A Comparative Analysis of Deep Learning Models for Power Quality Disturbance Classification

The increasing integration of digital technologies and communication systems into conventional power grid infrastructure has increased power quality disturbances (PQDs). These disturbances can harm grid operations, power equipment, and end-user devices. Automated PQD classification is seen as an effective solution to this problem in the power grid. This paper presents the results of a comprehensive comparative study that successfully identified the most effective deep learning method for classifying power quality disturbances (PDQs). The comparative analysis involved various deep learning models, including ResNet-18, CNN+LSTM, ResNet-50, LSTM, RCNN, and DNN. Our dataset comprises a diverse set of PQDs, including voltage sags, voltage swells, harmonics, flicker, transient events, and more, with a total of 207000 training samples, 23000 validation samples, and 25000 testing samples. The results of our comparative analysis demonstrated that the ResNet-18 model outperformed the other models in terms of PQD classification, achieving a testing accuracy of 99.80%. Our study contributes to understanding deep learning-based PQD classifiers, emphasizing the role of model architecture, dataset characteristics, and computational requirements in determining the most effective approach. Our proposed test bench for PQD classification also serves as a valuable resource for future research in this domain.