Deep Directed Information-Based Learning for Privacy-Preserving Smart Meter Data Release

Abstract

Smart Meters (SMs) are an important component of smart electrical grids, but they have also generated serious concerns about privacy data of consumers. In this paper, we present a general formulation of the privacy-preserving problem in SMs from an information-theoretic perspective. In order to capture the casual time series structure of the power measurements, we employ Directed Information (DI) as an adequate measure of privacy. On the other hand, to cope with a variety of potential applications of SMs data, we study different distortion measures along with the standard squared-error distortion. This formulation leads to a quite general training objective (or loss) which is optimized under a deep learning adversarial framework where two Recurrent Neural Networks (RNNs), referred to as the releaser and the attacker, are trained with opposite goals. An exhaustive empirical study is then performed to validate the proposed approach for different privacy problems in three actual data sets. Finally, we study the impact of the data mismatch problem, which occurs when the releaser and the attacker have different training data sets and show that privacy may not require a large level of distortion in real-world scenarios.

Review of results on smart‐meter privacy by data manipulation, demand shaping, and load scheduling

“…This way, the directed information would not put a higher emphasis on information leakage from the past outputs and balances all leakages equally. In the context of smart grid privacy, directed information has been recently used in [125].…”

Privacy-Cost Management in Smart Meters Using Deep Reinforcement Learning

“…To protect consumers' sensitive information, two main families of SM data privacy enabling techniques have been proposed in the literature: i) SM data manipulation methods [2]- [5]; and ii) demand shaping methods [6]- [12]. In the first family, the SM data are processed and manipulated, e.g., by adding random noise, before sharing that with the UP.…”

Privacy-Preserving Adversarial Network (PPAN) for Continuous non-Gaussian Attributes

“…An information-theoretical approach based on distorting useful data was proposed in [1]- [5] where using the Kullback-Leibler(KL) divergence, the privacy risk was quantified as the mutual information between private attribute X and released/shared data Z. In [1] for the Gaussian attributes, they used a numerical procedure (a modification of the steepest descent algorithm) to solve this privacy-distortion trade-off.…”

“…The reason could be this fact that generally there is no closed-form solution for the privacy-distortion trade-off. In practical applications, there are several cases (such as those related to smart meters [5]) where attributes come from a continuous range but with the non-Gaussian distribution. Therefore, the natural challenge raised here is that how the PPAN model can be applied and evaluated in these cases?…”