•  
  •  
 

Keywords

low voltage in distribution networks, multi-scale time-frequency feature, wavelet convolution, Transformer model, multi-label classification

Abstract

A cause analysis model based on multi-scale time-frequency features and global dependency modeling is proposed to improve the accuracy and robustness of cause classification and address the complex and diverse causes of low voltage in distribution networks and the non-stationary nature of signal characteristics.The model consists of two feature extraction branches：the wavelet convolution branch employs one-dimensional discrete wavelet convolution to extract multi-scale time-frequency features and capture the local dynamic variations of the signals；the Transformer branch converts the time-series data into a recursive graph and models global temporal dependency through the self-attention mechanism.Subsequently，local and global features are integrated through feature fusion and input into a multi-label classifier to complete the cause classification task.Finally，on the low-voltage dataset of a distribution dataset from a certain city in China，simulation analysis of the model is carried out.Simulation analysis shows that the model achieves a classification accuracy of 98.74%.Compared to traditional methods，this model effectively combines local and global features and significantly improves cause classification capability，demonstrating excellent performance and robustness in handling non-stationary signals.The model can provide a reliable solution for the accurate analysis of low voltage problems in distribution networks.

DOI

10.19781/j.issn.1673-9140.2026.04.006

First Page

61

Last Page

74

Download

Share

COinS