Resilient Strategies Against Cyberattacks in Network Control Systems

Abstract

Large scale modern control systems involve the interconnection of system components using a communication network. The presence of a network can make these systems vulnerable to cyber attacks, thus compromising overall system’s performance and stability. Cyber attacks in control systems has been a constant topic of research for over a decade, with concentration on three main issues; namely, attack detection, resiliency of the control system, and state estimation under cyber attacks. In this study, we address real-world challenges within this field and aim to enhance the current state of resiliency strategies in network control systems.

In the first part, we study input-to-state stability of nonlinear systems under DoS attack. More specifically, our goal is to obtain a relationship between ISS and DoS attack parameters. We propose a novel model-based dual-mode sampling approach which, depending on the attack intervals, intermittently switches between event-triggered and periodic sampling. We show that the combined model-based state prediction, packetized data transmission, and event-triggered sampling can attenuate the effects of DoS attack on stability.

In the next part of this research, our interest is in the study of one of the most critical forms of deception attacks, known as zero-dynamics attacks \\AM{(ZDAs)} in sampled-data systems. This type of attack excites the internal dynamics of the system resulting in unobservable, stealthy, deviation of the states when the internal system dynamics is non-minimum phase. In this part, our interest is in mitigating the effect of ZDAs in nonlinear sampled-data systems using the multi-rate approach. Our approach consists of analyzing the dissipativity property in the zero-dynamics part of the system and finding conditions on the sampling rate that neutralize the attack. We show that, under some mild conditions, using a multi-rate approach provides a secure nonlinear system against ZDAs by preserving the minimum-phase property.

Then, we consider the practical limitation of the network control system and design a secure control framework that takes advantage of the asynchronous sampling in event-triggered schemes and embeds a sense of sampling zeros dynamics into the triggering threshold. Therefore, the triggering instants happen in such a way that stabilize the zero-dynamics of the sampled-data system. In this way, the event-based controller not only addresses the stability problem and network limitation but also provides a solution to the unstable sampling zeros issue. Thus, aside from all the aforementioned benefits, our proposed method also serves as a self-defence mechanism that confronts ZDAs whenever the system is targeted by an adversary.

Finally, we propose a new method that aims to compensate for the performance loss observed with the previous approach. We develop a model-based event-triggered control setup consisting of a novel triggering condition with an inference-based control rule using a nonlinear model. The key point is to exploit the concept of asynchronous (nonuniform) sampling inherent in the event-triggered mechanism as the main solution. We employ the multiple Lyapunov functional method and determine conditions on the switching signal that produce the desired result. Finally, we analyze the stability of the overall system using Lyapunov theory and discover conditions on the event-triggered parameters and inference-based control law that satisfy the stability criteria and render the zero-dynamics minimum-phase. As a result, \\AM{ZDA}s become ineffective and not a viable option to a malicious attacker.