钻采工艺 ›› 2021, Vol. 44 ›› Issue (3): 5-9.DOI: 10.3969/J. ISSN.1006-768X.2021.03.02

• 钻井工艺 • 上一篇    下一篇

基于K-Means聚类算法的沉砂卡钻预测方法研究

苏晓眉1,张涛1 ,2,李玉飞3,卿玉3,李玉梅1 ,2   

  1. 1北京信息科技大学信息与通信工程学院  2北京信息科技大学“高动态导航技术北京市重点实验室”  3中国石油川庆钻探工程有限公司钻采工程技术研究院
  • 出版日期:2021-05-25 发布日期:2021-05-25
  • 作者简介:苏晓眉(1995-),女,硕士,主要研究方向为智慧感知与信息处理。地址:(100101)北京市朝阳区北四环中路35号北京信息科技大学,电话:18201112416,Email:182011124616@ 163 . com
  • 基金资助:
    国家自然科学基金面上项目“基于模型预测控制理论与状态机架构的控压钻井压力控制方法研究”(编号:51374223);北京市属高校高水平教师队伍建设支持计划———青年拔尖人才培育计划(编号:CIT&TCD201804057);北京市教委一般项目“井下工程参数测量与数据分析方法研究”(编号:KM201811232011);北京信息科技大学师资补充与支持计划项目“干热岩基础岩石力学参数测试研究”(编号:5112011131);中国石油集团公司重大科技项目“油气井溢流预警监测与井喷预防控制技术研究与应用”(编号:2019D-4618)。

Research on the Sticking Prediction Method Based on K-Means Clustering Algorithm

SU Xiaomei1, ZHANG Tao1,2, LI Yufei3, QIN Yu3, LI Yumei1,2   

  1. 1. School of Information and Communication Engineering, Beijing Information Science and Technology University, Beijing100101, China; 2. Beijing Key Laboratory of High Dynamic Navigation Technology, University of Beijing Information Science & Technology, Beijing 100088, China; 3 . CCDC Drilling & Production Engineering Technology Research Institute, Guanghan, Sichuan 638100, China
  • Online:2021-05-25 Published:2021-05-25

摘要: 井眼不清洁造成卡钻是常见的钻井复杂工况之一,在卡钻发生前,部分井下工程参数特征会表现出异常。为了掌握卡钻前的工程参数特征变化规律,及时调整钻井参数,可避免卡钻发生。提出了一种基于主成分分析(PCA)算法和 K-Means聚类算法的卡钻工况预测模型。该模型利用 PCA算法对冀东油田某井卡钻工况前的井下近钻头实测工程参数进行降维处理,再利用 K-Means聚类模型对降维后的数据进行训练测试,最后利用 Calinski-Harabaz指数对聚类质量进行评估。结果表明,正常工况数据和卡钻前工况数据的聚类中心相距较远,两类数据得到有效分类,且 Calinski-Harabaz指数值高,表明聚类质量高,K-Means聚类算法能够有效地分析近钻头工程参数测量数据,根据聚类分析结果可及时对卡钻工况进行预警,减少卡钻工况的发生。

关键词: 卡钻, 钻井复杂工况, PCA, K-Means, 近钻头工程参数

Abstract:

Sticking is one of the common complex drilling conditions. Before the occurrence of sticking, some downhole engineering parameters will show abnormal characteristics. In order to master the variation rule of engineering parameters before sticking and adjust drilling parameters in time, the sticking can be avoided. This paper presents a prediction model of stuck drilling conditions based on PCA and K-Means clustering algorithm. In this model, PCA is used to reduce the dimension of the measured engineering parameters of downhole near the bit before the stuck condition in Jidong Oilfield, and K-Means clustering model is used to train and test the reduced dimension data. Finally, Calinski-Harabaz index is used to evaluate the clustering quality. The results show that the clustering centers of normal condition data and the working condition data before sticking are far from each other, the two types of data are effectively classified, and the Calinski-Harabaz index value is high, indicating the clustering quality is high. K-Means clustering algorithm can effectively analyze the measurement data of near bit engineering parameters. According to the results of clustering analysis, it can give early warning to the stuck condition in time and reduce the stuck occurrence.

Key words: sticking, complex drilling condition, PCA, K-Means, near bit engineering parameters