基于改进变分模态分解和Hilbert变换的变压器局部放电信号特征提取及分类

摘要

针对现有局部放电(PD)信号特征提取方法存在的不足,提出一种基于变分模态分解(VMD)和Hilbert变换(Hilbert-VMD)的特征提取方法,并提出一种双阈值筛选法来确定VMD算法中的分解模态数.首先,根据PD信号功率谱,采用双阈值筛选法确定VMD算法中的分解模态数;其次,采用VMD算法对PD信号进行分解,得到数个有限带宽的固有模态分量(BLIMFs);然后,对各模态分量进行Hilbert变换并线性叠加后得到PD信号的Hilbert时频谱,并计算各模态分量的边际谱;最后,根据各模态分量的边际谱提取PD信号频域内的特征量,并用支持向量机(SVM)对所提取的特征量进行分类.实验结果表明,对试验环境下和现场实测两种环境下的PD信号,采用该文方法提取得到的特征量均具有较高的正确识别率,充分说明该特征提取方法可以有效提取PD信号特征.对于噪声较大的实测信号,采用该方法得到的正确识别率并未明显降低,说明该方法具有较好的噪声鲁棒性.此外,该文所提Hilbert-VMD方法也为PD信号提供了一种新的时频分析方法.%Feature extraction of partial discharge (PD) signals is one of the key step in the analysis of PD signals.Owing to the shortcomings of existing feature extraction methods,a novel method based on the variational mode decomposition (VMD) and Hilbert transform (Hilbert-VMD) was put forward in this paper.Meanwhile,a dual threshold method was proposed to determine the number of decomposition modes.Firstly,the number of decomposition modes was obtained by using the dual threshold method according to power spectra of PD signals.Secondly,the known PD signals were decomposed by VMD and several band-limited intrinsic mode functions (BLIMFs) were extracted.Then each BLIMFs was processed by Hilbert transform and the marginal spectrum of each BLIMFs was calculated.Finally,the features of PD signals in the frequency domain were extracted based on the marginal spectrum of each BLIMFs.In order to verify the effectiveness of the proposed feature extraction method,PD signals in laboratory environment and field measured were processed by Hilbert-VMD and HilbertHuang respectively and support vector machine (SVM) classifiers were utilized for pattern recognition.Compared with the feature extraction method based on Hilbert-Huang,the feature extracted by the proposed method have a higher correct recognition rate.The experimental results show that the proposed method can effectively extract the features of PD signals.The correct recognition rate of field measured signals using the proposed method is not significantly reduced and it is proved that the proposed method has better noise robustness.In addition,the Hilbert-VMD also provides a new time-frequency analysis method for PD signals.