General Model Based on Artificial Neural Networks for Estimating the Viscosities of Oxygenated Fuels

Abstract

Oxygenated fuel is a promising alternative fuel for engines because of the advantage of low emission. In this work, a general model based on back-propagation neural networks was developed for estimating the viscosities of different kinds of oxygenated fuels including esters, alcohols, and ethers, whose input variables are pressure, temperature, critical pressure, critical temperature, molar mass, and acentric factor. The viscosity data of 31 oxygenated fuels (1574 points) at temperatures ranging from 243.15 to 413.15 K and at pressures ranging from 0.1 to 200 MPa were collected to train and test the back-propagation neural network model. The comparison result shows that the predictions of the proposed back-propagation neural network model agree well with the experimental viscosity data of all studied oxygenated fuels using the general parameters (weight and bias). The average absolute relative deviations for training data, validation data, and testing data are 1.19%, 1.27%, and 1.30%, respectively.