Adaptive Fault Detection in Microgrids Using LSTM-Based Neural Networks

Safeguarding microgrids with decentralized generation presents challenges due to the reciprocal power flow and fluctuations in renewable energy sources. Conventional protection systems often fail to adapt to these dynamic conditions, resulting in unreliable operation. This work proposes an innovative methodology for the automatic detection and classification of faults, using a Long Short-Term Memory (LSTM) neural network. The LSTM network was selected for its proven ability to process time series data, allowing it to capture the complex transient signatures of faults, which is crucial for accurate analysis. The research utilizes an extensive set of synchrophasor data (PMU) obtained from detailed simulations of a microgrid model in the MATLAB/Simulink environment. This dataset includes a variety of fault scenarios, including line-to-ground, line-to-line, and three-phase faults. To prepare the data, signal processing techniques from the Signal Processing Toolbox are applied to extract relevant features. Subsequently, an LSTM neural network is designed and trained using the Deep Learning Toolbox to classify fault types with high precision. The results demonstrate that the proposed approach achieves high accuracy and robustness in identifying different types of faults. The methodology contributes to the advancement of adaptive protection systems, offering an intelligent and effective alternative to traditional methods, and reinforces the security and resilience of modern microgrids.