时间域航空电磁数据SVM反演方法

发布时间：2018-08-26 13:28

【摘要】：航空瞬变电磁法(ATEM)是一种以飞机为载体,电磁感应为勘探机理的航空物探方法。由于具有勘探深度较大、可大面积勘探、能克服复杂地形等优点,特别适合我国的地理国情。目前,航空瞬变电磁法已经在地质填图、矿产资源勘探和环境监测等领域获得了广泛应用。由于航空瞬变电磁数据属于宽频带信号,容易受到多种噪声的影响,特别是航空瞬变电磁数据量较大,影响因素较多,使得对其解释难度较大,总而言之,航空瞬变电磁信号处理与解释方法研究仍然是当前航空瞬变电磁法实用化的研究热点和难点。本文尝试引入机器学习的思想,使用支持向量机对大量地电模型及其电磁响应进行学习,实现航空瞬变电磁反演,提高反演精度;此外,为了获得高质量航空电磁数据,本文还研究了主成分分析及小波变换相结合的去噪方法以提高信噪比。具体来说,本文的主要研究内容如下:(1)主成分分析与小波变换结合的去噪方法。该方法通过主成分分析提取出航空瞬变电磁数据的主成分,然后对主成分进行小波分析,再用主成分重构电磁数据,达到抑制噪声的目的。经加噪的正演模拟数据测试表明,该算法有较好的去噪能力,能够将数据信噪比提高10-14dB。经野外实测航空瞬变电磁数据测试发现,主成分与小波变换结合能够较好的抑制航空电磁数据在空间域和时间域上的随机噪声和高频噪声。此外,该方法在保幅去噪的同时,还具有很好的稳定性与较高的计算效率,能够满足野外实测数据现场处理的需求。(2)基于航空瞬变电磁原始数据的支持向量机反演。本文以航空瞬变电磁正演程序计算二层与三层地电模型的电磁响应,将其作为样本数据集合,把样本数据集分为两个子集,一个子集作为支持向量机反演的训练样本集,另一部分作为测试样本集。然后用两层地电模型的样本集进行支持向量机反演最佳参数组合进行分析,找到最佳的反演参数组合。最后用该参数组合分别对二层与三层地电模型的航空瞬变电磁数据进行支持向量机训练与反演。以二层地电模型为例,其反演结果的电阻率相对误差平均值为8.06%,深度的相对误差平均值为11.56%。(3)基于航空瞬变电磁数据主成分的支持向量机反演。由于电磁数据相邻道间的相关系数最高达到0.9,数据的冗余信息较多,通过主成分分析提取出航空电磁数据的特征成分,采用支持向量机找到特征成分与地质模型的属性间的映射关系。本文以航空瞬变电磁正演程序计算二层与三层地电模型的电磁响应,采用主成分分析获取数据主成分,将数据主成分作为样本数据集合,把样本数据集分为两个子集,一个子集作为支持向量机反演的训练样本集,另一部分作为测试样本集。对二层与三层地电模型正演数据主成分的支持向量机反演实验,发现其反演得到的结果与基于原始数据进行支持向量机反演结果基本一致,其效率相较于前者也得到了一定提高。
[Abstract]:Aeronautical transient electromagnetic method (ATEM) is a kind of aerogeophysical method with aircraft as carrier and electromagnetic induction as exploration mechanism. It is especially suitable for China's geographical conditions because of its advantages of large exploration depth, large area exploration and overcoming complex topography. Aerial transient electromagnetic (TEM) data are widely used in many fields, such as measurement and so on. Because it belongs to broadband signal, it is vulnerable to many kinds of noise. Especially, the large amount of TEM data and many influencing factors make it difficult to interpret them. In a word, the research of processing and interpreting methods of aviation TEM signal is still the current aviation. In this paper, we try to introduce the idea of machine learning and use support vector machine to study a large number of geoelectric models and their electromagnetic responses, so as to realize the aeronautical transient electromagnetic inversion and improve the inversion accuracy. In addition, in order to obtain high-quality aeronautical electromagnetic data, we also study the principal component analysis and small. In particular, the main contents of this paper are as follows: (1) Principal component analysis and wavelet transform combined denoising method. This method extracts the principal component of Aeronautical transient electromagnetic data by principal component analysis, then carries out wavelet analysis on the principal component, and reconstructs the electromagnetic number by principal component analysis. The forward simulation data test shows that the algorithm has better denoising ability and can improve the signal-to-noise ratio of the data by 10-14 dB. It is found that the combination of principal component and wavelet transform can restrain the aero-electromagnetic data in the spatial and temporal domains. In addition, this method has good stability and high computational efficiency, and can meet the needs of field data processing. (2) Support Vector Machine inversion based on the original data of Aeronautical transient electromagnetic. Electromagnetic response of geoelectric model is taken as sample data set, and the sample data set is divided into two subsets, one as training sample set of SVM inversion, the other as test sample set. Finally, the combination of parameters is used to train and invert the aero-transient electromagnetic data of two-layer and three-layer geoelectric models respectively. Taking the two-layer geoelectric model as an example, the average relative error of resistivity is 8.06%, and the average relative error of depth is 11.56%. (3) Based on aero-transient electromagnetic model Because the correlation coefficient between adjacent channels of electromagnetic data is up to 0.9 and the redundant information of data is more, the characteristic components of aero-electromagnetic data are extracted by principal component analysis, and the mapping relationship between the characteristic components and the attributes of geological model is found by support vector machine. The electromagnetic response of the two-layer and three-layer geoelectric model is calculated by the magnetic forward program. The data principal component is obtained by the principal component analysis. The data principal component is taken as the sample data set. The sample data set is divided into two subsets. One subset is used as the training sample set of SVM inversion, and the other part is used as the test sample set. The SVM inversion experiment of the principal component of forward data of geoelectric model shows that the inversion result is basically consistent with the SVM inversion result based on the original data, and its efficiency is also improved compared with the former.
【学位授予单位】：成都理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P631.326

【参考文献】