On modeling of bitumen/n-tetradecane mixture viscosity: Application in solvent-assisted recovery method

Abstract

Extremely high viscosity of bitumen is of great challenge in bitumen production, transportation, and processing. Therefore, different techniques such as dilution of bitumen by liquid solvents have been proposed for bitumen viscosity reduction. Accurate prediction of diluted bitumen mixture viscosity is a key parameter in solvent injection during hydrocarbon recovery. In this study, a simple correlation and a model in accordance with adaptive neuro-fuzzy inference systems joined with particle swarm optimization (ANFIS-PSO) algorithm have been developed for accurate estimation of bitumen/n-tetradecane mixture viscosity in comparison to other available mixing rules at different pressures, temperatures, and n-tetradecane mass fractions. For evaluation of the models efficiency, cumulative frequency test was employed to examine the data frequency over the whole databank. The ANFIS-PSO model exhibits the best performance with average absolute relative deviation of 2.44% and squared correlation coefficient (R2) of 0.9985. Moreover, it is proved that the suggested correlation has the best practical performance than the Lederer and other studied correlations in this work. In other words, the proposed correlation is easy-to-use and also, it is a more straightforward function of pressure, temperature, and solvent mass fraction than other literature correlations. Based on the sensitivity analysis, it is demonstrated that solvent concentration is the most influencing parameter in reducing bitumen viscosity. In accordance to the results of this study, the proposed techniques can serve as robust and reliable tools for efficient estimation of bitumen/n-tetradecane mixture viscosity.