<p>A new hybrid time–frequency analysis method known as Ensemble Local mode decomposition (ELMD) combined with detrended fluctuation analysis (DFA) is presented in this paper for accurate islanding detection for microgrids. ELMD decomposes microgrid disturbance signals into a number of product functions (PFs), from which the PF component containing the disturbance features is selected by a new robust Spectral Kurtosis Entropy (SKE) index. The selected PF is used as input to the detrended fluctuation analysis (DFA) method for extracting features such as low, medium and high values of scaling exponents (α) from microgrid disturbance signals. These features are used as input to a novel ensemble deep kernel random vector functional link network (edKRVFLN) for optimal model generalization, reduction of data reconstruction error due to the use of invertible kernel matrix, improved execution time and classification accuracy when compared to some deep neural networks. The detection method also encompasses other electrical disturbances like harmonic distortions, load switching, capacitor switching transients, voltage sag and swell, etc. for both islanded and grid connected modes of the microgrid. The real time capability of the proposed scheme is verified in the laboratory environment using hardware in a loop with MATLAB/Simulink interface.</p>