Abstract
•A hybrid deep learning model for FCC performance is developed.•The hybrid model reduces the prediction error by up to 9%.•The hybrid model outperformed the DNN model with limited data availability.•The hybrid model shows a significantly greater generalization ability.
Fluid catalytic cracking (FCC) is one of the most important processes in the renewable energy as well as petrochemical industries. The prediction and understanding of the FCC performance in a real industrial environment is still challenging, as this is a highly complex process affected by many extremely non-linear and interrelated factors. In this paper, a novel hybrid predictive framework for FCC is developed by integrating a data-driven deep neural network with a physically meaningful lumped kinetic model, powered by orders of magnitude greater number of high-quality data from a modem automated FCC process. The results show that the novel hybrid model exhibits best predictions with regards to all the evaluation criteria such as Mean Absolute Percentage Error, Pearson coefficient, and standard deviation. It indicates that the hybrid data-driven deep learning with mechanistic kinetics model creates a better approach for fast prediction and optimization of complex reaction processes such as FCC.