Power Quality Disturbances Classification Analysis Using Residual Neural Network

Abstract

Along with the Power Quality Disturbances (PQD) such as normal, harmonics, notch, transient, sag and swell that are due to load or electrical appliances continuously occurring in a power system, the supervised detection, and classification method is still in development progress to gain the ideal PQD classification method in order to improve the low power quality in a power system. Automatic detection and classification techniques such as deep learning algorithms are frequently preferred nowadays. Many researchers implement deep learning algorithms especially Convolutional Neural Network (CNN) architecture as a multiple PQD analysis using advanced CNN architecture namely Residual Neural Network (ResNet). To identify which ResNet architecture gives the best performance, two types of ResNet architecture; ResNet-18 and ResNet-50 are implemented. The results obtained and then compared with other CNN architectures such as basic CNN, Deep CNN (DCNN) and GoogLeNet. The results show that ResNet-18 outperforms other CNN architectures with achieved the best performance in terms of accuracy (95.77%), precision (73.73%), sensitivity (67.37%), specificity (97.29%) and F1-score (70.14%).