A methodology for training artificial neural networks for islanding detection of distributed generators

The anti-islanding protection of synchronous generators is typically performed by voltage and frequency relays. However, one of the main issues related to setting these relays is to identify and differentiate the magnitude and frequency variations of an islanding event from other disturbances that may occur along the system, such as severe load switching. By using an Artificial Neural Network (ANN) based algorithm, it is possible to recognize existent patterns on the distributed generator voltage waveform, which makes possible to obtain an accurate response about islanding events. However, the ANN training process is not so easy, because it involves important issues such as the definition of the ANN architecture, the size of data window, sampling rate and selection of a representative training set for the studied problem. In this context, this paper discusses the fundamental aspects for training an ANN used for islanding detection of synchronous distributed generators.