Physics and deep learning based methods for compositional reservoir simulation
Abstract/Contents
- Abstract
- Gas injection processes for enhanced oil recovery are modeled using numerical compositional simulation. These complex displacement processes entail the mass transfer of large numbers of components among multiple fluid phases. The phase behavior is described using thermodynamic equilibrium constraints. The solution of these constraints yields the phase state and the composition of the different fluid phases that share the pore space. The accuracy and efficiency of the phase behavior computations are essential for using compositional simulation for problems of practical interest. We extend our in-house research simulator (AD-GPRS) to model general three-phase fluid mixtures. Therefore, this simulator can describe the three-phase gas injection process with water existence. In the three-phase compositional simulation, every component is allowed to exist and transfer among hydrocarbon and aqueous phases. We validate the compositional simulation by comparison with a commercial simulator's results and refinement studies. We demonstrate the robust performance of the three-phase compositional simulator by extensive test cases. Thermodynamic calculations require extensive computations, which will be significantly increased as more components involved in the reservoir fluids. Therefore, the acceleration of compositional simulation is crucial to explore after the robustness is preserved. We develop two deep-learning based models for stability analysis and flash calculation separately, to reduce the cost of thermodynamic calculations. We systematically present the process of data generation, training, and model optimization, which unravels some of the mysteries surrounding the theory and application of data-driven methods in reservoir simulation. Then the robustness and accuracy of two deep-learning-based models are verified by various case tests. Besides, incorrectly labeled phases could cause phase oscillations during simulation, further leading to the failure in numerical convergence. The main challenge for currently published methods is to keep a balance between computational costs and correctness in compositional simulation. Here, we propose a mixed methodology based on deep-learning techniques and straightforward geometrical principles. We use several case studies to show that this method can ensure accuracy and efficiency at the same time. What is more, we pioneer the application of deep-learning-based techniques in compositional simulations to accelerate computational speed. Notably, we present a strategy to batch mixtures together as one whole input vector for a deep-learning-based model, referred to as the vectorization. This technique allows the phase module to determine phase behaviors for abundant grid cells simultaneously, eliminating the iteration process in traditional compositional simulation. Improvements in computational speed are examined by a series of challenging cases on immiscible and miscible displacements. With in-depth analyses of performances, we demonstrate the robustness and strength of deep-learning-based techniques in accelerating compositional simulation computations.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2020; ©2020 |
| Publication date | 2020; 2020 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Sun, Ruixiao |
|---|---|
| Degree supervisor | Tchelepi, Hamdi |
| Thesis advisor | Tchelepi, Hamdi |
| Thesis advisor | Kovscek, Anthony R. (Anthony Robert) |
| Thesis advisor | Pan, Huanquan |
| Degree committee member | Kovscek, Anthony R. (Anthony Robert) |
| Degree committee member | Pan, Huanquan |
| Associated with | Stanford University, Department of Energy Resources Engineering |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Ruixiao Sun. |
|---|---|
| Note | Submitted to the Department of Energy Resources Engineering. |
| Thesis | Thesis Ph.D. Stanford University 2020. |
| Location | electronic resource |
Access conditions
- Copyright
- © 2020 by Ruixiao Sun
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...