Distributed generation intelligent islanding detection using governor signal clustering

Abstract

One of the major protection concerns with distribution networks comprising distributed generation is unintentional islanding phenomenon. Expert diagnosis system is needed to distinguish network cut off from normal occurrences. An important part of synchronous generator is automatic load-frequency controller (ALFC). In this paper, a new approach based on clustering of input signal to governor is introduced. Self-organizing map (SOM) neural network is used to identify and classify islanding and non-islanding phenomena. Simulation results show that input signal to governor has different characteristics concern with islanding conditions and other disturbances. In addition, the SOM is able to identify and classify phenomena satisfactorily. Using proposed method, islanding can be detected after 200 ms.