The increasing penetration of distributed generators(DGs)brings significant uncertainty and noises to microgrid

which leads to the increasing difficulty of microgrid monitoring. The islanding detection devices may make misjudgment because they are prone to be interfered by grid disturbance

thus causing the consequence of DGs out of service. The islanding detection devices must be able to accurately distinguish islands from grid disturbance in a noise environment. This paper introduces the concept of deep learning based on the multi-scale high-order singular spectrum entropy(MSHOSSE)into islanding detection. And a novel deep learning framework combining empirical mode decomposition(EMD)and high-order singular spectrum entropy is proposed. As a signal processing method after EMD

the MSHOSSE combines multi-resolution high-order statistics analysis and spectrum analysis

and take the entropy as the feature to output. Then the intrinsic features of islanding and grid disturbance can be extracted for training and testing in the deep learning framework. The simulation results show that the proposed method can achieve accurate detection of islands

thus avoiding running out of DGs.