Local Data-Driven Fault Localization Method for Ring-Type DC Microgrid with Fewer Sensors

For DC microgrids (DCMGs) with short transient time, sharp current rise and weak overcurrent capacity, accurate line fault localization is crucial for scheduling maintenance tasks and accelerating system recovery. In this paper, a decentralized offline line fault localization method within the framework of fewer sensors is proposed. It does not rely on communication or the current and voltage of the faulty line, which contributes to reducing localization costs and improving protection reliability. The single-end converter bus-side voltage and positive pole current shared with the local controller are adopted. Firstly, a discrete wavelet transform (DWT)–based feature extraction method is adopted to extract the weak positive pole current characteristics of negative-to-ground (NPG) faults. Subsequently, an Att-GRU data augmentation method that integrates the attention mechanism (Att) and gate recurrent unit model (GRU) is presented to solve the problems of few available fault samples. Furthermore, an ELM-Adaboost localization model combining extreme learning machine (ELM) and adaptive boosting algorithm (Adaboost) is presented to capture the complex nonlinear relationship between fault features and fault location. Finally, the effectiveness of the proposed method has been verified by MATLAB/Simulink and experimental tests.