| Summary: | Power quality denotes the methodology employed to supply power and ground-sensitive equipment in a manner conducive to optimal equipment functionality. It encompasses a diverse range of challenges, including voltage fluctuations like sags and spikes, harmonics, transients, and voltage imbalances, collectively known as power quality disturbances (PQD). Recently, Convolutional neural networks (CNN) have been the most often utilized technique for PQD classification However, CNN's most recent version has not yet been implemented. Therefore, the objectives of this work are to use Fourier Transform as a feature extraction method in classifying and analyzing PQD using EfficientNet with 8 models (from B0 to B7) and to validate and evaluate the performance of the best EfficientNet model by comparing it with 1D-CNN and ResNet-50 architectures. The results show that EfficientNet B0 outperformed 1D-CNN and ResNet-50 in terms of accuracy (84.72%), precision (90%), recall (84.33%), and F1-score (86%), among other metrics. This research will help to improve the PQD classification system. © 2024 IEEE.
|
|---|