Aiden: Association-Learning-Based Attack Identification on the Edge of V2X Communication Networks

Abstract

In vehicle security, attack identification has been proposed to identify the compromised electronic control units (ECUs) of a vehicle. Fingerprinting methods using a variety of features have been widely applied to identify attacks. However, these methods only consider the features of an individual ECU, and ignore the logical association among different ECUs. This condition leads to high requirements in terms of feature measurements, and a great deal of useful information is lost to achieve identification. In this paper, an association-learning-based model, designated Aiden, is proposed to identify the compromised ECUs on the edge of V2X communication networks and without feature measurements. Experiments on a real vehicle show the effectiveness of the proposed model.