Anomaly detection and string stability analysis in connected automated vehicular platoons
Anomaly detection and string stability analysis in connected automated vehicular platoons
复制标题
DOI：
10.1016/j.trc.2023.104114
复制
发表时间：
2023-06
期刊：
影响因子：
--
通讯作者：
Yiyang Wang;Ruixuan Zhang;Neda Masoud;Henry X. Liu
中科院分区：
文献类型：
--
作者：
Yiyang Wang;Ruixuan Zhang;Neda Masoud;Henry X. Liu
在这项研究中，我们开发了一个全面的框架来模拟网络攻击对互联和自动化车队的安全性，安全性和首尾稳定性的影响。首先，我们提出了一个通用的队列动力学模型与异构的时间延迟，可能源于通信信道和/或车载传感器。基于所提出的动力学模型，我们开发了一个增广状态扩展卡尔曼滤波器（ASEKF）平滑传感器读数，并使用它与异常检测器检测传感器异常。具体来说，我们考虑两个检测器：一个参数检测器，χ 2检测器，和一个基于学习的检测器，一类支持向量机（OCSVM）。我们研究了车辆动力学模型（EKF和ASEKF）和检测器（χ 2和OCSVM）的所有组合的检测能力。此外，我们引入了一个新的概念，在字符串稳定性，即伪字符串稳定性，以衡量一个排的字符串稳定性下的网络攻击和模型的不确定性。我们证明了伪字符串的稳定性的一个排和它的检测率之间的关系，这使我们能够识别的关键检测灵敏度/召回，排的成员应该满足排保持伪字符串稳定。
In this study, we develop a comprehensive framework to model the impact of cyberattacks on safety, security, and head-to-tail stability of connected and automated vehicular platoons. First, we propose a general platoon dynamics model with heterogeneous time delays that may originate from the communication channel and/or vehicle onboard sensors. Based on the proposed dynamics model, we develop an augmented state extended Kalman filter (ASEKF) to smooth sensor readings, and use it in conjunction with an anomaly detector to detect sensor anomalies. Specifically, we consider two detectors: a parametric detector, the χ 2-detector, and a learning-based detector, the one class support vector machine (OCSVM). We investigate the detection power of all combinations of vehicle dynamics models (EKF and ASEKF) and detectors (χ 2 and OCSVM). Furthermore, we introduce a novel concept in string stability, namely, pseudo string stability, to measure a platoon’s string stability under cyberattacks and model uncertainties. We demonstrate the relationship between the pseudo string stability of a platoon and its detection rate, which enables us to identify the critical detection sensitivity/recall that the platoon’s members should meet for the platoon to remain pseudo string stable.