An artificial neural network for predicting and optimizing immiscible flood performance in heterogeneous reservoirs

Enhanced Oil Recovery (EOR) processes are used to recover additional oil left in place after primary recovery and water flooding stages. The prediction of their performance is of great importance in the selection and design of a certain EOR process and also during planning for oil production. This work presents an extension of an earlier attempt on the use of neural networks to predict reservoir performance in homogeneous reservoirs. The consideration of heterogeneity and its accompanying interactions with the fluid flow equations in a porous media is considered in this paper. The independent dimensionless groups that characterize the flow behavior in a heterogeneous media have been used as inputs to a neural network model in order to predict oil recoveries. Various neural network architectures have been considered, and the network that best mimics a reservoir numerical simulator was retained. The simulations of this network are compared to those obtained from the reservoir simulator. The effect of various dimensionless groups on the oil recovery is discussed and an optimization study is performed to determine from the prepared neural network the optimal conditions leading to the best oil recovery efficiencies. (C) 1998 Elsevier Science Ltd. All rights reserved.