新疆石油天然气 ›› 2026, Vol. 22 ›› Issue (3): 28-36.DOI: 10.12388/j.issn.1673-2677.2026.03.003

• 油气开发 • 上一篇    下一篇

嵌入物理模型的钻进工况油气侵表征方法

张彦龙, 李根生, 周赫, 尚有志, 尹志勇, 季勇, 孙正心   

  1. 1. 中国石油集团工程技术研究院有限公司,北京  102206;

    2. 油气资源与工程全国重点实验室,北京  102249;

    3. 中国石油大学(北京),北京  102249;

    4. 中国石油集团油田技术服务有限公司,北京东城 100007

  • 收稿日期:2025-05-28 修回日期:2026-07-19 接受日期:2026-09-01 出版日期:2026-09-15 发布日期:2026-09-15
  • 通讯作者: 李根生(1961—),1998年毕业于中国石油大学(北京)石油工程系油气钻井工程专业,博士,教授,中国工程院院士,长期从事水射流钻井与完井、智能钻完井与地热开采、复杂油气钻井提速等领域的研究工作。(E-mail)ligs@cup.edu.cn
  • 作者简介:张彦龙(1986—),2008年毕业于中国石油大学(北京)石油工程专业,在读博士,高级工程师,目前从事钻完井数字化、智能化等方面研究与推广。(E-mail)zhangyanlong01@cnpc.com.cn
  • 基金资助:

    1.国家科技重大专项“万米深井智能钻机与超高压井口装备”(2024ZD14018);

    2.中国石油集团公司科技专项“钻井安全控制软件”(2024ZZ46-05);

    3.中国石油集团公司科技专项“钻井工序优化数字孪生系统研发”(2025ZG06-02)

Physics-Embedded Characterization Method for Hydrocarbon Influx During Drilling

Zhang Yanlong,  Li Gensheng, Zhou He , Shang Youzhi , Yin Zhiyong , Ji Yong , Sun Zhengxin   

  1. 1. CNPC Engineering Technology R&D Company Limited,Beijing 102206,China;

    2. State Key Laboratory of Petroleum Resources and Engineering,Beijing 102249,China;

    3. China University of Petroleum (Beijing),Beijing 102249,China;

    4. China Petroleum Technical Service Corporation Limited,Beijing 100007,China

  • Received:2025-05-28 Revised:2026-07-19 Accepted:2026-09-01 Online:2026-09-15 Published:2026-09-15

摘要:

深层、超深层及非常规油气钻井中,异常压力导致的地层-井筒流体侵入频繁发生,每年造成超过2 000 d钻机日非生产时间,亟需高精度溢流早期识别技术。基于“钻遇异常压力层-油气侵入井筒-油气返出地面”三阶段机理,构建机理-数据混合驱动的三段式溢流识别模型;采用皮尔逊相关系数筛选敏感参数,并以随机森林算法识别异常压力层;引入井筒多相流物理模型约束工程参数,用长短期记忆网络(LSTM)预测气体在井筒内运移位置;基于出口流量与总池体积构建优化的分布式梯度提升库(XGBoost)模型,实现地面返出实时识别;依据293起溢流事故数据,建立分井型迟到时间特征体系,显著提升复杂井段提前识别能力。同时,通过全过程三段式框架、物理约束提升模型一致性及分井型特征增强泛化性,为溢流智能监测和早期预警提供了工程化可行路径。

关键词:

溢流早期识别, 嵌入物理模型, 智能报警, 钻井事故复杂, DeepSeek

Abstract:

In drilling of deep,ultra-deep,and unconventional oil and gas reservoirs,formation pressure anomaly leads to frequent formation-wellbore fluid influxes,resulting in annual non-productive time of over 2,000 rig-days. This highlights the urgent need for high-precision early kick detection methods. This study develops a mechanistic-data hybrid,three-stage kick identification framework based on the processes of “encountering abnormal pressure zones-fluid influx into the wellbore-surface return of oil and gas.” Sensitive parameters are selected using Pearson correlation coefficients and employed in a random forest model to identify abnormal pressure zones;multiphase flow physical models are used to constrain wellbore engineering parameters,and a long short-term memory (LSTM) network is used to predict the gas migration within the wellbore;surface return is monitored using an XGBoost model based on flow rate and total pit volume. A lag-time feature system is established from 293 kick events to improve early detection in complex wellbore intervals. The three-stage identification framework and the incorporation of physical constraints enhance model consistency,and the lag-time feature strategy improves generalizability. The proposed approach provides a practical and reliable solution for intelligent kick monitoring and early warning in drilling operations.

Key words:

early identification of overflow, PENN, smart alarm, accident and complexity during drilling, DeepSeek

中图分类号: