Abstract
Computational Stratigraphy (CompStrat) is a state-of-the-art earth-modeling method that captures the key heterogeneities in subsurface reservoirs through modeling of the detailed flow and sediment transportation processes in various depositional environments. The method is fully based on physics and generates high-resolution 3D earth models that are much more geologically realistic than those generated by traditional earth-modeling methods. It can accurately predict and preserve those spatially continuous but vertically thin and volumetrically insignificant layers, such as shale layers, thus enabling a much more accurate representation of natural reservoir connectivity.
In the past few years, CompStrat has been studied mainly within the earth science community and has yet been broadly applied in reservoir simulation research and practices. Our objective is to bridge this gap and allow this frontier technology to offer geologically realistic earth models for reservoir simulation to better understand how various geological features contribute and control subsurface flow patterns and performance, and subsequently leading to a better integration among earth modeling, flow simulation, and more reliable reservoir performance predictions.
CompStrat models often have large number of cells (hundreds of millions or more). A large proportion of them are related to thin shale layers. These thin cells can often cause convergence difficulties in reservoir simulations. We developed a grid coarsening method to drastically reduce the cell number and the simulation time with minimum altering of overall model connectivity characteristics. The method reduces the cell number by 85% to 93% and the simulation time by 94% to 99.4% with limited loss of accuracy for representative examples. Without this method, the simulation may take impractically long time to run for large models with complex multiphase flow dynamics.
The successful removal of the computational bottleneck enables the application of this frontier earth-modeling method in high-fidelity reservoir simulation. It also facilitates detailed understanding of the connection between geology and flow to offer valuable insight for reservoir modeling, production forecast uncertainty analysis, and history matching. We developed a method to label, evaluate, and rank geological features based on their influence on flow performance, with shale layers being the specific focus. The labeling is performed semi-automatically and the evaluation and ranking is done efficiently with a reduced-physics solver. The result is statistically consistent across multiple realizations.
