Probabilistic Neural Network Based Islanding Detection in Distributed Generation

This article proposes a probabilistic neural network based islanding detection technique for distributed generation considering multiple parameters in order to secure the detection of islanding for any possible network topology. This approach utilizes different parameters (features), such as rate of change of frequency, rate of change of voltage, rate of change of power, rate of change of frequency over power, total harmonic distortion (current), and rate of change of power factor, derived at the target distributed generation location and fed to the probabilistic neural network for automatic islanding detection. While comparing with other techniques, such as the radial basis function neural network and decision tree, the probabilistic neural network based technique is found to be highly effective in islanding detection. The proposed algorithm has been tested for simulation model, as well as the real-time digital simulator, and results indicate that the proposed method can reliably detect islanding conditions in the power distribution network with multiple distributed generation interface.