Cyber attacks and defenses against nonlinear state estimation in cyber-physical systems

Abstract

Nowadays, researchers have found many new cyber attacks in the cyber-physical system (CPS), as well as lots of attack-detection methodologies. However, there still are some attacks cannot be caught by existing detection techniques (e.g. bad data detection and machine learning based methods). To study these attacks in CPS is quite important in the research of system security. In this thesis, we focus on cyber attack analysis and the security mechanisms in the nonlinear system, which is more widely used in real world, but draw less attentions (than the linear system) in recent years.

In the first part of this thesis, we aim to apply two types of cyber attacks (proposed in the linear system) to the nonlinear system: false data injection (FDI) attack, and data framing (DF) attack. To implement the FDI attack, we propose an easy-to-implement approach: the adversary does not need the knowledge of system states (which are difficult to obtain, even for insiders). To our knowledge, there are no research results that are able to give an attack to any buses without the knowledge of system states. Furthermore, we study the mechanisms of DF attack, and aim to implement the first attack strategy in the nonlinear system. Our technique is to maximize the expected energy of the normalized residuals on one hand, and put the residuals of the adversary meters as small as possible at the same time. We show that these attacks can be successfully launched in nonlinear CPS.

In the second part of this thesis, we explore the techniques of detecting the cyber attacks in nonlinear systems. Many traditional works, e.g. bad data detection (BDD) and machine learning (ML) based methods, have been proposed for detecting cyber attacks. However, most of these methods do not work well for nonlinear systems. In our work, we propose a ``first difference" aware machine-learning (FDML) method to detect the attacks. We will show that the ``first difference" aware features used in our system are very powerful for detecting cyber attacks. Also, we have proposed another detection method using sparse optimization for determining the measurement signals. The simulation results show the effectiveness of these two defense techniques. We hope that the works presented in this thesis could provide more insights for researchers to come up with a more robust and sophisticated solution to avoid more cyber attacks.