Modelling of Density of Organic Compounds Using QSPR Approach

The density of organic compounds is very important for the design of industrial plants,pipelines, and pumps [1]. However it is not always possible to find experimental values of thisproperty for the compounds of interest in the literature, and experimental measurements arelengthy and costly [2]. The two most commonly used methods of calculating PVT properties areequation-of-state (EOS) and PVT correlations. The EOSs are computationally demanding, inparticular for complex fluids, where they require adequate knowledge and representation of themolecular interactions. On the other hand, PVT correlations involve simple mathematicalcomputations and they only require readily available experimental data for a small number ofrepresentative compounds [3]. Therefore, predictive methods are generally employed in thissituation. Quantitative structure-property relationship (QSPR) study is one of the most widelyused methods to estimate various physical and chemical properties using some chemicalstructure based parameters [4]. In this work, we propose a QSPR approach in order to model thedensity of organic compounds over a wide range of temperature and pressure. The structuralgroup method was used to select the most important descriptors of compounds structure.Levenberg –Marquardt artificial neural network (ANN) was used to link molecular structuresand density data. The data set was randomly divided into three data sets: training, validation andtest set. After training and optimization of the ANN parameters, the performance of the modelwas investigated by the test set. The result indicates that optimized model can simulate therelationship between the selected descriptors and the density accurately.

کلیدواژه ها:

Organic Compounds ، Artificial Neural Network (ANN) ، Density ، Quantitative Structure-Property Relationship (QSPR)