A Data-Based Continuous and Predictive Viscosity Model for the Oil-Surfactant-Brine Microemulsion Phase

Abstract

This study presents a computationally produced data-based model/correlation that can accurately estimate the magnitude and predict the peaks of microemulsion viscosity at dynamic reservoir conditions. Equilibrium molecular dynamics (MD) simulation is used on a decane-SDS-brine interfacial system to generate a dataset of viscosity values as a function of different temperatures, surfactant concentrations, and salinities. The viscosity testing and training data are computationally measured using the Einstein relation of the Green-Kubo formula. Several machine learning (ML) based regression algorithms, including K-nearest Neighbors (KNN), Support Vector regression (SVR), Multivariate Polynomial Regression (MLPR), Light Gradient Boosting Machine (LGBM), and Decision Tree (DT), are used to train the model. The SVR regression provides the best performputaance for our model compared to other methods with an R2 (0.978 and 0.963 for train and test data, respectively) and mean absolute error value (0.059 and 0.072 for train and test data, respectively). The chosen model is then used to predict microemulsion viscosity for different reservoir conditions. The proposed model aims to accurately estimate microemulsion viscosity at dynamic reservoir conditions with variable input parameters such as pressure, temperature, brine salinity, and surfactant concentration, enabling accurate estimation and prediction of the transport properties of reservoir fluids and present phases at reservoir conditions, which is key to achieving maximum recovery during chemical EOR.