A Modeling Method of FCCU Reactor-regenerator System using NARX

The nonlinear autoregressive model with exogenous variables (NARX) by Levenberg-Marquardt(LM) algorithm is applied to model the FCCU reactor-regenerator system. NARX is a dynamic neural network, its delay term and output feedback structure features are suitable for modeling complex systems. We consider the regenerator pressure and riser temperature as the system output, which determines the system stability and product distribution. At the same time, we select the regenerator smoke gas flow and the riser catalyst flow as system inputs. The simulation results show that the NARX model can effectively describe the practical dynamic characteristic of the FCCU reactor-regenerator system, at the same time it has better predictive and generalization ability than the traditional artificial neural network (ANN) model.