Get started
Work

Prediction of permeation-related hydrogen losses in pipelines based on machine learning

Year: 2025
Type: article
Abstract: Green hydrogen plays a crucial role in the transition toward a sustainable energy matrix, with transport through pipelines representing an important part of this transition. Fiber-reinforced polymers hold promise for pipe construction but are gas-permeable, as are other candidate materials for barrier layers, such as high-density polyethylene. Consequently, feasibility studies on hydrogen pipeline infrastructures face the challenge of accounting for permeation-caused losses. Herein, machine learning is used to predict permeation-related hydrogen losses in polymer pipelines. Data generated using Aspen HYSYS and validated with MATLAB simulations are used to construct two machine learning models, one based on correlation and the other based on an artificial neural network. The correlation-based model is established considering similar models describing pressure drops in pipelines and exhibits a relative error of 5% for the training dataset. The neural network–based model considers the complex relationships between the input variables and therefore captures intricate patterns and exhibits an exceptional predictive ability (Pearson correlation coefficient = 0.99999). The strengths and limitations of the two models are revealed using comparative analysis. The correlation-based model enables rapid estimation, whereas the neural network–based model offers superior accuracy and potential for real-time applications. These findings underscore the transformative potential of machine learning in advancing hydrogen pipeline safety and efficiency, establishing it as a valuable tool for real-world applications involving polymer-based pipelines.
DOI: 10.1016/j.egyr.2025.04.048
Language: English
Open Access status: gold
Estimated APC paid: $3,040
OpenAlex ID: W4410087870
FWCI: 0.414
Cites: 63
Cited by: 2
Related to: 10