Islanding fault detection based on data-driven approach with active developed reactive power variation

The fluctuation of grid frequency can cause a pseudo islanding phenomenon (PIP) which cannot be avoided by the conventional detection method and the false alarm would be made. In order to reduce the reactive power variation (RPV) which is used by the conventional active islanding detection method (IDM) and avoid the false alarm, a novel data-driven IDM combines adaptive back propagation neural network data fusion and support vector machine (ABPS) is presented by taking PIP into consideration. The feature datasets (frequency, RPV) are selected to detect the islanding event. The proposed method involves two steps: first, k-means cluster and logic operation techniques are used to process the offline data for training ABPS; second, the online testing data are classified into two categories by ABPS: islanding and non-islanding. Compared with the conventional active IDM, the proposed method could detect islanding event by using less RPV, and the false alarm caused by PIP is significantly avoided. Then, a case study of the single-phase-inverter is used to prove the proposed method is more capable of islanding detection than the conventional active IDM. (c) 2019 Elsevier B.V. All rights reserved.