Attack Detection and Estimation for Cyber-Physical Systems by using Learning Methodology
Networked control systems (NCSs) are considered as a special form of cyber-physical systems (CPS) where the cyber-layer is coupled with the physical system through a wired or wireless network. In NCSs, both the communication links and the physical system are vulnerable to a variety of attacks, and hence, it is of utmost importance to detect them and mitigate their effect on the system. The network and physical system dynamics will become stochastic and uncertain due to packet losses and random delays from the network. Moreover, the overall NCS can become unstable with an increase in delays and packet losses due to attacks. In this chapter, both cyber- (or network) and physical system dynamics of an NCS are considered linear initially and then nonlinear. In both cases, by using a state-space representation of the communication network, controllers are designed to stabilize the traffic flow in the network and the NCS by using Q-learning in the case of a linear NCS and neural networks in the case of a nonlinear NCS. The attack detection and estimation are accomplished by using residual signals that are generated from novel adaptive observers. Also, the detectability conditions are derived and the maximum delay and packet loss bound are given before the NCS becomes unstable. Simulation results are given.
H. Niu et al., "Attack Detection and Estimation for Cyber-Physical Systems by using Learning Methodology," Artificial Neural Networks in Engineering Applications, pp. 107-126, Elsevier, Jan 2019.
The definitive version is available at https://doi.org/10.1016/B978-0-12-818247-5.00018-6
Mechanical and Aerospace Engineering
Electrical and Computer Engineering
Intelligent Systems Center
International Standard Book Number (ISBN)
Book - Chapter
© 2019 Elsevier, All rights reserved.
01 Jan 2019