【摘要】 为了提高三电平逆变器复杂开路故障诊断的准确率，提出了一种应用“改进自适应噪声完备集合经验模态分解-模糊熵（ICEEMDAN-FE）”和“支持向量机（SVM）”结合的三电平逆变器故障诊断方法。首先，检测信号选取三相负载电压，为降低特征向量的维数，对三相负载电压进行Concordia变换，转换为α-β相电压；然后，通过ICEEMDAN算法提取α-β相电压的特征，得到不同尺度的内禀模态函数（IMF），再利用主成分分析（PCA）降维剔除IMF虚假分量；最后，计算优选的IMF的模糊熵均值作为特征向量，输入到多分类SVM中进行训练分类，进而实现对二极管中点箝位型（NPC）三电平逆变器的故障诊断。仿真试验结果表明，该方法能够有效识别多种开路故障模式，具有抗噪性能强，诊断速度快，诊断精度高等优点。更多还原

【Abstract】 In order to improve the accuracy to diagnose complex open-circuit faults for three-level inverters, a new fault diagnosis method of three-level inverters was proposed, combining improved complete ensemble empirical mode decomposition with adaptive noise-fuzzy entropy(ICEEMDAN-FE) and support vector machine(SVM). First, the detection signal is supplied at three-phase load voltage, which was converted into α-β phase voltage by Concordia to reduce the dimension of the eigenvector. Second, the characteristics of the α-β phase voltage were extracted by the ICEEMDAN algorithm to generate the intrinsic modal functions(IMFs) at multiple scales.Then the principal component analysis(PCA) was conducted for dimensionality reduction to remove the false component of IMFs. Finally, the optimized mean FE of the IMFs was used as the eigenvector input into the multi-class SVM for training classification, to further enable fault diagnosis of the diode midpoint clamped(NPC) three-level inverters. The simulation results show that the proposed method can effectively identify a variety of open-circuit failure modes, and has the advantages of strong noise resistance, fast diagnosis and high diagnostic accuracy.更多还原