초록 |
This study presents a hybrid physics-informed neural network (PINN) and data-driven surrogate model (SM) for simultaneous inference and system identification of CO2 methanation in an isothermal fixed-bed reactor (IFB). The hybrid surrogate model include (1) governing equations, (2) boundary and initial conditions, (3) reaction kinetics, and (4) observation data obtained from an experiment or computational fluid dynamics (CFD) for the identification of the network parameters (w and b ) and unknown physical parameters. The unknown physical parameters are learned simultaneously with w and b of the artificial neural network. Considering the fluidized-bed reactor for CO2 methanation to be isothermal, the CFD and experimental data are combined with an isothermal plug-flow reactor governing equations into the hybrid surrogate model. To overcome the stiffness of the ODEs system, various hyperparameters such as network structure, training strategy, and sequential optimizers selection are tested. This hybrid surrogate model gives the ability to automatically discover and represent unknown parameters and weights in the isothermal reactors. |