Islanding Detection with Data Mining Methods - A Comparative Study

There is a worldwide trend towards the integration of renewable energy in the form of distributed generation, leading to the formation of microgrids. Connection of these sources introduces new issues in the operation and management of distribution systems. An important issue is that of islanding, where the microgrid remains energized locally while isolated from the main grid. It is important to detect this islanding event quickly and accurately in order to prevent possible damage to the DG and load that remains connected to the DG side after islanding. This paper explores the problem of timely islanding detection by machine learning techniques. Several cases of islanding and non-islanding are simulated on the IEEE-13 bus distribution system. Different types of DGs are connected to the system and disturbances are introduced. We consider the use of the voltage, frequency and their rate of changes at the Point of Common Coupling (PCC) as features for event detection using classifiers. These features are extracted from the simulation results and used to train and test several types of classifiers. It is shown that the random forest classifier detects the islanding with a high level of accuracy and within a reasonable amount of time after the occurrence of the disturbance.