Loading...
Multiphysical simulation of CO₂ enhanced oil recovery in unconventional reservoirs: from fundamental physics to simulator development
Wang, Shihao
Wang, Shihao
Citations
Altmetric:
Advisor
Editor
Date
Date Issued
2019
Date Submitted
Collections
Research Projects
Organizational Units
Journal Issue
Embargo Expires
Abstract
The economic recovery of unconventional reservoirs is demanding yet challenging. Traditional enhanced oil recovery (EOR) techniques face difficulty when being applied to unconventional reservoirs. In this PHD study, we aim to deepen the understanding of the transport mechanisms and the phase behavior in unconventional reservoirs. Moreover, we aim to develop numerical tools to simulate the multiphysical process occurring in the carbon dioxide enhanced oil recovery (CO2-EOR) process. Based on the developed simulator, we plan to investigate the potential of applying cold CO2 flooding in unconventional reservoirs. We have developed a novel transport model for the gas slippage effect of multicomponent gases in unconventional reservoirs. The model is essentially a second-order slippage boundary condition, based on kinetic theory of gases. The model has been validated with respect to both physical experimental data and numerical simulation data and has shown accuracy up to 90%. We have developed a compositional reservoir simulator named MSFLOW-CO2, which adopts integral finite difference method to simulate the coupled thermal-hydraulic-mechanical processes during CO2-EOR. In our method, the governing equations of the multiphysical processes are solved fully coupled on the same unstructured grid. An algebraic multiscale linear solver is adopted to speed up the non-isothermal simulation. In order to simulate the phase behavior of the three-phase system, a flash calculation module based on direct minimization of Gibbs energy is implemented in the simulator. We have investigated the impact of cold CO2 injection on injectivity as well as phase behavior. We conclude that cold injection is an effective way to increase the injectivity in tight intact reservoirs. We have also observed and studied the temperature decreasing phenomenon near the production well, known as the Joule-Thomson effect, induced by expansion of in-situ fluids. Moreover, we aim to speed up compositional modeling by stochastic training. In traditional compositional simulators, the phase behavior is simulated using flash calculation techniques on each grid block within each iteration, which results in a huge amount of computational time. In this work, based on the developed standalone flash calculation module, we have trained two fully connected neural networks, namely the phase classification and the concentration determination, using Keras to conduct proxy flash calculation and have implemented the proxy module into the simulator. The networks are trained with stochastic gradient method optimizer and are used as the preconditioner of the physical flash calculation module. The accuracy of the phase classification step is above 99%. The accuracy of the concentration determination step is up to 98%. Therefore, the overall accuracy of our proxy flash calculator is above 97%. With the implementation of the proxy flash calculation module as a preconditioner, the number of iterations of the flash calculation module is reduced by more than 50%. To sum up, the novelty of this work lies in the transport mechanism in unconventional reservoirs, the multiphysical simulation framework, the flash calculation module, and the proxy simulation module.
Associated Publications
Rights
Copyright of the original work is retained by the author.