Privacy Preserving Predictive Analytics with Smart Meters

Abstract

Smart meter data analysis provides key insights about energy demand and usage patterns for efficient operation of power generation and distribution companies. The increase in modern communication bandwidth enables smart meters to transmit the data to a corresponding utility company at hourly update rates or faster. Analysing such large amount of data often requires a high performance cloud computing environment. However, using such environment may lead to exposure of energy consumption patterns of individual households, with the potential consequence of damaging privacy breaches. To mitigate the risk of a privacy breach, this paper proposes a secure linear regression model for smart meter data analytics, based on a Partially Homomorphic Encryption algorithm. In the proposed method, the primary variable; here, the power reading, is encrypted. The statistical coefficients are then computed directly from the cyphertext using integer mappings. With this approach, a computationally feasible linear regression is achievable without compromising a detailed household energy usage profile. Simulation experiments are conducted that demonstrate the performance of proposed method with respect to accuracy and computational complexity.