Machine learning-based ensemble framework for event identification and power quality disturbance analysis in PV-EV distribution networks

Effective analysis and classification of operational events in distribution networks (DNs), particularly those involving photovoltaic (PV) systems and electric vehicle charging stations (EVCSs), are essential for mitigating potential disturbances. This paper introduces a robust ensemble framework designed for power quality disturbance (PQD) analysis and event classification within DNs. The methodology begins with an enhanced empirical wavelet transform (EEWT), which incorporates spectral trends and window functions to accurately decompose PQDs caused by various events. These decomposed signals are then analyzed for amplitude and frequency characteristics using a mean sliding window-improved Hilbert transform (IHT). Based on these decompositions and inherent periodic features, a scale and cycle-based feature set, including time-dependent spectral features (TDSF), is formulated to differentiate between events. This feature set is subsequently classified using a light gradient boosting machine (LightGBM) to ensure precise event identification. The proposed approach is validated on a modified IEEE 13-node DN integrated with PV systems and EVCSs, simulating scenarios such as synchronization, outages and islanding. Under various noise conditions, the average accuracy of event identification reaches 99.33%, significantly outperforming other benchmark methods. Furthermore, the method's effectiveness is verified through real-time hardware-in-the-loop simulation, achieving an event identification accuracy of 98.33%. The results demonstrate that the proposed framework exhibits enhanced robustness and lower computational complexity compared to existing state-of-the-art methods.