Deep learning for online cascading failures prediction: a comparison of graph neural networks and feed forward neural networks
Metadata only
Date
2022Type
- Other Conference Item
ETH Bibliography
yes
Altmetrics
Abstract
Past events have revealed that widespread blackouts are mostly a result of cascading failures in the power grid. A real-time detection of precursors to cascading failures will help operators take measures to prevent their propagation. Currently, the well-established probabilistic and physics-based models of cascading failures offer low computational efficiency, allowing them to be used as offline tools. In this work, we have developed a data-driven methodology for online estimation of the risk of cascading failures. We utilize an AC physics-based cascading failure model to generate a cascading failure dataset. The dataset covers a large set of grid states labeled as either safe or unsafe. Each sample of the dataset includes a contingency set, bus and branch conditions. We use the synthetic data to train deep learning architectures, namely Feed-forward Neural Networks (FNN) and Graph Neural Networks (GNN). With the development of GNNs, improved performance is achieved with graph-structured data. Furthermore, GNNs can generalize to graphs of different sizes. Since the power grids are complex networks, they can be represented by graphs, making it convenient to use of GNNs. Indeed, the GNN dataset carries information on the grid topology, in addition to bus and branch states. A comparison between FNN and GNN is made and the GNNs inductive capability is tested when switching from one test grid to another. Show more
Publication status
publishedBook title
Book of Extended Abstracts for the 32nd European Safety and Reliability Conference (ESREL 2022)Pages / Article No.
Publisher
Research PublishingEvent
Subject
Cascading failure; Transmission grid; Deep learning; Graph Neural Networks (GNNs); Power gridOrganisational unit
09452 - Sansavini, Giovanni / Sansavini, Giovanni
Notes
Conference lecture held on August 30, 2022More
Show all metadata
ETH Bibliography
yes
Altmetrics