Title: Deep learning-based power quality disturbance detection and classification in smart grid
Authors: Hengshuo Liang; Wei Yu; Cheng Qian; Yifan Guo; David Griffith; Nada Golmie
Addresses: Department of Computer and Information Sciences, Towson University, Towson, MD 21252, USA ' Department of Computer and Information Sciences, Towson University, Towson, MD 21252, USA ' Department of Computer Science and Information Technology, Hood College, Frederick, MD 21701, USA ' Department of Computer and Information Sciences, Towson University, Towson, MD 21252, USA ' National Institute of Standards and Technology (NIST), Gaithersburg, MD 20899, USA ' National Institute of Standards and Technology (NIST), Gaithersburg, MD 20899, USA
Abstract: This paper aims to enhance the detection of power quality disturbances (PQDs) in smart grid (SG) systems using four deep learning (DL) models: Vgg19, ResNet50, MobileNetV3-L, and ViT-H. We have developed a method to convert 1D PQD signals into 2D grayscale matrices for better representation in DL models. Our evaluation considers two PQD recognition scenarios: scenario I (abnormal detection) and scenario II (all-types classification). Our data indicates that most models can identify normal signals in the abnormal detection task with the original 1D dataset. However, they struggle to distinguish common features among 11 types of abnormal signals, affecting detection performance. Using 2D greyscale matrices improves performance for most models, except Vgg19. The ViT-H model outperforms others in the all-types classification task using the 1D dataset. All models show notable improvements with the 2D dataset, with the ViT-H model consistently leading the pack.
Keywords: deep learning; DL; power quality disturbances; PQD; smart grid; SG; convolutional neural network; CNN; vision transformer; ViT; greyscale matrices.
DOI: 10.1504/IJSNET.2024.142518
International Journal of Sensor Networks, 2024 Vol.46 No.3, pp.161 - 175
Received: 22 Jun 2024
Accepted: 03 Jul 2024
Published online: 05 Nov 2024 *