A Method for Rapidly Predicting Recovery of Multi-Layer Oilfields Developed by Water-Flooding

In order to quickly and accurately predict the recovery of oilfields developed by water flooding, a method for rapidly predicting oil recovery was established based on a back propagation (BP) neural network optimization algorithm with consideration of factors influencing the recovery, such as reservoir characteristics and fluid properties. Firstly, geological models for numerical reservoir simulation were constructed according to the geological characteristics and fluid properties of Penglai 19-3 Oilfield. Four key factors including coefficient of permeability variation, oil viscosity, net to gross ratio of oil layers, and production pressure differential were selected, with each factor defined into five levels. 625 groups of reservoir simulation cases were analyzed numerically, and a database indicating the relationship between the oil recovery of the cases and the influencing factors was established. Secondly, an artificial neural network (ANN) method for rapidly predicting oil recovery was set up based on BP neural network and optimization theory. Finally, 500 groups of data were selected as the algorithm training set, and 125 groups of data were tested for recovery predicting. The test result showed that the predicted oil recovery of the tested data had a relative error ranging from −2.91% to 5.07% with an average relative error of 0.16%, which met the requirement for engineering accuracy. The method for rapidly predicting recovery of multi-layer oilfields developed by water-flooding provides a new technical approach to rapidly predict the recovery of Penglai 19-3 Oilfield and other similar oilfields.