Zeolite-catalytic pyrolysis of waste plastics: Machine learning prediction, interpretation, and optimization

Converting waste plastics into renewable energy through zeolite-catalytic pyrolysis is a promising strategy for combating plastic pollution and supplanting conventional fossil fuels, thereby facilitating emission mitigation. However, it is still challenging to comprehensively decipherer this conversion process and screen efficient catalysts for diverse plastic feedstocks to improve the yield and quality of the resultant liquid fuel. This work explored the importance and correlations of factors in zeolite-catalytic pyrolysis and aided the zeolite screening and optimization of operational conditions for improving the oil yield and quality via machine learning (ML)-based interpretation and inverse design. The results indicated that the Extreme Gradient Boosting model developed from the complied dataset after feature selection exhibited the best performance (testing R2 of 0.85 and 0.87) for predicting the yields of liquid oil and gasoline-range (C5-C12) hydrocarbons among three tree-based algorithms. The ML-based interpretation showed that the polyethylene ratio in plastic feedstock, the reaction temperature, specific surface area, and Si/Al ratio of zeolites were the top-four important features, and their impacts on the yields of liquid oil and C5-C12 hydrocarbons were discussed in detail. A maximum oil yield of 80.85 % was achieved from ML-based inverse design, and the corresponding optimal inputs from the model could guide the experimental investigation. It showed that a high oil yield of 87.82 % was obtained from experiment that was even higher than the model result with a small error of −7.93 %. This work provides a novel ML-based approach to understand the zeolite-catalytic pyrolysis of waste plastics and improve the yield and quality of liquid oil for sustainable energy production.