AI-driven Fault Detection and Isolation in Smart Grids: A Review and Lightweight Framework Proposal

Abstract

Smart grids integrate renewable generation, distributed loads, and intelligent control to improve grid reliability and efficiency. Fault Detection and Isolation (FDI) are critical functions for maintaining the stability, especially under the increased variability from distributed energy resources (DERs) and microgrids.. Conventional protection (e.g., overcurrent relays) struggles with dynamic fault currents and evolving topologies, motivating AI-based methods. This paper surveys deep learning (DL) techniques for smart grid FDI, including convolutional and recurrent networks, autoencoders, graph neural networks (GNNs), and hybrid models. We categorize common fault types (phase-to-ground, phase-to-phase, series, high-impedance, and cyber-physical faults) and review state-of-the-art DL applications for each. Finally, we propose a conceptual Lightweight FDI Framework that leverages edge computing and federated learning to enable scalable detection across centralized and decentralized grid architectures. The framework emphasizes real-time processing, model simplicity, and adaptability. The paper concludes with a discussion of key challenges (data scarcity, evolving grid conditions, cybersecurity, interpretability) and future research directions.

Keywords — Smart grid; fault detection; fault classification; deep learning; convolutional neural networks; recurrent neural networks; autoencoders; graph neural networks; edge computing; federated learning.