Safe Control of Cyber-Physical Systems under False Data Injection Attacks

Li, Zhouchi

Etd

Safe Control of Cyber-Physical Systems under False Data Injection Attacks

Public Deposited

The control of cyber-physical systems relies on real-time measurements from distributed sensors to perceive the environment and make decisions. When low-cost sensors are deployed unattended in open environments, they become inviting targets for malicious adversaries. Malicious adversaries may attempt to degrade the system performance by introducing false data into the sensor measurements resulting in suboptimal decisions. More seriously, the injected false data may drive the system states into an unsafe region. In this thesis, we consider safe and resilient cyber-physical systems operating under false data injection attacks. The goals of the systems are synthesizing controllers to satisfy some critical properties in the adversarial environment and minimize the cost function in the benign scenario. We first investigate the linear quadratic Gaussian reference tracking problem. There are multiple possible compromised sensor sets, each of which corresponds to a different attack scenario. The controller of the system needs to guarantee safety and reachability properties without knowing which attack scenario occurs. We derive the controller as the solution to a quadratically constrained quadratic problem. We further develop the controller in the special case, in which there is only one possible set of compromised sensors. We then study a special category of false data injection attacks, named LiDAR spoofing attack, that disturbs the object detection algorithms of the systems by injecting adversarial points in the point cloud of the LiDAR sensors physically. We develop a new approach to detect and mitigate the LiDAR spoofing attack based on LiDAR sensor measurements shared by neighbor agents. In the proposed system, a Fault Detection, Identification, and Isolation module detects and classifies the attacks, and updates the unsafe region for the agent. A safe controller guarantees the safety of the system based on the updated unsafe region.

Creator