A Dual Representation for Multiscale Fracture Network Characterization and Modeling
As shown by field studies, the size distribution of natural fractures generally follows a power or exponential law. Large fractures are sparsely distributed, often cut across several stratigraphic units and act as major flow pathways. Conversely, small fractures densely populate the reservoir, are often confined to specific beds and mainly affect the reservoir anisotropy and large scale heterogeneity. For characterizing the reservoir connectivity and understanding its dynamic behavior, large features (conductive faults, fracture clusters) should therefore ideally be modeled explicitly while smaller fractures can efficiently and accurately be represented by effective properties.
We have developed a new technique in which fracture statistical attributes such as density and orientation measured along well paths and interpolated in 3D using seismic and structural information are represented by a set of regionalized random variables at the reservoir scale. These variables are described by probability distribution laws and can be embedded into various geometrical supports (maps, 3D grids), defining a continuous fracture model (CFM). Discrete fractures interpreted on borehole or seismic images can also be integrated.
The CFM can then be converted, in part or in full, to a discrete fracture network (DFN) through an object-based stochastic simulation. However, since an excessive number of discrete objects (typically tens of billions) would be required to discretize the whole fracture network and retain the high frequency vertical heterogeneity, the proposed solution consists in simulating explicitly only those fractures assumed to have a significant impact on the reservoir connectivity. The limit between DFN and CFM is described by a fracture size parameter, below which are the small fractures modeled by the CFM, whereas the larger fractures are represented through the DFN. The size is defined by an arbitrary threshold that can vary in space (e.g. as a function of the distance to wells). The final fracture network model is thus composed of the combination of both CFM and DFN visualized, edited and upscaled simultaneously. The CFM upscaling and its calibration to dynamic flow data are performed by arithmetic operations on random variables.
We describe the implementation of the dual model and compare it against a purely discrete model, highlighting its advantages in terms of memory footprint, processing time, accuracy and scalability, especially with large models.
AAPG Search and Discovery Article #90090©2009 AAPG Annual Convention and Exhibition, Denver, Colorado, June 7-10, 2009