Development of a surrogate artificial neural network for microkinetic modeling: case study with methanol synthesis

Abstract

Microkinetic models allow the description of complex reaction kinetics but require high computational costs, hindering their combination with detailed reactor models. In this contribution, a methodology to develop a surrogate artificial neural network (ANN) was proposed and demonstrated for methanol synthesis on Cu/Zn-based catalysts. The resulting model accurately reproduces the simulations of the original microkinetic model, reducing the computational costs by orders of magnitude. In the developed methodology, the ANN learns only the kinetics of the global reaction rates, thereby decreasing model complexity and computational costs while ensuring thermodynamic consistency. In addition, an improved activation function for the ANN was designed in this work to minimize computational costs and to smooth out calculations. The proposed approach creates a bridge to integrate microkinetics into applications in the field of reaction engineering, such as reactor design, process optimization, and scale-up.