Anomaly Detection and Prevention in Smart Grid Transmission System Using Machine Learning

Modern power transmission networks are evolving rapidly with the growing integration of communication, automation, and control technologies. While this transformation enhances real-time monitoring and operational flexibility, it also introduces new challenges in ensuring reliability, stability, and cybersecurity. Transmission lines remain prone to physical anomalies such as faults due to their long-distance exposure, while increasing digitalization has made them vulnerable to cyber anomalies like False Data Injection Attacks (FDIAs) and Denial-of-Service (DoS) attacks. Conventional protection schemes, based on predefined thresholds and static models, struggle to adapt to these evolving threats, highlighting the need for intelligent, learning-based solutions capable of accurately identifying and localizing system disturbances. To address these challenges, this research develops a series of machine learning–based frameworks to improve anomaly detection, isolation, and localization in transmission systems. Unsupervised learning methods identify operational anomalies without labeled data, while a data-driven Optimal Phasor Measurement Unit (PMU) Placement approach enhances network observability. Hybrid learning-based FDIL techniques provide accurate fault localization under diverse operating conditions. For cyber anomaly detection, a multilabel learning framework identifies and localizes FDIAs across multiple sensors, improving the precision of attack localization. A complementary forecasting-based approach maintains observability and detection reliability under simultaneous FDIA–DoS conditions, ensuring system continuity during partial data loss. These frameworks are validated through simulations and Hardware-In-the-Loop (HIL) experiments on benchmark transmission systems with varying sizes and topologies, demonstrating their reliability, scalability, and resilience.