Model-Based Distributed Strategies for Detection and Isolation of Covert Cyber-Attacks in Networked Systems

This thesis project investigates the cyber-security problem for linear interconnected systems in a distributed fashion. Starting from existing results on the detectability of covert cyber-attacks in a single agent of the network, the work addresses the isolation task, proposing different algorithms based on the algebraic properties of the interconnection matrices of each local neighborhood. Moreover, the detection problem is extended to the scenario of multiple agents simultaneously attacked. The whole theoretical analysis focuses on large-scale systems subject to bounded process and measurement disturbances. All the proposed methodologies can be implemented by using only local information available at each subsystem, and are endowed with a suitable threshold to avoid false alarms due to the action of noise. Finally, numerical simulations on a simple data center model are given, showing the effectiveness of the introduced techniques in detecting and isolating covert cyber-attacks.