新疆石油天然气 ›› 2026, Vol. 22 ›› Issue (1): 26-32.DOI: 10.12388/j.issn.1673-2677.2026.01.003

• 油气勘探 • 上一篇    下一篇

卡钻风险无监督诊断与处置措施检索推荐系统

吕泽昊1,李桢2,魏风奇1,纪国栋3,李令东4,陈伟峰3   

  1. 1.中国石油油气和新能源分公司,北京东城   100007;2.中国石油大学(北京),北京昌平  102249;3.中国石油工程技术研究院,北京昌平  102249;4.中国石油勘探开发研究院,北京海淀  100083
  • 收稿日期:2025-11-18 修回日期:2026-01-10 接受日期:2026-01-13 出版日期:2026-03-10 发布日期:2026-03-10
  • 通讯作者: 李桢(2003-),中国石油大学(北京)软件工程专业,在读硕士,目前从事智能钻完井技术研究。(E-mail)13618488509@163.com
  • 作者简介:吕泽昊(1993-),2019年毕业于中国石油大学(北京)油气井工程,博士,高级工程师,现从事钻完井工程技术研究与推广。(E-mail)1003154924@qq.com
  • 基金资助:
    中国石油天然气集团有限公司科研项目“深地油气顶部驱动钻井装置关键零部件项目”(TC240HAJ8-173)。

Unsupervised Diagnosis and Response Recommendation System for Stuck Pipe Risks

LV Zehao1,LI Zhen2,WEI Fengqi1,JI Guodong3,LI Lingdong4,CHEN Weifeng³   

  1. 1.CNPC Oil,Gas & New Energies Company,Dongcheng District 100007,Beijing,China;2.China University of Petroleum (Beijing),Changping District 102249,Beijing,China;3.CNPC Engineering Technology R&D Company Limited,Changping District 102249,Beijing,China;4.PetroChina Research Institute of Petroleum Exploration & Development,Haidian District 100083,Beijing,China
  • Received:2025-11-18 Revised:2026-01-10 Accepted:2026-01-13 Online:2026-03-10 Published:2026-03-10

摘要: 卡钻是钻井过程中常见的风险,严重制约安全高效钻井。传统的阻卡诊断与卡钻处置方法大多基于专家经验、机理模型或监督机器学习算法,存在诊断滞后、过拟合虚警高、主观性强和适配性低等问题。基于无监督学习与K近邻算法,提出了卡钻风险诊断与处置案例推荐方法,并形成软件系统。该系统分为卡钻风险诊断与卡钻事故处理两大核心模块。卡钻风险诊断模块采用孤立森林算法(Isolation Forest)对井下实时数据中的核心工程参数的异常变化趋势进行监控与诊断;卡钻事故处理模块则采用基于KD-Tree+KNN的案例推理算法检索卡钻案例决策库,将相似度最高的前三个案例数据返回到客户端辅助工程师处理卡钻事故。基于上述研究,设计并实现了基于案例推理的井下卡钻智能诊断系统。测试表明,本系统在事故前5 min左右成功发出预警,并快速检索出相似度最高的3个历史处置案例,并推送到界面,为工程师提供了直接的决策支持。

关键词: 卡钻预测, 案例推理, 孤立森林, KD树

Abstract: Stuck pipe,a common risk during drilling operations,severely restricts safe and efficient drilling. Traditional methods for diagnosing and handling stuck pipe incidents mostly rely on expert experience,mechanistic models,or supervised machine learning algorithms and suffer from issues such as diagnostic lag,high false alarms due to overfitting,strong subjectivity,and low adaptability. This paper proposes a method for diagnosing stuck pipe risks and recommending cases of countermeasures based on unsupervised learning and K-Nearest Neighbors (KNN),which has been implemented as a software system. The system comprises two core modules:stuck pipe risk diagnosis and stuck pipe incident handling. The stuck pipe risk diagnosis module employs the Isolation Forest algorithm to monitor and diagnose abnormal trends in key real-time downhole engineering parameters. The stuck pipe incident handling module utilizes a Case-Based Reasoning (CBR) algorithm based on KD-Tree and KNN to retrieve cases from a stuck pipe decision case library,returning the data of three most similar cases to the client-side to assist engineers in handling stuck pipe incidents. Based on the aforementioned,an intelligent downhole sticking diagnosis system based on case-based reasoning was designed and implemented. Tests show that the system successfully issues warnings approximately 5 minutes before an incident occurs,quickly retrieves three most similar historical handling cases,and pushes them to the interface. This developed system provides direct decision support for engineers.

Key words: stuck pipe prediction, case-based reasoning (CBR), isolation forest, KD-Tree

中图分类号: