Power System Transient Stability Prediction in the Face of Cyber Attacks: Employing LSTM-AE to Combat Falsified PMU Data

Abstract

Phasor measurement units (PMUs) are essential instruments in delivering real-time data crucial for monitoring the dynamics of power systems. They are widely used in transient stability prediction (TSP), significantly contributing to the effective maintenance of power systems post-contingency stability. The accuracy and reliability of data derived from PMUs are crucial for the effective execution of TSP. However, the PMU data is at risk of being compromised by false data injection (FDI) attacks. Such vulnerabilities could lead to a significant degradation in the reliability of the data, potentially resulting in the misdirection of algorithms tailored for TSP. In response to this challenge, this article presents a resilient approach for TSP capable of functioning effectively under FDI attacks. Utilizing a long short-term memory autoencoder (LSTM-AE), our proposed method is engineered to proficiently capture and learn the normative spatial and temporal correlations and patterns present in time-series PMU data, across both steady-state and transient operational states. Consequently, this approach facilitates the algorithmic reconstruction and rectification of PMU measurements that have been compromised due to FDI, thereby upholding the robustness of the TSP process in the face of cyber threats. The performance of the proposed method is validated using the IEEE 39-bus system, subjected to a wide array of scenarios. This rigorous testing demonstrates the algorithm's robustness and effectiveness in maintaining accurate TSP in scenarios where the integrity of PMU data is professionally compromised to avoid easy detection or reconstruction.