Artificial neural network for the correlation and prediction of surface tension of refrigerants

Abstract An Artificial Neural Network model is proposed for the calculation and prediction of the surface tension of refrigerants along the liquid-vapor interphase. A total amount of 2879 data for 76 refrigerants were used for training and testing the network model. After considering different architectures, that one including three layers with 10 neurons in the inner layer, 10 neurons in the hidden layer, and one neuron in the output layer, using four dependent variables (the reduced temperature, critical temperature, critical pressure, and acentric factor) was found to give the best results. Overall mean absolute percentage deviation of 1.64% was found. Six corresponding-states based models were considered for comparison, the lowest overall deviation being 3.7%. Computer files needed to predict new values are given as Supplementary Material.