TY - JOUR N2 - The active distribution network (ADN) represents the future development of distribution networks, whether the islanding phenomenon occurs or not determines the control strategy adopted by the ADN. The best wavelet packet has a better time-frequency characteristic than traditional wavelet analysis in the different signal processing, because it can extract better and more information from the signal effectively. Based on wavelet packet energy and the neural network, the islanding phenomenon of the ADN can be detected. Firstly, the wavelet packet is used to decompose current and voltage signals of the public coupling point between the distributed photovoltaic (PV) system and power grid, and calculate the energy value of each decomposed frequency band. Secondly, the network is trained using the constructed energy characteristic matrix as a neural network learning sample. At last, in order to achieve the function of identification for islanding detection, lots of samples are trained in the neural network. Based on the actual circumstance of PV operation in the ADN, the MATLAB/SIMULINK simulation model of the ADN is established. After the simulation, there are good output results, which show that the method has the characteristics of high identification accuracy and strong generalization ability. L1 - http://czasopisma.pan.pl/Content/114118/PDF/01_AEE-2019-4_INTERNET.pdf L2 - http://czasopisma.pan.pl/Content/114118 PY - 2019 IS - No 4 EP - 703–717 DO - 10.24425/aee.2019.130678 KW - active distribution network KW - islanding detection KW - neural network KW - solar distributed generation KW - wavelet pocket transform A1 - Xi, Zhongmei A1 - Zhao, Faqi A1 - Zhao, Xiangyang A1 - Peng, Hong A1 - Xi, Chuanxin PB - Polish Academy of Sciences VL - vol. 68 DA - 2019.12.02 T1 - Research on islanding detection of solar distributed generation based on best wavelet packet and neural network SP - 703–717 UR - http://czasopisma.pan.pl/dlibra/publication/edition/114118 T2 - Archives of Electrical Engineering ER -