Development of soft computing methods to predict moisture content of natural gases

• Water content of natural gas was accurately predicted using novel network-based fuzzy logic models.
• The new models were tested and validated using 420 data.
• The models show excellent agreement with experimental data.
• Radial basis function (RBF) neural network provided more accurate results than other models.

In this paper, several numerical models have been presented for predicting the water content of natural gases in equilibrium with liquid water. Machine learning approaches including multilayer perceptron (MLP) neural network, radial basis function (RBF) neural network, and least squares support vector machine (LSSVM) algorithm have been utilized for precise determination of water content of natural gases.The presented models work for pressures up to 69 MPa and temperatures between 298.15 and 450.15 K as well as acid gas mole fractions up to 0.4. With accordance to the error analysis results it was found that the proposed LSSVM, RBF, and MLP models reproduce targets with the average absolute relative deviations (%AARD) being less than 2.8%, 4.1%, and 7.7%, respectively. Coefficients of determination values of the developed models are found to be greater than 0.99, illustrating good association of the predictions with corresponding reported data in the literature.

Figure optionsDownload as PowerPoint slide