In the current common island detection method there is still diffidulty of deterring effectively the threshold of the electrical characteristic

which leads to the generation of detection of blind area. Some active detection methods also have a certain impact on power quality. Based on this

a kind of intelligent island detection method based on the XGBoost algorithm is proposed. The grid-connection of photovoltaic system and island operation model are set up through Matlab/Simulink simulation and the relevant electrical feature quantities are extracted to form a database. Then

the XGBoost algorithm is used to generate a binary classification prediction model. The division of threshold by this method is determined by algorithmic intelligence in accordance with the operation of the system itself

which improves the difficulty of threshold setting of traditional island detection methods. At the same time

the comprehensive judgement of multiple electrical quantities makes the detection blind area not easy to overlap with each other

which improves greatly the detection accuracy without affecting power quality. Finally

the verification is performed by the use of simulation and 5 common intelligent detection methods are compared. It is shown by the result that the method has certain advantages and reliability.