Development of a new model to predict gas viscosity using artificial neural networks

Prediction of viscosity of natural gas is an important parameter in the oil and gas industry as it has a major effect on gas engineering calculations including reservoir recovery, fluid flow, deliverability, and storage. Gas viscosity is determined directly by experiment, but if unavailable, predicted by empirical correlations. An accurate prediction of natural gas viscosity is needed in the appropriate design and operation of equipment in industrials and processing. In this work, by means of an artificial neural network which is a branch of Artificial Intelligence and a powerful tool for prediction, gas viscosity of hydrocarbon mixtures is predicted in a wide range of pressure and temperature. Therefore, different networks with variable layers and neurons in each layer are trained and tested by different training algorithms. Then, the network with minimum error values and maximum rate of convergence is chosen as the best network for gas viscosity prediction. Results showed that the network with two hidden layers trained by the Levenberg-Marquardt training algorithm has an average absolute error of 0.001 for the training and validation data set. This network is chosen as the best network for prediction of gas viscosity using reduced temperature, reduced pressure and gas density as input parameters.