Machine learning analyses of low salinity effect due to surface-charge alteration in carbonate porous media

Surface charge alteration is a microscopic mechanism responsible for wettability alteration toward a less-oil-wet state and increasing the recovery factor in low salinity water flooding (LSWF). However, it is difficult to understand the exact relationship between different controlling parameters such as oil, water, rock properties and operational parameters on rock surface charge in carbonates. The goal of this work is to investigate the effects of oil, brine, rock and operational parameters on surface charge of reservoir rocks by applying non-linear machine learning models such as Multilayer Perceptron (MLP) and Radial basis function (RBF) neural networks. To do this, about 900 datasets from zeta potential measurements were used. A sensitivity analysis has been performed in this study in order to identify the best model structure by finding the optimal number of hidden layers for these models. The greatest accuracy was obtained using 19 and 17 neurons in the hidden layer for RBF and MLP model, respectively. Also, both models demonstrated the highest prediction efficiency with the trainbr and trainlm algorithms. Results at training data size of 80% with the optimized models, showed that the root means square error (RMSE) and coefficient of determination (R ² ) of the models were 2.90, 0.96 (RBF), and 5.91, 0.81 (MLP), indicating that the RBF model was more efficient than the MLP model. According to the statistical analysis results, it can be deduced that proper agreements exist between the actual and predicted results of zeta potential in the train and test phases of RBF model. Hence, it is proposed that the developed model can be a useful tool for a quick estimate of the rock zeta potential in a wide range of oil, water, rock properties and operational parameters.