Abstract

This paper presents a new islanding detection algorithm for distributed generation. The method contains a multi-layer feed forward artificial neural network structure with tapped time delays and a feedback from output which is generally called Recurrent Neural Networks (RNN) for short-term memory. The inputs of the proposed structure are generator speed deviations with a window of 60 ms time interval. The speed deviation data are produced by using EMTP simulating program for different loading, switching operation, and network conditions. Simulation studies showed that the RNN-based algorithm detects islanding situation faster than the frequency and rate of change of frequency based algorithms. The new algorithm is also stable in loading conditions when parallel operation with the utility.

Key words: Recurrent neural network, islanding protection, power system protection.