Prediction Of The Product Yields From Heavy Oil Catalytic Cracking Process By Lumped Kinetic Model Combined With Neural Network

Modeling methods for the complex reaction system of fluid catalytic cracking(FCC)were reviewed in this paper,especially illustrating lumped kinetic model.Based on FCC reaction mechanisms and characteristics of maximizing iso-paraffins(MIP)process,the reaction network of an 8-lump kinetic model including saturates,aromatics,asphaltenes+resins,diesel,gasoline,liquefied petroleum gas,dry gas and coke was developed,and the corresponding lumped kinetic model equation was deduced.Then 47 kinetic parameters for the model were calculated respectively by genetic algorithm(GA),particle swarm optimization(PSO),and simulated annealing algorithm(SA)with a large amount of industrial data,which show good consistence with the reaction mechanism of heavy oil catalytic cracking.The average relative errors between calculated values and actual values of products are all less than 5%.The three optimization algorithm show good adaptability to multi-parameter optimization problem,however,GA is superior to PSO and SA in both average errors and error distribution in general.In order to decrease the prediction errors of products distribution by the model,a 14-7-5 type of BP neural network model was established,in which main properties of raw materials and catalysts,and operation conditions were selected as input variables,errors of diesel,gasoline,liquefied gas,dry gas and coke between the predicted value and actual value were selected as output variables.The results show that the prediction accuracy of the products distribution can be further improved by the hybrid model,which provides a new direction for simulation and optimization of heavy oil catalytic cracking.