MF-CANN: A Novel Anomaly Detection Method for Connected and Automated Vehicles

The development of connected and automated vehicles(CAVs) has improved the efficiency of transportation systems, but they are vulnerable to cyber attacks, leading to traffic paralysis and accidents. It is worth mentioning that existing methods often rely on a single type of time-domain data from vehicle sensors, resulting in the misjudgment of anomalies. To mitigate potential wrecks of data anomalies caused by cyberattacks or data failures, a novel hybrid model for vehicle anomaly detection, i.e., multi-information fusion convolutional attention neural network (MF-CANN), is proposed. Specifically, the multivariate information fusion strategy is used in MF-CANN to balance the importance of time and frequency domain information in vehicle sensor data. Subsequently, the proposed dual self-attention method is organically combined with the Convolutional Neural Network (CNN) to enhance the feature extraction ability of vehicle abnormal data. Experimental evaluations show MFCANN's advanced performance, with an average accuracy of 97.02%, indicating its potential for robust anomaly detection in intelligent transportation systems (ITS).