Cyber Security of Traffic Signal Control Systems with Connected Vehicles

Abstract

Our world is becoming increasingly connected through smart technologies. The same trend is emerging in transportation systems, wherein connected vehicles (CVs) and transportation infrastructure are being connected through advanced wireless communication technologies. CVs have great potential to improve a variety of mobility applications, including traffic signal control (TSC), a critical component in urban traffic operations. CV-based TSC (CV-TSC) systems use trajectory data to make more informed control decisions, therefore can accommodate real-time traffic fluctuations more efficiently. However, vehicle-infrastructure connectivity opens new doors to potential cyber attacks. Malicious attackers can potentially send falsified trajectory data to CV-TSC systems and influence signal control decisions. The benefit of CV-TSC systems can be realized only if the systems are secure in cyberspace. Although many CV-TSC systems have been developed within the past decade, few consider cyber security in their system design. It remains unclear exactly how vulnerable CV-TSC systems are, how cyber attacks may be perpetrated, and how engineers can mitigate cyber attacks and protect CV-TSC systems. Therefore, this dissertation aims to systematically understand the cyber security problems facing CV-TSC systems under falsified data attacks and provide a countermeasure to safeguard CV-TSC systems. These objectives are accomplished through four studies.

The first study evaluates the effects of falsified data attacks on TSC systems. Two TSC systems are considered: a conventional actuated TSC system and an adaptive CV-TSC system. Falsified data attacks are assumed to change the input data to these systems and therefore influence control decisions. Numerical examples show that both systems are vulnerable to falsified data attacks.

The second study investigates how falsified data attacks may be perpetrated in a realistic setting. Different from prior research, this study considers a more realistic but challenging black-box attack scenario, in which the signal control model is unavailable to the attacker. Under this constraint, the attacker has to learn the signal control model using a surrogate model. The surrogate model predicts signal timing plans based on critical traffic features extracted from CV data. The attacker can generate falsified CV data (i.e., falsified vehicle trajectories) to alter the values of critical traffic features and thus influence signal control decisions.

In the third study, a data-driven method is proposed to protect CV-TSC systems from falsified data attacks. Falsified trajectories are behaviorally distinct from normal trajectories because they must accomplish a certain attack goal; thus, the problem of identifying falsified trajectories is considered an abnormal trajectory identification problem. A trajectory-embedding model is developed to generate vector representations of trajectory data. The similarity (distance) between each pair of trajectories can be computed based on these vector representations. Hierarchical clustering is then applied to identify abnormal (i.e., falsified) trajectories.

In the final study, a testing platform is built upon a virtual traffic simulator and real-world transportation infrastructure in Mcity. The testing platform integrates the attack study and defense study in a unified framework and is used to evaluate the real-world impact of cyber attacks on CV-TSC systems and the effectiveness of defense strategies.