Localized Privacy Protection in a Cooperative LQG Control System with Multiple Sensors

In this paper, we study the privacy preservation problem in a cooperative LQG control system with multiple sensors. We consider the user-server system, which consists of two parts, a user and a server. The user owns the plant, and it has
$N$
sensors to measure its states. In order to compute the optimal LQG control inputs, it employs the server who has computation ability. To achieve the cooperation, the user needs to provide certain private information about the states. We propose a novel method to protect the user's privacy, where a localized privacy scheme is employed, so that the server can only obtain “incorrect” and deviated knowledge of the private information. Specifically, the privacy scheme is achieved by adding noises to the measurements of the states to be sent to the server. Meanwhile, we present a privacy metric which quantifies the privacy. Then, we formulate an optimization problem to analyze the tradeoff between the privacy preservation performance and the LQG control loss. This problem can be efficiently solved by numerical methods.