Process Simulation, Optimization, And Property Control Of FCCU Main Fractionator

In China, Fluid Catalytic Cracking Unit (FCCU) which is the most important secondary process in refineries is the major supply of fuels such as gasoline and diesel. With the goal of producing more gasoline, a mechanical model of FCC main fractionator is established to simulate the fractionator with the help of Aspen Plus. According to the simulation result, the impact of operating conditions on the yield of gasoline is conducted. Meanwhile, a hydraulic analysis of the column trays is implemented using the vapor and liquid flow profiles as well as the physical properties provided by the simulation.Gasoline endpoint is an important control variable of the overhead product, so controlling the gasoline endpoint is a major approach to assure the quality of the gasoline. Nowadays, there are mainly two methods to measure gasoline endpoint: the first method is analyzing off-line in every 4 or 8 hours, this will generate large time lags; the other one is analyzing on-line with the help of on-line analyzer, but the expensive analyzer is very hard to maintain while it also generates a couple of minutes time lags. Thus neither of these two methods could provide real-time information for the quality control. The advent of soft sensor effectively mitigate the trouble referred above. However, the soft sensors of gasoline endpoint are established mainly using the empirical method which are difficult to correlate with mechanisms and difficult to adapt the abrupt change of the operating conditions. In order to explore a feasible solution, in the thesis, with the analysis of the pseudo-components of gasoline, two mechanical models of gasoline endpoint are proposed. One model is to correlate endpoint with pseudo-components about the endpoint, while the other one is to correlate endpoint with gasoline-diesel overlapping pseudo-components. It is testified that both of the models tally well with the assay data. In addition, two hybrid soft sensors are established combining BP neural network with the two endpoint mechanical models respectively. Compared with the result of BP neural network empiric model, the hybrid models achieves certain improvements.