Abstract & Overview
This paper presents an enhanced and practically validated islanding detection framework for grid-connected solar PV systems, integrating Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and a Pattern Recognition Neural Network (PANN). The method processes negative sequence voltage signals at the Point of Common Coupling (PCC) to extract intrinsic mode functions (IMFs), mitigating mode mixing and improving signal fidelity. Significant IMFs are selected based on their power percentile, and three statistical features—maximum value, standard deviation, and entropy—are extracted and normalized before classification by the PANN. Unlike prior studies, this work extends evaluation to zero-power mismatch scenarios, noisy environments, and load-switching conditions, providing practical validation of real-time detection performance. Simulation results demonstrate a classification accuracy of 98.6% with a detection time of 0.1806 seconds, complying with IEEE 1547 standards. The proposed approach exhibits robust and reliable islanding detection across diverse operating conditions, significantly reducing the non-detection zone (NDZ) and enhancing the safety and reliability of modern distribution systems.