Machine learning model to predicting synergy of ultrasonication and solvation impacts on crude oil viscosity

The growing energy demand is attracting a lot of interest in heavy crude oil. While various conventional methods have been employed to reduce the viscosity of crude oil, the challenges associated with these techniques have driven researchers to seek economical and environmentally sustainable alternatives. In this context, combining traditional methods, such as solvent treatment, with ultrasonic radiation presents a promising, yet complex solution. In this study, we develop a machine learning-based algorithm to rigorously predict the synergistic effects of ultrasonication and solvation on crude oil viscosity. This research is divided into two parts. First, we conducted experimental measurements of crude oil viscosity following treatment with ultrasound and/or n-heptane solvent. It was found that the optimum irradiation period for samples containing 0% to 16% n-heptane was 8 min. Notably, after 8 min of treatment, the viscosity decreased by over 34% for samples with 16% n-heptane and by more than 47% for those with 0% n-heptane. When the n-heptane concentration was increased to 22% and 30%, the required sonication time increased by 2 to 10 min. In the former case, viscosity was reduced by more than 50%, while in the latter, it decreased by over 48%. In the second part of the study, a Machine Learning (ML) model was developed using the experimental data. In particular, a Random Forest Regressor (RFR) model was applied, with results demonstrating high reliability based on RMSE and R² values for training (3.3395, 0.9764), validation (3.0166, 0.9602), and testing (2.4778, 0.9557). Considering the significance of features, n-Heptane (0.54) and irradiation time (0.45) were the key predictors, with n-heptane showing slightly greater impact on error reduction. The application of this model to additional datasets from other oil fields shows significant promise for future research and practical implementation.