Hybrid connectionist model determines CO₂-oil swelling factor

Area of Science:

• Petroleum Engineering
• Reservoir Simulation
• Artificial Intelligence in Energy

Background:

• Understanding crude oil and CO2 interactions is vital for designing and simulating CO2-based enhanced oil recovery (EOR).
• CO2 dissolution in crude oil causes oil swelling, a phenomenon influenced by temperature, pressure, and oil composition.
• Oil swelling factor is inversely proportional to residual oil saturation, making its precise estimation critical for EOR efficiency.

Purpose of the Study:

• To develop a predictive model for accurately calculating the CO2-oil swelling factor.
• To optimize the model using a genetic algorithm for enhanced precision in EOR simulations.

Main Methods:

• Utilized a least-squares support vector machine (LS-SVM) predictive model.
• Employed a genetic algorithm to optimize LS-SVM hyperparameters (gamma and sigma^2).
• Validated the model using available experimental data for CO2-oil swelling factor estimation.

Main Results:

• The LS-SVM model achieved a high coefficient of determination (R^2 = 0.9953).
• The model demonstrated a low mean squared error (MSE = 0.0003).
• The developed method offers a straightforward and accurate approach with negligible uncertainty.

Conclusions:

• The optimized LS-SVM model provides a precise and reliable method for determining the CO2-oil swelling factor.
• This predictive approach can be integrated into commercial reservoir simulators, especially when experimental data is scarce.
• Accurate CO2-oil swelling factor estimation enhances the design and simulation of CO2-based EOR processes.