Multi-Level Damage-Aware Graph Learning for Resilient UAV Swarm Networks

Unmanned aerial vehicle (UAV) swarm networks leverage resilient algorithms to address communication network split issues and restore connectivity. However, existing graph learning-based resilient algorithms face over-aggregation and non-convergence problems caused by uneven and sparse topology under massive damage scenarios. To alleviate these problems, we propose a novel Multi-Level Damage-Aware Graph Learning (ML-DAGL) algorithm, which generates recovery trajectories by mining information from destroyed UAVs. We first introduce a Multi-Branch Damage Attention (MBDA) module, which forms a sequence of multi-hop Damage Attentive Graphs (mDAG) with different ranges of receptive fields. Each mDAG links only remaining and damaged nodes to ensure a more even degree distribution for mitigating over-aggregation, and utilizes multi-hop dilation to establish more links for sparse topology enhancement. To resort to the mDAG, we propose a Dilated Graph Convolution Network (DGCN), which generates the optimal recovery trajectories with theoretically proven convergence under massive damage cases. Simulation results show that the proposed algorithm can guarantee the connectivity restoration under large swarm and damage scales, while significantly expediting the recovery time by 75.94% and improving the topology uniformity after recovery.