Scholarlink

<p>Rock drillability has a direct impact on drilling efficiency, project costs, and the selection of drill bits. Conventional methods for estimating drillability depend heavily on laboratory core experiments and empirical formulas, which suffer from difficult data acquisition, high costs, and poor timeliness. This paper investigates the Xihu Sag of the East China Sea and proposes an interpretable, intelligent drillability prediction framework that integrates six machine learning algorithms and twelve diverse input variables. We conducted Spearman correlation analysis to identify twelve key features drawn from logging, drilling, and rock‑mechanical datasets. Model performance was evaluated using six metrics—such as MSE and R²—and compared across algorithms. We then applied the Rank Sum Ratio (RSR) method for overall assessment, identifying CatBoost (RSR = 0.694) as the top performer for practical forecasting. Finally, we explored the global interpretability of the chosen model through its decision‑tree structure, quantifying each input variable’s contribution and elucidating its influence mechanisms. Compared to traditional approaches, our method delivers substantial gains in accuracy, applicability, and efficiency. Crucially, it transcends the “black‑box” limitation of conventional machine‑learning models, enhancing the transparency and reliability of drillability predictions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predicting offshore downhole rock drillability in complex geological conditions using explainable machine learning and Multi-Source data

  • 期刊论文

  • 发表日期：

    2025年9月25日

查看全文
  • Huayang Li,
  • Quanyou Liu,
  • Jiaao Chen,
  • Chi Zhang,
  • Dantong Liu,
  • Zehui Shi,
  • Xinyi Wang

摘要

Rock drillability has a direct impact on drilling efficiency, project costs, and the selection of drill bits. Conventional methods for estimating drillability depend heavily on laboratory core experiments and empirical formulas, which suffer from difficult data acquisition, high costs, and poor timeliness. This paper investigates the Xihu Sag of the East China Sea and proposes an interpretable, intelligent drillability prediction framework that integrates six machine learning algorithms and twelve diverse input variables. We conducted Spearman correlation analysis to identify twelve key features drawn from logging, drilling, and rock‑mechanical datasets. Model performance was evaluated using six metrics—such as MSE and R²—and compared across algorithms. We then applied the Rank Sum Ratio (RSR) method for overall assessment, identifying CatBoost (RSR = 0.694) as the top performer for practical forecasting. Finally, we explored the global interpretability of the chosen model through its decision‑tree structure, quantifying each input variable’s contribution and elucidating its influence mechanisms. Compared to traditional approaches, our method delivers substantial gains in accuracy, applicability, and efficiency. Crucially, it transcends the “black‑box” limitation of conventional machine‑learning models, enhancing the transparency and reliability of drillability predictions.