Evaluating Privacy Measures for Load Hiding

Vadim Arzamasov,Klemens Böhm

2024-08-13

Abstract:In smart grids, the use of smart meters to measure electricity consumption at a household level raises privacy concerns. To address them, researchers have designed various load hiding algorithms that manipulate the electricity consumption measured. To compare how well these algorithms preserve privacy, various privacy measures have been proposed. However, there currently is no consensus on which privacy measure is most appropriate to use. In this study, we aim to identify the most effective privacy measure(s) for load hiding algorithms. We have crafted a series of experiments to assess the effectiveness of these measures. found 20 of the 25 measures studied to be ineffective. Next, focused on the well-known "appliance usage" secret, we have designed synthetic data to find the measure that best deals with this secret. We observe that such a measure, a variant of mutual information, actually exists.

Cryptography and Security,Information Theory

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the privacy issues arising from smart meters in smart grids when measuring household - level power consumption. Specifically, researchers have designed a variety of load - hiding algorithms to manipulate the measured power consumption data in order to protect user privacy. However, there is currently no consensus on which privacy measure is most suitable for evaluating the effectiveness of these algorithms.

To determine the most effective privacy measurement method, the author carried out the following work:

1. **Compile the types of secrets in the literature**: The author sorted out various types of secret information (such as household appliance usage, living habits, etc.) that can be detected through load curves mentioned in the literature, and observed that so far, the relationship between load - hiding algorithms and privacy measures has not been systematically studied.
2. **Propose general requirements**: For 25 popular privacy measurement methods, the author proposed natural requirements and tested whether these measurements meet these requirements. The results show that only five measurements passed at least six tests.
3. **Measurement consistency analysis**: The author compared the behaviors of different privacy measurements and found that only two measurements showed similar behaviors and could be used interchangeably.
4. **Evaluation of the hiding effect of specific secrets**: The author used synthetic data to evaluate the ability of privacy measurements in quantifying the hiding effect of load - hiding algorithms on specific secrets (such as household appliance usage).
5. **Estimator selection**: For the two better - performing measurements (MIi and H), the author compared the effects of the commonly - used histogram estimator and the alternative nearest - neighbor - based estimator, and found that the latter performed better in some cases.

Finally, the author found that a variant of mutual information (MIi) is the most effective privacy measurement method, especially in evaluating the hiding effect of load - hiding algorithms on the "household appliance usage" secret. In addition, the author also explored the problem of the best estimator selection for calculating MIi, pointing out that the nearest - neighbor estimator may be superior to the commonly - used histogram estimator.

In summary, this paper aims to identify the measurement method most suitable for evaluating the privacy protection effect of load - hiding algorithms through a series of experiments and analyses, and provides a reference framework for future research.