Abstract

This study presents a novel hybrid model designed to predict six key process variables in an industrial-scale fluid catalytic cracking (FCC) riser reactor: vacuum gas oil (VGO) conversion, outlet temperature, and yields of gasoline, LCO, light gases, and coke. The methodology integrates computational particle-fluid dynamics (CPFD) with artificial intelligence, using Barracuda Virtual Reactor 22.0® to solve first principle model (FPM) equations. Based on 216 simulations incorporating CREC-UWO catalytic kinetics, a comprehensive dataset was generated to train a machine learning algorithm. A multiple output feedforward neural network (FNN) was utilized to correlate reactor feeding conditions with output parameters, achieving an average regression coefficient of 0.83 and an overall RMSE of 1.93. This research demonstrates that combining CPFD simulations with ML significantly enhances predictive accuracy and process optimization, offering substantial advancements for industrial FCC operations.