Research on Early Fault Identification of Cables Based on the Fusion of MTF-GAF and Multi-Head Attention Mechanism Features

• Published in: IEEE access Vol. 12; pp. 81853 - 81866
• Main Authors: Wu, Hao, Tang, Dan, Cai, Yuan, Zheng, Chaowen
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: AlexNet; Algorithms; Artificial neural networks; Cable early faults; Cables; Circuit faults; Classification; Damage; DenseNet; Fault detection; Fault diagnosis; fault identification; Faults; Feature extraction; Filtering algorithms; GAF; MTF; Neural networks; Time series analysis
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2024.3401254

The frequent occurrence of cable early faults can lead to permanent failure of cables, making the power grid damaged and unable to work normally. To avoid the cable early faults causing great damage to the power grid operation, in this paper, we propose a research method for cable early fault identification based on the fusion of Markov Transition Field (MTF)-Gramian Angular Field (GAF) and multi-head attention mechanism features to accurately identify the cable early faults. Firstly, the fault data are preprocessed by the least mean square algorithm optimized by the adaptive gradient method; then the preprocessed one-dimensional data are converted into two-dimensional (2D) images by using MTF and GAF, respectively, and then the two types of images are fused to serve as the input of the classification network; finally, a hybrid neural network for cable early fault identification composed of a deep convolutional neural network and dense convolutional network is established, the hybrid neural network is improved by using group convolution and Ghost convolution, and the output features of the hybrid neural network are fused and classified through the mechanism of multi-head attention, and the output results of the cable early fault identification are output. At the same time, the classification results of cable early faults are visualized using the t-distributed Stochastic Neighbor Embedding (t-SNE) method to visually observe the classification effect of the hybrid neural network. The experimental results show that the algorithm has a high recognition rate for cable early fault classification, and the least mean square algorithm optimized by the adaptive gradient method is more noise-resistant compared with other optimization methods.