Total organic carbon content estimation for mixed shale using Xgboost method and implication for shale oil...

Area of Science:

• Geochemistry
• Petroleum Geology
• Machine Learning Applications

Background:

• Accurate estimation of total organic carbon (TOC) is crucial for evaluating shale oil potential.
• Mixed carbonate and siliciclastic shale formations present unique challenges for TOC prediction.
• The Hashan area in the Junggar Basin is a key region for shale oil exploration.

Purpose of the Study:

• To estimate TOC in mixed carbonate and siliciclastic shale using the Xgboost method.
• To enhance TOC estimation accuracy by applying Principal Component Analysis (PCA) for dimensionality reduction.
• To identify potential shale oil exploration targets based on improved TOC distribution mapping.

Main Methods:

• XGBoost (Extreme Gradient Boosting) machine learning algorithm for initial TOC estimation.
• Principal Component Analysis (PCA) for reducing well log data dimensionality from five to two dimensions.
• Isopleth mapping of TOC distributions to delineate exploration targets.

Main Results:

• Initial XGBoost model showed moderate TOC estimation performance (R²=0.54), attributed to a small dataset.
• PCA dimensionality reduction significantly improved the correlation coefficient between predicted and measured TOC to 0.68.
• Redrawn TOC distribution maps revealed two distinct shale oil exploration targets within the Fengcheng Formation.

Conclusions:

• The combined XGBoost and PCA approach provides a reliable method for TOC estimation in the study area.
• The identified exploration targets offer valuable insights for shale oil exploration and development.
• This methodology serves as a robust example for similar geological settings and hydrocarbon exploration studies.