Modeling of Thermal Cracking of Heavy Liquid Hydrocarbon: Application of Kinetic Modeling, Artificial Neural Network, and Neuro-Fuzzy Models

Abstract

To predict the main product yields of thermal cracking of heavy liquid hydrocarbon, four models, kinetic, artificial neural networks (ANN), neuro-fuzzy (NF), and polynomial, were developed. The models investigated the influence of COT, steam ratio, and feed flow rate on product yields at the reactor tube outlet. A semimechanistic kinetic model based on free radical chain reactions was developed using experimental results. This semimechanistic kinetic model contains 148 reactions for 43 species. An objective function was defined to optimize the kinetic parameters. For the artificial intelligence systems, a three-layer perceptron neural network with back-propagation (BP) training algorithm and Sugeno inference system were used. To compare the accuracy of artificial intelligence method, another empirical method based on response surface methodology was also developed. Finally, the models were compared to experimental data, and a comparison between the results of kinetic model, designed ANN, and NF was also carried out by analysis of variance (ANOVA) calculation. The obtained results demonstrate that these four models are in good agreement with experimental data, while the ANN and NF models show better results than do the kinetic model and polynomial model.