Neural Network Optimization In Fccu Simulation To Control The Application Of A Preliminary Study

Two typical models, relevancy-model based on Blandmg equation and lumping kinetic model based on lumping theory are used in Fluidized Catalytic Cracking(FCC). When they are used to a unit, it is needed to adjust the model parameters according to the unit. The model using artificial neural network has the characteristics of self-teaching and self-adjusting, and can handle problems involving data that are imprecise or 搉oisy?as well as those that are highly nonlinear and complex. In this paper, model to predict the yields of FCC based on the theory of artificial neural network is built. It can predict the yields of FCC accurately according to the operating condition. When it is applied in simulation and optimization of practical unit, adjusting of the model can be done automatically and continually according to the practical data of unit. Accordingly, it can dynamically tail after the practical unit, and realize the real-time optimization of unit preferably. Compared with the method of using strict mechanism-model to realize optimization, it is more convenient and feasible, and has satisfied precision. In this paper, three different optimization methods are used to optimize the neural network. Coordinate-rotating and NPSOL method(SQP) have the same optimized result. It is also proved that the model can reflect the law of FCC.