An inverse technique for developing models for fluid flow in fracture systems using simulated annealing

Abstract

One of the characteristics of flow and transport in fractured rock is that the flow may be largely confined to a poorly connected network of fractures. In order to represent this condition, Lawrence Berkeley Laboratory has been developing a new type of fracture hydrology model called an “equivalent discontinuum” model. In this model we represent the discontinuous nature of the problem through flow on a partially filled lattice. This is done through a statistical inverse technique called “simulated annealing.” The fracture network model is “annealed” by continually modifying a base model, or “template,” so that with each modification, the model behaves more and more like the observed system. This template is constructed using geological and geophysical data to identify the regions that possibly conduct fluid and the probable orientations of channels that conduct fluid. In order to see how the simulated annealing algorithm works, we have developed a synthetic case. In this case, the geometry of the fracture network is completely known, so that the results of annealing to steady state data can be evaluated absolutely. We also analyze field data from the Migration Experiment at the Grimsel Rock Laboratory in Switzerland.