Prediction of the product yield from catalytic cracking (MIP) process by an 8-lump kinetic model combined with neural network

Artificial Intelligence Methods Applied to Catalytic Cracking Processes

“…The team of professor Ouyang from East China University of Science and Technology has carried out a lot of work in analyzing FCC process by using BP neural networks combined with GA [55][56][57] . Fang [58] selected 19 variables, including feed oil properties, catalyst properties, and conditions, as the neural network inputs, and the yields of liquefied gas, gasoline, diesel oil, and coke as the neural network outputs.…”

“…Bollas et al [66,67] proposed a hybrid model combining FCC mechanism and neural networks, and found that the hybrid model can improve the prediction accuracy better than the simple mechanism model and the simple neural networks model by comparing with the industrial data from Greek refineries. Liu [68] took the MIP process as the research object, constructed an eight-lumped reaction network and calculated the product distribution. The input layer of the BP neural network of 14 variables, including the main raw material properties, catalyst properties, and conditions, are selected, and 5 variables, including the error between the predicted value and the industrial actual value of the yield of diesel, gasoline, liquefied gas, dry gas, and coke calculated by the lumped kinetic model are used as the output of the BP neural network.…”

Analysis of Heat and Mass Transfer and Parametric Sensitivity in an Experimental Fixed-Bed Reactor for the Catalytic Cracking of Heavy Hydrocarbons Based on Modeling and Experiments

“…On the other hand, semiempirical models, which have simple calculations, cannot be used for these reactions. , Therefore, the lumping methodology has been utilized for the complex reaction systems involved in the catalytic cracking process. In this way, similar components according to their boiling point and molecular characteristics are categorized into several cuts or lumps. , Many lumping models for catalytic cracking have been proposed and classified into two groups: (1) components are lumped based on the molecular structure characteristics of hydrocarbons, such as the four-lump kinetic model of Ojong et al, the seven-lump kinetic model of Fukuyama and Terai, the eight-lump model of Wang et al., the twelve-lump model of Zong et al., and the seventeen-lump model of Singh et al; (2) components are lumped based on the boiling point or molecular weight range, such as the four-lump kinetic model of Dave et al, Ancheyta-Juárez et al, and by Shayegh et al, the five-lump kinetic model of Sánchez et al and Ancheyta-Juárez et al, , the six-lump model of Ancheyta and Sotelo, Sadighi et al, Xiong et al, John et al, and by Seyed Asaee et al, the seven-lump model of Heydari, and the eight-lump model of Sani et al, You et al, and Sun et al In addition to these two types, in several studies, lumped kinetic models based on the chemical structure and boiling points are mixed, such as the ten-lump kinetic model proposed by Du et al and the nine-lump model by Ebrahimi et al and by Zhang et al Generally, the type of model is selected based on the purpose of the research.…”