Hypergraph Neural Network Based EAE Model for Power Grid Dispatching and Control

Extracting arguments from historical operational ticket data is crucial for automating power grid dispatching and control. However, Event Argument Extraction (EAE) in grid operational tickets poses significant challenges for existing models due to its complexity and heterogeneity. The distance between these arguments and the event triggers can also be considerable. To address these problems, we propose a novel hypergraph neural network-based EAE (HNN-EAE) model for power grid dispatching and control. We design a joint encoding approach that integrates graph and hypergraph structures to generate feature representations of sentences. Features are extracted using graph convolution and hypergraph convolution techniques. Our model effectively captures global interactions among crucial trigger words and entity mentions. We validate our model using the SW-EVENTS datasets from the Southwest Branch of the State Grid Corporation of China. Comprehensive analyses from these experiments demonstrate the superiority of HNN-EAE over state-of-the-art models.