MUST: Multi-Scale Structural-Temporal Link Prediction Model for UAV Ad Hoc Networks

TL;DR

This work tackles link prediction in highly dynamic and sparse UAV ad hoc networks (UANETs) where route information is unavailable. It introduces MUST, a multi-scale structural-temporal model that extracts micro-, meso-, and macro-scale features via a weighted GAT, community pooling, and global pooling, then models temporal evolution with a stacked LSTM, and predicts future links through a sparsity-aware loss with end-to-end training. MUST achieves state-of-the-art performance on simulated UANET datasets across four mobility models, with substantial gains in AUC and especially AUPRC under sparse conditions, and ablation confirms the critical role of multi-scale features, particularly micro-scale cues. The approach offers robust, end-to-end dynamic link prediction for UANETs and has potential applicability to broader dynamic-network prediction tasks and adversarial analyses in UAV contexts.

Abstract

Link prediction in unmanned aerial vehicle (UAV) ad hoc networks (UANETs) aims to predict the potential formation of future links between UAVs. In adversarial environments where the route information of UAVs is unavailable, predicting future links must rely solely on the observed historical topological information of UANETs. However, the highly dynamic and sparse nature of UANET topologies presents substantial challenges in effectively capturing meaningful structural and temporal patterns for accurate link prediction. Most existing link prediction methods focus on temporal dynamics at a single structural scale while neglecting the effects of sparsity, resulting in insufficient information capture and limited applicability to UANETs. In this paper, we propose a multi-scale structural-temporal link prediction model (MUST) for UANETs. Specifically, we first employ graph attention networks (GATs) to capture structural features at multiple levels, including the individual UAV level, the UAV community level, and the overall network level. Then, we use long short-term memory (LSTM) networks to learn the temporal dynamics of these multi-scale structural features. Additionally, we address the impact of sparsity by introducing a sophisticated loss function during model optimization. We validate the performance of MUST using several UANET datasets generated through simulations. Extensive experimental results demonstrate that MUST achieves state-of-the-art link prediction performance in highly dynamic and sparse UANETs.

Figures (6)

• Figure 1: An illustration of the link prediction problem in UANETs, where $G_1, G_2, \ldots, G_{l-1}$ represent the historical topology snapshots, and the links in $G_l$ are the target to be predicted.
• Figure 2: (a) The overall framework of our proposed model, MUST, comprises three main components: a multi-scale feature extraction module, a temporal feature extraction module, and a feature decoding module. The multi-scale feature extraction module is further divided into (b) a micro-scale feature extraction sub-module, which employs a weighted GAT, (c) a meso-scale feature extraction sub-module, where meso-level pooling is applied, and (d) a macro-scale feature extraction sub-module, where macro-level pooling is performed. The temporal feature extraction module leverages (e) a LSTM network to capture the dependencies within temporal topology sequences. Finally, the feature decoding module utilizes (f) a two-layer fully connected neural network to generate the prediction results.
• Figure 3: AUC vs. Sampling interval for all methods across the four mobility models. MUST consistently outperforms the baseline methods across all sampling intervals.
• Figure 4: AUPRC vs. Sampling interval for all methods across the four mobility models. MUST consistently outperforms the baseline methods across all sampling intervals.
• Figure 5: AUC vs. Flight speed for all methods across the four mobility models. MUST consistently outperforms the baseline methods across all flight speeds.