Integrated Deep Learning-Based Detection and Robust State Observer Mitigation of Time-Synchronization Attacks: A WADC Case Study

Wide-area damping controllers (WADCs) have become increasingly prevalent in modern power grids to deal with large-scale disturbances. The necessary measurement data for WADCs are acquired from geographically scattered phasor measurement units (PMUs) and transmitted through IEEE C37.118 communication channels to phasor data concentrators (PDCs) for being augmented and time-aligned. In such a system, time-synchronization attacks (TSAs) on PMUs and PDCs can compromise PMU data collection, leading to dropout in critical measurements, undermining the performance of WADCs, and threatening system stability. To tackle this issue, this paper first formulates an optimal TSA model that exploits the vulnerability of PDC time-alignment strategies to target WADC modal observability. Afterward, to counter TSAs targeting WADCs, this paper proposes an integrated TSA detection and mitigation scheme. The TSA detection scheme is a convolutional neural network (CNN) model which captures the temporal correlation of data quality information at the regional PDC level to identify and localize TSAs while distinguishing them from communication network incidents. Based on the output of the deep CNN model, an extension to the collected IEEE C37.118 data stream is proposed to indicate the occurrence of TSA in each PMU. If this extension is enabled for any PMU, the WADC switches to the proposed robust state observer mitigation scheme, which is adaptive based on the location of the detected TSA. The observer is designed using a set of linear matrix inequality (LMI) conditions derived based on the Lyapunov stability theory. Simulation results on Two-Area Kundur, IEEE 39-Bus, and 14-Generator Model of the Southeastern Australian Power System demonstrate the impact of the TSA and verify the effectiveness of the proposed detection and mitigation schemes.