基于深度特征聚类和RNN的电网故障诊断

针对传统方法难以利用大量时序数据和无标签数据对电网进行故障诊断的问题,提出了基于深度特征聚类和循环神经网络(RNN)的电网智能故障诊断方法.该方法首先利用卷积神经网络搭建起特征提取器来提取时序数据的高层特征,然后对提取的特征进行半监督聚类,为无标签样本获得对应的标签,从而可以确定无标签样本所属的故障类别并加以利用;然后...

Fault Diagnosis of Power Grid Based on Deep Feature Clustering and Recurrent Neural Network

Aiming at the problem that traditional methods are difficult to diagnose power grid faults with a large amount of time series data and unlabeled data,an intelligent fault diagnosis method of power grid based on deep feature clustering and recurrent neural network(RNN) is proposed in this paper.Firstly,the feature extractor constructed by convolutional neural network is used to extract the high-level features of the time series data,and semi-supervised clustering is implemented for all the features extracted from the data,so that the corresponding labels can be obtained for the unlabeled samples.The fault category to which the unlabeled sample belongs can be determined and used.After different types of fault samples are sampled by the oversampling algorithm,a classifier constructed by RNN is used for classification and recognition.Finally,the data simulation shows that the proposed method can accurately assign the correct category labels to unlabeled samples,has great fault diagnosis performance for multiple fault types and has a higher recognition accuracy than common classification methods.The proposed method provides a reference for the practical application of deep feature clustering and RNN in fault diagnosis of power grid.