Considering Time Correlation in the Estimation of Privacy Loss for Consumers with Smart Meters

Abstract

Global electricity smart meter roll-out has brought about serious privacy risks for consumers. The masking of consumer consumption using rechargeable batteries has been studied as a means of protecting consumer privacy. One metric used to measure the effectiveness of such approaches is the empirical mutual information (MI), whose computation requires the estimation of both consumer load and grid-visible load distributions. These distributions have previously been modelled as independent and identically distributed (i.i.d.), or as stationary first-order Markov processes for simplicity. However, consumer load statistics are time-varying in nature, and have inherent intertemporal dependencies. Consequently, the empirical MI based on the stationarity assumption lacks accuracy, resulting in the risk of underestimating the information leakage. In this paper, we propose using features to characterise the change in consumer demand, modelling them as feature-dependent first-order Markov processes to better approximate the actual privacy-loss. Results indicate that this approach is more accurate than i.i.d. models, and in certain cases may be a better empirical estimate of MI compared to stationary first-order Markov models.