Related Experiment Video
Updated: May 12, 2026
07:31
Realistic Membrane Modeling Using Complex Lipid Mixtures in Simulation Studies
Published on: September 1, 2023
Development of a Rapid Deep-Learning-Assisted Multiphase Multicomponent Numerical Simulation Protocol.
Luecheng Fang1, Qian Sun2,3, Qiang Xu4
1School of Energy Resources, China University of Geosciences (Beijing), Beijing 100083, China.
ACS Omega
|May 11, 2026
Summary
This study introduces a deep-learning workflow to accelerate compositional reservoir simulations by replacing slow flash calculations. The AI model significantly enhances simulation speed and accuracy for CO2 injection and other complex fluid behaviors.
Area of Science:
- Petroleum Engineering
- Computational Science
- Artificial Intelligence
Background:
- Compositional reservoir simulations offer high accuracy but are computationally expensive.
- Black-oil models are faster but less reliable for phase-equilibrium-sensitive problems like CO2 injection.
Purpose of the Study:
- To develop a deep-learning workflow to replace conventional iterative flash calculations in compositional reservoir simulations.
- To improve the computational efficiency and maintain accuracy of reservoir simulations.
Main Methods:
- A deep-learning model was developed to perform classification (phase stability) and regression (equilibrium ratios).
- The trained model was integrated into a compositional simulator and tested on benchmark cases, including phase-envelope reconstruction and gas injection problems.
Main Results:
- Phase-envelope reconstruction showed mean relative errors below 1% for liquid fraction and phase compositions.
- Production rates in a gas-injection problem had mean relative errors below 0.2%.
- Significant efficiency gains were observed, with flash calculations up to 517% faster and phase-stability testing up to 2464% faster.
Conclusions:
- The deep-learning workflow effectively replaces iterative flash calculations, offering substantial speedups.
- The method maintains thermodynamic consistency and high accuracy, making it suitable for complex reservoir simulation scenarios.
- This approach enables more practical application of accurate compositional simulations for enhanced oil recovery and carbon sequestration.
