机器学习在钻柱振动识别与预测中的研究进展  

Research progress of machine learning in drill string vibration recognition and prediction

作　　者：汪海阁 高博 郑有成[3] 赵飞[1] 崔猛 丁燕 邢世旺 WANG Haige;GAO Bo;ZHENG Youcheng;ZHAO Fei;CUI Meng;DING Yan;XING Shiwang(CNPC Engineering Technology R&D Co.,Ltd.,Beijing 102206,China;China University of Petroleum-Beijing,Beijing 102249,China;PetroChina Southwest Oil&Gasfield Company,Chengdu,Sichuan 610051,China)

机构地区：[1]中国石油工程技术研究院有限公司 [2]中国石油大学(北京) [3]中国石油西南油气田公司

出　　处：《天然气工业》2024年第1期149-158,共10页Natural Gas Industry

基　　金：国家重点研发计划课题“复杂地层智能化破岩机理与导向控制方法”(编号:2019YFA0708302);中国石油天然气集团有限公司科技项目“智能钻完井控制理论与关键模型研究”(编号:2023ZZ06)。

摘　　要：钻柱振动是影响钻井效率、钻柱失效、井眼稳定和钻井安全的主要因素,复杂振动的早期识别对于缓解井下工具受损、提高生产时间至关重要。为此,充分调研国内外机器学习在钻柱振动识别方法方面的研究成果,从数据获取角度对钻柱振动识别与预测方法进行了全面分析,对比研究了各算法模型的框架、特征参数和测试效果,系统评估了各算法模型的优缺点,并对未来振动识别与预测的发展方向提出思考。研究结果表明:①机器学习算法可以从大量振动数据中学习和提取特征来建立模型,对振动进行分类和预测,通过不断优化算法和模型,提高钻柱振动识别与预测的准确性和可靠性;②随着数据采集和处理技术的不断进步,地面与井下多源数据融合方法将多种数据共同分析处理,可以最大程度地发掘地面和井下数据特征,有望成为解决井下问题的重要途径;③随着钻井工程与人工智能技术的不断融合与发展,振动缓解与钻井提速联合优化,将为钻井工程提供更为可靠的指导和决策。结论认为,机器学习在钻柱振动识别与预测方面的应用和发展进一步缓解了超深井井下钻柱振动这一复杂问题,提高了钻井工程的效率和安全性,推进了钻井过程的高效化和智能化发展步伐。Drill string vibration is a major factor influencing drilling efficiency,drill string failure,borehole stability and drilling safety.Early recognition of complex vibration is crucial to mitigating downhole tool damage and improving production time.In this paper,domestic and foreign application achievements of machine learning in drill string vibration recognition are fully investigated,and drill string vibration recognition and prediction methods are analyzed comprehensively from the perspective of data acquisition.Then,different algorithm models are comparatively studied in the aspects of framework,characteristic parameters and test effects,and their advantages and disadvantages are evaluated systematically.Finally,some ideas on the future development direction of drill string vibration recognition and prediction are put forward.And the following research results are obtained.First,by learning from abundant vibration data and extracting the characteristics,machine learning algorithm can establish a model for vibration classification and prediction.The accuracy and reliability of drill string vibration recognition and prediction can be improved by optimizing the algorithm and the model continuously.Second,with the continuous development of data acquisition and processing technologies,the surface and downhole multi-source data fusion method can analyze and process multiple data simultaneously,so it can identify surface and downhole data characteristics as much as possible,and it is predicted to be an important solution to downhole problems.Third,the continuous integration and development of drilling engineering and AI technology promotes vibration alleviation and ROP improvement jointly and provides reliable guidance and decision for drilling engineering.In conclusion,the application and development of machine learning in drill string vibration recognition and prediction further alleviates downhole drill string vibration in ultra-deep wells,improves the efficiency and safety of drilling engineering,and promotes th

关 键 词：钻柱振动 深层 油气钻井 机器学习 分类算法 识别 预测 

分 类 号：TE92[石油与天然气工程—石油机械设备]