|
Abstract
|
The growing release of greenhouse gases from hydrocarbon use has intensified the demand for clean and sustainable energy solutions. Hydrogen, with its high energy output and eco-friendly combustion byproducts, stands out as a promising candidate. However, storing hydrogen, especially for vehicles, poses significant challenges due to the safety risks of high-pressure systems. Metal-organic frameworks (MOFs) offer a potential solution, enabling hydrogen storage at lower pressures through their porous structures. Yet, achieving optimal volumetric storage density while balancing material design and operational conditions remains a critical issue. This study presents an advanced predictive model for hydrogen storage (HS) capacity, leveraging a dataset of 14,544 synthesized MOF porous crystals and 10 key material properties. Machine learning (ML) techniques, including Linear Regression (LR), Artificial Neural Networks (ANN), Random Forest (RF), Support Vector Regression (SVR), LSBoost, and their hybrid versions combined with Particle Swarm Optimization (PSO), were applied to model hydrogen uptake across 18 operational conditions (OCs). The models were rigorously evaluated using metrics such as R2, RMSE, MSE, and MAE, demonstrating exceptional accuracy in predicting HS performance. The results underscore the significant impact of thermodynamic factors (pressure and temperature), material density, pore size, and surface characteristics on hydrogen adsorption in MOFs. Hybrid ML models, particularly ANN-PSO and RF-PSO, outperformed traditional methods, delivering more precise and reliable predictions. Additionally, this study introduces a novel approach by averaging multiple training runs and testing various data percentages, ensuring the robustness and consistency of the models. These findings highlight the transformative potential of ML-driven models in optimizing hydrogen storage processes, offering a pathway to safer and more efficient hydrogen fuel cell systems. By integ
|