双通道融合的DC-DC电路软故障诊断  

Soft fault diagnosis of DC-DC circuits with dual channel fusion

作　　者：王力 WANG Li(Anhui University of Science and Technology,Huainan 232001 China)

机构地区：[1]安徽理工大学电气与信息工程学院,安徽淮南232001

出　　处：《新余学院学报》2022年第5期22-31,共10页Journal of Xinyu University

摘　　要：针对DC-DC电路故障诊断中存在特征提取和诊断精度的问题,提出了一维卷积神经网络和双向门控逻辑单元双通道融合的DC-DC电路故障诊断方法。卷积神经网络(CNN)端到端的优势使其善于提取数据空间中的局部重要特征,双向门控循环单元(BiGRU)提取信号在时间维度上的特征具有优势。结合CNN和BiGRU的优势,同时提取DC-DC电路信号的特征,并且融合为新的特征向量,输入到分类层进行故障识别。该方法可以自适应从原始电压信号提取到更为全面的特征,经实验证明,所提方法诊断的准确率达到99.92%。For the problems of feature extraction and diagnostic accuracy in DC-DC circuit fault diagnosis,a two-channel fusion of one-dimensional convolutional neural network and bi-directional gated logic unit is proposed for DC-DC circuit fault diagnosis.The end-to-end advantage of convolutional neural network(CNN)makes it good at extracting locally important features in the data space,and bidirectional gated recurrent unit(BiGRU)has the advantage of extracting features of signals in the time dimension.Combining the advantages of CNN and BiGRU,the features of DC-DC circuit signals are simultaneously extracted and fused into a new feature vector that is input to the classification layer for fault identification.The method can adaptively extract more comprehensive features from the original voltage signal,and the accuracy of the proposed method for diagnosis has been experimentally demonstrated to be 99.92%.

关 键 词：电路故障诊断 双通道融合 一维卷积神经网络 双向门控循环单元 

分 类 号：U226.81[交通运输工程—道路与铁道工程]