Abstract

One main challenge in reservoir modeling is the computational costs of flow simulation. Although computer power has significantly increased in the past few decades, effective upscaling to reduce computation time in heterogeneous reservoirs still remains a challenge.

This research demonstrates the results of a comprehensive study on common upscaling techniques for porosity and permeability. This work aims at providing guidelines on which upscaling methods are superior for different depletion/production processes. The processes investigated include straight depletion, water injection, miscible and immiscible gas injection.

Several fine grid heterogeneous models were developed using actual field data and heterogeneity was evaluated using Lorenz and Dykstra Parsons Coefficients. Those different models were used to generate benchmark results for comparison with coarser models developed at different levels of coarsening and with different upscaling algorithms. Several scenarios were investigated in order to identify which upscaling algorithms result in the closest match with the fine model results. The results of coarse models were treated statistically to quantify their deviation from the fine models results and rank the upscaling algorithms accordingly. It was found that the depletion process affects the choice of the appropriate upscaling algorithm.

The coarse grid models lose some accuracy with increasing the upscaling ratio, as suspected. Also we found that several upscaling algorithms work equally well in depletion processes. However, only flow-based upscaling techniques (especially with open boundary condition) worked well for water injection processes. For miscible and immiscible gas injection processes, the use of flow based upscaling techniques is generally a must, while the flow-based upscaling technique with closed boundary conditions shows slightly better results than other flow-based upscaling techniques.

For miscible gas injection processes, in particular, the use of which upscaling technique is sensitive to the heterogeneity index as well specially for models with Lorenze Coefficient higher than 0.5.

You can access this article if you purchase or spend a download.