A General Surrogate Model for CO₂ Flooding Dynamic Prediction Based on Dimensionless Features and Implicit Time.

Area of Science:

• Petroleum Engineering and Reservoir Simulation
• Artificial Intelligence in Energy
• Carbon Capture, Utilization, and Storage (CCUS)

Background:

• Traditional numerical simulations for CO2 flooding are computationally intensive due to complex phase interactions and nonlinear flow.
• High computational costs limit the efficiency of optimizing large-scale reservoir development schemes.
• Existing models struggle with generalization across different reservoir geological scales.

Purpose of the Study:

• To develop a general surrogate model for efficient and high-precision prediction of cumulative oil production and CO2 storage capacity in CO2 flooding reservoirs.
• To address the computational burden and generalization difficulties in traditional reservoir simulations.
• To provide a computationally efficient tool for CCUS-Enhanced Oil Recovery (EOR) development assessment.

Main Methods:

• Proposed a Transformer architecture-based surrogate model.
• Constructed a dimensionless feature system using Pore Volume (PV) normalization to decouple reservoir geological scales.
• Integrated implicit time parameterization and Bayesian adaptive optimization, replacing physical time with cumulative fluid injection volume.

Main Results:

• Achieved R-squared values of 0.9986 for oil production and 0.9968 for CO2 storage predictions across 280,000 samples.
• Demonstrated significant computational efficiency, achieving a speedup of nearly 60 times compared to numerical simulations for a 5-year prediction.
• The model exhibits time independence and excellent generalizability across different reservoir scales.

Conclusions:

• The proposed Transformer-based surrogate model effectively mitigates the computational bottleneck of long-cycle reservoir simulations.
• The dimensionless feature system and continuous time mapping enhance model generalizability and physical universality.
• This approach offers robust technical support for rapid assessment and optimization in CCUS-EOR development.