Using artificial neural networks in naphthas RVP modeling

Artificial neural networks (ANNs) based on a feedforward architecture and a back-propagation on technique alternately using negative and conjugate gradients, have been used to create a naphtha Reid vapor pressure model of process data from an HYSIM process simulation. The accuracy of the model is to within 1.185×0025; of the target values. Some important measures are proposed to produce simulated operation cases of the stabilizer to provide training and testing data. Procedures were also developed to improve network testing performance and to avoid oscillatories during this network learning procedure in which improved convergence. These measures can be expanded to any chemical engineering process model using ANNs.