Davide Rubinetto
Surrogate Model for Fluid-Dynamic Simulation and Hydrogen Blending Tracking in Natural Gas Distribution Grids.
Rel. Pierluigi Leone, Marco Cavana. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Energetica E Nucleare, 2026
Abstract
The injection of hydrogen into natural gas distribution networks represents a viable strategy for decarbonizing the European energy sector. However, the variability of the thermophysical properties of the mixture and the fluctuations in demand require advanced tools to monitor gas quality and network integrity. Traditional fluid dynamics solvers based on differential equations, although accurate, involve high computational costs that limit their applicability for operational control. This thesis proposes an innovative computational framework applied to the real distribution network of the city of Riccione. In the first phase, a steady-state physical solver was validated in MATLAB, able to simulate multicomponent transport and track hydrogen propagation.
The validation with respect to the commercial software SAInt confirmed the accuracy of the model, with an R2 > 0.98 for nodal pressures and composition tracking. The model was employed to conduct a parametric analysis of the hydrogen concentration distribution across the network, varying one parameter at a time. The investigation specifically focused on the displacement of zero-flow points and/or mixing points, where hydrogen concentration shows the highest variations. While this approach enabled a detailed understanding of the underlying physical mechanisms, it did not allow for a generalization of the analysis to a fully multivariate framework. To overcome computational limitations, the research implemented a surrogate model based on Vector-Valued Kernel Regression (VVRR) techniques. The model was trained on a large dataset generated by Monte Carlo simulations, varying critical parameters such as supply pressure, pipe’s roughness, consumptions and hydrogen concentration at the sources, to verify its reliability under different operating conditions. For each set of inputs, the physical solver calculated the corresponding state of the network, creating the training and test datasets. The model was designed to accurately predict hydraulic quantities, which were then used to assess gas quality and analyse the nodal concentration of hydrogen in the entire infrastructure. Validation was performed by comparing the predictions with the outputs of the physical solver, quantifying the errors and using graphical tools such as scatter plots. The process was applied to three main case studies of increasing complexity: a single injection case with 4 variable parameters, a double injection case with 5 variable parameters and an advanced double injection scenario with 30 variable parameters, characterized by partially independent consumption clusters. The surrogate model showed excellent accuracy in predicting fluid dynamic parameters, allowing precise calculation of nodal concentration. Despite increasing complexity, the quality of the hydraulic prediction remained high, with only slight discrepancies in the concentration calculation, attributable to minimal uncertainties in the estimation of flow rates and consumption. The greatest differences occur near the interconnections between adjacent loops, where the flow direction is sometimes predicted incorrectly. However, given the low frequency of these outliers, the results confirm the suitability of the model for operational analysis. The analysis provides DSOs with a rapid and sufficiently effective predictive tool for managing future multi-component networks, offering new research opportunities in an area that is still relatively unexplored but could play a key role in the energy transition.
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Informazioni aggiuntive
Corso di laurea
Classe di laurea
Aziende collaboratrici
URI
![]() |
Modifica (riservato agli operatori) |
