图像中确定能有效表征断块的高阶特征;通过进一步搭建基于预训练Yolo模型的断层识别模型,输入实测-增强图像数据进行领域数据微调训练,从而建立断层智能识别模型;最后,将现场三维地震数据输入到训练好的断层智能识别模型中,提取被分割、识别、标注和计算的断层特征。以中国中部地区某储气库建设运营地块为例,该方法能在不过多依赖人工介入的情况下高效识别储层断层。本研究适用于地震勘探断层识别任务,能为储气库合理选址提供智能化解决方案。As an essential channel for the accumulation and migration of oil and gas in reservoirs,geological faults are significant indicators for evaluating reservoir characteristics and trap closure,and they are also a prerequisite for the selection of structural styles in gas storage reservoirs.Nevertheless,identifying faults from seismic image data heavily relies on expert knowledge,suffers from poor timeliness,and has strong ambiguity.In recent years,artificial intelligence methods represented by deep learning and large-scale model technologies have profoundly transformed the paradigms of traditional industrial tasks with their highly efficient nonlinear data analysis capabilities.Based on this,this paper proposes an intelligent fault identification method based on the enhanced domain data fine-tuning of the Yolo model.Firstly,to address the sparse onsite data,an image self-enhancement algorithm based on reinforcement learning is employed.Through the downstream task requirements-directed training and optimization algorithm,the optimal enhancement combination scheme of seismic volume images is achieved.Then,in accordance with the expert knowledge in the geological domain,the high-order features that can effectively represent fault blocks are determined in three-dimensional seismic images.By further establishing a fault identification model based on the pre-trained Yolo model and inputting the measured and enhanced image data for domain data fine-tuning training,the intelligent fault identification model is established.Finally,the on-site three-dimensional seismic data is input into the trained intelligent fault identification model to extract the fault features that have been segmented,identified,labeled,and calculated.This technique can effectively detect formation faults without heavily depending on human interaction,as demonstrated by the building and operation plot of a gas storage facility in central China.This study can offer clever solutions for the sensible placement of gas storage facilities and is relevant to

关 键 词：三维地震勘探 断层识别 深度学习 Yolo模型 储气库 

分 类 号：P631.84[天文地球—地质矿产勘探]