A data driven predictive viscosity model for the microemulsion phase

The changes in phase viscosity at the oil-brine interface due to surfactant addition are critical under practical reservoir conditions. This study develops a computational, data-driven model to accurately estimate and predict peak phase viscosity in microemulsion systems at dynamic environments. Using equilibrium molecular dynamics (MD) simulations, we investigate a decane-sodium dodecyl sulfate (SDS)-brine system, generating viscosity data as of temperatures, pressures, surfactant concentrations, and salinities. The data, computed via the Einstein relation and Green-Kubo formula, provides robust training and test datasets for model development. Various machine learning (ML) based regression algorithms are employed on our dataset to train and fit the model. This study aims to compare the accuracy and correlation coefficients of these models, selecting the most precise model for predicting microemulsion phase viscosity under diverse reservoir conditions. Support Vector Regression (SVR) outperformed other models with an R² of 0.978 and 0.963 and mean absolute errors of 0.059 and 0.072 for training and test datasets, respectively. Unlike traditional empirical viscosity correlations, this model incorporates physics-based relationships, enhancing its adaptability to varying reservoir conditions. The proposed model accurately predicts microemulsion phase viscosity, including peak viscosity locations, across pressures, temperatures, salinities, and surfactant concentration. This work facilitates precise viscosity estimation, improving recovery efficiency under reservoir conditions.