Scholarlink

<p>In this study, the swelling of Ranikot shale formation was analyzed using a water-based mud containing composite of MWCNTs and TiO<sub>2</sub> nanoparticles. The swelling of Ranikot shale was observed at different compaction pressure followed by modeling of swelling through different machine learning algorithms. To simulate the different overburden stresses in subsurface, four different pressures 2000, 4000, 6000 and 8000 psi were used to compact the Ranikot shale. Five models, Linear Regression, Decision Tree, Random Forest, Support Vector Regression and K-Nearest Neighbor were employed. The Random Forest Regression showed excellent performance at 2000 and 4000 psi (R<sup>2</sup> ≈ 0.999, MSE ≈ 0.000), but accuracy dropped at 8000 psi (R<sup>2</sup> = 0.926, R<sup>2</sup><sub>std</sub> = 0.105).While the Decision Tree Regression demonstrated strong predictive capability at 2000 psi (R<sup>2</sup> = 0.999), but lower accuracy and higher variability at 8000 psi (R<sup>2</sup> = 0.905, R<sup>2</sup><sub>std</sub> = 0.1317).Linear Regression performed poorly at 8000 psi (R<sup>2</sup> = 0.0274, MSE = -0.440), with the highest error and lowest accuracy across all models. Support Vector Regression (SVR)’s performance was good at 2000 psi (R<sup>2</sup> = 0.977), but less consistent at 8000 psi (R<sup>2</sup> = 0.894, R<sup>2</sup><sub>std</sub> = 0.073). Meanwhile the KNN Regression was strong at 2000 psi (R<sup>2</sup> = 0.999), but accuracy declined at 8000 psi (R<sup>2</sup> = 0.941, R<sup>2</sup><sub>std</sub> = 0.082).Overall, the Random Forest was found to be the most consistent, while Linear Regression performed the worst at higher pressures. This study studies the combined influence of MWCNTs/TiO<sub>2</sub> as a shale swelling inhibitor and uses machine learning models to model the swelling as a function of temperature and compaction pressure which has not been found in the previous literatures. This allows for a optimized inhibitor formulation and improved wellbore stability during drilling operations.</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

Analyzing machine learning algorithms in predicting Ranikot swelling at different compaction pressures in presence of carbon supported TiO2 water based mud

  • 期刊论文

  • 发表日期：

    2025年2月19日

查看全文
  • Faiq Azhar Abbasi,
  • Syed Mohammad Ali Shah,
  • Muhammad Mustafa,
  • Usman Humayun,
  • Mohsin Ayubi,
  • Zeeshan Ahmad Arfeen,
  • Shaine Mohammadali Lalji,
  • Syed Imran Ali,
  • Mujtaba Mateen

摘要

In this study, the swelling of Ranikot shale formation was analyzed using a water-based mud containing composite of MWCNTs and TiO2 nanoparticles. The swelling of Ranikot shale was observed at different compaction pressure followed by modeling of swelling through different machine learning algorithms. To simulate the different overburden stresses in subsurface, four different pressures 2000, 4000, 6000 and 8000 psi were used to compact the Ranikot shale. Five models, Linear Regression, Decision Tree, Random Forest, Support Vector Regression and K-Nearest Neighbor were employed. The Random Forest Regression showed excellent performance at 2000 and 4000 psi (R2 ≈ 0.999, MSE ≈ 0.000), but accuracy dropped at 8000 psi (R2 = 0.926, R2std = 0.105).While the Decision Tree Regression demonstrated strong predictive capability at 2000 psi (R2 = 0.999), but lower accuracy and higher variability at 8000 psi (R2 = 0.905, R2std = 0.1317).Linear Regression performed poorly at 8000 psi (R2 = 0.0274, MSE = -0.440), with the highest error and lowest accuracy across all models. Support Vector Regression (SVR)’s performance was good at 2000 psi (R2 = 0.977), but less consistent at 8000 psi (R2 = 0.894, R2std = 0.073). Meanwhile the KNN Regression was strong at 2000 psi (R2 = 0.999), but accuracy declined at 8000 psi (R2 = 0.941, R2std = 0.082).Overall, the Random Forest was found to be the most consistent, while Linear Regression performed the worst at higher pressures. This study studies the combined influence of MWCNTs/TiO2 as a shale swelling inhibitor and uses machine learning models to model the swelling as a function of temperature and compaction pressure which has not been found in the previous literatures. This allows for a optimized inhibitor formulation and improved wellbore stability during drilling operations.