A soft computing approach for prediction of P- ρ-T behavior of natural gas using adaptive neuro-fuzzy inference system

Density is an important property of natural gas required for the design of gas processing and reservoir simulation. Due to expensive measurement of density, industry tends to predict gas density through an EOS. However, all EOS are associated with uncertainties, especially at high-pressure conditions. Also, using sophisticated EOS in commercial software renders simulation highly time-consuming. This work aims to evaluate performance of adaptive neuro-fuzzy inference system (ANFIS) as a widely-accepted intelligent model for prediction of P-ρ-T behavior of natural gas. Using experimental data reported in the literature, our inference system was trained with 95 data of natural gas densities in the temperature range of (250-450)K and pressures up to 150 MPa. Additionally, prediction by ANFIS was compared with those of AGA8 and GERG04 which both are leading industrial EOS for calculation of natural gas density. It was observed that ANFIS predicts natural gas density with AARD% of 1.704; and is able to estimate gas density as accurate as sophisticated EOS. The proposed model is applicable for predicting gas density in the range of (250-450) K, (10-150) MPa and also for sweet gases, i.e., containing a low concentration of N2 and CO2.