Faster Detection of Microgrid Islanding Events using an Adaptive Ensemble Classifier

Abstract

A novel scheme for fast detection of islanding events in a microgrid is proposed in this paper. The scheme consists of a passive islanding detection process using an adaptive ensemble classifier and measurements of three-phase voltage at the distributed generation (DG) terminals. Initial features inputs to the classifier are obtained using the phase-space method. They are then used to train an ensemble classifier which comprises a set of randomized neural networks called extreme learning machine (ELM). The optimal parameter settings for each ELM are identified through evolutionary computation. An adaptive decision mechanism is designed to progressively adjust the decision time of the events classification, where its decision speed can be significantly improved over existing methods. Extensive tests are conducted on a radial distribution system with two identical DG units and the IEEE 33-bus with four DG units, considering various scenarios of islanding and non-islanding conditions, as well as the different types of DG. The simulation results show that the proposed scheme can make accurate decisions at appropriately earlier times, which achieves a good balance between efficiency and accuracy in detecting the islanding events.