基于信噪辨识的矿集区大地电磁噪声压制

为了避免形态滤波方法在大地电磁强干扰分离中的"过处理"、进一步保留大地电磁低频段的有用信息,提出基于信噪辨识的矿集区大地电磁噪声压制方法.首先,从信号处理的角度剖析矿集区典型强干扰与天然大地电磁微弱信号之间的定量辨识关系,利用形态分形维数和形态膨胀谱熵对大地电磁信号与强干扰进行信噪辨识.然后,结合形态滤波技术和阈值法,仅对辨识出明显不是天然大地电磁信号的异常波形进行噪声压制.最后,重构大地电磁有用信号,并对算法进行性能评价.仿真结果表明,形态分形维数和形态膨胀谱熵能较好地定量辨识大地电磁信号与强干扰,大地电磁信号中一些缓变化的低频信息得到了更为精细的保留;与形态滤波整体处理相比,本文所提方法获得的卡尼亚电阻率曲线更为光滑、连续,视电阻率值相对稳定,其结果更为真实地反映了测点本身所固有的大地电磁深部构造信息.     

数学形态滤波与大地电磁噪声压制

数学形态滤波是一种新型的非线性滤波方法.介绍了形态学滤波的基本原理,针对大地电磁信号表现出的非线性、非平稳性和非最小相位特性,综合结构元素特征我们提出一种基于数学形态滤波的大地电磁噪声压制方法.为了有效抑制目标信号中的噪声干扰及修正统计偏倚现象,通过选用合理的结构元素及形态开-闭、闭-开组合,将正、负结构元素级联构造组合广义形态滤波器对实测大地电磁信号进行噪声压制.实验结果表明,该方法是切实可行的,有效地剔除了大尺度干扰及基线漂移,较好地还原了大地电磁信号的原始特征,修正了标准形态算子所产生的统计偏倚现象,去噪精度高.该方法计算速度快,具有潜在优势,为矿集区海量大地电磁信号与强干扰的分离提供了一条新的途径,应用前景广阔.     

基于循环神经网络的大地电磁信号噪声压制研究

由于天然电磁场源信号微弱,观测数据极易受到噪声干扰,严重影响反演和解释结果.传统去噪方法依赖于人工对时间序列和功率谱的筛选,去噪效率低,主观性强.本文提出利用循环神经网络对大地电磁时域信号进行特征噪声的识别和提取,进而重构出去噪后的大地电磁信号.在对大地电磁时域信号进行大量分析的基础上,对噪声进行分类并搭建含噪信号数据库,利用该数据库训练了两个循环神经网络,并选取长短时记忆单元优化循环神经网络结构,分别实现含噪数据段筛选和噪声形态提取.对仿真和实测数据分别进行了测试,循环神经网络均能准确筛选出大地电磁信号中的噪声段,本方法在避免人为操作主观性的同时提高了工作效率,视电阻率和相位曲线质量得到明显改善.

     

基于匹配追踪和遗传算法的大地电磁噪声压制

针对匹配追踪计算量大、大地电磁数据处理效率低的问题,提出基于匹配追踪和遗传算法的大地电磁噪声压制方法.首先,利用Gabor原子构建过完备原子库,并对过完备原子库集合进行划分.然后,借助遗传算法的自适应性,快速搜寻最优匹配原子及所在位置.最后,运用最优匹配原子对待处理信号进行稀疏分解,重构有用信号.通过对计算机模拟的典型强干扰和矿集区实测大地电磁数据进行分析处理,实验结果表明,相对于匹配追踪和正交匹配追踪,文中所提方法能从过完备原子库中快速、自适应地选取最优匹配原子与不同噪声干扰类型高精度的匹配,极大地提升了计算效率;大地电磁时间域序列中的大尺度强干扰被有效剔除,视电阻率曲线更为光滑、连续,低频段的数据质量得到明显改善.     

基于自适应多分辨率奇异值分解的大地电磁数据处理

针对强电磁干扰极易掩盖微弱的大地电磁有用信号,本文结合奇异值分解在去噪方面的优越性,提出基于自适应多分辨率奇异值分解（Adaptive Multi-Resolution Singular Value Decomposition,AMRSVD）的大地电磁数据处理方法.首先对大地电磁数据构建Hankel矩阵,利用MRSVD得到不同分辨率的近似信号和细节信号;然后选用近似信号和细节信号的标准差差值,对大地电磁数据进行信噪辨识;接着结合MRSVD和相邻细节信号的标准差差值,提出先验信息未知情况下的AMRSVD法;最后对辨识出的强干扰运用AMRSVD去除噪声,重构有用信号.实验结果表明,该方法的处理效率高,能有效分离出相关性较强的噪声,时间序列和视电阻率-相位曲线均得到有效改善.

     

广义S变换及其在大地电磁测深数据处理中的应用

广义S变换是一种优于短时窗傅立叶变换和小波变换的时频分析方法,利用广义S变换能够准确定位大地电磁资料中存在的噪声,通过定义时频窗对噪声进行滤除,从而明显提高阻抗视电阻率与相位的估算质量.本文基于S变换和大地电磁测深资料处理的基本原理,研究了基于广义S变换的大地电磁测深资料的处理流程和方法.对理论模拟信号及实测大地电磁场时间序列数据的处理,证实了方法的有效性.     

Hilbert-Huang变换与大地电磁噪声压制

大地电磁信号具有非线性、非平稳、非最小相位特征,不符合以Fourier变换为基础的传统功率谱估计的基本要求. Hilbert-Huang变换是近年发展起来的处理非线性、非平稳信号的完全局部时频分析方法. 本文在简要介绍Hilbert-Huang变换基本原理与算法基础上,以实际数据分析为例,探讨了它在大地电磁信号处理及噪声压制中的应用. 提出利用Hilbert时-频能量谱对大地电磁信号进行时段筛选,以提高信号品质,增强数据处理的质量和资料的可解释性. 利用经验模态分解方法及其多尺度滤波特征,可以有效地分析MT信号中的噪声分布特征,并进行干扰压制.     

庐枞矿集区大地电磁测深强噪声的影响规律

天然大地电磁场信号微弱,极化方向随机,极易受电磁噪声污染.张量阻抗分析、远参考技术、Robust估计等对随机噪声和不相关噪声有比较好的压制效果,但对强的相关噪声目前还没有有效的压制方法.庐枞矿集区人烟稠密、工业发达,是我国著名的铁、铜、硫等矿产基地之一,区内强烈的工业、通讯、矿山、民用等电磁干扰严重污染了大地电磁测深数据.本文首先根据实测的电磁场时间域波形和卡尼亚电阻率测深曲线形态,挑选出基本未受噪声污染的测点(Y1650).然后利用数学形态学从受严重污染的电磁场时间序列中提取出类方波、三角波、阶跃、脉冲和充放电5种典型噪声的波形,并以不同的方式将这些噪声波形与Y1650的电磁场波形叠加,对比分析加噪后Y1650点电阻率和相位测深曲线的变化,进而归纳出典型强噪声对庐枞大地电磁测深资料的影响规律,为进一步压制强噪声和资料处理提供依据.     

Cancelling directional EM noise in magnetoteilurics1

The prospecting of densely urbanized areas by the measurement of magnetic and electric natural fields is severely hampered by electromagnetic (EM) noise. Active man-made EM noise sources can generally be considered fixed in space, thus affecting the magnetotelluric (MT) signals of a measuring site mainly along their polarization directions. Taking advantage of the impulsive nature of polarized EM noise, a time-domain directional noise cancelling (DNC) technique is proposed. The comparison of noisy data with data predicted, using a low noise reference signal or with data interpolated whenever no reference is available, allows the detection of the most likely noise sources with prevailing directional patterns using a Bayes's criterion. The DNC approach is general and can be adapted, depending on the reference signal used (single-site or remote-reference). In field data, hodograms of the prediction residuals basically confirm the directional noise model assumed in DNC. An example is presented in which the DNC technique has been applied to a single-site MT survey carried out in northern Italy, where the signal was heavily corrupted by noise with prevailing directional properties due to the densely urbanized area. MT apparent resistivities and phases obtained at the site of the survey before and after DNC are presented and discussed.     

Study of the Structure of the Shenzhen Fault Zone with the Magnetotelluric Method

The magnetotelluric sounding method was used to study the active fault in Shenzhen city.Four magnetotelluric profiles with a dense station interval were laid out across the Shenzhen fault zone.The remote reference technique was used in both data observation and processing to eliminate the electromagnetic noise near the survey sites,and relatively smooth apparent resistivity curves were obtained.TM mode data and two-dimensional inversion method as NLCG were used to obtain the electrical structures underground.According to the surficial geology survey results on regional strata and distribution of magmatic bodies and faults,the electrical structures at depths less than 2000m of each profile were interpreted.Two regional faults,the Henggang-Luohu fault and the Liantang fault,and seven local faults consistent with the electrical boundaries were verified or discovered.The strata in the survey area were then related with the change of resistivity.Electrical horizontal slices of depths less than 5000m indicate that the Henggang-Luohu fault and the Liantang fault bifurcate at the shallow part but tend to merge in the deep part.     

微处理机控制的全自动大地电磁测深仪

本文介绍了一种全自动大地电磁测深仪。仪器的研制工作是在法国地球物理研究中心G.Clerc博士领导下进行的。仪器的主要特点是采用CMOS微机自动操作,从而使之功耗低,自动化程度高和轻便灵活。这是新近发展的一种新型物探仪器设备。 本文讨论了该系统的基本原理、测量特点、测量灵敏度和噪声情况等。同时,也简要地讨论了信号的处理和控制等问题。     

实数编码遗传算法在大地电磁测深二维反演中的应用

作为全局非线性优化的新方法之一的遗传算法,近年来已从生物工程流行到大地电磁测深资料解释中.然而,大地电磁反演问题具有不适定性,解的非唯一性.通过结合求解不适定问题的Tikhonov正则化方法,本文采用实数编码遗传算法求解大地电磁二维反演问题.此算法在构建目标函数时引入正则化的思想,利用遗传算法求解最优化问题.常规的基于局部线性化的最优化反演方法易使解陷入局部极小值,而且严重的依赖初始模型的选择.与传统线性化的迭代反演方法相比,实数编码遗传算法能够克服传统方法的不足且能获得更好的反演结果.通过对大地电磁测深理论模型进行计算,结果表明:该算法具有收敛速度快、解的精度高和避免出现早熟等优点,可用于大地电磁资料解释.     

Estimation of the magnetotelluric impedance functions

The successful interpretation of magnetotelluric data demands well presented, and accurate, estimates of the impedance and tipper functions.In this paper, a method based on the singular value decomposition of the local data matrix, using well known results from multivariate statistical analysis, is proposed as an alternative to current least-squares methods. The advantage obtained is that error on all data channels is recognised, so that impedance estimates are not biased by the choice of “independent noise-free” fields.An important result of the method is an extensive and structured capability for the analysis of the frequency-dependent noise which could be used to improve the resulting impedance estimates.     

基于压缩感知重构算法的大地电磁强干扰分离

为压制大地电磁信号中的强人文干扰,提出一种基于压缩感知重构算法的大地电磁信号去噪方法.通过构建与常见典型强干扰相匹配而对有用信号不敏感的冗余字典原子,利用改进的正交匹配追踪算法,分离出大地电磁信号中的强干扰成分.为了验证所述方法的强干扰分离效果,首先通过在实测大地电磁信号中加入理想的强干扰信号进行了仿真分离实验,然后从大量实测数据中选取三种含有不同类型强干扰的时间域片段,用所述方法对实测数据中的强干扰进行分离,最后将所述方法应用于青海试验点以及庐枞矿集区某测点实测数据的综合处理.仿真实验结果表明,该方法在分离出强干扰的同时,能够较好地保留有用信号.实测数据处理结果表明,该方法能够有效压制强干扰,改善强干扰区大地电磁数据的质量.     

基于字典学习的音频大地电磁数据处理

音频大地电磁（Audio Magnetotelluric,AMT）信号常常受到持续性人文噪声影响,这类噪声使用远参考法和Robust阻抗估计等常规方法往往效果不佳.为此,本文从噪声的规律与特征出发,提出一种新的AMT数据处理方法.首先通过字典学习方法从观测数据中自主学习到人文噪声的特征结构,构建冗余字典,然后利用学习到的冗余字典,分离出AMT数据中的人文噪声.为验证方法的有效性,首先进行了合成数据的仿真试验,然后在四川凉山进行了针对性的野外试验研究,最后将本文方法应用于庐枞矿集区实测数据的处理.结果表明,本文方法能够快速、准确地分离出AMT信号中的人文干扰,保留有用信号,显著改善AMT数据质量.     

Multi-station superposition for magnetotelluric signal

Extending the observation time to increase the length of time series is the most effective but consuming way to improve the signal-to-noise ratio of magnetotelluric (MT) signal. Multi-station superposition technique at present proposed for magnetotelluric survey aims to attenuate random noises by multi-station synchronously observation at one site. The idea stems from seismic array stacking and is mathematically verified corresponding to a linear random noise model. The multi-station superposition can be done in time and frequency domains. A practical testing measurement was performed by setting three stations on vertexes of an equilateral triangle and a single station on centre of the triangle in desert area of the Tarim basin, western China. The results of the testing data demonstrate the effectiveness of the proposed scheme. The method has great potentials for long-period MT survey in dramatically reducing the observation time and suppressing the static shift caused by near-surface inhomogeneities.     

二维MT快速曲线对比反演方法的可行性研究

最近开发了一种针对二维大地电磁野外数据进行处理解释的新反演方法。该方法以加入阻抗相位信息的一维大地电磁连续介质曲线对比法为基础 ,把一维反演得到的电阻率和相位的数据集作为二维反演的初始模型 ,使用二维有限单元法做正演模拟。在程序的后继迭代中 ,深度方向上用一维反演修改模型的电阻率和深度值 ,沿测线方向由二维有限元作修改 ,反演结束可得到一个接近真实电性分布的电阻率数据集 ,并绘制成电阻率断面图。对模型的反演实验结果显示 ,该反演方法能够较真实地反映地下电性分布 ,而且避免了偏导数矩阵的计算 ,其原理简单 ,计算速度快 ,表明该反演方案是可行的     

利用K-SVD字典学习算法压制海洋大地电磁噪声

海洋大地电磁数据处理的结果会影响数据反演和解释的准确性.在浅水中, 由海浪引起的强电磁干扰和大地电磁场叠加在一起, 使得噪声所在频段内的视电阻率和相位曲线出现畸变.针对这种强干扰, 本文将K-SVD字典学习算法应用于海洋大地电磁信号处理中, 通过稀疏表示海浪感应磁噪声实现大地电磁磁场信号的重构, 并结合视电阻率信息进行相位校正.仿真数据和南黄海实测数据的处理结果表明, 该方法能够有效压制海浪感应磁噪声, 恢复后的视电阻率和相位曲线更加连续、光滑.

     

Depth determination from a non-stationary magnetic profile for scaling geology

The conventional spectral analysis method for interpretation of magnetic data assumes stationary spatial series and a white‐noise source distribution. However, long magnetic profiles may not be stationary in nature and source distributions are not white. Long non‐stationary magnetic profiles can be divided into stationary subprofiles following Wiener filter theory. A least‐squares inverse method is used to calculate the scaling exponents and depth values of magnetic interfaces from the power spectrum. The applicability of this approach is demonstrated on non‐stationary synthetic and field magnetic data collected along the Nagaur–Jhalawar transect, western India. The stationarity of the whole profile and the subprofiles of the synthetic and field data is tested. The variation of the mean and standard deviations of the subprofiles is significantly reduced compared with the whole profile. The depth values found from the synthetic model are in close agreement with the assumed depth values, whereas for the field data these are in close agreement with estimates from seismic, magnetotelluric and gravity data.     

Near-source noise suppression of AMT by compressive sensing and mathematical morphology filtering

In deep mineral exploration, the acquisition of audio magnetotelluric (AMT) data is severely affected by ambient noise near the observation sites; This near-field noise restricts investigation depths. Mathematical morphological filtering (MMF) proved effective in suppressing large-scale strong and variably shaped noise, typically low-frequency noise, but can not deal with pulse noise of AMT data. We combine compressive sensing and MMF. First, we use MMF to suppress the large-scale strong ambient noise; second, we use the improved orthogonal match pursuit (IOMP) algorithm to remove the residual pulse noise. To remove the noise and protect the useful AMT signal, a redundant dictionary that matches with spikes and is insensitive to the useful signal is designed. Synthetic and field data from the Luzong field suggest that the proposed method suppresses the near-source noise and preserves the signal well; thus, better results are obtained that improve the output of either MMF or IOMP.