Information-theoretic privacy in smart meters

Abstract

Smart grids promise to enhance drastically the efficiency of today's power networks. One of the key components of smart grids is the smart meter, which allows to monitor a user's electricity consumption with much higher resolution compared to conventional energy meters. However, the high resolution of smart meter measurements also entails serious privacy implications for the users, as it makes easier to distinguish the power signature of single appliances from the aggregate household consumption. This would lead an attacker, which may be a thief, a surveillance agency, or the utility provider itself, to gain insights into users' activities and behaviors. In this dissertation we analyze several privacy-preserving techniques that protect users' privacy without diminishing the utility of smart grids. We adopt physical resources that are increasingly available at the users' premises, such as renewable energy sources and rechargeable batteries, and use them to minimize the information leaked about a user's electricity consumption, as well as the cost of energy. We deploy information-theoretic tools to characterize the fundamental limits of smart meter privacy, measuring privacy via mutual information, and characterizing single-letter expressions for the information leaked when considering infinite and zero-capacity rechargeable batteries. These scenarios represent lower and upper bounds on the privacy performance of more realistic settings with finite-capacity rechargeable batteries. When considering a finite-capacity battery, we express the information leakage as an additive quantity, and cast the problem as a stochastic control problem, which can be solved by dynamic programming. We also propose more empirical privacy-preserving strategies, testing their performance against real smart meter measurements and time of use pricing tariffs. In particular, we measure privacy as the squared difference between the smart meter measurements and a target profile, which we consider as a completely private power profile, and characterize the optimal trade-off between privacy and cost of energy.