A REVIEW ON AI BASED FAULT AND ANOMALY DETECTION IN POWER SYSTEMS

Abstract

Power systems are becoming more complex because of large-scale integration of renewable energy, the widespread generation of electricity from various sources, electric vehicle charging networks, advanced power electronic converters, and significant automation through digital substations and smart grid technologies. These changes create new operational conditions with less system inertia, two-way power flows, and highly variable generation patterns. This leads to non-linear, rapidly changing, and data-heavy disturbances that are hard to detect and isolate using traditional relay-based protection methods. Conventional protection systems, which rely on preset thresholds, fixed setups, and basic network models, often have trouble with high levels of inverter-based resources, layout changes, and communication delays. In this context, artificial intelligence offers strong data-driven methods that can learn complex system behavior directly from data. It can uncover hidden traits, identify types and locations of faults, spot early and developing anomalies, and assist with smart, adaptable decision-making almost in real time. This review outlines key AI techniques, including machine learning, deep learning, fuzzy logic, and reinforcement learning. It examines how recent studies have used these methods for fault detection, diagnosis, classification, and anomaly monitoring in transmission, distribution, and microgrid environments. Additionally, this paper addresses practical issues such as data quality, limited labels, class imbalance, model generalization across different network setups, computational demands, and constraints for real-time use in digital relays and edge devices.