Optimization of nickel-based catalysts for hydrogenolysis of light paraffins using machine learning

Abstract

In this study, optimization of nickel-based catalysts for the hydrogenolysis of hydrocarbons using machine learning methods. A comprehensive dataset comprising 419 experiments was compiled on the basis of literature data, focusing on key parameters such as the catalyst composition, support type, surface area, reduction temperature, and reaction conditions. We employed a random forest model to predict the reaction rates, achieving a mean absolute error of 0.37 and an R² value of 0.76. Two Ni/Al₂O₃ and Ni/TiO₂ catalysts were synthesized following the model recommendations on the optimal synthesis parameters for Ni-based catalysts. Experiments demonstrated excellent agreement between predicted and observed reaction rates and, moreover, these catalysts appeared to be more active than all other Ni-based catalysts, considered at the model learning. This means that such a model is capable not only of interpolating existing information, but also of creating improved catalysts to some extent exceeding the current level of scientific achievements, though still remaining strictly within the existing knowledge set. The proposed approach may be helpful to researchers when they start working with a reaction or a catalyst which is unfamiliar to them. Another possible research application of the proposed approach is classification of catalysts, including indication of potentially promising new approaches.