Dynamic Adaptation Gain for Threat Discrimination
Kangkang Zhang, Kaiwen Chen, Marios M. Polycarpou, Thomas Parisini · 2024
This paper proposes an adaptive-observer-based threat discrimination method for systems with disturbances, aiming to identify the occurring threat type: component faults or cyber attacks. Stealthy attacks are typically exponentially decaying with time and only slightly alter the system outputs, the effects of which can be easily inundated by non-attacker-incurred disturbances. To this end, an integrator is applied to the system output to retain the effects of stealthy attacks. Compared to the classical integral-type adaptive observers with a constant adaptation gain, a dynamic adaptation gain is exploited to provide additional degrees of design freedom for frequency-domain loop-shaping. This allows to apply distinct threat/disturbance-to-residual gains in the frequency intervals to which the threats and the disturbances belong, respectively, thereby improving the threat discrimination performance. A numerical example to demonstrate the effectiveness of the proposed methodology is presented.