【摘要】 针对级联H桥七电平逆变器不同故障表现相似程度高以及浅层分类器难以应对高维特征输入而制约故障诊断准确性的问题,该文提出一种基于多特征融合CNN的级联H桥七电平逆变器故障诊断策略。首先,采集原始三相电流信号,并结合参考电流信号求取电流偏差信号;其次,多尺度主元分析(multi-scale principal component analysis,MSPCA)算法通过将变分模态分解与主元分析相结合筛选故障信息存在的尺度分量,并将得到的各尺度分量直接重构作为高维时域特征输入,对得到的各尺度分量进行希尔伯特黄(HHT)变换,提取边际谱作为高维时频域特征输入;最后,将上述两种特征作为双通道CNN模型的输入进行训练,建立最终的多特征融合CNN故障诊断模型。结果表明:所提方法的故障诊断准确率达到95%,相较于单一特征与浅层分类器相结合的故障诊断策略,具有更高的识别率和更强的适应性,可为基于电信号的高相似度故障的分类识别提供一定参考。更多还原

【Abstract】 Aiming at the problems of high similarity in different fault performance of cascaded H-bridge sevenlevel inverter and the difficulty of shallow classifiers to cope with high-dimensional feature inputs, which restricts the accuracy of fault diagnosis, this paper proposes a fault diagnosis strategy for cascaded H-bridge seven-level inverter based on multi-feature fusion CNN. Firstly, the original three-phase current signal is collected, and the current deviation signal is obtained by combining the reference current signal. Secondly, the multi-scale principal component analysis(MSPCA) algorithm, which combines variational mode decomposition and principal component analysis, is used to select the several component from original signal.On the basis of above, the Hilbert-Huang transform(HHT) marginal spectrum was used to extract the highdimensional fault features from the selected component as time-frequency features, and the high-dimensional fault features reconstructed from selected component at various scales were used as time-domain features.Finally, The above two features are used as the input of the two-stream CNN model to establish the final multifeature fusion CNN fault diagnosis model.The results show that the fault diagnosis accuracy of the proposed method is 95%, which has higher recognition rate and stronger adaptability than the fault diagnosis strategy combined with single feature and shallow classifier. The research in this paper can provide a reference for fault diagnosis of the cascaded H-bridge seven-level inverter.更多还原