Artificial neural networks modeling for hydrocarbon gas viscosity and density estimation

Abstract

Natural gas is a naturally occurring petroleum product and one of the major fossil energy sources. It is composed of a complex mixture of hydrocarbon compounds and a minor amount of inorganic compounds. Hydrocarbon gas properties of viscosity and density are of great importance for gas engineering calculations. These properties are measured experimentally but if unavailable, they can be predicted through different correlations. This work is aimed at developing new models for gas viscosity and gas density using generalized regression neural (GRN) networks. A large database of experimental measurements were gathered from the literature and used to develop and test the models. The database consists of gas composition, measured viscosity and density, temperature, pressure, and compressibility factor of different hydrocarbon gases and pure and impure gas mixtures containing up to pentane plus fractions and small concentrations of non-hydrocarbon components. A total of 4445 experimental measurements were used in this study constituting of 1853 pure gases and 2592 gas mixtures. Two neural nets were trained and tested separately to predict gas viscosity and gas density. Viscosity is predicted as a function of gas density, pseudo reduced pressure, and pseudo reduced temperature while density is predicted as a function of molecular weight, pseudo reduced pressure, and pseudo reduced temperature. The two neural networks were trained and validated using a set of 800 data points chosen randomly from the collected data set. The developed networks were blind tested using a total of 3645 data points. The networks prediction was validated and their efficiencies were tested against some other correlations. The comparison indicates a better performance for the developed neural networks compared to the conventional tested correlations with an average absolute error of 3.65% and 4.93% for gas viscosity and gas density nets, respectively.