Data-Driven Privacy-Enhancement for Cyber-Physical Systems: A Joint-Design Method of Controller and Artificial Noise

The openness and sharing of cyber-physical systems (CPSs) promote privacy to be an increasing concern of cyber security. In view of the fact that the system dynamics of CPSs are usually high-dimensional, coupled and unknown, this paper studies the privacy enhancement of CPSs by optimizing the injection direction of artificial noises with data-driven technologies. However, as the excitation signals, artificial noises and their direction matrix are given in advance. Therefore, optimizing the noise direction matrix will lead to the re-excitation of CPSs, which is a significant challenge for the data-driven privacy-enhancement. To solve this problem, a searchable virtual noise space is introduced to jointly design the controller gain and noise injection direction with the data-driven parametrization method. Then, the privacy of monitored data is optimally enhanced to prevent eavesdroppers from obtaining real data, while ensuring that the influence of the noises on the system performance remains below a specified level. Finally, the proposed method is applied to a VTOL aircraft model, and the simulation results show that the privacy of monitored data is greatly enhanced under a slight loss of the control performance.