Neural network modeling of trans isomer formation and unsaturated fatty acid changes during vegetable oil hydrogenation

A multi-layer neural network model with back-propagation training algorithms was designed to predict total trans isomer content, as well as oleic acid, linoleic acid and linolenic acid during vegetable oil hydrogenation. Eight variables including reaction temperature, H2 pressure, catalyst concentration, mixing rate, iodine value, and initial unsaturated fatty acid contents including oleic, linoleic, and linolenic acid have strong effects on forming trans isomer which is produced during vegetable oil hydrogenation. So the eight variables were considered as independent variables and used as inputs to the Artificial Neural Network (ANN) model. The neural network was trained, tested, and evaluated by use of a large number of experimental data obtained from a pilot-plant hydrogenation reactor and using experimental data of a published paper the network generalization was evaluated. Experimental data were statistically compared with predicted results such that the network predictability was assessed. Statistical assessments showed a very good agreement of predicted and observed results.