Leveraging AI for accurate prediction of hydrogen density (in pure/mixed Form): Implications for hydrogen energy transition processes

Abstract

The transition to sustainable energy is critical to addressing climate change and growing energy demands. Hydrogen, as a clean energy carrier, supports this transition by stabilizing grids and integrating renewables. Accurate prediction of hydrogen's thermophysical properties, particularly gas density, is crucial for operational safety, efficiency, and hydrogen-dependent processes like transportation, conversion, and utilization. This study develops an artificial intelligence framework to predict hydrogen density in pure form and mixtures with gases such as methane, nitrogen, and carbon dioxide across varying pressure, temperature, and molecular weight conditions. Using 3336 experimental data points, advanced machine learning models—including Decision Tree, Random Forest, Adaptive Boosting, Multilayer Perceptron (MLP), and K-Nearest Neighbors—were applied. The MLP model demonstrated the highest accuracy (R² = 0.9956, NRMSE = 1.4147 %). Feature importance analysis identified molecular weight as the most influential factor, followed by pressure, while temperature showed a negative correlation. These findings highlight the potential of AI-driven methods to enhance hydrogen technologies, contributing to efficiency and reliability in hydrogen processes. This research provides valuable insights for advancing clean energy systems and supporting the global shift toward a sustainable energy future.