Adaptive Mesh Refinement and Coarsening for Compositional Reservoir Simulation
Abstract
This dissertation introduces a new method to create adaptive mesh refinement and coarsening in compositional reservoir simulation. The methodology targets individual cells for refinement based on forecasted compositional fronts calculated using streamlines and the analytical convection-dispersion transport equation. Quadtree decomposition determines the optimal spatial discretization across the simulation grid using dynamic and static reservoir properties. Application of the new approach results in improved computational performance without compromising the accuracy of phase behavior.
Current dynamic gridding implementations have rigid schemes, posing two major limitations: cell refinement size is a pre-determined input value and compositional maps from the previous time step define the refinement region. This solution leads to suboptimal modeling due to time-lagging refinement and lack of grid adaptability in heterogeneous reservoirs and/or fast-moving compositional fronts. The new methodology overcomes these limitations by combining streamline and particle trajectory to forecast the injection front location and adapt grid sizes in advance.
Tracking compositional variations starts by calculating fluxes for all cells using the finite-difference solution. Next, Pollock’s tracing method allows reducing the 3-dimensional model into a series of 1-dimensional streamlines, while the convection-dispersion equation forecasts future compositions, shape, and location of injection front along each streamline trajectory. Finally, quadtree decomposition analyzes the homogeneity of dynamic and/or static properties (e.g., composition, pressure, permeability, facies) to determine if a volume can be represented by a single gridblock or if it requires refinement to preserve spatial details. Grid discretization is dynamic over time, refining cells requiring high-resolution and/or coarsening those with low variation.
A mechanistic model with CO2 injection served to evaluate the methodology. The fluid was modeled with five pseudo-components and the Peng-Robison equation of state with volume translation to improve volumetric predictions. The new approach reduced the total number of cells required to model miscible injection by continuously creating adaptive grids that represent the advancement and shape of the injection front. Results showed a reduction in computational cost between 30-63% over a static fine grid without compromising the representation of compositional mixing phenomena and production forecast.
Subject
Adaptive mesh refinementlocal grid refinement
compositional reservoir simulation
streamline
convective-dispersive equation
Citation
Gonzalez Abad, Karin Gabriela (2016). Adaptive Mesh Refinement and Coarsening for Compositional Reservoir Simulation. Doctoral dissertation, Texas A & M University. Available electronically from https : / /hdl .handle .net /1969 .1 /159028.