Development of new gas condensate viscosity model using artificial intelligence

Research output: Contribution to journalArticlepeer-review

148 Downloads (Pure)


Accurate estimation of gas condensate fluid properties is a challenging task due to evolving the condensate liquid from the gas phase below the saturation pressure. Among the fluid properties viscosity of condensate liquid has the largest prediction uncertainty. The existing literature methods cannot cope with non-linearity and physics of gas condensate mixture (transition from single phase to two-phase) below the saturation pressure. Hence, in this study a novel and accurate condensate viscosity correlation as a function of pressure (P), temperature (T) and solution gas to oil ratio (Rs) was developed. For this purpose comprehensive data source of 1368 experimental data points acquired from open literature has been used. For developing new condensate viscosity correlation an Artificial Intelligence (AI) method known as Takagi – Sugeno – Kang (TSK) fuzzy algorithm was utilized. The accuracy of the developed correlation was compared with five previously published literature models. The superiority of new correlation over existing literature models is confirmed by statistical parameters of least root mean square error (RMSE) of 0.0194, mean average error (MAE) of 0.0163 and average absolute relative deviation percentage (AARD%) of 7.123. The proposed condensate viscosity correlation is valid for pressure rang of 0.25 – 75.84 MPa), temperature range of 303 – 443.15°K and Rs of 41.96 – 13496 scf/STB.
The proposed correlation can be used as an alternative approach to existing models for accurate gas condensate reservoir simulation studies.
Original languageEnglish
Article numberJKSUES581
JournalJournal of King Saud University, Engineering Sciences
Early online date17 Nov 2021
Publication statusE-pub ahead of print - 17 Nov 2021


Dive into the research topics of 'Development of new gas condensate viscosity model using artificial intelligence'. Together they form a unique fingerprint.

Cite this