Optimal Innovation‐Based Stealthy Attacks in Networked Control Systems With Asymmetric Information

ABSTRACT This paper studies an optimal innovation‐based stealthy attack strategy for networked control systems (NCSs) with asymmetric information. Different from previous literature on NCSs with a single controller, in the considered NCSs consisting of two controllers where Controller 1 merely uses its own observations to design strategies and Controller 2 uses not only its own observations but also the observations of Controller 1 to perform decisions which brings forth the asymmetric information. Meanwhile, the packet dropout happens in the channel of Controller 1 and the attack occurs in the communication channel of Controller 2 where the innovation signal is tampered. In order to avoid the detector, the Kullback–Leibler divergence (KLD) is employed as a stealthy measure, enabling the attacker to execute strict or relaxed stealthy attacks. The object is to make the quadratic control cost maximized and the attack cost minimized which leads to a non‐convex optimization problem. In virtue of the singular value decomposition, the optimal strict stealthy attack strategy is derived. Moreover, to gain a higher‐attack cost, the optimal relaxed stealthy attack strategy is presented by transforming the non‐convex optimization problem into the convex optimization problem. Numerical examples are given to show the effectiveness of the proposed algorithms.