Scholarlink

<p>Predicting volume of shale and porosity in structurally complex deltaic environments like the Niger Delta remains challenging for conventional inversion techniques. This study applies machine learning regression models to integrate multi-attribute 3D seismic data and well logs for improved prediction of reservoir properties in the Akaso Field, eastern Niger Delta. Petrophysical evaluation of Sand Unit C from six wells yielded porosity values of 0.11–0.34, shale volume of 1–11%, and water saturation of 4–30%. Ten seismic attributes including Acoustic Impedance, Reflection Strength, Instantaneous Frequency, and Dominant Frequency were extracted, normalized, and screened using Variance Inflation Factor. Porosity was computed from density–neutron crossplots, while volume of shale index was derived using k-means clustering. Six regression algorithms (Linear, Ridge, Lasso, SVR, Random Forest, and XGBoost) were trained using 80:20 train–test split and 10-fold stratified cross-validation with hyperparameter optimization and SHAP-based feature importance analysis. XGBoost achieved the highest accuracy with R² = 0.91 for porosity and R² = 0.83 for volume of shale, significantly outperforming SVR and Random Forest. SHAP analysis identified Reflection Strength, Acoustic Impedance, and Instantaneous Frequency as strongest porosity predictors, while Dominant Frequency, Hilbert Transform, and Instantaneous Phase best predicted volume of shale. Predicted porosity and volume of shale maps accurately delineated hydrocarbon-bearing sand bodies and aligned with well observations. This study demonstrates that integrating multi-attribute seismic data with optimized ML regression significantly improves volume of shale and porosity prediction in faulted deltaic settings. The combined use of accuracy metrics and SHAP-based interpretability provides a transferable framework for reducing subsurface uncertainty and supporting data-driven reservoir characterization in the Niger Delta.</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

Multi attribute predictions of volume of shale and porosity from seismic data over Akaso Field, Niger Delta using machine learning regression analysis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月19日

查看全文
  • Edeye Ejaita,
  • Francis Omonefe

摘要

Predicting volume of shale and porosity in structurally complex deltaic environments like the Niger Delta remains challenging for conventional inversion techniques. This study applies machine learning regression models to integrate multi-attribute 3D seismic data and well logs for improved prediction of reservoir properties in the Akaso Field, eastern Niger Delta. Petrophysical evaluation of Sand Unit C from six wells yielded porosity values of 0.11–0.34, shale volume of 1–11%, and water saturation of 4–30%. Ten seismic attributes including Acoustic Impedance, Reflection Strength, Instantaneous Frequency, and Dominant Frequency were extracted, normalized, and screened using Variance Inflation Factor. Porosity was computed from density–neutron crossplots, while volume of shale index was derived using k-means clustering. Six regression algorithms (Linear, Ridge, Lasso, SVR, Random Forest, and XGBoost) were trained using 80:20 train–test split and 10-fold stratified cross-validation with hyperparameter optimization and SHAP-based feature importance analysis. XGBoost achieved the highest accuracy with R² = 0.91 for porosity and R² = 0.83 for volume of shale, significantly outperforming SVR and Random Forest. SHAP analysis identified Reflection Strength, Acoustic Impedance, and Instantaneous Frequency as strongest porosity predictors, while Dominant Frequency, Hilbert Transform, and Instantaneous Phase best predicted volume of shale. Predicted porosity and volume of shale maps accurately delineated hydrocarbon-bearing sand bodies and aligned with well observations. This study demonstrates that integrating multi-attribute seismic data with optimized ML regression significantly improves volume of shale and porosity prediction in faulted deltaic settings. The combined use of accuracy metrics and SHAP-based interpretability provides a transferable framework for reducing subsurface uncertainty and supporting data-driven reservoir characterization in the Niger Delta.