Hybrid modeling of methanol to olefin fluidized bed reactor corrected by artificial neural network

Abstract

Hybrid modeling of fluidized bed reactor (FBR) remains as a challenging issue. A novel hybrid model was developed for the real-time optimization and control of FBR in the methanol to olefin (MTO) process. Specifically, a pseudo-homogeneous moving bed reactor (PHMBR) model was proposed as an approximation of the FBR, and then a correction model based on the artificial neural network (ANN) was adopted to compensate errors. A data expansion method based on the bootstrap algorithm was used to supplement insufficient training data for the ANN. Two hybrid models with deviation correction and factor correction were compared in order to attain the better predictive performance. The results showed that the hybrid model of MTO-FBR with factor correction could predict FBR outlet parameters with solution time below 0.03 s, RMSE below 2.85, and MRE below 7.63 %. The proposed modeling approach is expected to provide a new strategy for other reaction processes.