Xinjiang Oil & Gas ›› 2022, Vol. 18 ›› Issue (1): 14-20.DOI: 10.12388/j.issn.1673-2677.2022.01.002

• OIL AND GAS EXPLORATION • Previous Articles     Next Articles

Intelligent Prediction for Rate of Penetration Based on Support Vector Machine Regression

SONG Xianzhi1,2PEI Zhijun1WANG Pantao1ZHANG Gonglingyan1YE Shanlin1#br#   

  1. 1. Petroleum Engineering College,China University of Petroleum (Beijing),Beijing 102249,China; 2. State Key Laboratory of Oil & Gas Resources and Prospecting,Beijing 102249,China;

  • Online:2022-03-25 Published:2022-03-25

基于支持向量机回归的机械钻速智能预测

宋先知1,2裴志君1王潘涛1张宫凌燕1叶山林1
  

  1. 1.中国石油大学(北京)石油工程学院,北京 102249; 2.油气资源与探测国家重点实验室,北京 102249;
  • 作者简介:宋先知(1982-),2010年毕业于中国石油大学(北京)油气井工程,博士学t位,教授、博士生导师,主要从事智能钻完井工作。(Tel)15210242339,(E-mail)songxz@cup.edu.cn
  • 基金资助:
    国家重点研发计划(变革型技术关键科学问题)项目复杂油气智能钻井理论与方法2019YFA0708300)、中石油战略合作科技专项项目物探、测井、钻完井人工智能理论与应用场景关键技术研究ZLZX2020-03-03)联合资助

Abstract: Tarim Basin is rich in deep oil and gas resources. With the increase in well depth,however,formation drillability decreases
and abrasiveness increases,resulting in low rate of penetration (ROP) and high drilling cost. Therefore,there is an urgent need for drilling optimization technology. ROP prediction is a key technology to optimize drilling. Accurate ROP prediction provides important basis for drilling parameter optimization and drilling tool selection. This study establishes several intelligent prediction models based on decision tree regression algorithm,random forest regression algorithm,support vector machine (SVM) regression algorithm and deep neural network respectively,by using the mud logging data acquired in real time from the drilling field,which are then compared and analyzed in aspect of root-mean-square error,R square,maximum error and relative error,so as to select the optimum prediction model for ROP prediction. The results show that SVM prediction model is superior to other models regarding prediction accuracy and stability.
Key words:rate of penetration;machine learning;support vector machine;neural network

Key words: rate of penetration, machine learning, support vector machine, neural network

摘要: 塔里木盆地深层油气资源丰富,但随着井深的增加地层可钻性降低、研磨性升高,导致机械钻速低,钻井成本高等难题,亟需钻井优化技术。机械钻速预测是优化钻井的关键技术之一,准确的机械钻速预测可以为钻井参数优化、钻井工具优选等提供重要依据。利用钻井现场可实时获得的录井数据,基于决策树回归算法、随机森林回归算法、支持向量机回归算法和深度神经网络分别建立了机械钻速智能预测模型。从均方根误差、R平方、最大误差和相对误差四个方面进行对比分析,从而优选最优的机械钻速智能预测模型。结果表明支持向量机回归模型的预测精度、稳定性均优于其他模型。

关键词: 机械钻速, 机器学习, 支持向量机, 神经网络

CLC Number: