基于卷积神经网络的开关柜局部放电故障识别

Partial discharge fault identification of switchgear based on convolutional neural network

目前的局部放电故障的分类算法大多是浅层学习算法,人工提取的特征直接影响分类结果。与浅层学习算法相对,深度学习具有更深的架构,能够自动从样本中学习特征,卷积神经网络是典型的深度学习算法。本文旨在研究卷积神经网络在开关柜局部放电的应用,证明深度学习架构能够有效提高识别率。本实验共采集正常和故障两种可听声信号,将以上两类声音信号进行提取特征后,分别放入SVM模型和卷积神经网络中进行分类。实验结果表明,卷积神经网络比传统的SVM提高了声音识别的准确度。

The current classification algorithms for partial discharge faults are mostly shallow learning algorithms, and the features extracted manually directly affect the classification results. In contrast to shallow learning algorithms, deep learning has a deeper architecture that automatically learns features from samples. Convolutional neural networks are typical deep learning algorithms. This thesis aims to study the application of convolutional neural network in partial discharge of switchgear, and proves that deep learning architecture can effectively improve the recognition rate. In this experiment, two kinds of audible signals are collected, which are normal and faulty. After extracting the above two types of sound signals, they are respectively classified into SVM model and convolutional neural network for classification. The experimental results show that the convolutional neural network improves the accuracy of voice recognition compared with the traditional SVM.