Hydrogen and syngas production from methane via thermochemical conversion: a mini review of artificial neural network applications

Abstract

This mini-review examines the application of artificial neural networks (ANNs) in hydrogen and syngas production from methane via thermochemical conversion processes, including steam methane reforming, dry reforming, methane pyrolysis and partial oxidation of methane. These processes involve complex interactions between thermodynamics, reaction kinetics and transport phenomena. In particular, gaps in the understanding of reaction kinetics limit the effectiveness of conventional modelling approaches. This review highlights how ANNs, particularly multilayer perceptron (MLP) and radial basis function (RBF) networks, provide efficient data-driven alternatives for modelling thermodynamic equilibrium, reaction kinetics, catalyst deactivation (coking and poisoning) and reactor performance. Reported studies demonstrate that ANNs can achieve high predictive accuracy, often outperforming simplified empirical models while significantly reducing computational cost. The influence of key operating variables, such as temperature and feed composition, is identified as critical for process performance. Recent advances emphasize hybrid approaches integrating ANNs with first-principles models, computational fluid dynamics, and microkinetic simulations, including physics-informed and physics-enhanced neural networks. Despite promising results, challenges remain related to limited datasets and model interpretability.