HyperKGE: Hyperedge-based Knowledge Graph Embedding for Satellite Threat Prediction

In the increasingly complex aerospace environment, the growing number of satellites makes it critical to monitor and predict inter-satellite threat events such as conjunctions. Traditional methods rely on target recognition and tracking based on imagery and motion parameters, but these overlook the structural relationships between satellites and their orbits. In reality, such threat events often arise from higher-order interactions within the satellite-orbital network.To address this gap, we construct a satellite orbital knowledge graph centered on threat events. We model the complex semantics between satellites and their orbits through knowledge graph embedding techniques to predict potential threat events. We introduce a domain-driven hyperedge construction approach that captures high-order semantic relationships via metapaths, enabling integration of diverse information while maintaining semantic coherence. Additionally, we propose a soft contrastive learning mechanism that improves robustness and discrimination by enhancing the contrastive learning process in a complex semantic space. Experiments show that our method significantly outperforms existing approaches, offering a powerful new solution for satellite threat prediction and deeper insights into the structure of space object interactions.