Abstract
The increasing adoption of DC microgrids, driven by the integration of renewable energy sources and the need for efficient power systems, necessitates advanced fault detection mechanisms. Traditional fault detection methods, such as Fourier Transform (FT) and Discrete Fourier Transform (DFT), are limited by their assumptions of signal stationarity and inadequate time-localization capabilities, particularly in detecting high-resistance faults. This research paper investigates the integration of Long Short-Term Memory (LSTM) networks with the Hilbert-Huang Transform (HHT) model to address these limitations. The proposed LSTM-HHT approach leverages LSTM's ability to capture long-term dependencies and time-series patterns, along with HHT's proficiency in analysing non-linear and non-stationary signals. The integrated model is implemented and tested using MATLAB Simulink to evaluate its performance in practical DC microgrid scenarios. Results demonstrate that the LSTM-HHT approach significantly enhances fault detection accuracy and reliability, particularly for high-resistance faults that are challenging to identify with traditional methods. The empirical validation in simulated environments highlights the model's effectiveness in accurately detecting and localizing faults, thereby improving the stability and safety of DC microgrids. This research contributes to the development of more resilient and intelligent fault detection systems, supporting the broader adoption of DC microgrids and the transition to sustainable energy systems. The findings underscore the potential of combining advanced signal processing techniques with machine learning to overcome the inherent limitations of conventional fault detection methods, paving the way for further innovations in microgrid management.