Prediction of petroleum viscosity from molecular weight and density

165 crude oils with viscosity, density, and molecular weight variation in the range 0.54 - 24135cP; 0.746 - 1.016 g/cm3; 117-579 g/mol respectively were examined for viscosity prediction using eight available in the literature models and three more, developed in this work models. The best empirical model was that of Sinha et al., 2020 with % AAD (absolute average deviation) = 18.2 %, The ANN (artificial neural network) model for the data set of the 165 crude oils outperformed the empirical correlations with % AAD = 17.7 %. 93 crude oils with viscosity, density, molecular weight, and SARA composition data variation in the range 2.3 - 23 000cP; 0.819 - 0.992 g/cm3; 179-579 g/mol; Sat.: 26.0-79.3 %; Aro:11.9-52.8 %; Res.: 2.5-30.9; Asp.:0.1-19.6 % respectively were also examined for viscosity prediction by the available in the literature empirical correlations and another new developed empirical correlation that includes besides molecular weight and density, the crude oil saturate content. The best empirical model was that developed in this work with saturate content inclusion, that showed % AAD = 23.8 %. The ANN model for the data set of 93 crude oils again outperformed the empirical correlations with % AAD = 18.8 %. The most accurate model predicting viscosity was found the new developed in this work model on the base of a reference viscosity at a particular temperature and molecular weight with % AAD = 2.5 %.

Automated summary

This study predicts the viscosity of 165 crude oils with varying viscosity, density, and molecular weight, using eight existing models and three new models developed in this study. The results show that the artificial neural network (ANN) model performs better than the empirical correlations. Additionally, the study examines the viscosity prediction of 93 crude oils with varying viscosity, density, molecular weight, and SARA composition using existing empirical correlations and a new empirical correlation that also considers the crude oil saturate content. The results indicate that the model developed in this study, incorporating the saturate content, has the highest prediction accuracy.