Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Huan Lin, Chenguang Zhu, Lianghui Ding, Lin Wang, Feng Yang

TL;DR

This work tackles connectivity restoration in UAV swarm networks under massive damage by proposing ML-DAGL, a two-stage graph-learning framework that explicitly leverages information about destroyed nodes. The method combines a Multi-Branch Damage Attention (MBDA) module, which constructs multi-hop, damage-aware graphs to mitigate over-aggregation and sparsity, with a Dilated Graph Convolution Network (DGCN) that operates on bipartite mDAGs to generate recovery trajectories; convergence of the bipartite graph convolutions is theoretically established. The authors prove an upper bound on recovery time and demonstrate, through extensive simulations, that ML-DAGL achieves guaranteed connectivity restoration (R_c = 1) across varying damage scales and swarm sizes, while significantly reducing recovery time and producing more uniform recovered topologies compared with state-of-the-art baselines. The approach is designed to be scalable and practically deployable with fixed input dimensions and modest online computation, thanks to pre-trained models and a fixed-dimension representation for any damage pattern within a given swarm size.

Abstract

Unmanned aerial vehicle (UAV) swarm networks leverage resilient algorithms to restore connectivity from communication network split issues. However, existing graph learning-based approaches face over-aggregation and non-convergence problems caused by uneven and sparse topology under massive damage. In this paper, we propose a novel Multi-Level Damage-Aware (MLDA) Graph Learning algorithm to generate recovery solutions, explicitly utilizing information about destroyed nodes to guide the recovery process. The algorithm first employs a Multi-Branch Damage Attention (MBDA) module as a pre-processing step, focusing attention on the critical relationships between remaining nodes and destroyed nodes in the global topology. By expanding multi-hop neighbor receptive fields of nodes to those damaged areas, it effectively mitigating the initial sparsity and unevenness before graph learning commences. Second, a Dilated Graph Convolution Network (DGCN) is designed to perform convolution on the MBDA-processed bipartite graphs between remaining and destroyed nodes. The DGCN utilizes a specialized bipartite graph convolution operation to aggregate features and incorporates a residual-connected architecture to extend depth, directly generating the target locations for recovery. We theoretically proved the convergence of the proposed algorithm and the computational complexity is acceptable. Simulation results show that the proposed algorithm can guarantee the connectivity restoration with excellent scalability, while significantly expediting the recovery time and improving the topology uniformity after recovery.

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

TL;DR

This work tackles connectivity restoration in UAV swarm networks under massive damage by proposing ML-DAGL, a two-stage graph-learning framework that explicitly leverages information about destroyed nodes. The method combines a Multi-Branch Damage Attention (MBDA) module, which constructs multi-hop, damage-aware graphs to mitigate over-aggregation and sparsity, with a Dilated Graph Convolution Network (DGCN) that operates on bipartite mDAGs to generate recovery trajectories; convergence of the bipartite graph convolutions is theoretically established. The authors prove an upper bound on recovery time and demonstrate, through extensive simulations, that ML-DAGL achieves guaranteed connectivity restoration (R_c = 1) across varying damage scales and swarm sizes, while significantly reducing recovery time and producing more uniform recovered topologies compared with state-of-the-art baselines. The approach is designed to be scalable and practically deployable with fixed input dimensions and modest online computation, thanks to pre-trained models and a fixed-dimension representation for any damage pattern within a given swarm size.

Abstract

Unmanned aerial vehicle (UAV) swarm networks leverage resilient algorithms to restore connectivity from communication network split issues. However, existing graph learning-based approaches face over-aggregation and non-convergence problems caused by uneven and sparse topology under massive damage. In this paper, we propose a novel Multi-Level Damage-Aware (MLDA) Graph Learning algorithm to generate recovery solutions, explicitly utilizing information about destroyed nodes to guide the recovery process. The algorithm first employs a Multi-Branch Damage Attention (MBDA) module as a pre-processing step, focusing attention on the critical relationships between remaining nodes and destroyed nodes in the global topology. By expanding multi-hop neighbor receptive fields of nodes to those damaged areas, it effectively mitigating the initial sparsity and unevenness before graph learning commences. Second, a Dilated Graph Convolution Network (DGCN) is designed to perform convolution on the MBDA-processed bipartite graphs between remaining and destroyed nodes. The DGCN utilizes a specialized bipartite graph convolution operation to aggregate features and incorporates a residual-connected architecture to extend depth, directly generating the target locations for recovery. We theoretically proved the convergence of the proposed algorithm and the computational complexity is acceptable. Simulation results show that the proposed algorithm can guarantee the connectivity restoration with excellent scalability, while significantly expediting the recovery time and improving the topology uniformity after recovery.

Paper Structure

This paper contains 40 sections, 2 theorems, 63 equations, 9 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. System model
  4. UAV Communication Model
  5. Graph Topology of USNET
  6. Network Connectivity
  7. Problem Formulation
  8. Approach
  9. Overall Framework of ML-DAGL
  10. Multi-Branch Damage Attention
  11. Multi-hop dilation branches
  12. Damage attentive graph
  13. Dilated Graph Convolution Network
  14. Structure of DGCN
  15. Bipartite GCO
  16. ...and 25 more sections

Key Result

Proposition 1

In the metric space of position matrices $\{\bm{X}_{gco}\}\subseteq\mathbb{R}^{N\times 2}$, the B-GCO with $\bm{A}_{dag}^k$ is a contraction operation when $0<\epsilon\leq\frac{1}{\|\bm{A}_{dag}^k\|_\infty}$, and there exists and only exists one position matrix $\bm{\bar{X}}_{gco}$ such that Especially, when B-GCO takes $\bm{X}_{in}$ as the input, $\bm{\bar{X}}_{gco}\equiv[\bm{p}_c,\bm{p}_c,...,\

Figures (9)

  • Figure 1: Overall framework of the proposed ML-DAGL algorithm.
  • Figure 2: Structure of the proposed MBDA module.
  • Figure 3: Structure of the proposed DGCN.
  • Figure 4: Statistical results of resilient algorithms under varying damage scenarios with swarm scale $N=200$.
  • Figure 5: Results of the case study of different algorithms.
  • ...and 4 more figures

Theorems & Definitions (2)

  • Proposition 1
  • Proposition 2