Scholarlink

<p>Shear wave velocity (V<sub>S</sub>) plays a vital role in geophysical characterization, particularly within geologically complex and unconsolidated sediments. Despite its importance, direct V<sub>S</sub> logging is often limited by operational and economic constraints. In this study, Logging While Drilling (LWD) data from gas hydrate-bearing sediments of Area B in the Krishna–Godavari Basin, obtained during the National Gas Hydrate Program Expedition-02 (NGHP-02), are analysed. Out of 12 investigated wells, V<sub>S</sub> logs are available for only seven. To address this gap, a feature selection approach identified P-wave velocity (V<sub>P</sub>) and depth (mbsf) as the most significant predictors. These features were employed to train four machine learning regression models Ordinary Least Squares Regression (OLR), Elastic Net Regression (ENR), Stochastic Gradient Descent Regression (SGDR), and Support Vector Regression (SVR). The dataset from wells with existing V<sub>S</sub> logs was merged and partitioned into training (70%) and testing (30%) subsets. Model training incorporated 10-fold cross-validation for hyperparameter optimization. All models exhibited comparable predictive performance (R² ≈ 0.85) on the test data. The OLR model was further applied to estimate V<sub>S</sub> in wells lacking log measurements, and the results showed good consistency with V<sub>S</sub> values derived from pressure core analysis. Additionally, pressure core measurements established a strong correlation between V<sub>S</sub> and triaxial shear strength. Based on this relationship, a quadratic function was developed and subsequently applied to predict triaxial shear strength across all 12 wells using the modelled V<sub>S</sub> values.</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 shear wave velocity through machine learning regression and its role in triaxial shear strength assessment

  • 期刊论文

  • 发表日期：

    2025年11月25日

查看全文
  • Pradeep Kumar Yadav,
  • Uma Shankar

摘要

Shear wave velocity (VS) plays a vital role in geophysical characterization, particularly within geologically complex and unconsolidated sediments. Despite its importance, direct VS logging is often limited by operational and economic constraints. In this study, Logging While Drilling (LWD) data from gas hydrate-bearing sediments of Area B in the Krishna–Godavari Basin, obtained during the National Gas Hydrate Program Expedition-02 (NGHP-02), are analysed. Out of 12 investigated wells, VS logs are available for only seven. To address this gap, a feature selection approach identified P-wave velocity (VP) and depth (mbsf) as the most significant predictors. These features were employed to train four machine learning regression models Ordinary Least Squares Regression (OLR), Elastic Net Regression (ENR), Stochastic Gradient Descent Regression (SGDR), and Support Vector Regression (SVR). The dataset from wells with existing VS logs was merged and partitioned into training (70%) and testing (30%) subsets. Model training incorporated 10-fold cross-validation for hyperparameter optimization. All models exhibited comparable predictive performance (R² ≈ 0.85) on the test data. The OLR model was further applied to estimate VS in wells lacking log measurements, and the results showed good consistency with VS values derived from pressure core analysis. Additionally, pressure core measurements established a strong correlation between VS and triaxial shear strength. Based on this relationship, a quadratic function was developed and subsequently applied to predict triaxial shear strength across all 12 wells using the modelled VS values.