Flow-Based Upscaling of Fractured Porous Media Using a Discrete Fracture and Matrix Model and a Fracture Merging Algorithm

Abstract

Abstract Many geo-energy related applications involve predicting the behavior of fluid flow in fractured subsurface reservoirs. Naturally fractured carbonate reservoirs are particularly important for being a major source of the world's hydrocarbon production. These reservoirs are also currently being considered as potential CO2 storage sites that will support net zero emissions goal. Simulation of flow in fractured reservoirs is a challenging task that typically involves upscaling the effective permeability of the fracture network and matrix into continuum models that consider the reservoir scale. The most accurate way to obtain such upscaled permeability for fracture networks is to perform single-phase flow simulations in statistical realizations of the fracture network using three-dimensional unstructured grids and explicit modelling of fractures. This step can be computationally challenging for highly dense fracture networks due to the difficulty in meshing the fractures and the rock matrix. Here, we present a method to reduce the complexity of the fracture network while still preserving the behavior of its effective permeability. Our approach involves a fracture merging algorithm that reduces the number of fractures allowing for faster meshing and upscaling. The fracture merging algorithm uses three different similarity metrics: fracture orientation, fracture area and distance between fractures. These metrics are used to identify similar fractures that can be merged into one single fracture with increased permeability. The upscaling algorithm to obtain the effective permeability of a grid cell containing a fracture network relies on flow simulations in three-dimensional unstructured meshes. We applied our method to different sub-networks extracted from a stochastically generated fracture network of a Brazilian Pre-Salt carbonate reservoir. We found that the average permeability of all fractures of the resulting fracture network increases with merging intensity, i.e., with decreasing the number of fractures, while the resulting upscaled effective permeability for the network remains in the same order of magnitude. This shows that the flow-based upscaling workflow including the merging algorithm leads to a significant reduction of complexity of fracture networks and consequently their 3D unstructured meshes while maintaining the structural and topological features that account for the fracture network effective permeability. Our proposed method is simple to implement and relies only on geometrical properties of the fractures. Other machine-learning based models have been proposed to achieve similar simplification of fracture networks, however, they are not easily incorporated into existing reservoir simulation tools and codes like the method presented in this work. Moreover, such previously published approaches do not consider flow in matrix and thus haven't been tested in scenarios where the matrix also contributes to flow.