基于机器学习的水平井地质导向技术研究及应用  

AI-Based Horizontal Well Geological Steering Technology and Applications

作　　者：王虹雅 龚斌 孙雄伟 孙雪冬 赵凤坤 牟宣 杨京华 方军龄 WANG Hongya;GONG Bin;SUN Xiongwei;SUN Xuedong;Zhao Fengkun;MOU Xuan;YANG Jinghua;FANG Junling(PetroChina Coalbed Methane Company Limited,Beijing 100028,China;Zhonglian Coalbed Methane National Engineering Research Center Company Limited,Beijing 100095,China;School of Earth Resources China University of Geosciences(Wuhan),Wuhan,Hubei 430074,China;Tracy Energy Technologies Company Limited,Hangzhou,Zhejiang 310000,China)

机构地区：[1]中国石油煤层气有限责任公司 [2]中联煤层气国家工程研究中心有限责任公司 [3]中国地质大学(武汉)资源学院 [4]特雷西能源科技股份有限公司

出　　处：《钻采工艺》2025年第1期238-246,共9页Drilling & Production Technology

基　　金：国家重点研发计划项目“地质资源精准开发风险预测的大数据智能分析技术及平台建设”(编号:2022YFF0801202)。

摘　　要：非常规油气藏具有储层非均质性强、甜点薄等特征,水平井开发模式不仅扩大了泄油气面积,而且增加了单井储层动用程度。传统地质导向技术主要凭借专家经验对井下采集数据进行综合判断、指导钻进,整套流程表现出判定标准不统一、耗费时间长、主观性强的痛点,因此,文章基于机器学习算法建立了地质分层智能预测模型和水平井段轨迹状态预测模型,两种模型预测准确度分别达到98.7%和92.33%。通过采集地下工程参数和地质参数信息数据,训练得到钻井过程中垂直井段地质分层预测模型和水平井段井眼轨迹预测概率模型,从数据层面降低了现场对人工经验的依赖,实现了复杂地层的智能识别分类,量化了井眼轨迹与目的层位的位置关系,大幅提升钻井效率和储层钻遇率,显著提高复杂油气藏单井产量和采收率。该地质导向技术在鄂尔多斯东缘深部煤层气大宁-吉县区块进行了试点应用,结果显示,系统显著提升了气藏研究的效率与工程决策的有效性,为全面了解和掌握气藏资源潜力、全力推进智能化在科研生产中的深度应用提供了有力的支撑。Unconventional oil and gas reservoirs are characterized by strong reservoir heterogeneity and thin sweet spot zones.Horizontal wells can effectively increase the length of wellbore trajectory within the reservoir,not only enlarging the drainage area but also facilitating subsequent reservoir modifications,thereby enhancing the degree of efficiency of single-well reservoir.Traditional geosteering technology relies on expert experience to make comprehensive judgment on downhole collection data and guide drilling,and the whole process shows the problems of inconsistent judgment standards,long time consumption and strong subjectivity.Based on machine learning algorithms,a intelligent prediction model for geological stratification and a trajectory prediction model for horizontal well are established in this paper,and the prediction accuracy of the two models reaches 98.7%and 92.33%,respectively.By collecting dwonhole drilling and geological parameter information and data,the geological stratification prediction model of vertical well and the wellbore trajectory prediction model of horizontal well are trained during the drilling process,which reduces the dependence on manual experience in the field from the data level,realizes the intelligent identification and classification of complex formations,quantifies the position relationship between the wellbore trajectory and the target formation,greatly improves the drilling efficiency and reservoir-encountered rate,and significantly improves the single well production and recovery rate of complex oil and gas reservoirs.This geosteering technology platform has been piloted in Da Ning JiXian County block of the deep coalbed methane reservoirs on the eastern edge of the Ordos Basin.The application results have shown that the system significantly improves the efficiency of gas reservoir research and the effectiveness of decision-making,providing robust support for a comprehensive understanding and mastering the potential of gas reservoir resources and the full effort to promote

关 键 词：地质导向技术 智能算法 储层钻遇率 地质分层 井眼轨迹 

分 类 号：TE3[石油与天然气工程—油气田开发工程]