New ANN-Based Method for Islanding Detection in Distribution Systems with PV Generation

Anti-islanding protection of distributed generators (DG) is typically performed by conventional schemes based on measurements of voltage magnitude and frequency. Nevertheless, these schemes pose challenges regarding the definition of threshold values for differentiating islanding conditions from other disturbances that may occur in distribution systems, such as voltage sag or swell. In this context, schemes based on Artificial Neural Networks (ANNs) can be useful to identify patterns in voltage signals, making it possible to precisely differentiate islanding events from any other events. However, in general, the ANN training process is not simple, since it involves the following definitions: suitable neural network topology; data window length; sampling rate; and a representative training set for the analyzed power grid. Therefore, this paper presents a methodology for training and testing an ANN MLP-type for islanding detection of photovoltaic (PV) DG. A real scenario representing the Federal University of ABC (Brazil) was modelled, thus allowing to develop and test the proposed ANN-based solution with practical data. The results show that ANNs are an interesting alternative to perform the islanding detection of PV DG.