An Adaptive Intelligent Strategy for Efficient Fault Detection and Localization in Hybrid Microgrid

Fault detection and protection is one of the challenging tasks in a power system, especially when integrated with microgrids. This is due to frequent changes in topology and variations in the short-circuit level, which affect the overcurrent grading of the relays. However, machine learning (ML) has been found to be effective in such scenarios. This article proposes an adaptive intelligent fault detection and classification method that dynamically integrates three learning models, adjusting their contributions based on performance under various conditions. This approach simplifies the system by utilizing novel data labeling for fault line detection and localization with a light gradient boosting machine (LightGBM) model, thus reducing complexity and response time. The current, measured as data input, is decomposed using wavelet packet decomposition (WPD). The standard deviation and energy are calculated from the wavelet coefficients, which serve as features for training the models. The proposed method effectively addresses challenges in hybrid microgrids, achieving: 1) 99.35% accuracy in fault detection and classification and 2) 99.99% accuracy in identifying faulty lines and their locations. It offers a precise and adaptable solution for simulated data, outperforming conventional protection strategies.