Estimation of activity coefficients at infinite dilution of halocarbons in water and organic compounds in hydrofluoroparaffins using neural networks

Infinite dilution activity coeffcents ( 1) of 19 halocarbons in water and 18 organic compounds in 5 hydrofluoroparaffins over a temperature range of 291 333K have been estimated by the method of neural nets. The network was trained using 351 data points covering 18 halocarbons in water and 17 organic compounds in 4 hydrofluoroparaffins, and generalization test was carried out using the remaining 92 data points covering one halocarbon in water, 17 organic compounds in a hydrofluroparaffin, and an organic compound in 5 hydrofluoroparaffins. Seven inputs are included in the network, namely the temperature, dipole moment, molar refraction and critical pressure of the solute and solvent. The network represented the data with an overall average absolute deviation of 11.8% estimated on the basis of the values of 1. The result obtained could be considered reasonable in view of the wide variations in the values of the input properties and 1 values and the varied nature and size of compounds treated. Among the several variables studied, dipole moment contributed to the correlation the most, followed by the critical pressure. Further study on the applicability of MLR (multilinear regression) method revealed the overwhelming superiority of the neural network over MLR.