A Lightweight Unsupervised Intrusion Detection Model for In-Vehicle Edge Computing Based on FlexRay

Aiming at the security protection challenges faced by FlexRay bus in the multi domain fusion architecture of intelligent connected vehicles, this study proposes a lightweight unsupervised intrusion detection model for vehicle edge computing scenarios. To address the physical layer vulnerabilities, protocol layer priority inversion vulnerabilities, and resource constraints in traditional in vehicle networks, a hybrid architecture is designed that integrates a feedforward sequence storage network and an autoencoder (FSMN-AE). The adaptive extraction of temporal features is achieved through a memory window sliding mechanism. This study proposes a lightweight unsupervised intrusion detection system for resource-constrained design in the vehicle network environment, which realizes efficient training directly in the vehicle. Experimental results demonstrate that an average detection accuracy of 0.99 and an F1-score of 0.98 are achieved on the ARM Cortex-A57 platform. Furthermore, when comparing models within the same functional domain and of equivalent complexity, the FSMN-AE in the engine control domain achieves approximately 91.3% parameter compression compared to the GRU-AE. The FSMN-AE in the body control domain reduces memory usage by about 76.6% relative to the LSTM-FCN, while the FSMN-AE in the advanced driver assistance domain shortens training time by roughly 74.3% compared to the LSTM-AE. These results highlight the significant advantages of this model for lightweight deployment in vehicle edge computing scenarios.