Machine learning based analysis of formic acid dehydrogenation for hydrogen production: Role of catalytic features and dehydrogenation reaction parameters on catalytic hydrogen production

Abstract

Formic acid (HCOOH, FA) has attracted attention as a liquid organic hydrogen carrier (LOHC) for the hydrogen economy due to its favorable hydrogen storage capability, liquid-phase operability, and potential for a storageproduction cycle via reversible carbon dioxide (CO2) hydrogenation and FA dehydrogenation (FAD). This study analyzes catalytic hydrogen production from FAD using machine learning. A dataset of 8458 data points from 847 experiments involving homogeneous and heterogeneous catalysts is used to predict turnover frequency (TOF), representing the hydrogen production efficiency from FA. Eight input features describing catalyst characteristics and reaction conditions are evaluated. Four advanced machine learning algorithms (Random Forest (RF), Extreme Gradient Boosting (XGBoost), CatBoost, and Gradient Boosting Regressor (GBR)) are trained, and Random Forest yields the best performance (RMSE: 12.71 training, 20.44 testing; R2: 0.9959 and 0.9893). Feature importance analysis revealed that the active catalytic phase is the most influential parameter affecting the amount of hydrogen generated per active site per unit time. The developed model provides a data-driven framework for analyzing the relationships between catalyst characteristics, reaction conditions, and FAD efficiency in hydrogen production systems.