Abstract: |
Stochastic discrete fracture networks (DFNs) are commonly created as models of natural fracture systems. Traditional methods for creating discrete fracture networks (DFNs) simulate fracture centroids and orientation randomly (using a Poisson process) from independent user defined distributions. DFNs created using these techniques do not necessarily reproduce fracture spacing and local fracture orientation distributions observed in outcrops, images or other data. This may lead to fracture networks with unrealistic spatial properties, such as fractures that are too close together or too far apart. Moreover, unlikely scenarios are usually apparent in the DFNs such as fractures from the same fracture set intersecting in an unrealistic manner. In this paper, natural fracture networks and DFNs created using traditional methods are compared by the number of fracture intersections measured and distributions of local fracture spacing and orientation. The analysis shows that some natural fracture patterns cannot be reproduced using the traditional methods. A new approach to DFN simulation is presented. The proposed algorithm works by simulating a large pool of fractures and finding a subset that minimizes an objective function. An initial DFN is generated by selecting a random subset with the correct intensity from the pool of fractures. Then fractures are randomly added or removed from the DFN and an objective function is evaluated. The objective function is calculated as the sum of the difference between target and actual distributions of local fracture spacing, orientation, intensity and the number of intersections. Changes to the DFN are accepted if they decrease the objective function and rejected otherwise. Using the proposed approach, fracture networks that better reflect the true local fracture orientations and spacing can be produced. A number of examples are shown. |
