Spatial-Temporal Graph Neural Network Enabled Security Situation Prediction for 6G Edge Networks

Abstract

Accurate security situation prediction remains a challenging issue for sixth generation (6G) edge networks, due to missing data in the security situation sequence and the spatial correlations of smart attacks. In this paper, we propose a spatial-temporal graph network-based scheme, in which a time-varying undirected graph is constructed to capture the spatial-temporal dependencies of security situations derived from attacks among different edge servers. To further improve the prediction accuracy under severe data missing cases, we propose a spatial-temporal graph network-based scheme, spatiotemporal graph attention network with missing data. Specifically, we design a novel graph attention network with missing data module as the spatial component to optimize the weight fusion mechanism to utilize information from adjacent nodes. Experimental evaluations on public datasets validate that our proposed method achieves the highest prediction accuracy for both complete and incomplete data scenarios, outperforming existing approaches in edge network security situation prediction.