چكيده به لاتين
There are important challenges and technical difficulties in the production and transportation of heavy crude oils due to high viscosity of this type of crude oils. Therefore, several viscosity reduction methods has been proposed by researchers. One of these methods is dilution method. In dilution method, the crude oil is mixed with a low viscosity fluid which is called diluent. The viscosity of crude oil/diluent blend depends on several parameters including crude oil and diluent physical properties, and the diluent to crude oil ratio. Viscosity is an important parameter in designing the crude oil extraction and transportation equipment. Therefore, the development of reliable models and correlations for the estimation of crude oil/diluent blend is important. The main aim of the present study was the development of data driven models using Genetic Algorithm (GA), Gene Expression Programming (GEP), Artificial Neural Network (ANN) and Support Vector Regression (SVR) for the estimation of viscosity of heavy crude oil/diluent blends. 831 data points on the viscosity of blends (i.e., 698 weight fraction based data (0.185–156,862 cSt) and 133 volume fraction based data (0.412–165,860 cSt) were obtained from the literature.
The prediction results for the weight fraction based models in terms of the absolute average relative error (AARE %) were 7.28, 11.00, 13.97 and 8.63 for GA, GEP, ANN and SVR developed models, respectively. The prediction results for the volume fraction based models in terms of the absolute average relative error (AARE %) were 8.73, 13.94, 10.36 and 6.02 for GA, GEP, ANN and SVR developed models, respectively. Therefore, the best developed models for mass fraction based data and volume fraction based data were GA-based correlations and SVR models, respectively.
