Coal-Oil Co-Refining Oil Cracking to Btex Over Different Morphology Mfi Zeolites：Structure-Activity Relationship Derived by Machine Learning

Abstract

The morphology’s transformation of MFI zeolites and its influence on the light liquid products derived from coal-oil co-refining (LCOCR) cracking was systematically studied. MFI zeolites with different morphologies were prepared in a composition of x Na2O/100 SiO2 (x= 8, 12, 15, 18, 21, 22, 23 and 30) with a di-quaternary ammonium-type surfactant (C22H15-N+(CH3)2-C6H12-N+(CH3)2-C6H13, C22-6-6). By X-ray powder diffractometer and scanning electron microscope, it was found that the morphology of the crystals changed significantly with increasing x values, and the average particle size of the catalyst changed in two stages: 6 μm to 0.8 μm, and 0.8 μm to 3.5 μm. The change of morphology is mainly attributed to temperature, Na+ concentration and pH. And then, fast pyrolysis-gas chromatography/mass spectrometry (Py-GC/MS) was adopted to explore the yield and selectivity of products from LCOCR cracking at different temperatures, which the selectivity and yield of BTEX can reach 9.70% and 4152.12 mg/kg by MFI-21 at 700 °C. The structure-activity relationship between MFI zeolites and the product selectivity was investigated by machine learning. The results of leave-one-out cross validation (LOOCV) show acceptable performance (R2> 0.6) for selectivity of alkanes, aromatic hydrocarbons (ArHs) and BTEX (benzene, toluene, ethylbenzene, xylene), and good performance (R2= 0.7226) for benzene. Then the result of that pore structure contributed more to the alkane and aromatic selectivity than acidity is sorted out by SHapley Additive exPlanation (SHAP).

Keywords: MFI zeolites, Modulation, Coal-oil co-refining products, catalysis, Machine learning