A hybrid simulation-machine learning proxy model for waterflood design optimization in the Bahariya Formation.

Area of Science:

• Petroleum Engineering
• Artificial Intelligence in Energy
• Reservoir Management

Background:

• Optimizing waterflooding is crucial for maximizing oil recovery in complex geological formations.
• Accurate prediction of oil recovery efficiency (RF%) is essential for effective reservoir development planning.
• Conventional simulation methods can be time-intensive for screening multiple waterflooding scenarios.

Purpose of the Study:

• To develop and validate a hybrid methodology combining machine learning (ML) and reservoir simulation for waterflooding optimization.
• To accurately predict oil recovery efficiency (RF%) across different injection patterns in the Bahariya Formation.
• To identify dominant reservoir parameters influencing oil recovery for various waterflooding strategies.

Main Methods:

• A hybrid approach integrating ML models (e.g., linear regression) with conventional reservoir simulation.
• Development of data-driven models to predict Ultimate Oil Recovery (UOR) for Peripheral, Staggered Line Drive (SLD), and 5-Spot injection patterns.
• Utilizing permutation importance analysis to quantify the influence of key reservoir parameters on recovery.

Main Results:

• Peripheral flooding demonstrated the highest UOR (44.7%), followed by SLD (39.4%) and 5-Spot (33.7%).
• ML models achieved high predictive accuracy with R² scores > 0.95 and low RMSE values.
• Residual oil saturation (Sor) was the most significant parameter (38-42%), with injection rate (WINJ) and API gravity showing pattern-dependent importance.

Conclusions:

• The hybrid AI-numerical framework offers an efficient method for rapid waterflooding scenario screening and optimization.
• Optimal waterflooding pattern selection is highly dependent on specific reservoir characteristics.
• Findings provide practical insights for field engineers to enhance injection strategies and increase recovery factors.