Privacy-aware Design of Distributed Networked Control Systems: A Directed Information Approach

 Distributed networked control systems (NCSs) benefit from high scalability, fault tolerance, and flexibility. As a result, they have been tightly integrated with many engineering systems, e.g., the smart grid, intelligent transportation systems, and smart buildings. Despite their significant benefits, NCSs are at risk of privacy breaches due to (i) their integration with information and communication technologies and (ii) widespread usage of third-party computing services, e.g., cloud computing. Privacy breaches may have significant consequences for the operators and users of NCSs, e.g., diminished reputation and loss of revenue. This project will address two important questions regarding the privacy of distributed NCSs: (i) How to quantify the privacy level of a distributed NCS? and (ii) How to optimally design a distributed NCS under privacy requirements? We first proposedirected informationas a novel privacy metric to quantify the privacy level of distributed NCS. Directed information is an information-theoretic concept that measures the flow of information between stochastic processes. Using this metric, we quantify the flow of private information in distributed NCSs and investigate the impact of system parameters, e.g., communication topology, on their privacy level. These results can be used to assess the privacy level of various distributed NCSs. We next develop computational frameworks for the jointlyoptimal design of privacy filters, estimators, and controllers in distributed NCSs.We cast the privacy-aware design of distributed NCSs as a decentralized optimal control problem wherein the objective is to optimize the closed-loop performance (estimation loss) subject to a privacy level captured by directed information. We will study the structural properties of the optimal decentralized privacy filters, estimators, and controllers using two different techniques: thecommon information approachand theperson-by- person approach.We then developefficient numerical algorithmsfor the optimal privacy-aware design of distributed NCSs and study theoptimal trade-offbetween privacy and closed-loop performance (estimation loss).  Finally, we integrate all the developed results andvalidate the framework using a real practical building automation application.Occupancy (the number of occupants in different building zones) is highly sensitive information. However, occupancy can be accurately inferred from the sensor measurements of the heating, ventilation, and air-conditioning (HVAC) systems. Using directed information, we first characterize the flow of private information in HVAC systems. We then study the optimal design of privacy filters and controllers for HVAC systems. We finallycompare and contrastthe project’s outcomes with state-of-the-art privacy protection solutions using a building simulator.