In order to select the optimum solvents of extractive distillation on the basis of molecular simulation

a mathematical model must be firstly established for describing the quantitative structure-property relationships (QSPR) of the solvents.In this paper the molecular connectivity indexes (MCI) were used to express two-dimensional molecular topological structure

which could indicate not only the kind and number of atoms or groups composing the solvent molecule but also the inter-connective characteristics among them.The major solvent properties

such as selectivity and solubility were the functions of infinite dilution activity coefficients

γ∞.Because of the extraordinary complexity between molecular structure and γ∞

an artificial neural network (ANN) model was used for the QSPR.An appropriate database was established and the back-propagation (BP) algorithm was used to train the ANN model

and the well-trained ANN model proved to be more accurate in calculating γ∞ than the prevailing UNIFAC method.