Learning-Based Attacks in Cyber-Physical Systems

Abstract

We introduce the problem of learning-based attacks in a simple abstraction of cyber-physical systems-the case of a discrete-time, linear, time-invariant plant that may be subject to an attack that overrides the sensor readings and the controller actions. The attacker attempts to learn the dynamics of the plant and subsequently overrides the controller's actuation signal, to destroy the plant without being detected. The attacker can feed fictitious sensor readings to the controller using its estimate of the plant dynamics and mimic the legitimate plant operation. The controller, on the other hand, is constantly on the lookout for an attack; once the controller detects an attack, it immediately shuts the plant off. In the case of scalar plants, we derive an upper bound on the attacker's deception probability for any measurable control policy when the attacker uses an arbitrary learning algorithm to estimate the system dynamics. We then derive lower bounds for the attacker's deception probability for both scalar and vector plants by assuming an authentication test that inspects the empirical variance of the system disturbance. We also show how the controller can improve the security of the system by superimposing a carefully crafted privacy-enhancing signal on top of the "nominal control policy." Finally, for nonlinear scalar dynamics that belong to the Reproducing Kernel Hilbert Space (RKHS), we investigate the performance of attacks based on nonlinear Gaussian-processes (GP) learning algorithms.

Research on a PSO-H-SVM-Based Intrusion Detection Method for Industrial Robotic Arms

“…Li et al [25] designed a two-loop covert attack on an industrial control system, using least squares support vector machines to construct a covert attack with an attack loop and a covert loop, and verified the effectiveness of the covert attack with a PLC system. Khojasteh et al [26] investigated a learning-based attack method to estimate the dynamics of a system through a nonlinear Gaussian process-based learning algorithm that attacked the control policy. Zhao et al [27] used subspace recognition technology to propose an attack method based on wrong data injection.…”

A Controllable False Data Injection Attack for a Cyber Physical System

“…(16) where Q, R, and G T +1 are positive definite matrices associated with state cost, input cost, and final cost respectively. The optimal control for this cost and resulting Riccati equation are given as in [8]…”

“…With the assumption of chi-square failure detection and perfect knowledge of a control system at the attacker, a sufficient and necessary condition for being perfectly attackable was provided [7]. When the attacker does not have full knowledge of the system, it may try to refine it through learning so that it can carry out more sophisticated attack [8]. The fundamental tradeoffs between the system performance and security were quantified through modeling a dynamical system with Markov decision process (MDP) which can be seen as partially observable MDP (POMDP) in the perspective of the attacker with limited access to the system [9].…”

“…Among many different CPS attacks, FDIA can be considered to be one of the most prominent attacks [5]. The FDIA injects a modified control input [13] or modified measurement [17] [18] or both [8] to degrade the performance of a control system. Even though FDIA was considered as a specific deception attack in the sense that the FDIA as the attack to the state estimator [6], MitM, FDIA, a replay attack and a deception attack [11] can be regarded as the attack of the same class.…”

“…Many of existing FDIAs can be considered as a covert attack which degrades the performance of the control system without being detected as an attack [8]. There can be many different ways to achieve the covert attack.…”