Automated machine learning-assisted analysis of biomass catalytic pyrolysis for selective production of benzene, toluene, and xylene

Biomass pyrolysis catalyzed by metal-modified zeolites presents a viable pathway for producing benzene, toluene, and xylene (BTX), yet the complexity of interactions among feedstock characteristics, processing conditions, and catalyst properties presents substantial optimization challenges. To address these complexities, this study employed advanced automated machine learning (AutoML) techniques, specifically the tree-based pipeline optimization tool (TPOT) and H2O AutoML, to model the catalytic pyrolysis process. These AutoML approaches demonstrated successful prediction capabilities, with R2 values exceeding 0.912 for BTX production. Through model explainability analysis, this paper elucidated the quantitative mechanisms linking feedstock, process, and catalyst characteristics to performance outcomes, as well as the structure-activity relationships of the catalysts. Operating parameters were found to have the greatest impact on BTX production, contributing 40.5%, followed by catalyst descriptors at 34.1% and feedstock characteristics at 25.4%. Additionally, by integrating the particle swarm optimization (PSO) algorithm, this study conducted reverse process design and catalyst screening, identifying Zn-modified zeolite as the optimal catalyst, with the optimal process conditions including a catalytic temperature of 550–650 °C, metal loading of 1–3 wt%, Si/Al ratio of 30–40, and feedstock H/C ratio of 1.4–1.6. Overall, this research underscored the potential of AutoML to advance catalytic pyrolysis technology by facilitating targeted product regulation.