Abstract：The islanding detection method is a necessary method for grid-connected PV systems，but，the

current islanding detection method is prone to misjudgement when encountering grid operation states such as highresistance short-circuit ground faults and large-capacity load casting. For this reason，an islanding detection method based on reactive power perturbation and featured kurtosis was proposed based on analyzing the difference between the power at the grid point in the islanding operation state and other grid operation states. The method firstly used the three-phase voltage amplitude offset rate at the grid point to realize the starting criterion of active reactive power injection in PV system. Then，the apparent power waveform at the grid point was obtained，and after the normalization operation was performed to obtain the normalized apparent power（NAP）waveform，variational modal decomposition algorithm was used to decompose NAP at 8 layers，and the time-frequency component at layer 5 was obtained as the feature detection component. And then the feature detection component was characterized by a featured kurtosis to achieve islanding detection. Finally，a typical PV grid-connected system was constructed using Matlab simulation platform，and the effectiveness of the proposed method was tested through the working conditions of islanding operation with different quality factor loads，different high-resistance short-circuit ground faults，large-capacity load switching，and lightning strike faults，and so on.

      Key words：photovoltaic system；islanding detection；high-resistance short-circuit ground fault；reactive power perturbation；variational modal decomposition（VMD）；featured kurtosis