Identification and Intelligent Prediction of Microscopic Residual Oil Distribution Based on the TransUNet Neural...

Area of Science:

• Petroleum Engineering
• Artificial Intelligence
• Geoscience

Background:

• Accurate classification of microscale remaining oil (<10 μm) is crucial for effective oilfield development.
• Traditional methods suffer from manual errors, low intelligence, and insufficient precision.

Purpose of the Study:

• To develop a deep learning-based method for precise recognition and classification of microscale remaining oil.
• To enhance feature extraction and classification performance for improved oilfield development strategies.

Main Methods:

• Utilized TransUNet as the backbone network for image segmentation.
• Incorporated image augmentation and a Transformer multihead attention mechanism.
• Applied the method to analyze remaining oil morphology under different displacement mechanisms.

Main Results:

• Achieved an overall classification accuracy of 94% for remaining oil.
• Identified effective recognition thresholds: 3.09 μm for film-like and 1.54 μm for drip-like remaining oil.
• Demonstrated that surfactant flooding yields droplet-shaped remaining oil, indicating superior displacement efficiency.

Conclusions:

• The proposed deep learning method offers high accuracy and precision in microscale remaining oil classification.
• Understanding the morphological distribution of remaining oil aids in formulating targeted oilfield development measures.
• Surfactant flooding shows a microscopic advantage in displacing remaining oil due to its droplet-like morphology.