Susan M.Schrader, The University of Texas of the Permian Basin, Robert S. Balch and Roger Ruan, New Mexico Tech Petroleum Recovery Research Center
Artificial neural networks (ANN) are computer programs designed to mimic the functioning of the human brain. ANNs can be designed to "learn" by reviewing a data set consisting of a set of known inputs and a corresponding set of desired outputs. Once an ANN has been trained, it can predict outputs given just the set of known inputs. While the applications are broad reaching, one valuable application of such tools is in petroleum exploration. In places where both exploration data (such as log, core or geophysical data) and production data are available, a network can be designed and trained and used to predict production given a similar suite of exploration data. This work will discuss design issues in exploration neural networks, explore available software and review two case studies where neural networks were used to predict total production for undrilled sites in two formations in the Permian Basin.