Energy-Efficient Reconstruction Method for Transmission Lines Galloping With Conditional Generative Adversarial Network

Abstract

Conductor galloping seriously threatens the safe operation of power systems and may lead to various damages such as wire fractures or tower collapses and large-scale grid breakdowns. Real-time galloping data are important in the mechanism and effect analysis of conductor dancing prevention; moreover, they are critical for verifying anti-galloping designs and developing galloping prevention plans. However, owing to the limitations of using sensors on cables, obtaining complete galloping data is an ill-posed and challenging problem. In this study, a novel curve reconstruction method using a conditional generative adversarial network (GAN), CR-CGAN, is proposed for fully synthesizing transmission line galloping curves. We use the modeling capabilities of the recently introduced GAN by imposing additional constraints to achieve full reconstruction of the galloping curves. Moreover, we introduce a novel design in the generator-discriminator pair for improved results and a new refined loss function to enhance details. The generator uses an autoencoder with skip connections, and the inception module is used to capture different scales of spatiotemporal correlation. The discriminator is designed to use global information to determine the reliability and smoothness of the reconstructed curve. The refined loss function is aimed at reducing artifacts introduced by the GAN and ensures better reconstruction quality. A single-degree-offreedom model is constructed to verify the effectiveness and feasibility of the proposed method. Simulation results demonstrate that the proposed method can accurately reconstruct galloping curves with limited use of sensors, thus meeting the energy efficiency demands of the monitoring system.

Enhancing Secure Data Transmission in IoT via Advanced Conditional Generative Adversarial Network and Encryption Techniques

“…Another study proposed a faster FHE technique that required fewer ciphertext refreshes. This method, based on the Ducas and Micciancio (DM) technique, seemed to have a lower computational cost compared to the GSW and DM methods [14].…”

Automatic pixel-level detection method for concrete crack with channel-spatial attention convolution neural network

“…The image size is recovered layer by layer in the decoder stage and merged with the feature map generated in the encoder stage to obtain the output feature map finally. The network is previously proposed for dealing with the cell segmentation problem of biomedical images, 32 and then has been well performed in image-to-image translation, 33 reconstruction of astronomical transients, 34 transmission galloping curves 35 and seismic data. 36 Inspired by the success of biomedical segmentation, some researchers have already used this network for crack detection.…”