An Industrial Data-Based Model to Reduce Octane Number Loss of Refined Gasoline for S Zorb Process

Abstract

S Zorb process is one of the main technologies for deep desulfurization of gasoline from fluid catalytic cracking (FCC) process, which by the process will also cause some research octane number (RON) loss of gasoline. Establishing a data-driven model with data mining technologies to optimize production is one of the development directions in petrochemical field. Based on the industrial data from a 1.20 Mt/a S Zorb unit in China in recent three years, 422 modeling samples and 22 modeling variables were screened out and then three data-driven models were established by back propagation neural network (BPNN), radial basis function neural network (RBFNN) and generalized regression neural network (GRNN) to predict RON of refined gasoline (r-RON). The results show that the BPNN model has the best prediction effect and generalization ability. Genetic algorithm (GA), particle swarm optimization algorithm (PSO) and simulated annealing algorithm (SA) in combination with the BPNN model respectively were used to optimize the operation variables to reduce the r-RON loss. The results indicate that the optimized performance of PSO-BPNN model is best because of its largest reduction in r-RON loss at 48.55%. The validity of the PSO-BPNN model was verified in the S Zorb unit and the research methods to establish a data-driven model for reducing r-RON loss are also worthy of reference for other S Zorb units.