摘要
针对二极管箝位式三电平逆变器的直流母线电容的老化现象进行故障诊断,文中提出一种基于遗传算法优化VMD参数的小波能量谱特征提取方法。变分模态分解(VMD)的分解个数和惩罚因子数会对分解效果产生影响。首先,采用遗传算法求出线电压信号VMD分解的最佳参数。然后,分别采用VMD和互补集合经验模态分解(CEEMD)将线电压故障信号分解成多个固有模态分量,再对固有模态分量(IMF)进行小波三层分解,根据分解系数求出各IMF的小波能量,作为故障特征。最后构建支持向量机(SVM)诊断模型,进行诊断实验。实验结果表明:VMD较CEEMD能够更好地将故障信号分解;基于遗传算法优化VMD小波能量谱方法能有效提取逆变器25种电容故障信号的故障特征,与CEEMD小波能量谱方法相比,提取的故障特征更加明显,有效提高了故障诊断的准确度。
In allusion to the aging phenomenon of DC bus capacitance in diode clamped three⁃level inverter,a wavelet energy spectrum feature extraction method that variational modal decomposition(VMD)parameter is optimized based on genetic algorithm is proposed for the fault diagnosis.The number of decompositions and penalty factors of VMD can affect the decomposition effect.The best parameters of VMD of line voltage signal are obtained by means of the genetic algorithm.The VMD and complementary ensemble empirical mode decomposition(CEEMD)are used respectively to decompose the line voltage fault signal into multiple inherent mode functions(IMF),and then the wavelet three⁃layer decomposition for the IMF is performed.According to the decomposition coefficient,the wavelet energy of each IMF is calculated as the fault feature.The support vector machine(SVM)diagnosis model was constructed to carry out the diagnosis experiment.The experimental results show that VMD can decompose the fault signal much better than what the CEEMD can do.The wavelet energy spectrum method to conduct VMD optimization based on genetic algorithm can effectively extract the fault features of 25 kinds of capacitor fault signals of inverter,and the extracted fault features are more obvious than those of the wavelet energy spectrum method of CEEMD.It effectively improves the accuracy of fault diagnosis.
作者
唐嘉瑞
帕孜来·马合木提
TANG Jiarui;PAZILAI Mahemuti(Xinjiang University,Urumqi 830047,China)
出处
《现代电子技术》
2021年第24期112-118,共7页
Modern Electronics Technique
基金
国家自然科学基金项目:风力发电系统并网逆变器智能故障诊断方法研究(61364010)。
关键词
三电平逆变器
直流母线电容
故障诊断
特征提取
变分模态分解
信号分解
诊断模型
three⁃level inverter
DC bus capacitance
fault diagnosis
feature extraction
variational mode decomposition
signal decomposition
diagnosis model
