Intelligent Fault Detection in Power Grids Using Deep Learning Algorithms

Modern power systems are increasingly complex and interconnected, demanding intelligent, adaptive solutions for fault detection and operational resilience. Traditional methods often depend on linear approximations or computationally demanding simulations, which struggle to capture the nonlinear, time-varying dynamics of real-world power networks. To address these challenges, this study proposes an advanced framework that integrates spectral graph theory with deep learning, offering a more efficient and accurate approach to fault detection. Central to the framework is the Spectral-Temporal Energy Flow Graph (STEF-Graph), which models spatial and temporal interactions within the grid. By applying spectral filters in the graph’s eigenbasis, the system learns high-level spatial features reflective of the network topology. Concurrently, Gated Recurrent Units (GRUs) process temporal sequences, enabling robust predictions of system states under dynamic conditions. To ensure practical viability, a projection operator is applied to enforce physical constraints such as voltage limits, line capacity, and stability requirements. This ensures all predictions remain operationally feasible. Based on these predictive insights, the Graph-Informed Energy Adjustment Mechanism (GEAM) is introduced to guide control decisions. GEAM balances localized efficiency with system-wide coordination using a graph-regularized penalty, ensuring both adaptability and coherence. STEF-Graph and GEAM jointly constitute an integrated two-module framework that boosts fault detection precision while strengthening the grid’s adaptability to disruptions and fluctuations. This research contributes to the development of scalable, data-driven methodologies aligned with the evolving demands of modern energy systems, and reflects the journal’s emphasis on innovative computational methods for complex infrastructures.