PowerGraph

We have created a comprehensive graph dataset that models power flow (PF), optimal power flow (OPF), and cascading failure events in power systems. To generate the dataset for PF and OPF, we utilized MATPOWER, and for the cascading failure events, we employed a physics-based cascading failure model called Cascades. This model simulates the propagation of failures within a power grid, resulting in unmet demand (DNS). Each power grid state is represented as a graph, where buses (including loads and generators) form the nodes and branches (including transmission lines and transformers) form the edges.

For PF and OPF, we treat the problem as a regression task to predict electrical quantities at the node level. In contrast, for cascading failure analysis, we assign graph-level labels based on the outcomes of the physics-based model. This dual approach allows our dataset to support a variety of tasks, including node regression, graph-level multi-class classification, binary classification, and regression. Additionally, we provide ground-truth explanations for the cascading failure analysis, enabling our dataset to serve as a benchmark for evaluating GNN explainability models in graph-level tasks.

• The PF and OPF dataset is in 'dataset_pf_opf' for the IEEE24, IEEE39, UK, IEEE118 and Texas bus systems
• The Cascading failure dataset is in 'dataset_cascades' for the IEEE24, IEEE39, UK, and IEEE118 bus systems. The raw Cascading failures dataset for the Texas dataset is in 'raw.7z'.