Control Strategy of Energy Storage System for Islanding Detection Based on Load Analysis With Machine Learning

This paper introduces an islanding detection method using machine learning for load analysis to facilitate a seamless transition of the energy storage system for an intentional islanding scenario. In the proposed method, islanding condition is detected through the frequency variation caused by an intentional reactive power mismatch. The degree of frequency variation is determined by the load characteristics and the ratio of
${d}-{q}$
currents of the inverter. Although certain load conditions are included in the associated standards for testing islanding detection, they may not be realistic in practical grid conditions. Therefore, the adequate ratio of
${d}-{q}$
currents of the inverter is determined by machine learning-based load analysis of a manufacturing facility, utilizing its actual power data. The proposed islanding detection method can be implemented in the control board of the inverter because the results of the machine learning process can be transferred as several parameters. The validity of the proposed method in real-time scenarios was assessed via diverse hardware-in-the-loop simulations. Even under harsh load conditions, the islanding detection time of the proposed method is less than 31 ms.