Sensor attack detection and identification for cyber-physical systems: A data-driven approach

This paper investigates the problem of sensor attack detection and identification in cyber-physical systems, leveraging the advantage of zonotopes in dealing with stochastic properties. Unlike previous research that relies on system dynamics knowledge to infer safety boundaries and monitoring schemes, the proposed approach is geared toward addressing the challenges posed by unknown system dynamics, attack strategies, and attack locations. Firstly, we analyze the feasibility of using zonotopes for attack detection and deduce the necessary information quantity and observation window length requirement for effective detection. Subsequently, a zonotopes-based algorithm is proposed for computing the over-approximated reachable set and measurement set from noisy data. Then, an attack detection and identification strategy based on the predicted measurement set is developed. To reduce the computational complexity of the presented detection method, a column truncation algorithm is proposed. The effectiveness of the proposed method is validated through numerical simulations.