Classifying reservoir facies using attention-based residual neural networks.

Area of Science:

• Petroleum Geoscience
• Machine Learning
• Reservoir Characterization

Background:

• Accurate reservoir facies classification is crucial for efficient resource extraction and reservoir characterization.
• Traditional methods struggle with the complexity and heterogeneity of well-log data.
• Existing deep learning models may not fully capture hierarchical data representations and contextual dependencies.

Purpose of the Study:

• To develop and validate a novel deep learning framework for enhanced reservoir facies classification.
• To address the limitations of traditional and existing machine learning approaches in handling heterogeneous well-log data.
• To improve the accuracy of identifying facies boundaries and lithological variations in complex geological formations.

Main Methods:

• Developed a deep learning framework with an architectural approach distinct from single-stream or non-residual designs.
• Trained and evaluated the framework using well-log measurements from eight diverse geological settings.
• Conducted comparative experiments against conventional machine learning and state-of-the-art deep learning techniques.

Main Results:

• The proposed method achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.883.
• The method obtained an Area Under the Precision-Recall Curve (AUPRC) of 0.502, outperforming existing techniques.
• Demonstrated superior ability to concentrate on key geological features and preserve hierarchical data representations.

Conclusions:

• The novel deep learning framework significantly enhances reservoir facies classification accuracy.
• The method effectively addresses data heterogeneity and contextual dependencies in well-log data.
• The robust and reproducible performance makes it viable for real-world reservoir characterization and field development planning.