Title: Fault transient waveform recognition in actual transmission lines using multi-scale convolution and lightweight channel attention DenseNet

Authors: Nailong Zhang; Jie Chen; Chao Gao; Xiao Tan

Addresses: State Grid Jiangsu Electric Power Co. Ltd., Research Institute, Nanjing, 210000, China ' State Grid Jiangsu Electric Power Co. Ltd., Research Institute, Nanjing, 210000, China ' State Grid Jiangsu Electric Power Co. Ltd., Nanjing, 210000, China ' State Grid Jiangsu Electric Power Co. Ltd., Research Institute, Nanjing, 210000, China

Abstract: Ensuring the proper functioning of power transmission lines is crucial for society and everyday life. Accurate fault identification methods are essential for line maintenance and inspection. Existing methods based on simulated data overlook the complexity of real-world fault data, which limits their reliability. This study presents a method for identifying the data types of real transmission line fault transient waveforms using a multi-scale convolutional channel attention DenseNet. The method consists of two main components: data preprocessing and fault recognition. The imbalance of fault samples is addressed using the synthetic minority over-sampling technique (SMOTE), and the Gramian angular field (GAF) method converts complex time series data into image data. Multi-scale convolution further extracts fault feature information, while a lightweight channel attention block ReLU-alpha-sigma activation map (RASimAM) enhances discriminative capacity. Experimental results demonstrate that the method effectively addresses sample imbalance issues and performs well on small datasets. It accurately identifies fault types and exhibits excellent generalisation performance.

Keywords: transmission lines; fault identification; deep learning; DenseNet; multi scale; channel attention.

DOI: 10.1504/IJCSM.2024.139084

International Journal of Computing Science and Mathematics, 2024 Vol.19 No.4, pp.366 - 382

Received: 29 Dec 2023
Accepted: 02 Feb 2024
Published online: 12 Jun 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article