Comparison of Z-Factor Correlations Using a Large PVT Dataset with Emphasis on the GERG-2008 EOS

The gas compressibility factor is an important property in reservoir simulation studies. It is directly linked to the gas formation volume factor and the gas density thereby impacting wellhead injection pressure, reservoir voidage, injectivity, as well as the tendency for gas gravity override to occur in the reservoir. ADNOC's PVT database contains experiments on almost 2,000 samples, of which more than 100 have been subject to advanced gas injection experiments. Z-factor data have been compiled from the liberated gas during DV experiments as well as from CCE experiments on reservoir gases, injection gases, and swollen fluid mixtures. Several of these mixtures are very rich in H2S, whereas pressure and temperature are in the range of 14.7-14,500 psia and 80-365 °F, respectively. We test several different methods for predicting the Z-factor, such as the industry-standard Hall-Yarborough method, in combination with various models for pseudo-critical pressure and temperature and including correction for non-hydrocarbon components. Other methods tested include the GERG-2008 model, considered to be state-of-the-art for predicting physical properties for well-described gas mixtures, as well as the well-known Peng-Robinson cubic equation of state. Based on close to 10,000 data points in our database, the GERG-2008 model typically predicts the Z-factor to be within 2% of the measured value, which is on par with the experimental uncertainty. However, for some rich gas condensate mixtures, the model gives larger errors because its parameters are only tuned to compositions with components up to C10. This is to our knowledge the first time that the GERG-2008 EOS has been compared to standard Z-factor correlations for such a large number of data points. If compositional information is available, we recommend using either the GERG-2008 model or the Hall-Yarborough model with pseudo-critical properties provided by Kay (1936). When compositions are not available, we find that the Standing correlation is more accurate than the Sutton model, also for sour mixtures.