The artificial neural network-based two-phase equilibrium calculation framework for fast compositional reservoir simulation of CO2 EOR

Abstract

Injecting CO₂ into the reservoir is an essential method for Carbon Capture, Utilization, and Storage (CCUS) and enhanced oil recovery (EOR). However, due to the complex phase behavior between CO₂ and hydrocarbons, the reservoir simulation of the injection process becomes time-consuming. To expedite phase equilibrium calculations (PECs) involved in CO₂-EOR, we have developed an artificial neural network (ANN)-based PECs framework comprising the 1P-stability and 2P-flash models, which replaced traditional single-phase stability analysis and two-phase flash calculations. Additionally, We proposed a straightforward method for generating training points tailored to CO₂-EOR production characteristics. Specific settings are placed on the two models to ensure a 100% correct solution, including predefining criteria to filter the stability model output and utilizing the flash model output as the standard algorithm initial value. We have enhanced the ANN-based models to integrate seamlessly with the compositional simulator. Four published fluids were selected to test this framework by implementing the standalone PECs, and one fluid was used for the simulation of CO₂-EOR. The ANN-based framework can save up to 80% of time on phase equilibrium calculations, resulting in a 40% reduction in simulation time compared to the conventional algorithm. In summary, the newly developed ANN-based PECs framework shows great potential to accelerate the reservoir simulation for CO₂-EOR, which helps design the program of CCUS by injecting CO₂ into the reservoir.