AI and physics together for the security of energy infrastructures

Energy infrastructures: Enrico Zio, professor at the Department of Energy of Politecnico di Milano, is among the authors of a research published in npj Artificial Intelligence (Nature Portfolio Journal).

The study, titled “Towards parameter identification in pipeline hydraulics: integrating data-driven discovery and knowledge embedding,” proposes an innovative approach to improve parameter identification and hydraulic transient simulation in pipelines.

The work addresses one of the most critical challenges in pipeline management: accurately monitoring hydraulic transients, sudden changes in pressure and flow that represent the moments of greatest vulnerability of these infrastructures.

The authors introduce a method capable of integrating data-driven models and physical knowledge of hydraulic behavior, overcoming the limits of purely statistical approaches or black-box neural networks.

The developed algorithm combines:

• physical principles of traditional hydraulic models, 
• neural networks and techniques for data-driven equation discovery, 
• a system for dynamic parameter updating, adapted to real operational conditions.

The tests reported in the study show a significant improvement in the accuracy of flow and pressure simulations, with an advantage particularly noticeable during transients.