Surrogate-Assisted Optimization of Highly Constrained Oil Recovery Processes Using Classification-Based Constraint...

Area of Science:

• Petroleum Engineering
• Computational Science
• Optimization Theory

Background:

• Real-world operations, particularly in petroleum engineering, frequently encounter complex, high-dimensional problems with numerous constraints.
• Optimizing these processes, such as waterflooding, is computationally expensive and challenging due to the vast number of operational limitations.

Purpose of the Study:

• To develop an efficient and feasible optimization methodology for highly constrained reservoir simulations.
• To reduce the computational burden associated with optimizing complex waterflooding processes.

Main Methods:

• Utilized a Feedforward Neural Network (FFNN) as a surrogate model for objective quantification (Net Present Value - NPV).
• Implemented a classification-based constraint handling technique to transform numerous constraints into a binary feasible/infeasible problem.
• Trained models offline using reservoir simulation data (Egg and UNISIM models) and employed model-based and search-based optimizers in the online phase.

Main Results:

• Achieved high predictive accuracy for the surrogate model (R² > 0.98) with 20,000 bottom-hole pressure (BHP) settings and NPV outputs.
• The classification-based constraint handling effectively distinguished feasible from infeasible reservoir settings.
• The integrated approach significantly reduced computational load while ensuring optimized and feasible solutions.

Conclusions:

• The proposed methodology offers a computationally efficient and effective solution for optimizing highly constrained waterflooding operations.
• Integrating accurate surrogate models with classification-based constraint handling leads to superior results compared to deterministic approaches, maximizing economic gain.