A New Islanding DetectionTechnique Based on Passive Parameter using a Combination of Artificial Neural Network and Evolutionary Programming Algorithm

To protect distributed generation (DG) from the harmful impact of islanding, DG connection codes worldwide require that all islanded DGs to be disconnected immediately after the formation of islanding. Islanding can be detected using three main types of detection techniques i.e local, remote and computational intelligent. The computational intelligence technique is the most recent approach that can produce almost zero non-detection zones. This paper presents the development of a new islanding detection technique for a synchronous-type DG based on the most sensitive passive parameters using the Artificial Neural Network (ANN)-Evolutionary Programming (EP). The most sensitive parameter was selected from 16 passive parameters using the sensitivity analysis. The analysis was conducted based on the response of each parameter when the test system was subjected with different types of islanding and non-islanding events. The selected parameter was then applied as an input for ANN-EP. EP was used to optimally tune the ANN parameter to accurately classify the islanding phenomenon. Large numbers of training and testing data sets were recorded from a simulation study that was conducted on an 11 kV distribution test system. A performance comparison between ANN and ANN-EP in classifying islanding and non-islanding events was performed. The results showed that the ANN-EP had outperformed ANN in terms of accuracy of classificatio