Accelerated Screening of Zeolites for Methanol-to-Propylene Conversion Using Machine Learning with Interpretable Selectivity Function.

Methanol-to-propylene (MTP) is one of the important propylene production technologies that is commonly achieved by the use of zeolite catalysts. However, effectively regulating the selectivity toward propylene remains challenging due to the insufficient understanding of the principles behind selectivity control. In this study, we proposed an interpretable function to assess the propylene/ethylene (P/E) selectivity based on a thorough investigation of the MTP selectivity control mechanism and implemented a machine learning (ML) method for high-throughput screening of zeolite frameworks for MTP. Density functional theory and grand canonical Monte Carlo simulations have been used to build the ML model. The P/E selectivity predictions are supported by the reported and our own experiments. We chose Mobil Five (MFI) (rank 374) and Deca-dodecasil 3R (DDR) (rank 5869) for experimental validation. Experimental results showed that the Ga-MFI catalyst achieved a P/E ratio of 12.3 and a remarkably longer operational lifespan of 121.3 h in the MTP process, surpassing previously reported values. Additionally, the marked superiority of Ga-MFI over Ga-DDR, especially in terms of P/E ratios and lifespans, further underscores the effectiveness of our zeolite screening strategy.