Internet of Vehicles (IoV) in high-density scenarios relies on spatial correlations between vehicles to ensure secure communication. However, malicious nodes frequently exploit spatial consistency vulnerabilities to launch attacks (e.g., Sybil attacks, position falsification). Existing methods exhibit poor detection performance due to neglecting spatial dependencies. This study proposes a hybrid architecture integrating Graph Neural Networks (GNN), diffusion models, and Long Short-Term Memory (LSTM): GNN captures vehicle-to-vehicle spatial interactions, LSTM models temporal dynamics, and the diffusion model enhances robustness against adversarial attacks. Experiments on the VeReMi dataset show that the model achieves an average F1-score of 0.975, with the precision of Sybil attack detection improved by 15% compared to the VeReMi baseline.
Yang et al. (Mon,) studied this question.