Semantic-Aware Streaming Learning for Anomalous Event Detection in Power Grids

The growing adoption of smart grid technologies is turning power grids into data-rich environments, where continuous sensor streams capture complex and evolving operational states. Anomalous event detection in this context is particularly challenging due to the structural complexity of the grid and the non-stationary nature of sensor data. To address these challenges, we propose a semantic-aware approach that integrates sensor stream events with contextual information of the power grid, a Knowledge Graph (KG) as in International Electrotechnical Commission’s standards. Our method employs Chimera, a semantic data analytics platform, and models capable of incremental learning and adaptation. We publish as open-data a comprehensive data-stream. We comparatively evaluate four scenarios: (1) only streaming events; (2) streaming events enriched with explicit features from the KG; (3) streaming events enriched with graph embeddings of KG’s subgraphs; and (4) a combination of the last two. Experiments demonstrate the superior performances of models learned exploiting semantics in the KG.