| dc.description.abstract | Modeling unconventional reservoirs has been an active area of research in response to significant reserves in the U.S. Various analytical and numerical models incorporating relevant physics at varying fidelity levels have been proposed. A common approach to capture complex physics inherent in unconventional reservoirs is to run numerical reservoir simulations. However, fine spatio-temporal discretization is required to capture underlying nonlinear dynamics, resulting in long runtimes which are typically too computationally expensive to perform tasks like history matching and optimization. To address this challenge, the fast-marching method (FMM) based rapid simulation workflow has been proposed. The approach transforms original 2-D or 3-D problem into an equivalent 1-D problem along the diffusive time-of-flight (DTOF) coordinate representing travel time of pressure front propagation. The coordinate transformation enables significant savings in the computation time. To date, the powerful FMM-based simulation workflow has found many applications in modeling unconventional reservoirs, including rapid history matching and optimization. However, the approach poses certain limitations in capturing the effects of gravity because the physical direction of gravity is not necessarily aligned with the 1-D DTOF coordinate. Also, the current FMM calculations are difficult to generalize for irregular grid systems, such as unstructured grids, despite their increasing use in modern reservoir simulators.
Three major research contributions are made in this dissertation. First, we present an extension of the FMM-based simulation workflow that can account for gravity. The effects of gravity and multiphase flow are captured by retaining original discretization (gridblocks) for hydraulic fractures during the coordinate transformation. Second, a new FMM framework based on a finite-volume discretization of the Eikonal equation is developed as a generalization of the FMM calculations for irregular grid systems. The power and efficacy of the approach are demonstrated through its application to a variety of examples with different types of irregular grids. Third, an efficient deep learning-based approach that can visualize subsurface images, such as pressure propagation, in near real-time is developed. The FMM-based simulation workflow developed in the previous chapters is used in place of full-physics simulation to efficiently generate training datasets. The workflow is further enhanced by incorporating an autoencoder to perform a dimensionality reduction. The resultant subsurface visualizations can be readily used for qualitative and quantitative characterization and forecasting of unconventional reservoirs. The novelty of the approach is the framework which combines the strengths of deep learning-based models and the FMM-based rapid simulation. | |
