Fault detection, classification, and protection in low-voltage direct current microgrid with multiple energy sources using hybrid bidirectional long short-term memory–decision tree approach

This paper delivers an advanced fault management framework for a mesh-type low-voltage direct current (LVDC) microgrid (MG) integrating photovoltaic, wind, battery energy storage systems, and an auxiliary AC grid source. The MG adopts a mesh interconnection topology to enhance reliability, fault tolerance, and power-sharing flexibility. A hybrid diagnostic scheme combining bidirectional long short-term memory (BiLSTM) networks with a Decision Tree (DT) classifier is proposed for intelligent fault detection, classification, and isolation. The BiLSTM module extracts temporal features from DC-link voltage and current signals, while the DT ensures fast classification of fault type and location. MATLAB/Simulink studies were conducted under pole-to-ground and pole-to-pole scenarios. The result confirms that the hybrid BiLSTM–DT outperforms conventional classifiers such as support vector machine, k-nearest neighbors, artificial neural network, and ensemble, achieving accuracy levels of 99.3%–99.8%, sensitivity and specificity above 99%, and an error rate as low as 0.2%. Fault was detected and classified within 0.2 s, with moderate training time of around 90–95 s, which ensures real-time applicability. Compared to the conventional method, where accuracy ranges between 74% and 95% with a higher error rate (up to 25.7%), the hybrid BiLSTM–DT demonstrates more superior precision (99.8%) and F1 score (99.6%–99.8%) while successfully isolating faulty sections without disturbing healthy nodes. These conclusions highlight the potential of the proposed framework for achieving more resilient, intelligent, and self-healing LVDC MGs in the future distributed energy system.