Forward Physics-Informed Neural Networks Suitable for Multiple Operating Conditions of Catalytic CO2 Methanation Isothermal Fixed-Bed

Abstract

This study aims to develop a forward physics-informed neural network (fPINN) suitable for multiple operating conditions, representing the plug-flow reactor (PFR) model of catalytic CO2 methanation in an isothermal fixed-bed (IFB). The fPINN was constructed by a fully connected feed-forward artificial neural network (ANN) and physical constraints including PFR governing equations, nonlinear reaction kinetics, and boundary conditions. The fPINN showed outstanding extrapolation performance for the PFR model. The speedup factor of fPINN overwhelmed the stiff ODE numerical solver when the number of spatial points became large. The fPINN can be used as a surrogate model for process optimization where multiple reactors and operating conditions are considered.