Catalyst development for the tri-reforming of methane (TRM) process by integrated singular machine learning models

Abstract

Tri-reforming of methane (TRM) is a promising technology for the simultaneous production of hydrogen and syngas with high energy efficiency (above 70%). However, catalyst design for TRM is challenging due to complex reaction kinetics and the need for catalyst materials with great stability and activity. Machine learning, particularly artificial neural networks (ANNs), has emerged as a powerful tool in catalyst development for the TRM process. More than 6000 data points were selected to build individual models for each reaction and later coupled into an ensembled model used to make predictions considering TRM experimental conditions. The reaction temperature input parameter was found to be the one with major relative importance (61.4%), contributing the most to changes in the CH₄ conversion %. Dry reforming of methane (DRM), steam reforming of methane (SRM), and partial oxidation of methane (POX) models observed errors (RMSE) of 3.44%, 2.20%, 1.61%, respectively, with the ensembled model having a maximum error of 4.48%. The newly devised artificial neural network (ANN) model demonstrates remarkable capability in accurately predicting CH₄ conversion for novel catalyst formulations in the TRM process, exhibiting minimal error deviation.