Modeling and Detection of Cyber-Attacks in UAV Swarms using a 2D-LWR Model and Gaussian Processes


This paper considers a class of cyber-attacks attacking a swarm of Unmanned Aerial Vehicles (UAVs). Our focus is on scenarios wherein an attacker may hack into a subset of vehicles in the swarm and create subtle changes in their parameters. These hacked vehicles (referred to as malicious vehicles) are subsequently able to modify the behavior of the overall swarm. The swarm comprising the mix of malicious and normal vehicles is modeled using a system of coupled Partial Differential Equations (PDEs) in a two-dimensional LWR model. We develop a methodology that combines Gaussian Processes (GP) with this two-species 2D PDE model, and use this method for detecting the presence of such malicious vehicles in the swarm. A Bayesian Optimization scheme is employed to determine the optimal choice of basis and kernel functions that constitute the GP. Simulation results demonstrate that this detection architecture performs successful detection of the malicious vehicles, and also their mode of attack on the traffic.