Abstract
Originally proposed for conventional reservoirs, the dual-continuum idealization is being applied directly to simulation in ultralow-permeability reservoirs. Recent gridding techniques can mesh the geometries of large discrete natural fracture networks (NFNs) into an unstructured grid to simulate fluid flow. Insufficient literature exists to help identify the tradeoffs in selecting one approach over another. This paper not only analyzes two approaches with respect to their underlying assumptions and applicability but also proposes a hybrid approach.
A commercial reservoir simulator supporting both dual-continuum and unstructured simulation grid capability was used to compare these approaches in two separate stages using CPU times and accuracy of results as metrics. First, small-scale reservoirs with simple two-dimensional (2D) fracture patterns were simulated to examine the impact of matrix permeability, fracture spacing, and fracture orientation. Second, drilling-spacing-unit-size reservoirs with networks of stochastically generated three-dimensional (3D) fracture surfaces were simulated to compare the effect of density and clustering of fractures. These models were also simulated using a hybrid approach, modeling one portion of the fracture network as discrete and the remaining as dual continuum.
Results show that although the dual-continuum technique can be very fast, it is not appropriate for all simulations. For example, extremely long transient periods in ultralow-permeability reservoirs raise doubts about the applicability of steady-state transfer functions in such models, whereas this assumption is considered acceptable at higher permeabilities. Conversely, discrete-fracture techniques, where the fracture geometry is resolved accurately in structured or unstructured mesh, are applicable to a broader set of problems but can become prohibitively slow for dense NFNs.
A variety of cases were explored that demonstrate the strengths and weaknesses of the two approaches. The proposed hybrid approach discussed in this paper offers a good balance between run times and the quality of results.
This paper offers practical insight into technique selection over a large class of problems. Furthermore, the proposed hybrid approach of combining dual and discrete NFN simulation in the same unstructured grid is novel but provides clearly demonstrated benefits. Stochastic generation of 3D NFNs, conditioned on secondary data, is used to control the clustering of natural fractures. Two distinct types of unstructured gridding methods, both supporting multi-point flux approximation, are used to mesh discrete NFNs.