Realistic modeling of fracture networks in a giant carbonate reservoir

Abstract Naturally fractured reservoirs (NFRs) account for a significant fraction of the world's petroleum reserves but pose significant challenges for reservoir characterization and simulation. In this work, we develop a novel workflow for the geologically realistic modeling of a fractured carbonate reservoir. We first train a neural network with variables such as porosity and structure to predict fracture densities everywhere within a sector. The neural network is conditioned to honor well fracture densities inferred from well-based data. A fracture network is then stochastically generated by specifying its distributions of fracture dimensions and orientation. We next create an unstructured tetrahedral mesh that conforms to the hundreds of fractures, which are explicitly represented in the resulting discrete fracture model (DFM). Unlike discrete fracture network (DFN) models, which only include a discrete representation of fractures, our DFM also accounts for fracture-matrix and matrix-matrix flow explicitly. We demonstrate the utility of our workflow by performing fully-compositional simulations of miscible gas injection and primary depletion on the DFM. Six hydrocarbon pseudocomponents are used to represent the miscible fluids and a connection-list-based simulator is employed in these simulations. To our knowledge, compositional simulations with realistic DFMs have not been previously reported. From our study, we find that matrix permeability, which is a key uncertainty in the characterization of NFRs, has a very significant impact on the recovery efficiency of both processes. It is also found that the presence of well-connected fractures could alter the relative attractiveness of these recovery processes. The DFM simulations are, however, computationally intensive and not well-suited for full-field studies. To address this issue, we apply the recently-developed multiple subregion (MSR) procedure to upscale the DFM to a generalized dualporosity representation. We show that this approach provides results in agreement with the DFM but with substantial computational speedups. The MSR upscaling procedure thus offers a promising means for performing fast and accurate simulations of NFRs. Finally, reference solutions from DFMs could be used to calibrate parameters for full-field single- and dual-porosity simulations. 1. Introduction The oil field of interest is an isolated carbonate platform that consists of a central area of flat-lying strata surrounded by a raised rim, beyond which lies a depositional slope. Each of these regions shows distinct reservoir properties due to differences in depositional facies and diagenetic effects, which impact reservoir quality and production characteristics. The central platform is relatively unfractured, comprising grainstone, packstone, and wackestone while the rim and slope regions are largely made up of boundstone with significant fracturing. Natural fractures pose significant challenges in regards to reservoir characterization and flow simulation of the field.