A neural network prediction model of fluid displacements in porous media

This paper presents the development and design of an artificial neural network that is able to predict the breakthrough oil recovery of immiscible displacement of oil by water in a two-dimensional vertical cross section. The data used in training the neural network was obtained from the results of fine-mesh numerical simulations. Several network architectures were investigated and trained using the back propagation with momentum algorithm. The neural network that gave the best predictive performance was a two-hidden layer network with 8 neurons in the first hidden layer and 8 neurons in the second hidden layer. This network also performed well against a cross validation test. The reservoir simulation data used so far in the training process was for a homogeneous reservoir, the more general case is still under investigation.