Conductive fracture identification using neural networks

The fluid flow properties of many petroleum reservoirs and hazardous waste sites in low-permeability rock are dominated by a subset of fractures in the rock mass. Identification of these significant conductive features is critical for numerical models of fluid flow and contaminant transport. A series of excavations at the Kamaishi test facility on the Island of Honshu, Japan, were used to test various methods of conductive fracture identification. Encouraging performance was obtained from a backpropagation neural network, which demonstrated an ability to learn and apply the non-linear relationships between the geologic input variables and the conductive state of individual fractures.