Random forest regression for catalyst performance prediction and validation of tri-reforming of methane (TRM)

Abstract

Carbon dioxide-reduced hydrogen can be synthesized through various methods such as dry-reforming (DRM), steam reforming (SMR), and partial oxidation (POX). Tri-reforming of methane (TRM) is a promising technology which combines all the above-mentioned processes for the simultaneous production of hydrogen and syngas with high energy efficiency. However, catalyst design for TRM is challenging due to complex reaction kinetics and the need for catalyst. This way, the use of data-driven models, such as random forest models (RFs), can be used to assist in the search of new material compositions, leading to a faster scale-up process. In this study, a comprehensive database was built, encompassing approximately 6000 data points and 6 input parameters collected from literature published between 2013 and 2022. Three different random forest models were developed and demonstrated to accurately predict the methane conversion using temperature, metal wt.%, and feed ratios as input features, being later incorporated into a singular model based of the reaction kinetics of TRM. All models achieved high accuracy, with R² values exceeding 93%. The reaction temperature input parameter, ranging from 623 to 1173 K, was identified as having the greatest relative importance (60.7%), significantly influencing the changes in CH₄ conversion.