Using data-driven models to simulate the performance of surfactants in reducing heavy oil viscosity

Area of Science:

• Petroleum Engineering
• Chemical Engineering
• Data Science

Background:

• Unconventional heavy oil reservoirs are vital for global energy supply.
• Chemically enhanced oil recovery (EOR) uses surfactants to reduce heavy oil viscosity for easier extraction.
• Modeling oil-in-water (O/W) emulsion viscosity is key to optimizing EOR processes.

Purpose of the Study:

• To develop and optimize machine learning (ML) models for predicting O/W emulsion viscosity.
• To evaluate the performance of various ML algorithms in modeling emulsion viscosity.
• To identify key input parameters influencing emulsion viscosity through sensitivity analysis.

Main Methods:

• A dataset of 2020 experimental points was compiled from existing literature.
• Five ML algorithms (Adaptive Boosting, CNN, Ensemble Learning, ANN, Decision Tree) were employed.
• Algorithms were optimized using a Combined Simulated Annealing (CSA) method.
• Monte-Carlo sensitivity analysis was conducted to assess feature importance.

Main Results:

• The Artificial Neural Network (ANN) model demonstrated superior accuracy in predicting O/W emulsion viscosity.
• ANN achieved high performance metrics (e.g., R²=0.996, RMSE=0.0132) across the entire dataset.
• Sensitivity analysis identified critical factors affecting emulsion viscosity.

Conclusions:

• ML, particularly ANN, offers a highly accurate and efficient method for predicting O/W emulsion viscosity.
• This predictive capability can significantly reduce the reliance on time-consuming and expensive laboratory experiments.
• The developed ML models can expedite the optimization of surfactant-based EOR for heavy oil recovery.