AI Accelerated Optimization and Prediction of Key Performance Parameters in Catalytic Methane Decomposition

Abstract

The depletion of conventional fuels and the urgent need for clean and sustainable energy sources have driven research into H2 production through catalytic methane decomposition. This study aimed to use novel machine learning (ML) approaches to enhance and predict H2 yield using Artificial Neural Network (ANN), Ensembled Tree (ET), Gaussian Process Regression (GPR), Regression Tree (RT), and Support Vector Machine (SVM). A two-step approach involving feature selection and hyperparameter optimization was employed to enhance the models' performance for H2 yield. The coefficient of correlation (R2 ) and Root Mean Square Error (RMSE) were used to evaluate model performance. ET performed excellent with R2 of 0.929 (training) and 0.933 (testing) however GPR exhibited exceptional performance, achieving a perfect training R2 of 1.00 and low RMSE 0.00026. Furthermore, partial dependence plots (PDPs) were utilized to assess the impact of catalyst properties and reaction conditions on H2 yield. Temperature impacts H2 directly, while time shows an inverse relationship. Various catalysts and catalysts structure exhibited distinct behaviors, and the Average surface area demonstrated a direct linear relationship with H2 yield. These findings contribute to the understanding of catalytic methane decomposition and provide insights for optimizing H2 production processes using ML models. Given the depletion of conventional fuels, H2 has emerged as a crucial alternative energy source due to its clean and sustainable nature. The ability to accurately predict H2 yield using ML models opens new avenues for advancing H2 production technologies and meeting the growing global energy demand while mitigating environmental concerns.