Data‐based optimal Denial‐of‐Service attack scheduling against robust control based on Q‐learning

Abstract

Attack optimization is an important issue in securing cyber-physical systems. This paper investigates how an attacker should schedule its denial-of-service attacks to degrade the robust performance of a closed-loop system. The measurements of system states are transmitted to a remote controller over a multichannel network. With limited resources, the attacker only has the capacity to jam sparse channels and to decide which channels should be attacked. Under an  2 framework, a data-based optimal attack strategy that uses Q-learning is proposed to maximize the effect on the closed-loop system. The Q-learning algorithm can adaptively learn the optimal attack using data sniffed over the wireless network without requiring a priori knowledge of system parameters. Simulation results sustain the performance of the proposed attack scenario. KEYWORDS Cyber-physical systems, DoS attacks, multichannel transmission, Q-learning, robust control 5178 FIGURE 1 System architecture. DoS, denial of service [Colour figure can be viewed at wileyonlinelibrary.com] on perfect communication if there is no DoS attack, the work of Qin et al 17 further studies the problem of DoS attack strategy design over unreliable networks where the packet dropouts may also occur even in the absence of attacks.Note that the aforementioned results on the optimal DoS attack design are established based on the remote state estimation framework (degrade the estimation quality). To fill this gap from the aspect of closed-loop control performance, the works by Amin et al and Zhang et al 18,19 develop the optimal DoS attack scheduling against linear quadratic Gaussian (LQG) control. Further, the work reported by Zhang and Zheng 20 considers the problem of optimal attack power dispatch for CPSs within the LQG framework. In the work of Gupta et al, 21 a game theory approach is proposed to design the jamming attack strategy, which can destroy a finite-time linear quadratic control performance.Most previous attack strategies require a large amount of exact system information, including physical system parameters and the network topology. However, adversaries are often with limited resources, and as a result, the likelihood to access the complete knowledge of system parameters is either infeasible or very difficult to achieve. 22,23 Recently, some data-driven false data injection attack design schemes have been developed. 4,24 However, there are few studies undertaking an analysis of the data-driven DoS attacks. Moreover, the system robustness, which is the ability of a CPS against exogenous interference, is also an important security metric, which is concerned by both defenders and attackers and has not been analyzed yet.Motivated by these, this paper investigates the problem of DoS attack scheduling against robust control for a CPS with multichannel transmission (see Figure 1). The measurement signals of the CPS are clustered into multiple groups and sent via multiple wireless channels. Here, we assume that the DoS attacker has no knowledge of system parameters, ...

How Machine Learning Changes the Nature of Cyberattacks on IoT Networks: A Survey

“…The searches try to improve the previous works by using multi-channel support in [154]. Here the main objective of the jammer is to find the right channel before carrying out its attack.…”

“…This process learns the jammer's strategy and obtains the optimal anti-jamming policy with little information about the attacker. An et al circumnavigate this countermeasure thanks to the Q-learning algorithm taking into account three types of different rewards: a) the NACK when the communication is public, b) the change of power, and c) the detection of users switching channels [154]. They test their attack into three environments: a) a fixed user, b) a frequency hopping user and c) an anti-jamming user.…”

Static output feedback secure control for cyber‐physical systems based on multisensor scheme against replay attacks

“…3,4 Therefore, how to curb cyber attacks has become an urgent issue. [5][6][7][8][9] As we all know, cyber attacks can be mainly classified into three categories: denial of service (DoS) attacks, 10,11 deception attacks, 12,13 and replay attacks. 14 For instance, a resilient dynamic event-triggered controller was designed to cope with the DoS attacks.…”