A Physics-Informed Neural Network Approach Towards Cyber Attack Detection in Vehicle Platoons

Connected and Autonomous Vehicles (CAVs) are seen as a promising solution to reduce traffic congestion, improve passenger comfort and fuel economy. Although CAVs address such needs in an effective way, they are vulnerable to cyber attacks due to their extensive utilization of communication networks. In light of this problem, we present a cyber attack detection framework for a vehicle platoon based on physics-informed neural network (PINN) framework. The proposed algorithm exploits the physics based model of the platoon as well as limited available data to detect and distinguish cyber-attacks from various sources, namely, attacks affecting communication network and attacks affecting local vehicular sensors. Essentially, the PINN framework learns an uncertain parameter from the physics model and utilizes the learned parameter knowledge to infer attack scenarios. Finally, as shown through the simulation studies, the proposed algorithm is able to detect and distinguish various cyber attacks showing its potential.