| 引用本文: | 汪颖,孙建风,肖先勇,卢宏,杨晓梅.基于优化卷积神经网络的电缆早期故障分类识别[J].电力系统保护与控制,2020,48(7):10-18.[点击复制] |
| WANG Ying,SUN Jianfeng,XIAO Xianyong,LU Hong,YANG Xiaomei.Cable incipient fault classification and identification based on optimized convolution neural network[J].Power System Protection and Control,2020,48(7):10-18[点击复制] |
|
| 摘要: |
| 准确识别电缆早期故障是及时消除故障隐患的必要前提。提出基于卷积神经网络的电缆早期故障分类识别的方法，可从含恒定阻抗故障、励磁涌流、电容投切扰动的过电流信号中准确识别电缆早期故障。通过小波变换提取过电流信号特征，构建卷积神经网络，进行训练调整网络参数形成输入特征与类别编码之间的映射关系。为解决训练过拟合和学习效率的问题，通过修正损失函数和采用自适应学习率的方法优化卷积神经网络。仿真结果表明，所提方法能对过电流信号进行有效分类，准确识别电缆早期故障，具有较高的工程应用价值。 |
| 关键词: 电缆早期故障 卷积神经网络 深度学习 分类识别 修正损失函数 |
| DOI：10.19783/j.cnki.pspc.190581 |
| 投稿时间：2019-05-22修订日期：2019-08-22 |
| 基金项目:国家自然科学基金项目资助(51807126) |
|
| Cable incipient fault classification and identification based on optimized convolution neural network |
| WANG Ying,SUN Jianfeng,XIAO Xianyong,LU Hong,YANG Xiaomei |
| (College of Electrical Engineering, Sichuan University, Chengdu 610065, China) |
| Abstract: |
| It is necessary to identify the cable incipient faults in order to eliminate the hidden faults in time. This paper proposes a method for cable incipient fault classification and identification based on Convolution Neural Network (CNN). This method can identify the cable incipient fault from the over-current disturbance waveforms, including the waveforms of constant impedance fault, inrush current, capacitance switching disturbance waveform, and so on. The features of the over-current waveforms are extracted by wavelet transform, which are used as the input of CNN. By training the mapping relationship between input features and class coding, the parameter is chosen and the CNN is formed. CNN is optimized by modifying the loss function and adopting the method of adaptive learning rate, for solving the problem of over-fitting and learning efficiency. The simulation results show that the proposed method can classify the overcurrent signals effectively and identify cable incipient fault accurately, which is with high engineering application value. This work is supported by National Natural Science Foundation of China (No. 51807126). |
| Key words: cable incipient fault convolution neural network deep learning classification identification modified loss function |
