A neural network for predicting saturated liquid density using genetic algorithm for pure and mixed refrigerants

In this study, a new approach for the auto-design of a neural network based on genetic algorithm (GA) has been used to predict saturated liquid density for 19 pure and 6 mixed refrigerants. The experimental data including Pitzer's acentric factor, reduced temperature and reduced saturated liquid density have been used to create a GA-ANN model. The results from the model are compared with the experimental data, Hankinson and Thomson and Riedel methods, and Spencer and Danner modification of Rackett methods. GA-ANN model is the best for the prediction of liquid density with an average of absolute percent deviation of 1.46 and 3.53 for 14 pure and 6 mixed refrigerants, respectively. (C) 2008 Elsevier Ltd and IIR. All rights reserved.