The invention discloses a microgrid island detection method based on an extreme learning machine. The method is characterized by establishing a microgrid island detection module based on feature extraction and an ELM classification algorithm; selecting a part of historical data as initial data; and appropriately selecting algorithm features for ELM classifier training according to practical situation. Test result shows that the island detection method based on the ELM classification algorithm is good in island detection effect and small in monitoring blind area; and if increasing the number of training samples and feature quantity, precision of island detection can be improved further. Besides, the island detection method based on the feature extraction and the classification algorithm has no adverse influence on operation of an electric power system; and the detection method is good in dilution, can meet on-site application requirements and has engineering practical significance.