Large-scale vehicle anomaly attack behavior perception and real-time early warning based on federated learning and edge computing

This study proposes a collaborative perception and early warning framework integrating federated learning and edge computing, aiming to achieve real-time detection of large-scale abnormal vehicle attack behaviors. At the perception level, federated learning is used to achieve distributed abnormal behavior modeling and cross-node perception; at the early warning level, edge computing is used for data preprocessing to achieve hierarchical response. Experimental findings show that the introduced model achieves recognition accuracies of 93.1% and 93.8% on public datasets. For real-time early warning, the average latency is only 15.2ms, and the system throughput reaches 1250 requests/s. In real-world road scenarios, a recall rate of 97.2% is achieved, with false positive and false negative rates below 2.1% and 1.8%, respectively, validating the effectiveness of the strategy. This research provides a feasible path for anomaly detection in vehicle-to-everything systems and has practical significance for improving the security of intelligent transportation systems.