Multiscale Grid Design and Upscaling for a Full Field Tight Gas Reservoir Model on a Next Generation Simulator

Abstract

Reservoir management and simulation of tight gas reservoirs pose several interesting technical challenges. Accurate simulation estimates of the effectiveness of multi-stage fracturing, well rates, the interaction between multiple fractures, and between the reservoir and fractures requires high resolution gridding near the well. These predictions have been achieved in the past through the design of LGR (Local Grid Refinement) grids, or through the use of fractured well indices or cell permeabilities. The latter techniques are based upon steady state concepts and are less able to resolve well rate transients compared to the use of LGR grids. Decisions on well counts, spacing, and timing also depend upon the interaction between wells. These simulations face specific challenges due to the geology of these reservoirs: in addition to the permeability being low, the connectivity of these reservoirs is intermittent due to the three dimensional spatial patterns of sand deposition and subsequent diagenesis. If these reservoir models are over-averaged or not carefully upscaled, the barriers between sands and internal heterogeneity will be lost, leading to far too optimistic estimates of lateral continuity and under-estimates of infill requirements. The need for accurate flow simulation of these systems has led to the development of spatially adaptive upscaling techniques that honor the vertical discontinuities of the sands in the design of the coarsened simulation grid and which also preserves the heterogeneity within sands. In the current study, these two approaches have been combined, providing a model with both high resolution and computational efficiency. Starting with a high resolution 3D geologic model, the simulation grid is designed using refinement at the fractured wells, and successively coarsened away from the well, leading to an unstructured numerical problem. Away from the wells, the vertical coarsening utilizes the "pillar-based" adaptive coarsening of Zhou and King, 2011. Simulation is performed using a next generation unstructured commercial simulator. We will report upon the performance of the different techniques in terms of accuracy, computational efficiency, and how they are impacted by different implementations of the multiscale simulation grid design.