Machine Learning-based Fault Detection and Classification in microgrid

Abstract

Fault Detection and Classification plays a vital role in maintaining the reliability and stability of microgrids, especially as they incorporate renewable energy sources and become more decentralized. Microgrids face a wide variety of faults, such as short circuits, line-to-ground faults, and other disturbances, which can negatively affect system performance. Traditional fault detection methods have primarily focused on False Data Injection and cyber-attacks, emphasizing vulnerabilities in communication infrastructure. However, this study addresses current faults within the electrical network, focusing on system stability and real-time fault detection in the absence of communication-related errors. In this work, machine learning techniques are employed to enhance fault classification accuracy. Partial Least Squares is used for feature selection to extract relevant statistical features from real-time current data collected from various microgrid components. By optimizing these features and applying them to machine learning models, the approach overcomes the limitations of conventional fault detection methods. The results show a significant improvement in fault classification performance, with up to 10% higher accuracy compared to traditional methods. Additionally, the use of data from neighboring microgrid components boosts the model's robustness, adaptability, and performance under varying operational conditions, contributing to a more resilient microgrid. This research introduces an innovative approach to fault detection in microgrids by combining machine learning and feature optimization, offering a more accurate, reliable, and efficient solution to ensure continuous energy supply and improve system stability under different fault scenarios.