Abstract: Machine Learning was used to create a model for prediction of three property parameters (carbon dioxide conversion XCO2, selectivity of CO2 conversion into methane SCH4 and methane yield YCH4) in reaction of carbon dioxide methanation at nickel-containing catalysts. The model was learned using a set of 576 experiments with 16 feature parameters. The most important feature parameters appeared to be temperature T, Ni and SiO2 content. The Decision Tree was chosen to build an explainable model, which revealed two zones of experiments with high values of YCH4 and one zone of experiments with medium YCH4, all other experiments lead to low YCH4. Outside this range, the samples tend to demonstrate low conversion and selectivity values. The rules derived are not linear or obviously correlated. However, machine learning helped to form the main criteria. The resulting rules can be used for optimization of the catalyst composition as well as for selection of the promising catalysts. Verification procedure included selection of promising catalysts supported on SiO2, Al2O3, TiO2, and NixAly and prediction of their basic properties using developed Random Forest model. Then, these catalysts were synthesized and experimentally tested. It was shown that developed model provides quite adequate and rather accurate predictions for a major part of experimental parameters; therefore, the developed model may be considered as successfully verified and recommended for further application.

Cite: Motaev K. , Mikhailov I. , Molokeev M. , Azarapin N. , Grigoriev M. , Zhovanik I. , Zagoruiko A. , Elyshev A.
Application of Machine Learning in the Study of CO2 Methanation Reaction at Ni-Containing Catalysts
Chemical Engineering Science. 2026. V.321. 122912 :1-12. DOI: 10.1016/j.ces.2025.122912 OpenAlex

Dates:

Submitted:	Aug 4, 2025
Accepted:	Nov 1, 2025
Published online:	Nov 3, 2025

Identifiers:

OpenAlex:

W4415827115

Citing: Пока нет цитирований

Altmetrics: