Machine learning regression-CFD models for predicting hydrogen dispersion in a naturally ventilated area

• Published in: Volume 8: Ocean Renewable Energy
• Main Authors: Patel, Parth, Garaniya, Vikram, Baalisampang, Til, Arzaghi, Ehsan, Mohammadpour, Javad, Abbassi, Rouzbeh, Salehi, Fatemeh
• Format: Article in Journal/Newspaper
• Language: English
• Published: 2023
• Subjects: Hydrogen safety; Computational Fluid Dynamics (CFD); Machine learning regression; Low velocity hydrogen release; Natural ventilation; Arctic
• Online Access: https://researchers.mq.edu.au/en/publications/0abc4a00-e516-4e4d-a202-1e4d4f087a06; https://doi.org/10.1115/OMAE2023-104522; http://www.scopus.com/inward/record.url?scp=85173612885&partnerID=8YFLogxK

Hydrogen can diversify the primary energy supply as it offers several benefits in terms of reduced emissions and greenhouse gases. Although hydrogen can be a great option for energy generation at higher efficiency and minimal environmental impacts, leakage and dispersion are the challenges to establishing safe and sustainable hydrogen infrastructure. A comprehensive analysis consisting of computational fluid dynamics (CFD) and machine learning algorithms (MLAs) is conducted to study the leakage of hydrogen in a cuboid room with two vents located on the side wall (door vent) and roof. This study aims to identify the optimum dimensional relationship between leakage and ventilation position that can efficiently extract hydrogen from semi-confined spaces. Three MLAs, including eXtreme Gradient Boosting (XGBoost), Multilayer Perceptron (MLP), and k-Nearest Neighbours (k-NN), are adopted here. The results confirmed that the lower distance between the door vent to the ceiling and the roof vent to the leakage and the larger distance between the leakage and the door vent are found to be the most dominant factors to keep hydrogen volumetric concentration lower. XGBoosting outperforms all other regression models in the prediction of the flammable hydrogen cloud size, while k-NN and MLP performed well in the prediction of the critical time. The outcome of this study can be used to develop appropriate control measures and risk mitigation strategies.