Machine-Learning Approach for Forecasting Steam-Assisted Gravity-Drainage Performance in the Presence of...

Area of Science:

• Petroleum Engineering
• Machine Learning Applications
• Enhanced Oil Recovery

Background:

• Steam-assisted gravity drainage (SAGD) is crucial for heavy oil recovery.
• Incorporating noncondensable gases (NCG) can reduce steam consumption but complicates design.
• Optimizing NCG integration requires advanced predictive capabilities.

Purpose of the Study:

• To develop a machine-learning-based forecasting model for SAGD applications with NCG.
• To predict oil recovery and cumulative steam-oil ratio (CSOR) using experimental data.
• To identify key parameters influencing SAGD performance with NCG injection.

Main Methods:

• Performed scaled physical model experiments using limestone and heavy oil.
• Injected steam mixed with carbon dioxide (CO2) or n-butane (n-C4H10).
• Trained neural network models using experimental data (temperature, pressure, production) to predict oil recovery and CSOR.

Main Results:

• A 3-hidden-layer neural network model achieved high prediction accuracy (R²=0.98 for oil recovery, R²=0.95 for CSOR).
• Key influential parameters identified: pore volume injected, well separation, and prior CO2 saturation.
• Model performance aligned with experimental observations.

Conclusions:

• The developed machine learning model effectively forecasts SAGD performance with NCG.
• The model aids in making informed design decisions for SAGD processes incorporating NCG.
• Understanding parameter importance facilitates optimization of heavy oil recovery strategies.