Unified artificial neural network-group contribution method for predictions of normal boiling point and critical temperature of refrigerants and related compounds

Abstract

In this study, both normal boiling point and critical temperature of refrigerants and related compounds are predicted only from their molecular structures using a simple and unified Artificial Neural Network - Group Contribution Method. Identical 32 (including molecular mass) groups and methodologies have been used with 251 experimental data for TB and 132 experimental data for TC. In spite of its simplicity, the agreements between experimental and ANN predicted data for TB and TC are very good, better than most of the existing models. The percentage errors for training and test data sets are 2.4% and 3.7% and 2.8% and 5.7% for TB and TC respectively. The overall percentage errors for TB and TC are 2.8% and 3.7% respectively. A comparison of the proposed models with other models shows that for the class of compounds considered i.e., refrigerants and related compounds, this model predicts most accurately. These models can be conveniently used for any preliminary screening of compounds as alternative refrigerants or working fluids or for any other applications.