Artificial neural network for the prediction of physical properties of organic compounds based on the group contribution method

PÉREZ?CORREA, IGNACIO; GIUNTA, PABLO D.; FRANCESCONI, JAVIER A.; MARIÑO, FERNANDO J.

THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING

JOHN WILEY & SONS INC

Año: 2022

0008-4034

In the development and optimization ofchemical processes involving the selection of organic fluids, knowledge of thephysical properties of compounds is vital. In many cases, it is complex to findexperimental measurements for all substances, so it becomes necessary to have atool to predict properties based on the characteristics of the molecule. One ofthe most extensively used methods in the literature is the estimation bycontribution of functional groups, where properties are calculated using theconstituent elements of the molecule. There are several models published in theliterature, but they fail to represent a wide variety of compounds with highaccuracy and simultaneously maintain a low computational complexity. The aim ofthis work is to develop a prediction model for eight thermodynamic properties(melting temperature, boiling temperature, critical pressure, criticaltemperature, critical volume, enthalpy of vaporization, enthalpy of fusion, andenthalpy of gas formation) based on the group contribution methodology byimplementing a multilayer perceptron. 2736 substances were used to train theneural network, whose prediction capacity was compared with other referencemodels available in the literature. The proposed model presents errors rangingfrom 1% to 5% for the different properties (except for the melting point) whichimproves the reference models with errors in the range of 3%–30%. Nevertheless,a difficulty in the prediction of the melting point is detected which couldrepresent an inherent hindrance of this methodology.