Hybrid Modeling of Methanol to Olefin Reaction Kinetics Based on the Artificial Neural Network

Abstract

Methanol to olefin (MTO) is an important coalbased process to ensure olefin yield. The current reaction kinetics model as a crucial tool of process regulation lacks real-time, accuracy, and simplicity in describing the catalyst deactivation. To fill in the gap, a reaction kinetics hybrid model of MTO was developed by a machine learning approach. Specifically, a catalyst deactivation model was first built by an artificial neural network (ANN) and then integrated into a lump-based reaction kinetics model over a fresh SAPO-34 catalyst. Reaction kinetics parameters were estimated by the genetic algorithm. A priori knowledge was extracted by analyzing the deactivation experimental data and used in the training of ANN to enhance its extrapolability. The developed model of MTO agreed well with the kinetics experimental data considering the catalyst deactivation. The model provided a satisfactory predictive performance. Results show that a higher reaction temperature, longer reaction space time, and shorter catalyst residence time are beneficial to the yield of ethylene and propylene. The hybrid model developed in this work is expected to apply to the process control of MTO. The proposed hybrid modeling approach could be used as theoretical guidance for other catalytic reaction kinetics modeling.

Intelligent Soft Measurement of Catalyst Activity during a Polyethylene Reaction Based on Mechanistic Modeling and Deep Learning

“…Mechanism modeling is an effective method to obtain invisible data, so mechanism and data fusion based approaches have the ability to address the lack of invisible data data sets. For example, an ANN-based hybrid model of methanol-to-olefin reaction kinetics was developed to predict catalyst activity . A hybrid model was utilized to count the spatiotemporal variation parameters in the moving boundary problem .…”