Research Info

As global energy systems transition toward low-carbon solutions, hydrogen is emerging as a vital carrier for clean energy storage and transport. Precise knowledge of hydrogen’s properties is a key requirement for designing and operating storage and transport systems, particularly when it interacts with cushion gases like methane, carbon dioxide, and nitrogen. In this way, viscosity is key to flow behavior and safe hydrogen handling. This study introduces a machine learning framework to predict the viscosity of pure hydrogen, its binary and multicomponent mixtures with cushion gases, and the pure forms of these gases. A refined dataset of 3547 viscosity measurements was used. A new composite parameter, Beta (β), was developed to improve prediction accuracy. Six advanced machine learning algorithms; decision tree, Gaussian process regression, K-nearest neighbors, random forest, AdaBoosting, and multilayer perceptron were trained and evaluated through statistical and visual metrics. Among them, AdaBoost achieved the highest accuracy with an R2 of 0.9953 and a MAPE of 2.8875 %. Sensitivity analysis and SHAP plots identified Beta and pressure as the most influential variables. The model shows strong generalization and reliable trend prediction across various conditions, offering a robust and scalable tool for hydrogen storage and transport applications.