A hybrid neural network-first principles model for fixed-bed reactor

Abstract In the present work, we combine first principles, in the form of mass and energy balance equations, with artificial neural networks (ANNs) as estimators for some of the important process parameters in modeling a wallcooled fixed-bed reactor. Experiments were carried out in a pilot wall-cooled fixed-bed reactor with benzene oxidization to maleic anhydride as a working reaction to show the performance of the proposed hybrid models. Compared with the two-dimensional pseudo-homogeneous models, the hybrid models predict equally well, but are simpler in structure and therefore easier to meet the real-time requirements of control and on-line optimization.