Gaussian Process Regression based Fault Location in DC Microgrid

DC microgrid protection is a challenging task owing to its need for ultra-high-speed fault detection and isolation, which further makes fault location and system restoration even more challenging for limited data and time window. This paper, thus, proposes a Gaussian Process Regression (GPR) based fault location in DC microgrid using both supervised and semi-supervised learning modes. The proposed approach is also capable of estimating fault resistance. The data measured locally is sufficient to locate the fault accurately. Unlike other existing methods, the proposed approach does not require pre-processing of data, thereby decreasing the complexity and computational time involved in feature extraction. Also, the algorithm is independent of the DC microgrid operating condition. A test system is simulated in Real-Time Digital Simulator (RTDS), and the current measurements extracted are used in training and testing of GPR in MATLAB. The proposed method is tested for different test scenarios, such as high impedance fault (up to 10Ω), fault parameters, different sampling frequency, and noise levels. The obtained results prove the superiority of the proposed approach as compared to other existing approaches.