An Intrusion Detection System for the Internet of Vehicles Based on Multi-Level Hybrid Feature Analysis

The rapid advancement of Internet of Vehicles (IoV) technology has increasingly automated and intelligentized modern vehicles. The connection between vehicle networks and external networks has become much closer, significantly increasing the risks of intrusions and threats, making vehicle network security more critical. This paper propose an intrusion detection model based on multi-level hybrid feature analysis to identify intrusions and threats in the IoV. The model utilizes parallel Long Short-Term Memory (LSTM), and One-Dimensional Convolution (Conv1D) to extract global and local features from input data separately and then merges these features to provide comprehensive and integrated hybrid features. It employs a multi-head attention mechanism to extract relationships at different positions in the sequence from multiple perspectives, and an improved emporal Convolutional Network (TCN) to further capture long-distance dependencies in the sequence, thereby extending the model's awareness of long-term dependencies. Finally, malicious traffic is identified through multi-classification. The proposed model was evaluated on two publicly available IoV security datasets, CICIoV2024 and CICIDS2017, achieving accuracy rates of 100% and 99.821 %, respectively. These findings underscore the effectiveness of our proposed model.