EXPERIMENTAL VALIDATION OF INTERPRETABLE MACHINE LEARNING PREDICTIONS FOR CH4 AND CO2 CONVERSIONS AND H2/CO RATIOS IN DRY REFORMING OF METHANE CATALYSTS

Abstract

Conventional dry reforming of methane (DRM) catalyst development is plagued by rapid deactivation (50-60% CH₄ loss in conversion in 5h) from coke formation and the lack of consistent benchmarking of results from literature in disparate pretreatments. This research presents an experimentally validated interpretable machine learning (IML) framework with unprecedented predictive accuracy (R²=0.94, RMSE=5.2%) on 17 synthesized catalysts screening on virtual formulations with 99.998% efficiency. A 6,067-datapoint strategy of 132 studies was narrowed down with the help of SHAP feature selection (T reaction=45 GHSV=32 dominance) and CatBoost optimization (depth=7, lr=0.09) to produce design guidelines over complex compositional hyperspace (22 metals, 24 supports). The highest value Ni₃Pr₂Ca₅Al₂O₃ catalyst achieves a breakthrough 240 min stability - CH₄ performance 10.7% (52.9->47.2%) and CO₂ 8.0% (64.2->59.1%) vs Ni/Al₂O₃ benchmarks deactivation 50-60%, 4x slower deactivation (0.045%/min) while H₂/CO approximately 1.25 Fischer-Tropsch synthesis. Interaction volumetrics of three dimensions (SHAP) identify the best operating window (750-800 C, GHSV 70-90 mg/min/mL, Ni 5-8 wt%) and ZrO₂ PDP shows phase-dependent behavior (best 12 wt%). Pr/Y basicity (+22% reverse gasification) and CaO synergy (+12%) are mechanistically justified; coke resistance was tested by TPO (less than 15 wt% carbon). This closed-loop of IML and experiment integration addresses the fatal flaws of catalysis ML: black box opacity, validation gap, and narrow scope, provides 80X screening speedup compared to trial and error, without compromising perfect rank order prediction for Ni/Ru/Ce/Fe systems. The validated design principles [Ni (3-5 wt%)-Pr/La (5-8 wt%)-Al₂O₃, 800 C pretreatment, (GHSV<90) are beyond the ability of noble metals to remain stable at 1/100th of the cost, making DRM to be the first carbon-neutral syngas between biogas and making IML the ultimate catalyst accelerator of rational thermal discovery.