A universal microkinetic-machine learning bimetallic catalyst screening method for steam methane reforming

Key Words:VOLCANO CURVE; DESCRIPTOR

Abstract:With the development of first-principles computing and artificial intelligence, catalyst screening have taken remarkable steps toward machine learning (ML). However, it is difficult to explore universal features and de-scriptors for catalyst screening due to the high spatial freedom of catalytic materials and the lacking kinetic insights of catalytic activity metrics. In this paper, a universal microkinetic-machine learning method (Mkml) is proposed to screen bimetallic catalysts. First, microkinetic models are built based on reaction information to search for descriptors and activity ranges. Then, datasets consisting of elemental properties and adsorption site encoding are evaluated by ML to predict the reference formation energy of the descriptors to optimize the microkinetic model. Finally, activity, stability and price factors are considered to screen competitive catalysts. Mkml was applied to steam methane reforming (SMR) to prove its high accuracy and low cost. The perform best ML is XGBoost with a 0.973 R2 score, and the feature importance also provides a valuable reference for catalyst design. 48 promising candidates are selected by Mkml from the 5000 catalytic database, which not only ac-celerates experimenters' discovery of efficient and cheap SMR catalysts, but also guides the search way in a larger materials space.

Volume:311

Issue:-

Translation or Not:no