A novel investigation based on the tree‐based machine learning methods on rheological behaviour of waxy crude oils

Abstract Flow assurance issues concerning wax precipitation in crude oil transportation pipelines make it necessary to predict the flow behaviour at different operating conditions. This work attempted to model measured shear stress and viscosity of waxy crude oils using tree‐based machine learning methods and consider wax content, additives, and solvent concentration as input parameters of models. Amongst all implemented techniques, the Extra trees model performed as a potential estimator in waxy oils rheology studies. Two models were run using the final values collected during hyperparameter tuning for shear stress and viscosity. Results show a root mean squared error value of 14.15 and a coefficient of determination ( R 2 ) of 0.998. RMSE of shear stress training dataset was reduced to 9.50 from 35.46 by adjusting hyperparameters of the model. The assessed techniques encompass decision tree, extra trees, gradient boosting, light gradient boosting machine, linear regression, random forest, ridge regression, and XGBoost. Also, in this research, three rheological models of power law, Dekee and Casson models have been used to correlate apparent viscosity values, and it was concluded that Dekee and Casson models have shown an acceptable match with experimental data. Also, the rheological behaviour of three crude oils in the absence/presence of flow improvers was investigated, and it was concluded that ethylene‐vinyl acetate (EVA) copolymer has performance in changing the flow behaviour from non‐Newtonian to Newtonian even at temperatures below WAT. Moreover, the addition of small quantities of asphaltene solvents, such as toluene, can improve the viscosity of crude oil with high wax content.