EXTENDED ABSTRACT: Hydrocracking technology effectively transforms inferior heavy oil into clean fuel oils and organic chemical feedstocks, which is a pivotal bridging technology facilitating the transition from reffning to the chemical industry while promoting high-quality, low-carbon development. The increasingly diversiffed and cost-competitive market imposes heightened demands on petroleum reffning processing. A rational design of catalytic materials and catalysts tailored for the targeted transformation of petroleum molecules is essential. Hydrocracking catalysts are bifunctionalities with hydrogenation and cracking, typically comprising with metal components, acidic constituents, matrices, additives, among other materials. Due to exceptional stability, appropriate pore structure, and tunable acidity, zeolites are usually chosen as acidic components in both experimental and industrial research. However,the zeolite structures are highly complex,and meanwhile the main characterizes such as pore structure, Bronsted acid content, acid strength are correlation. With advancements in theoretical research alongside high-throughput computing technologies and artiffcial intelligence (AI), the paradigm for catalyst development is progressively shifting towards theoretical frameworks combined with computational simulations driven by data analytics. Data-driven approaches leverage vast datasets accumulated from traditional experiments along with AI methodologies such as machine learning to model relationships between catalytic material properties and their corresponding reaction performances effectively establishing multifactorial structure-activity relationships even when reaction mechanisms remain ambiguous while predicting optimal material characteristics through modeling efforts thereby expediting high-performance catalytic material development.In this study, a series of Y molecular sieve samples were selected as acidic components and the corresponding catalysts NiMo/Al2O3-Y were prepared, employing n-C10 and tetralin as model compounds assessing their hydrocracking reaction performances. The physicochemical property of zeolites were systemeticaly characterized through XRF, XRD, SEM, BET, NH3-TPD and Py-IR. A comprehensive database cataloging the property proffles was established allowing correlation analyses. Finally, the prediction models were based upon random forest algorithms and the optimized zeolite were obtained. The evaluation results show that the model predicted results are in good agreement with the experimental results.

Keywords: hydorcracking; catalysts; zeolite,structure-activity relationship,machine learning