Smart Meter Data Masking Using Conditional Generative Adversarial Networks

Abstract

We present a novel two-stage smart meter (SM) data masking technique. In the first stage, data masking is carried out at an individual SM using a light-weight approach. Subsequently, the data are transmitted to a third party (TP), where the second-stage data masking is carried out using a conditional generative adversarial networks (CGAN)-based module. The second-stage masked data have the same statistics as those of the actual data; however, their individual values are different. The data are then transmitted to the energy supplier (ES) which uses them to calculate their statistics or aggregate them. The actual SM data are protected from eavesdropper, as well as the TP and ES. We compare the performance of the proposed method with existing GAN-based and Gaussian mixture model (GMM)-based techniques by calculating the mean absolute error (MAE) between the actual mean and standard deviation of the consumer data and those calculated from the second-stage masked data. The results show the lowest MAE obtained with the proposed technique for the estimated mean and standard deviation are at least three times less compared to the MAE values for the GAN-based approach, and between two to three times less than those obtained for the GMM-based method.