Prediction of natural gas density using only three measurable properties: intelligence and mathematical approaches

Abstract

ABSTRACT Accurate determination of natural gas (NG) density is a very important issue in the NG custody transfer. In the conventional methods for density calculation such as laboratory methods and equation of state (EoS), the temperature, pressure, and NG composition are required as input parameters. However, measuring NG composition is a complicated and costly procedure. To overcome this problem, two novel approaches are proposed to calculate density without the need to measure NG composition. In these approaches, speed of sound, pressure, and temperature as three simple measurable properties are introduced as input variables. The main approach is developed based on the artificial neural network (ANN). Moreover, a mathematical correlation is also developed as the alternative approach. The results of these two approaches are validated by comparing them with experimental data. The validation results show that the average absolute percent deviation (AAPD) and root mean square error (RMSE) is 1.94% and 2.88 for the ANN approach and are 2.54% and 3.82 for the correlation approach. The results show that the ANN approach has high precision and the correlation approach has acceptable accuracy. On the other hand, the density calculations using these approaches have a significant error at low temperature and high pressure.