Faster Reservoir Simulations with Unstructured Grids and Property Upscaling

What is it about?

Computer models play a vital role in planning how oil, gas, and carbon dioxide move underground. These simulations help engineers decide where to drill wells, estimate future production, evaluate carbon storage projects, and reduce financial and environmental risks. However, modern geological models often contain millions of small cells, making simulations extremely time-consuming and computationally expensive. This study presents a new way to make these simulations much faster without significantly reducing their accuracy. Instead of using the original highly detailed model everywhere, the proposed method automatically identifies which parts of the reservoir require high resolution and which regions can safely be represented using larger cells. Areas close to wells or containing highly variable rock properties remain detailed, while more uniform regions are simplified. This produces an unstructured computational grid that preserves important geological features while greatly reducing the size of the numerical problem. The new grid generator was integrated into the UTCOMPRS compositional reservoir simulator and tested using several challenging benchmark models, including the UNISIM-I reservoir, a Brazilian pre-salt proxy model, the SPE10 benchmark, and the Sleipner carbon storage project. These case studies represent realistic industrial problems with different levels of geological complexity. The results demonstrate that the proposed approach reproduces production forecasts and pressure behavior with excellent agreement compared to fine-scale models and a commercial reservoir simulator, while substantially reducing computational cost. In some cases, the simulations became several times faster while maintaining reliable accuracy. The method also reduces memory requirements for appropriately selected coarsening levels, making large-scale simulations more practical. By combining automatic grid generation, selective upscaling of reservoir properties, and unstructured finite volume simulation, this work provides a practical tool that allows engineers and researchers to perform high-fidelity reservoir simulations more efficiently, enabling faster engineering decisions and more effective use of computational resources.

Featured Image

Photo by Markus Spiske on Unsplash

Photo by Markus Spiske on Unsplash

Why is it important?

Several aspects distinguish this work from previous approaches. Rather than simply converting structured geological models into unstructured meshes, the proposed methodology automatically performs selective grid coarsening while preserving fine resolution only where it is most important for flow behavior. The workflow combines a user-defined clustering strategy, the Dykstra-Parsons heterogeneity coefficient, modified Cardwell-Parsons permeability upscaling, and automatic handling of complex geological features such as pinch-outs and degenerate cells within a single integrated framework. The timing of this work is particularly relevant because reservoir simulation is increasingly being applied to extremely large and complex models, including both traditional hydrocarbon reservoirs and carbon capture and storage (CCS) projects. These applications demand numerical methods capable of reducing computational costs without sacrificing predictive accuracy. The proposed methodology directly addresses this challenge by enabling efficient simulations on realistic field-scale models while remaining compatible with existing compositional reservoir simulators.

Perspectives