Decision tree analysis on the performance of zeolite-based SCR catalysts

Abstract

Urea-based selective catalytic reduction (SCR) is a promising method for removing NOx emissions. In the urea-based SCR method, zeolite-based catalysts are popularly used owing to their applicability over a wide range of temperatures compared to other catalysts. However, they still have several drawbacks such as inferior performance at low temperatures and thermal instability. This has subsequently led to numerous studies and experiments on the development of superior catalysts. While substantial experimental data exist on zeolite-based catalysts, extracting useful information from the data with a simple literature search is difficult owing to not only the amount of data but also complex correlations between feature, including preparation variables such as doped-metal loading and operational variables such as reaction temperature. Recently, extracting insights from a large database has become possible by utilizing machine learning tools. Among them, the decision tree can extract insights on the synthesis of the catalysts as the results derived from the models are intuitive and easy to interpret, unlike those from conventional discriminant machine learning models. In this study, classification models are obtained by training decision tree models with literature data on zeolite-based urea SCR catalysts, particularly the Beta and ZSM-5 types, and several experimental heuristics are extracted from the derived models.