DC Microgrid Islanding Detection Method Based on RUSBoost Algorithm

With the widespread integration of distributed power sources, DC microgrids (DCMGs) have become an important component of the new smart grid. Detecting unintentional islanding, defined as the inadvertent disconnection of distributed generators (DGs) from the utility grid, is a significant challenge for DC microgrids. When in near-zero power mismatch, the traditional passive over/under voltage islanding detection method will enter the non-detection zone (NDZ), and the active islanding detection method will compromise power quality due to the injection of disturbance signals. This paper proposes a passive islanding detection method based on Random Under Sampling Boost (RUSBoost) for DC microgrids. Initially, this method selects and extracts effective electrical feature metrics during DC microgrid islanding event occurrences, followed by the collection of historical grid operation data. The RUSBoost algorithm from machine learning (ML) is employed to train and create a model for classifying islanding events. This method divides the islanding detection issue as a binary classification issue, enabling precise differentiation between the grid-connected and islanding states. This method achieves passive detection without NDZ and has the advantages of an automatic threshold setting, fast detection speed, and high accuracy. Simulation and experimental results demonstrate that this method can detect unintentional islanding quickly and precisely.