Sybil attack detection and traceability scheme based on temporal heterogeneous graph attention networks

In the development and application of cooperative driving technology, Sybil attacks pose a serious threat to vehicle safety. Although existing detection schemes can identify erroneous information from Sybil nodes, they cannot prevent ongoing attacks and struggle to accurately trace their sources. The high concealment and intermittent message silences of attack sources are the root causes of this challenge. To address this, This paper propose a Sybil attack detection and tracing scheme based on a temporal heterogeneous graph attention network. Our method deeply integrates graph-structured data capturing vehicle behaviors, spatiotemporal characteristics, and dynamic traffic flow changes, and leverages graph attention to model complex interaction patterns among vehicles. This enables precise Sybil detection and physical tracing even during silent attack intervals. Experimental results on the VeReMi-Extension dataset demonstrate that our scheme achieves a Sybil node detection accuracy of 99.89% and successfully traces over 85% of attack source vehicles — a 50% improvement in tracing recall compared to existing approaches — effectively mitigating the threat of Sybil attacks. Notably, this work fills the existing research gap in tracking the physical locations of Sybil attackers.