Machine learning-driven product prediction and process optimization for catalytic pyrolysis of polyolefin plastics

Abstract

Pyrolysis offers a sustainable pathway to valorize plastic waste into value-added products, yet its experimental exploration remains costly and time-intensive. In this study, five machine learning models were developed to predict, interpret, and optimize both thermal and catalytic pyrolysis of polyolefin plastics, utilizing 511 data points collected from 63 articles published between 2006 and 2023. Among these models, extreme gradient boosting regression exhibited the best performance in product yield prediction. Feature importance and partial dependence analyses identified the impact of reaction temperature and key catalytic parameters (loading, acid properties, and pore structure) of zeolite-based catalysts on gas and oil yields. These features significantly affect the accessibility of active sites and molecular diffusion, and thus the occurrence of secondary reactions. The results showed that selecting an appropriate catalyst with proper loading, and optimizing the reaction temperature can effectively regulate the catalytic pyrolysis process to achieve the desired product distribution. At moderate reaction temperatures (∼450 °C), microporous zeolite-based catalysts with moderate acidity promoted gas production, while lower temperatures (<400 °C), higher acidity, and larger pore sizes favored oil yields. This work provides valuable mechanistic insights into the catalytic pyrolysis of polyolefins and offers guidance for process optimization and catalyst design.