Improving resilience of cyber physical power networks against Time Synchronization Attacks (TSAs) using deep learning and spline interpolation with real-time validation

The integration of high-speed communication networks and synchrophasors into smart grids has significantly improved real-time monitoring and control accuracy. However, the increased reliance on communication infrastructure has also heightened the vulnerability of the power networks to cyber intrusions. Synchronized phasor data and GPS time-stamping used by Phasor Measurement Units (PMUs), make them prime targets for cyber intrusions. Among the different types of intrusions in smart grids, Time Synchronization Attacks (TSA), because of their impact and easier execution, are widely employed by intruders to disrupt grid operations. Such attacks aim at spoofing GPS signals, thereby altering voltage and current phasor information across the network. The same leads to malfunction of the operations executed at the control center. The present work aims to develop a secured and resilient mechanism against TSAs in smart grids. In this regard, a two-stage mechanism based on deep learning and spline interpolation is proposed. The first stage employs an LSTM-based classifier to detect TSAs in the cyber layer. Post-TSA detection, the second stage uses spline interpolation to filter out malicious data. The filtering allows for the restoration of the actual pre-TSA data acquired from PMUs. The proposed TSA detection and correction scheme has been validated extensively across various TSA scenarios on IEEE 9, 14, and 57 bus systems. Majority of the reported works of TSA detection have been validated using offline numerical simulations, which have limitations in replicating practical TSA dynamics. To address the same, the proposed scheme has been validated using a real-time testbed comprising of a digital simulator, real PMU, GPS receiver, and a data acquisition module with a communication interface.