Deep learning and machine learning methods for detecting false data injection cyber-attacks on smart grid phasor measurement units

Compared to regular power systems, smart grids offer various advantages. However, they also present significant cybersecurity challenges, particularly in terms of data integrity and security. Phasor Measurement Units (PMUs) play a crucial role in addressing these challenges by providing real-time monitoring and management of the grid through synchronized, high-resolution measurements, thus improving its reliability. Unfortunately, hackers can exploit PMUs by manipulating the vital information they provide. False data injection attacks, which rely on incorrect data input into PMUs, pose a serious threat as they can compromise the efficiency and reliability of the entire power grid. Therefore, secure operations in smart grids depend on effective false data detection and mitigation. In this research, we focus on developing a comprehensive solution powered by both machine learning and deep learning methods to recognize false data intrusion attacks on smart grids. This involves utilizing supervised machine learning models trained on power system attack-labeled data. Tree-based techniques such as Random Forests and Extra Trees are initially employed for binary classification, categorizing PMU data as either natural or compromised. Subsequently, multi-class classification is performed to identify no events, natural events, or active attacks. Furthermore, we explore deep learning architectures like Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and CNN-LSTM hybrids, which are specifically designed for time-series data and can effectively detect and capture subtle temporal patterns. The primary objective of this research is to thoroughly evaluate the state-of-the-art machine and deep learning techniques on real PMU data, leading to a significant improvement in attack detection accuracy and ultimately protecting the security and resilience of critical energy supply infrastructure.