基于改进VMD-CNN的电缆短路电流预测研究

打开文本图片集

中图分类号：TB9 文献标志码：A 文章编号：1674-5124（2025）08-0137-10

Abstract： The power system is becoming increasingly complex，and cable short-circuit current is mixed with different noise components，making data prediction and extraction more difficult. Convolutional neural network （CNN），as a multilayer perceptron that can recognize two-dimensional shapes，can extract and identify the short-circuit current. However， there are many aperiodic components in the short-circuit current， with strong nonlinearity，which willead to long recognition time and large error.The traditional empirical mode analysis （EMD） data signal preprocessing method is no longer suitable for such complex scenarios， and attention mechanism （SE） can reduce the comprehensiveness of internal signals，forming local optima.So this article proposes an improved variational mode decomposition （VMD） method that can effectively separate irrelevant signals，achieve accurate extraction of the modal components （IMF）of the inherent signal，and filter the final extracted signal to reduce noise interference without reducing the information of the data. Using the sum of the extracted IMF components as input to the convolutional network （CNN） effectively improves recognition accuracy and reduces redundant time. Finally，analyze the causes of cable short-circuit layer leakage and apply the improved VMD-CNN method to specific cable short-circuit scenarios. According to the experimentalresults， the effectivenessof the proposed improved method is much greater than that of traditional EMD methods and SE attention mechanisms.

Keywords： short circuit prediction; variational mode decomposition; signal filtering; convolutional neural network;attention mechanism

0 引言

我国电缆发展较为迅速，已经可以自主研发出高压交直流电缆。（剩余11616字）