Enhancing V2X Security Through Combined Rule-Based and DL-Based Local Misbehavior Detection in Roadside Units

Abstract

In this paper, we address the limitations of existing deep learning (DL) methods for local misbehavior detection (LMBD) in vehicle-to-everything (V2X) communication systems by proposing an approach that combines rule-based and DL-based techniques. Conventional DL-based methods at roadside units (RSUs) struggle with forwarding basic safety messages (BSMs) received from every vehicle to centralized locations and preprocessing them, which leads to considerable time delays. To overcome these challenges, our approach leveraged multi-access edge computing (MEC) connected to RSU to decentralize the processing workload, considerably reducing latency and resource consumption. Specifically, we implemented a system where RSUs directly receive and forward BSMs to the MEC server, bypassing traditional deduplication and sorting processes at the centralized server. However, due to the fixed locations of RSUs, they often receive only truncated sequences of BSMs from passing vehicles, which necessitates LMBD on these incomplete datasets. To mitigate the performance degradation of DL-based anomaly detection in truncated sequences, we integrated a rule-based method performed for single or two consecutively received BSMs. Simulation results demonstrated that this combined rule-based pre-screening with DL analysis effectively improves the overall detection performances.

Keywords

• Vehicle-to-everything (V2X)
• misbehavior detection (MBD)
• roadside unit (RSU)
• misbehavior report (MBR)
• multi-access edge computing (MEC)
• deep learning