Prediction of solubility for polycyclic aromatic hydrocarbons in supercritical carbon dioxide using wavelet neural networks in quantitative structure property relationship

In this study, a wavelet neural network (WNN) model is proposed to predict the solubility of naphthalene, biphenyl, fluorene, phenanthrene and triphenylene in supercritical carbon dioxide (SC-CO2), over a temperature range of 308–333 K and a pressure range of 80–135 bar for the first time. The WNN model was constructed in quantitative structure property relationship (QSPR) using six descriptors consisting of temperature, pressure, volume of the molecule, highest occupied molecular orbital (HOMO), dipole moment and number of double bonds in the molecules. These descriptors are selected, in a stepwise manner, from many different descriptors using multiple linear regression (MLR) method. The capability of the model was evaluated by plotting experimental values of solubility against the predicted values by the model for the prediction set. The large correlation coefficient 0.996, large value of F, 1947, and a small standard error of 0.087 reveals the capability of the model.