HiSTGNN: Hierarchical Spatiotemporal Graph Neural Network for Multi-UAV Survivability Assessment

Multi-unmanned aerial vehicle (UAV) cooperative systems demonstrate exceptional scalability and robustness in high-risk missions. Despite these advantages, assessing UAV survivability remains challenging due to the complex and heterogeneous dynamics involved in interactions between UAV swarms and threat sources. To address this issue, we propose a hierarchical spatio-temporal graph neural network (HiSTGNN). This framework introduces a novel mechanism for decomposing and hierarchically modeling dynamic heterogeneous interaction graphs, explicitly capturing both direct and indirect interaction effects on UAVs. This approach facilitates multi-level feature extraction and heterogeneous information fusion to accurately represent complex interactions. The framework further integrates a recurrent-convolutional fusion module to improve dynamic spatial feature extraction, alongside a parameter-sharing mechanism that enhances model efficiency and generalization. Experimental results demonstrate that HiSTGNN significantly outperforms baseline methods in assessment accuracy and adaptability.