Comparative Study in Determining Features Extraction for Islanding Detection using Data Mining Technique: Correlation and Coefficient Analysis

A comprehensive comparison study on the datamining based approaches for detecting islanding events in a power distribution system with inverter-based distributed generations is presented. The important features for each phase in the island detection scheme are investigated in detail. These features are extracted from the time-varying measurements of voltage, frequency and total harmonic distortion (THD) of current and voltage at the point of common coupling. Numerical studies were conducted on the IEEE 34-bus system considering various scenarios of islanding and non-islanding conditions. The features obtained are then used to train several data mining techniques such as decision tree, support vector machine, neural network, bagging and random forest (RF). The simulation results showed that the important feature parameters can be evaluated based on the correlation between the extracted features. From the results, the four important features that give accurate islanding detection are the fundamental voltage THD, fundamental current THD, rate of change of voltage magnitude and voltage deviation. Comparison studies demonstrated the effectiveness of the RF method in achieving high accuracy for islanding detection