Knowledge-Enhanced Event Prediction of Carrier-Based Aircraft Emergencies

Shipboard emergencies can easily induce secondary events because of any time and space transition. Carrier-based aircraft emergencies have complexity, tight coupling and strong real-time characteristics. Therefore, it is of great significance to reasonably predict the development trend of the incident in the emergency treatment to reduce the harm. Due to the sudden and random nature of naval battlefield accidents, some incidents are not covered by the historical case base. Traditional retrieval approaches to event prediction using case knowledge are limited by the sparsity of events and the cold-start problem. To address these challenges, this paper proposes a multi-graph neural network for event development trend prediction, which disassembles events into event features and introduces the shipboard aircraft domain knowledge graph. We produce a data set around the event features of carrier-based aircraft on the simulation platform. The results show that the graph neural network with knowledge graph is superior in modeling the higher-order semantic association between event features. Compared to state-of-the-art baseline methods, our approach exhibits significant improvements.