Abstract

The Smart Grid (SG) is an upgraded, intelligent, and a more reliable version of the traditional Power Grid due to the integration of information and communication technologies. The operation of the SG requires a dense communication network to link all its components. But such a network renders it prone to cyber attacks jeopardizing the integrity and security of the communicated data between the physical electric grid and the control centers.
One of the most prominent components of the SG are Wide Area Monitoring Systems (WAMS). WAMS are a modern platform for grid-wide information,
communication, and coordination that play a major role in maintaining the stability of the grid against major disturbances.
In this thesis, an anomaly detection framework is proposed to identify False Data Injection (FDI) attacks in WAMS using different Machine Learning (ML) and Deep Learning (DL) techniques, i.e., Deep Autoencoders (DAE), Long-Short Term Memory (LSTM), and One-Class Support Vector Machine (OC-SVM). These algorithms leverage diverse, complex, and high-volume power measurements coming from communications between different components of the grid to detect intelligent FDI attacks. The injected false data is assumed to target several major WAMS monitoring applications, such as Voltage Stability Monitoring (VSM), and Phase Angle Monitoring (PAM). The attack vector is considered to be smartly crafted based on the power system data, so that it can pass the conventional bad data detection schemes and remain stealthy. Due to the lack of realistic attack data, machine learning-based anomaly detection techniques are used to detect FDI attacks. To demonstrate the impact of attacks on the realistic WAMS traffic and to show the effectiveness of the proposed detection framework, a Hardware-In-the-Loop (HIL) co-simulation testbed is developed. The performance of the implemented techniques is compared on the testbed data using different metrics: Accuracy, F1 score, and False Positive Rate (FPR) and False Negative Rate (FNR). The IEEE 9-bus and IEEE 39-bus systems are used as benchmarks to investigate the framework scalability. The experimental results prove the effectiveness of the proposed models in detecting FDI attacks in WAMS.