Design and Implementation of a Machine Learning Based Intelligent Grid Fault Detection and Prediction Platform

Traditional grid fault detection methods struggle with real-time response and accuracy in complex modern power systems. To address this, we propose an intelligent platform integrating XGBoost for precise fault classification and LSTM networks for temporal fault prediction. Crucially, we introduce an adaptive adjustment mechanism that dynamically optimizes the system: it continuously monitors grid data and model performance, employs online learning to update parameters (e.g., learning rate, hidden units) in real-time, and intelligently switches between XGBoost (for stable states) and LSTM (for fluctuating states) based on grid conditions. Experimental results demonstrate the platform's effectiveness: XGBoost achieves <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 8. 2 \%}$</tex> short-circuit detection accuracy, while LSTM enables 18 -second early warnings. The adaptive mechanism further boosts prediction accuracy by 7 % and reduces response time, significantly enhancing grid safety and stability management. This integrated approach provides a robust solution for intelligent grid operation and proactive maintenance.