Assist.Lec. ALI EMAD<br /><br />An Intelligent Fault Detection System for Islanded Microgrids Using Advanced Deep Learning Techniques<br /><br />Abstract:<br />Islanded microgrids face major challenges in detecting faults quickly and accurately due to their complex structure and the variability of loads and energy sources. This paper presents an intelligent fault detection system based on advanced signal processing and deep learning methods to enhance detection performance. The approach begins by converting voltage and current signals into two-dimensional spectrograms, capturing essential time and frequency domain information. These spectrograms are then processed by the OpenL3 network for feature extraction. To avoid overfitting and reduce computational load, Linear Discriminant Analysis (LDA) is used for dimensionality reduction.<br /><br />The refined features are fed into a hybrid model combining Gated Recurrent Units (GRU) and an Attention mechanism. GRUs are effective for handling temporal sequences, while the Attention layer helps the model focus on the most relevant parts of the signal that may indicate faults.<br /><br />Results show that the proposed system can accurately detect various types of faults, including rare and complex ones, even in noisy and dynamic environments. It improves microgrid reliability, reduces maintenance costs, and supports more efficient and resilient energy management.<br /><br />Keywords:<br />Islanded microgrids, fault detection, spectrogram, OpenL3, GRU, Attention mechanism, deep learning, signal processing<br />university of al mustaqbal the first university in Iraq