Machine learning for predicting catalytic ammonia decomposition: An approach for catalyst design and performance prediction

• Published in: Journal of energy storage 2024-06, Vol.89, p.111688, Article 111688
• Main Authors: Guo, Wenjuan, Shafizadeh, Alireza, Shahbeik, Hossein, Rafiee, Shahin, Motamedi, Shahrzad, Ghafarian Nia, Seyyed Alireza, Nadian, Mohammad Hossein, Li, Fanghua, Pan, Junting, Tabatabaei, Meisam, Aghbashlo, Mortaza
• Format: Article
• Language: English
• Subjects: Ammonia decomposition; Catalyst properties; Gradient boost regression; Hydrogen formation; Machine learning; Reaction conditions
• ISSN: 2352-152X; 2352-1538
• DOI: 10.1016/j.est.2024.111688

Ammonia, a cost-effective hydrogen carrier, holds the potential for hydrogen production through decomposition, where catalysts play a pivotal role in lowering the decomposition temperature. However, identifying suitable catalysts involves expensive and time-consuming experiments. Machine learning (ML) emerges as a powerful solution to address challenges in catalytic ammonia decomposition. This study focuses on creating an ML model to predict ammonia decomposition. A comprehensive database is compiled and statistically analyzed to discern correlations between descriptors and responses. Employing random forest regression, support vector machine, and gradient boost regression models, the study models the ammonia decomposition process as a function of catalyst properties and reaction conditions. Feature importance analysis evaluates the influence of descriptors on responses. The results unveil a robust positive correlation between ammonia decomposition and reaction temperature. Improved ammonia decomposition and hydrogen formation rates are achievable with a total metal loading below 20 %. The gradient boost regression tree model exhibits satisfactory performance during testing (R2 > 0.85, RMSE