钻井模型与机器学习耦合的实时卡钻预警技术  

Real-time Pipe Sticking Early Warning Technology Based on Coupling of Drilling Model and Machine Learning

作　　者：李紫璇 张菲菲 祝钰明 吴墨染 刘泳敬 于琛 Li Zixuan;Zhang Feifei;Zhu Yuming;Wu Moran;Liu Yongjing;Yu Chen(School of Petroleum Engineering,Yangtze University;Leak Resistance&Sealing Technology Research Room of National Engineering Laboratory of Petroleum Drilling Technology;Hubei Provincial Key Laboratory of Oil and Gas Drilling and Production Engineering;PetroChina Qinghai Oilfield Company;No.1 Drilling Engineering Company,PetroChina Bohai Drilling Engineering Co.,Ltd.;No.3 Drilling Engineering Company,PetroChina Bohai Drilling Engineering Co.,Ltd.;Engineering Technology Research Institute of CNPC Bohai Drilling Engineering Co.,Ltd.)

机构地区：[1]长江大学石油工程学院 [2]油气钻井技术国家工程实验室防漏堵漏研究室 [3]油气钻采工程湖北省重点实验室 [4]青海油田分公司 [5]渤海钻探第一钻井分公司 [6]渤海钻探第三钻井分公司 [7]渤海钻探工程技术研究院

出　　处：《石油机械》2022年第4期15-21,93,共8页China Petroleum Machinery

基　　金：国家自然科学基金项目“大位移井钻井过程中动态岩屑运移与钻柱受力耦合机理研究”(51874045);湖北省自然科学基金杰出青年基金项目“页岩气大位移井动态井眼清洁机理及智能监测算法研究”(2019CFA093)。

摘　　要：在现场钻井过程中,由于未考虑井眼清洁状况,导致基于钻井模型的卡钻分析模型的应用受限,只能达到卡钻后识别,并不能实现真正意义上的卡钻预警。为此,基于瞬态岩屑运移模型和改进的摩阻扭矩模型,结合邻井历史录井数据,使用贝叶斯优化算法对钻井模型进行训练,使钻井模型更适应当前区块,提高卡钻事故预测结果。采用钻井模型与机器学习相耦合的方法,提出了基于录井数据的实时卡钻预警技术。通过实时录井数据的输入,采用训练后的模型对卡钻风险参数进行实时监测。实例分析结果表明,该模型能够捕捉到卡钻是否发生,同时能够对要发生的卡钻进行分类,并在事故发生前给出预警信号。该方法将井眼清洁加入到卡钻风险预测,可为钻井作业提供更为全面的决策支持,帮助工程技术人员及时采取有效措施避免卡钻发生,缩短非生产时间。During drilling operation, the application of the pipe sticking analysis model based on drilling model is limited for not considering wellbore cleaning, and the model can only identify a pipe sticking that has occurred, in other words, it cannot realize an early warning of pipe stick in true sense. In this paper, based on the transient cuttings transport model and the improved torque and drag model, combined with the historical logging data of adjacent wells, the Bayesian optimization algorithm was used to train the drilling model, so as to make the drilling model more suitable for the current block and improve the prediction results of stuck pipe. Moreover, by coupling the drilling model and machine learning, a real-time pipe sticking early warning technology based on logging data was proposed. With the real-time logging data as input, the trained model was used to monitor the pipe sticking risk parameters in real time, and the pipe sticking risk was predicted depending on the development trend to ensure a warning is made once the pipe sticking risk is high. This method adds wellbore cleaning to the prediction of pipe sticking risk, so that it can provide more comprehensive decision support for drilling operation and help engineering technicians to take effective measures in time to avoid pipe sticking and reduce non-productive time.

关 键 词：实时卡钻监测 钻井模型 机器学习 贝叶斯优化 时序数据分析 

分 类 号：TE242[石油与天然气工程—油气井工程]