Machine learning knowledge-driven Pt-based catalyst design library development for selective oxidation of polyalcohol

Abstract

Laborious first-principles calculations and trial-and-error experimentation often fail to meet the demands of rational and efficient catalyst development. This paper introduces an approach that integrates costly labeled data with process reaction mechanisms for catalyst formulation in the high-value conversion of glycerol. We developed an innovative system framework, POCOM, which simultaneously generates the optimal process superstructure and operating conditions to achieve peak conversion rates and desired product specifications. We synergistically combined reaction mechanisms, machine learning, process optimization, and data generation techniques, encapsulating them into a cutting-edge software system specifically designed for catalyst formulation in glycerol selective oxidation. In this process, we identified a previously unreported Pt-ZnO catalyst formulation. The catalyst, with 1.8 wt% Pt and 0.4 wt% ZnO, demonstrated exceptional performance, achieving a glycerol conversion rate of 88% and a glyceric acid selectivity of 80%. This study offers groundbreaking insights and robust data support for the rational design of glycerol oxidation catalysts.