【作者】 陈绪荃；

【导师】 赵文光；

【作者基本信息】 华中科技大学 ， 结构工程， 2013， 博士

【摘要】 在人类走向文明这一进程中，雷电绝大多数扮演着给人类带来灾害这一角色。而随着人类的现代科技文明程度越来越高，雷电给人类带来的危害也越来越大。因此，对雷电的监测和预警将变得越来越重要。世界各国的雷电研究人员分别从雷电的发生机理、雷电的物理特征、雷电定位、以及雷电防护等等各个方面来进行闪电研究。在“十一五”国家科技支撑计划“雷电灾害监测预警关键技术研究及系统开发”的资助下，本文对雷电信号的时频分布特征以及多尺度特征进行了研究。同时在雷电定位过程中多源信号同时发生的情况下，对TOA和DOA的分选进行了研究。这些不管是对雷电的物理特征分析，还是对提高雷电定位精度都是很有意义的。本文通过理论分析、实验数据分析以及试验数据仿真从以下几个方面进行研究，得到了一些成果，它们分别如下：（1）给出了信号时频分析原理和独立分量分析原理。给出了傅里叶变换、小波变换的基本原理，指出了它们在分析非平稳非线性信号中的一些缺陷。基于此，进一步给出处理非平稳非线性信号的另外一种方法，经验模式分解。并指出经验模式分解方法在处理非平稳非线性信号时所具有的优势。（2）研究了基于小波变换雷电信号的多尺度特征分析。比较了雷电信号的傅里叶频谱图与小波谱图，得到小波谱图能显示闪电回击放电过程多尺度特征。进一步指出不同的小波基使得闪电小波谱图的多尺度特征不一致，建议使用与闪电波形比较类似的小波基函数进行分析。（3）首次研究了基于EEMD的雷电信号的多尺度特征分析。指出了小波分解的分量不具备物理意义的特点。使用EEMD对闪电回击场进行多尺度分析，得到EEMD分解的分量具有物理意义，分解趋势项对应着闪电静电场作用。同时结合分形分析，表明其它各分量分别对应着闪电不同尺度放电通道的特征。此外，研究了闪电回击场信号在使用EMD分解时产生模态混叠的原因，而EEMD能够消除模态混叠现象。最后得到，基于EEMD的希尔伯特黄谱能够更好的显示闪电回击场的多尺度特征。（4）引入一种测向去交叉算法进行雷电测向去交叉分选。给出了空间谱估计测向基本原理，利用实验数据对这种基于波形相似特征的测向去交叉分选仿真。得到该方法能很好的对DOA定位中的测向虚假点进行排除。此外，研究了DOA偏差对分离波形相关性的影响，得到在分离的波形的相关性对DOA偏差不是很敏感。进一步，得到噪声水平对DOA估计得到MUSIC方法对噪声具有很好的鲁棒性。同时研究了噪声水平对分离信号波形的影响，得出该方法对微弱信号也具备交叉去分选能力。（5）提出了一种基于ICA的雷电信号分选算法。指出利用TOA对多个源信号同时发生时进行定位时TOA获取存在的问题。以LMA系统为例，进一步分析了多枝状同时发生时不能准确定位出各枝状通道的原因。利用FastICA方法对多源信号进行分离，从而得到各个源信号波形，在利用分离信号的波形的TOA来进行定位可以解决这个问题。讨论了VHF信号的非高斯特征，得到小样本情况下是非高斯的，满足ICA求解基本模型。同时比较了混合信号波形的TOA与各源信号的波形的TOA。考虑噪声的影响，提出一种基于小波去噪的FastICA雷电信号分选方法，研究了分离信号的小波去噪分析，得到经去噪的信号具有比未经去噪的信号具有和源信号更好的相关性。更多还原

【Abstract】 In the process when mankind going towards the civilization, the lightning almostalways play a role that brought the disasters and harm to mankind. With the developmentof modern scientific technology, it will bring more and more disasters and harm to themankind. Therefore, the monitoring and early warning of the lightning will becomeincreasingly important. The researchers from the worldwide investigate the lightningfrom different aspects, such as the mechanism of the occurrence of lightning, thephysical characteristics of lightning, lightning location, lightning protection and so on.Funded by the “Elevnth Five” national key technology program “lightning disastermonitoring and early warning key technology research and system development”, thetime-frequency distribution of lightning signal characteristics and multi-scalecharacteristics were studied. In the case of when multi-source signals occurring at thesame time, the sorting of TOA and DOA were studied. It is very significant for both thephysical characteristics of the lightning and lightning location accuracy.Through the theoretical analysis, experiment data analysis and experiment datasimulation test, some results from the following aspects are obtained, and they are asfollows:(1) The the signal time-frequency analysis principle and independent componentanalysis principle were given, so as the the basic principle of Fourier transform andwavelet transform. Based on that, another non-stationary nonlinear signal method,empirical mode decomposition, was introduced. Still we pointed out that the advantagesof the empirical mode decomposition method in dealing with non-stationary nonlinearsignal.(2) The multi-scale feature of the lightning signal based on wavelet transform wasinvestigated. The Fourier frequency spectrum and wavelet spectra of the lightning returnstroke signals were compared, and the results show that the wavelet spectra can show themulti-scale features of the lightning discharge process. Additionally, choosing a differentwavelet base will lead to the different spectra. It is recommended to use the wavelet basiswhich is similar to the lightning waveforms.(3) Multi-scale features of the lightning signal based on EEMD were first study.Pointed out that the components of the wavelet decomposition do not have physicalmeaning. After analyzed using EEMD, the components have the physical meaning. Thetrend item corresponds to the lightning electrostatic field. Combined with fractal analysis, the other component corresponds to the Lightning different scale characteristics of thedischarge channel. In addition, the reason that the mode mixing occurring when useEMD to analyze to lightning return stroke was investigated, and EEMD can eliminatemode mixing phenomenon. Last, it is concluded that EEMD-based Hilbert-Huangspectrum can give better display of the lightning multi-scale characteristics.(4) A method of DOA crossing sorting method was introduced to the lightning DOAcrossing sorting. The principle of spatial spectrum estimation was introduced. Thesimulation using experiment data was conducted with this method, and the results wereobtained that the method can eliminated the DOA false point. In addition, the relationshipbetween the DOA deviation and the correlation of the separated waveforms wasinvestigated, and the results were obtained that the correlation of the waveforms was notsensitive to the DOA deviation. Further, it was concluded that the DOA estimated withMUSIC algorithm is robust to noise. Still the effects of the noise levels on the separatedwaveforms were investigated and it is concluded that this crossing sorting method canalso be used to weaken signal crossing sorting.(5) A method of lightning signal sorting based on ICA was proposed. The problemthat using TOA to locate the lightning when several source signals occurring at the sametime was discussed. Taking LMA system as an example, the reason that it is unable tolocate the several branch channels was given. Based on that, the several source signalscan be separated by using FastICA algorithm, and the problem can be solved when usingthe TOA of the separated waveforms. Then the non-Gaussian characteristics of the VHFlightning signals was discussed, and the results can be obtained that the VHF lightningsignals were non-Gaussian in the case of that when the samples were small, whichsatisfies the ICA solving model. At the same time, the TOA of waveforms between themixed and the separated was compared. In addition, a FastICA sorting method of thelightning signal based on wavelet de-nosing was also proposed. The wavelet de-nosing ofthe separated waveforms was investigated, and it is concluded that the de-noisedseparated signals can have a better correlation to the source signals than the withoutde-noised.更多还原