AI-Based Fault Detection and Classification in Smart Distribution Grids

Abstract

The reliable operation of smart distribution grids depends heavily on the timely detection and accurate classification of faults. Traditional fault detection methods, such as impedance and threshold-based analysis, provide essential diagnostic insights but are limited under noisy and dynamic operating conditions. This study proposes a hybrid framework combining MATLAB/Simulink-based modeling with Artificial Intelligence (AI) to enhance fault detection and classification in smart distribution grids. Fault scenarios including single line-to-ground (SL_G), line-to-line (L_L), double line-to-ground (LL_G), three-phase (LLL), and no-fault conditions were simulated to generate a labeled dataset of 9201 samples. The dataset was then used to train two AI models: a Multilayer Perceptron (MLP) and an Extreme Gradient Boosting (XGBoost) classifier. The MLP achieved the highest classification accuracy of 96.3%, while XGBoost reached 95.5% with reduced training complexity and faster execution. Comparative analysis demonstrated that AI-based approaches significantly outperform traditional methods in accuracy, adaptability, and computational efficiency. These results highlight the potential of integrating AI into modern power system protection schemes to achieve rapid, accurate, and cost-effective fault detection.