关键词: Carbon monoxide, Dynamic modeling, Hybrid model, Reaction kinetics, Semi-supervised learning

Abstract: Ethylene glycol (EG) plays a pivotal role as a primary raw material in the polyester industry, and the syngas-to-EG route has become a significant technical route in production. The carbon monoxide (CO) gas-phase catalytic coupling to synthesize dimethyl oxalate (DMO) is a crucial process in the syngas-to-EG route, whereby the composition of the reactor outlet exerts influence on the ultimate quality of the EG product and the energy consumption during the subsequent separation process. However, measuring product quality in real time or establishing accurate dynamic mechanism models is challenging. To effectively model the DMO synthesis process, this study proposes a hybrid modeling strategy that integrates process mechanisms and data-driven approaches. The CO gas-phase catalytic coupling mechanism model is developed based on intrinsic kinetics and material balance, while a long short-term memory (LSTM) neural network is employed to predict the macroscopic reaction rate by leveraging temporal relationships derived from archived measurements. The proposed model is trained semi-supervised to accommodate limited-label data scenarios, leveraging historical data. By integrating these predictions with the mechanism model, the hybrid modeling approach provides reliable and interpretable forecasts of mass fractions. Empirical investigations unequivocally validate the superiority of the proposed hybrid modeling approach over conventional data-driven models (DDMs) and other hybrid modeling techniques.

Key words: Carbon monoxide, Dynamic modeling, Hybrid model, Reaction kinetics, Semi-supervised learning