School of Architecture, Civil and Environmental EngineeringOpen OpportunitiesThe project is a collaboration between EPFL IMOS Lab and Belimo exploring Deep Learning algorithms for commissioning parameters of Heating, Ventilation, and Air Conditioning systems - Engineering and Technology
- Master Thesis
| Pumped-storage hydropower plants are the most established technology for large-scale energy storage by far. Due to their active participation in grid power-frequency control, they often operate under dynamic conditions, which results in rapidly varying sensor measurements. Predicting these dynamically changing sensor measurements is essential for comprehending the underlying sensor and machine conditions to detect anomalies and faults to ensure the reliable operation of the connected power grid and to identify faulty and miscalibrated sensors.
In this project, we want to explore Graph Neural Networks (GNNs) for that purpose. GNNs have been increasingly applied in the broader power grid environment. However, despite the model’s successes, many challenges exist in applying them effectively in practice. One key challenge is that GNNs can only be effectively applied if an appropriate graph is provided. However, it remains unclear how to determine the optimal graph in complex industrial processes involving multiple systems. One approach is to convert schematic diagrams into graphs, another is to learn the graph from the data. Nonetheless, this problem has not been thoroughly researched, particularly when it comes to fusing sensor data of combined multi-systems (systems of systems) that lack a standard network structure.
In collaboration with the Swiss Federal Railways Company (SBB CFF FFS), we are exploring ways to enhance Graph Neural Networks (GNNs) for large-scale industrial multi-systems. We are testing our methods in a real-world case study of a pumped-storage hydropower plant.
- Engineering and Technology
- Collaboration, Lab Practice, Master Thesis
|
|