Abstract:As a typical application supported by the Internet of Thing (IoT), smart grids, which are critical complex cyber-physical infrastructures, are facing increasing cybersecurity threats. One major substantial cybersecurity threat to smart grids is false data injection attacks (FDIAs), which could bypass bad data detectors so as to disrupt the operation of power grids. In this work, we consider the locational detection problem for FDIAs, namely, detecting the presence of FDIAs and locating the compromised buses, by utilizing smart grid data (either power injection measurements or system state estimates). Regarding smart grid data residing on the inherent underlying graph structures of power grids, we model smart grid data as non-Euclidean graph signals. We correspondingly develop the FDIA detector based on the Graph Convolutional Attention Network (GCAT) for the locational detection of FDIAs, by considering the underlying graph topology of a power grid. The proposed FDIA detector leverages the graph attention mechanism, which could elastically assign the graph shift operators in the GCAT, to enhance the locational detection performance. Integration of the proposed FDIA detector in a non-invasive way could endow existing power systems with the capability of locational detection of FDIAs. Illustrative simulation results demonstrate the superior locational detection performance of the proposed GCAT-based FDIA detector, compared to the other state-of-the-art FDIA detectors.

Locational Detection of False Data Injection Attacks in Smart Grids: A Graph Convolutional Attention Network Approach