AdaptEdge: Targeted Universal Adversarial Attacks on Time Series Data in Smart Grids

In recent years, deep learning has become an indispensable tool for various applications, including the analysis and operations of smart grids. These models have shown remarkable success in tasks like power quality disturbance (PQD) classification, enhancing the expansion and efficiency of power infrastructure. However, the susceptibility of deep learning models to adversarial attacks has raised concerns about their reliability and security in critical environments. In this paper, we address this vulnerability by investigating targeted universal adversarial perturbations, specifically in the context of time series data. For the first time, we introduce a novel algorithm called Adaptive Edge (AdaptEdge), which effectively deceives deep learning models working with time series data while balancing the trade-off between fooling rate and imperceptibility of the attack to human observers. Our research represents a significant advancement in adversarial attacks on deep learning models for time series data and emphasizes the need for robust and resilient deep learning models in PQD classification for smart grids. Our results demonstrate a fooling rate of up to 90.78%—the highest achieved thus far—while maintaining an optimal signal-to-noise ratio of 3dB and ensuring signal integrity. This underscores the effectiveness and reliability of our proposed method.