Novel analytical neural network modeling based on the corresponding state principle for determining the speed of sound in hydrocarbons

The main objective of this study is to propose a simple and accurate analytical model, directly derived from artificial neural network modeling, to predict the speed of sound in hydrocarbons is presented. The proposed model considers variables and properties readily available (pressure, temperature, critical properties and acentric factor). A total of 1000 experimental data points of speed of sound for 20 hydrocarbon substances were used in this study. These data were randomly divided into a training set (800 data points), a testing set (150 data points), and a prediction set (50 data points). The results were compared with experimental data and with values reported by other available estimation methods. The importance of the proposed analytical model is that, by keeping the good predicting capabilities of the trained network, new applications do not need to run the network software again but only apply a straightforward analytical model. Results show that the new analytical model gives low deviations and can be used with confidence in thermodynamic and engineering calculations.