ANN and RSM Modelling and Optimization of Paraffins and Aromatics in Crude Oil Distillation Products’ Properties in Iraq

Abstract

Back-Propagation neural networks, as well as RSM-DOE techniques, were used to predict the properties of various compositions of Iraqi oil, which were presented in this study. Paraffin and Aromatics’ effect on petroleum properties, e.g., yield, density, calorific value, and other essential properties, were studied. The input-output data to the neural networks were obtained from existing local refineries in Iraq. Several network activation functions to simulate the hydrocracking process were tested and compared. the network function that gave satisfactory results in terms of convergence time and accuracy was adopted. The data were divided into training and testing parts. The results of the trained artificial neural network models for each one of the tested functions have been cross-validated with the experimental data. The network that compared well against this new set of data (i.e. testing data), with an average percent error always less than 3% for the various products of the hydrocracking unit were chosen for the study. Aromatics showed to have more profound effect on the Octane number at low concentrations of paraffin, while, for specific gravity and calorific value they have similar effects. As for boiling points and sulfur contents, aromatics have almost no effect at lower levels of paraffin.