Nour Muhammad, Elsayed Said K, Mahmoud Omar
Department of Petroleum Engineering, Faculty of Petroleum and Mining Engineering, Suez University, Suez, 11252, Egypt.
Department of Petroleum Engineering, Faculty of Engineering and Technology, Future University in Egypt (FUE), Cairo, 11835, Egypt.
Sci Rep. 2024 Nov 4;14(1):26624. doi: 10.1038/s41598-024-76910-z.
With the increased directional drilling activities in the oil and gas industry, combined with the digital revolution amongst all industry aspects, the need became high to optimize all planning and operational drilling activities. One important step in planning a directional well is to select a directional tool that can deliver the well in a cost-effective manner. Rotary steerable systems (RSS) and positive displacement mud motors (PDM) are the two widely used tools, each with distinct advantages: RSS excels in hole cleaning, sticking avoidance and hole quality in general, while PDM offers versatility and lower operating costs. This paper presents a series of machine learning (ML) models to automate the selection of the optimal directional tool based on offset well data. By processing lithology, directional, drilling performance, tripping and casing running data, the model predicts section time and cost for upcoming wells. Historical data from offset wells were split into training and testing sets and different ML algorithms were tested to choose the most accurate one. The XGBoost algorithm provided the most accurate predictions during testing, outperforming other algorithms. The beauty of the model is that it successfully accounted for variations in formation thicknesses and drilling environment and adjusts tool recommendations accordingly. Results show that no universal rule favors either RSS or PDM; rather, tool selection is highly dependent on well-specific factors. This data-driven approach reduces human bias, enhances decision-making, and could significantly lower field development costs, particularly in aggressive drilling campaigns.
随着石油和天然气行业定向钻井活动的增加，再加上所有行业领域的数字革命，对优化所有钻井规划和作业活动的需求变得很高。规划定向井的一个重要步骤是选择一种能够以经济高效的方式完成钻井的定向工具。旋转导向系统（RSS）和容积式泥浆马达（PDM）是两种广泛使用的工具，各有明显优势：RSS在井眼清洁、防卡和总体井眼质量方面表现出色，而PDM则具有通用性和较低的运营成本。本文提出了一系列机器学习（ML）模型，用于根据邻井数据自动选择最佳定向工具。通过处理岩性、定向、钻井性能、起下钻和下套管数据，该模型预测即将施工井段的时间和成本。将邻井的历史数据分为训练集和测试集，并测试了不同的ML算法以选择最准确的算法。在测试期间，XGBoost算法提供了最准确的预测，优于其他算法。该模型的优点在于它成功地考虑了地层厚度和钻井环境的变化，并相应地调整工具推荐。结果表明，没有普遍规则表明RSS或PDM更具优势；相反，工具选择高度依赖于具体井的因素。这种数据驱动的方法减少了人为偏差，增强了决策制定，并可以显著降低油田开发成本，特别是在激进的钻井作业中。
