Scholarlink

<p>Mineralogy plays a critical role in understanding subsurface formations and enhancing wellbore stability. This paper introduces a novel prediction method for mineralogical composition that builds upon prior integrated approaches by combining high-fidelity core data, rock cuttings, and detailed drilling and fluid parameters in a support vector machine (SVM) learning model. The model was developed using 1000 data points collected from 19 wells drilled with KCl-PHPA-polyol mud in the Cambay Basin, India. In total, 71 mineral composition classes across 10 lithologies were defined, and the model achieved an impressive prediction accuracy of 90.2%. In contrast with existing deep neural network (DNN)-based approaches that rely on indirect data sources such as logs or generic core summaries, this paper introduces a SVM framework trained on categorical groupings of mineral assemblages derived from similarity in modal mineral composition, rather than individual minerals or quantitative weight fractions. These groupings were established using spectroscopically validated mineralogical compositions obtained through scanning electron microscopy with energy-dispersive X-ray spectroscopy and X-ray diffraction. A comparative analysis with the random forest algorithm demonstrated the superior performance of the SVM model. The complex interplay between lithology, mineralogy, depth, and rate of penetration was explored, significantly enhancing the understanding of drilling dynamics. Such comprehensive data integration aids in understanding rock–fluid interactions, enabling more informed drilling decisions, and optimizing operational efficiency. This significant enhancement in prediction capability paves the way for more sustainable and cost-efficient resource extraction in the petroleum industry. While the methodology was developed for the Cambay Basin, its adaptability to geologically distinct regions warrants further investigation to assess broader applicability.</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 Complex Mineralogy in Oil and Gas Drilling: A Support Vector Machine Learning Model Using Integrated Core and Operational Data

  • 期刊论文

  • 发表日期：

    2025年12月14日

查看全文
  • Raunak Gupta,
  • Uttam K. Bhui

摘要

Mineralogy plays a critical role in understanding subsurface formations and enhancing wellbore stability. This paper introduces a novel prediction method for mineralogical composition that builds upon prior integrated approaches by combining high-fidelity core data, rock cuttings, and detailed drilling and fluid parameters in a support vector machine (SVM) learning model. The model was developed using 1000 data points collected from 19 wells drilled with KCl-PHPA-polyol mud in the Cambay Basin, India. In total, 71 mineral composition classes across 10 lithologies were defined, and the model achieved an impressive prediction accuracy of 90.2%. In contrast with existing deep neural network (DNN)-based approaches that rely on indirect data sources such as logs or generic core summaries, this paper introduces a SVM framework trained on categorical groupings of mineral assemblages derived from similarity in modal mineral composition, rather than individual minerals or quantitative weight fractions. These groupings were established using spectroscopically validated mineralogical compositions obtained through scanning electron microscopy with energy-dispersive X-ray spectroscopy and X-ray diffraction. A comparative analysis with the random forest algorithm demonstrated the superior performance of the SVM model. The complex interplay between lithology, mineralogy, depth, and rate of penetration was explored, significantly enhancing the understanding of drilling dynamics. Such comprehensive data integration aids in understanding rock–fluid interactions, enabling more informed drilling decisions, and optimizing operational efficiency. This significant enhancement in prediction capability paves the way for more sustainable and cost-efficient resource extraction in the petroleum industry. While the methodology was developed for the Cambay Basin, its adaptability to geologically distinct regions warrants further investigation to assess broader applicability.