A new artificial neural network based method for islanding detection of distributed generators

• This paper presents an artificial neural network based method for islanding detection.
• The algorithm depends only on the voltage samples measured at the DG terminals.
• A data selection procedure has been proposed to train the ANN more effectively.
• A detailed discussion about the data sampling rate to feed has also been conducted.
• The method robustness has been verified by applying it to several non-islanding events.

This paper presents an artificial neural network (ANN) based method for islanding detection of distributed synchronous generators. The proposed method takes advantage of ANN as pattern classifiers. It is capable of identifying the islanding condition based on samples of the voltage waveform measured at the distributed generator terminals only, which is an important advantage over other ANN-based anti-islanding methods. Moreover, the proposed method is robust against false operation. In order to create a training data set for the ANN, a data selection procedure has been proposed, so that the ANN could be trained more effectively, which has contributed positively to the good performance of the method. The concept of the time-performance region has been introduced to assess the method performance, as well as the non-detection zones. A detailed discussion about the data sampling rate to feed the proposed method has also been conducted, so that the computational burden can be faced as an important factor to assess its performance.