Prediction of Wettability Alteration Using the Artificial Neural Networks in the Salinity Control of Water Injection in Carbonate Reservoirs

Abstract

Artificial Neural Networks (ANN) applications have grown exponentially in all areas of science and technology. The advantages are its versatility, speed and ability to aggregate information, perform predictions of a given set of data. These attributes attract the petroleum industry, which often depends on laboratory analysis or numerical simulation to estimate various reservoir behaviors. This research, aims to predict the relative permeability curves with wettability alteration effect, given a concentration of the ionic composition in water injection. For this, machine learning methods were applied. An analytical algorithm was developed that incorporated the effect of wettability alteration, generating the database for the training process. Two different networks were applied: (i) Self-Organizing Maps - SOM and (ii) Neural Net Fitting – NNF. The forecast data of the networks are compared with calculated for analytical results. This ANN performs a good forecast of data tested (NNF with R-squared results around 90%). The analyses confirm effects on relative permeability of oil and water with salt control, indicating wettability alteration (WA). These tests were able to confirm that the applied methodology is capable to predict, using ANN, results of several laboratory tests.