Prediction of Solubility of Food Oleoresins during Supercritical CO2 Extraction using Artificial Neural Networks

SEBASTIÁN ODDONE; SANDRO GOÑI; EMMANUEL PURLIS; LUCÍA POLLIO; JUAN A. SEGURA

Journal of Food Technology

Grace Publications Network

Año: 2004 vol. 2 p. 319 - 324

1684-8462

In this paper a backpropagation artificial neural network was implemented in order to predict the solubility of several chemical substances during supercritical extraction processes. The predictive models reported in literature can be classified into two groups: a) semi-empirical models, and b) empirical models. The semi-empirical models are derived from thermodynamic laws and take into account the non-linear behavior of the fluids; whereas the empirical models are based on non-linear regression in least square sense from experimental data sets. Due to its simplicity, the most common empirical model applied in literature is the Chrastil density-based model. However, this model produces relatively high errors when it is used to predict solubility of food oleoresins. In this work we have developed a predictive neural-network-based model and have concluded that the error in prediction is significantly low with respect to the error of both semi-empirical and empirical mathematical models.