Abstract
Maintaining a smooth power supply is mandatory for sensitive equipment like medical devices, sophisticated computing machines, etc. Sporadic variations in voltages, voltage sags, voltage harmonics, short period shortage in voltages and temporary incidents introduce disturbances in voltages. The performance of electrical power networks significantly influences by these random and dynamic events which can cause some problems in obtaining sustainable energy supply. The identification of the disturbances affecting the power quality and the fast interpretation of them are considered as the significant tasks when addressing these disturbances. Those disturbances usually are associated with the integration of generators that are operated by renewable energy, and any non-linear features caused from the connected load. Especially in DC microgrids there exist a high number of renewable energy sources, energy efficient loads and energy storage systems. Hence the hybrid power system, that incorporates the renewable energy sources into the utility networks, postures challenges concerning the power quality. Therefore, power quality disturbance detection is a critical task for a resilient and sustainable smart grid. Different time-frequency based approaches will be employed for power quality detection and grid synchronization of renewable energy sources. This paper will focus on three main modules. The efficient signal acquisition/processing, features extraction, and classification. Voltage signals are decomposed using Flexible Analytic Wavelet Transform (FAWT) to extract features and machine learning algorithms are used to detect the power quality disturbances related to different events during the integration of renewable energy sources. In this study, real-time power signal waveform recordings are classified, and 95.33% accuracy is achieved by Bagging with rotation forest ensemble classifier.