Cyberattack Detection for a Class of Nonlinear Multi-Agent Systems Using Set-Membership Fuzzy Filtering

This paper studies cyberattack detection in discrete-time, leader-following, nonlinear, multi-agent systems subject to unknown but bounded (UBB) system noise. The Takagi-Sugeno (T-S) fuzzy model is used to approximate the nonlinear systems over the true value of the state. A new method is developed for simultaneous distributed cyberattack detection and leader-following consensus control. The approach is based on a fuzzy set-membership filtering technique that consists of two steps: a prediction step and a measurement update step. An estimation ellipsoid set is found by updating the prediction ellipsoid set with the current sensor measurement data. A criterion is provided to detect cyberattacks that inject malicious signals into sensor data. The criterion is based on the intersection between the ellipsoid sets. If there is no intersection between the prediction set and the estimation set of an agent at the current time instant, a cyberattack on its sensors is declared. Recursive algorithms for solving the consensus protocol and calculating the two ellipsoid sets for detecting attacks are proposed. Moreover, a recovery mechanism that can mitigate the adversarial effects of cyberattacks is introduced. Simulation results are provided to demonstrate the effectiveness of the proposed method when replay attacks occur on the sensor data.