Analysis of Data-Driven Detection and Localization of Cyberattacks on Faulty Electric Vehicle Platoons

Connected autonomous vehicle (CAV) platooning enhances the perception and decision-making abilities of autonomous vehicles to improve traffic efficiency. However, intervehicle communications introduce vulnerabilities to cyberattacks, which can lead to unsafe conditions. To this end, data-driven CAV intrusion detection and classification have been the subject of recent work. Concurrently, adoption of electric vehicles (EVs) is growing to meet pressing environmental targets, and the dynamics of EVs offer many advantages that qualify them over conventional vehicles for platooning applications. We identify a need to examine the cyberphysical security of CAEVs, where EV faults occur simultaneously with cyberattacks and the modified dynamics of a faulty vehicle can affect the performance of an attack classifier. In this paper, we explicitly model EV motor dynamics and simulate platooning behavior in healthy, under-attack, under-fault, and simultaneous attack and fault scenarios. We propose the use of an attention-based attack classifier and compare its performance to traditional machine learning architectures in simultaneous attack and fault scenarios. Our results, based on platoon velocity data, demonstrate that the attention-based classifier outperforms other methods in most scenarios except for EVs operating with low battery state-of-charge. We highlight the need for further research on attack classifiers robust to physical fault scenarios.