QSPR molecular approach for representation/prediction of very large vapor pressure dataset

Farhad Gharagheizi (1) , Ali Eslamimanesh (2) , Poorandokht Ilani-Kashkouli (3) , Amir H. Mohammadi (2) , Dominique Richon (2)

1 Department of Chemical Engineering
2 CEP/Fontainebleau
3 Department of chemical engineering

Reliable estimation of vapor pressure is of great significance for chemical industry. In this communication, the capability of the Quantitative Structure-Property Relationship (QSPR) technique is studied to represent/predict the vapor pressure of pure chemical compounds from about 55 to around 3040. K. Around 45,000 vapor pressure values belonging to about 1500 chemical compounds (mostly organic ones) at different temperatures are treated in order to present a comprehensive, reliable, and predictive model. The sequential search mathematical method has been observed to be the only variable search method capable of selection of appropriate model parameters (molecular descriptors) regarding this extremely large data set. To develop the final model, a three-layer artificial neural network is optimized using the Levenberg-Marquardt (LM) optimization strategy. Through the developed QSPR model, the absolute average relative deviation of the represented/predicted properties from the applied data is about 7% and squared correlation coefficient is 0.990. In addition, the outliers of the model are identified using the Leverage Value Statistics method.

Mots clés: Computational chemistry; Optimization; Phase equilibria; Quantitative structure-property relationship; Thermodynamics process; Vapor pressure; Appropriate models; Data sets; Large data; Levenberg-Marquardt; Mathematical method; Molecular approach; Molecular descriptors; Optimization strategy; Predictive models

Domaines: Génie des procédés

Farhad Gharagheizi, Ali Eslamimanesh, Poorandokht Ilani-Kashkouli, Amir H. Mohammadi, Dominique Richon. QSPR molecular approach for representation/prediction of very large vapor pressure dataset. Chemical Engineering Science, 2012, 76, pp.99-107. ⟨10.1016/j.ces.2012.03.033⟩. ⟨hal-00796931⟩