Rigorous prognostication of natural gas viscosity: Smart modeling and comparative study

The current study plays a major role in modeling natural gas viscosity in terms of several operating parameters including pseudo-reduced properties and molecular weight through radial basis function neural network (RBFNN), least-squares support vector machine (LSSVM), and multilayer perceptron neural network (MLPFNN). As it known, an important feature of any comprehensive modeling is the application of a large database for model development. Therefore, more than 3800 gas viscosity data points were used for modeling. For upgrading the efficiency of the abovementioned predictive tools, four optimization algorithms including levenberg-marquardt (LM), coupled simulating annealing (CSA), Bayesian regularization (BR), and scaled conjugate gradient (SCG), were integrated with them to find the optimal models' parameters during prediction analysis. Consequently, it was understood that among the all suggested tools in this study, the MLP-LM and then MLP-BR are the most accurate models for estimating gas viscosity with root mean square error (RMSE) of 0.001 and 0.002, respectively. Comparison of the MLP-LM and MLP-BR with previously published models in literature demonstrates their higher prediction capability, with less numbers of input parameters (without needing any density data), than the existing literature models. Based on the sensitivity analysis, it is concluded that the molecular weight is the most affecting variable on the viscosity prediction. Finally, the suggested tools in this study can be of great value for effective estimation of gas viscosity in simulating both upstream and downstream natural gas processes.