Predictive modeling of hydrogen production and methane conversion from biomass-derived methane using machine learning and optimisation techniques

Abstract

The growing demand for cleaner and more efficient energy solutions has necessitated the development of biomass conversion techniques for hydrogen production. Thermocatalytic methane decomposition produces hydrogen and solid carbon directly from methane without CO₂ emission. However, there is the need to optimise this process for better efficiency and improved hydrogen production from biomass sources. In this study, the integration of various machine learning algorithms with Bayesian optimisation, firefly algorithm, Levenberg-Marquardt, and differential evolution techniques were investigated for hydrogen production via thermocatalytic methane decomposition. The key process parameters studied include calcination temperature (450–600 °C), time of calcination (3–8 h), specific surface area (5.4–249 m²/g), and pore volume (0.03–0.48 cm³/g); reduction temperature (500–700 °C), time of reduction (1–5 h), and catalyst weight (0.05–1.00 g). The Bayesian-optimized CatBoost regressor model, with an R² of 96.3% and an RMSE of 0.064 showed the best performance. For the prediction of methane conversion, the Support Vector Regressor (SVR) model optimised with Firefly showed the best performance among other models with an R² value of 95.5% and root mean squared error (RMSE) of 0.070. CatBoost regressor predicted hydrogen yield of 87% close to the actual yield of 86%. The predicted methane conversion using the firefly-optimized support vector machine regressor was 72%, with the actual conversion being 68%. Model-to-feature relationship studies showed that catalyst weight and calcination time were the strongest predictors of hydrogen yield and methane conversion volume. The study hence established the great opportunity of integration of machine learning models with optimisation techniques in attempts to improve the prediction of hydrogen yield and methane conversion in processes for hydrogen production.