基于EDA的ICA方法及其在重力固体潮地球物理信息分析中的应用

发布时间：2018-01-15 18:02

本文关键词：基于EDA的ICA方法及其在重力固体潮地球物理信息分析中的应用　出处：《昆明理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着仪器行业技术的发展,以及人类对地球更加完备的认识,越来越多的科学家认为地震是可以进行预报的,因此对研究地震前兆信息成为当下一个热点。通过对固体潮信号的分析来提取地震前兆信息也成为一种重要的途径,固体潮主要分为地倾斜固体潮和重力固体潮,而重力固体潮能够反映地球重力的变化,能够对地球内部构造的反演和将来运动的预测提供一种最为直观的方法。重力固体潮信号中包含大量的谐波分量,其谐波分量的产生都是由地球自转、月球和太阳的引力作用产生的。在对重力固体潮信号分析时,只有将其中包含的谐波分量分离进行分析,才能应用于地球的构造的反演或者是地球动力学等课题应用。本文介绍了一种重力固体潮正交分解模型,该模型能够将重力固体潮信号中包含的日波、半日波和长周期波分离在三维正交空间的三个方向上。为了实现信号的分解,本文引入了独立成分分析算法对其进行分解处理分析。独立成分分析算法(Independent Component Analysis,ICA)是一种为了解决盲源分离问题发展而来的算法,其主要解决的是源信号中包含多个独立成分进行将各独立成分进行分离的问题,计算过程由优化计算和选取合适的目标函数构成的。传统的ICA算法都表现出对其中涉及的解混矩阵的初始值权值敏感,而不能达到预期的分离效果。本文提出利用分布估计算法来进行优化计算ICA算法中的解混矩阵,以求得到更优更适合本课题的解混矩阵。分布估计算法(Estimation Distribution Algorithm,EDA)是从遗传算法的基础上发展而来的,该算法舍弃了遗传算法中的交叉和重组步骤,改用群体自身的概率模型进行不断更新,达到群体最优的。本文提出用EDA算法和ICA算法结合去处理问题,并在仿真实验中进行了验证,得到了分离效果更佳的实验结果,验证了本算法的可行性。后利用基于EDA的独立成分分析算法对重力固体潮理论值进行分析计算,既验证了重力固体潮正交分解模型的正确性,还从中提取出了三类谐波分量,分别为日波、半日波和长周期波,且这些分离出的独立成分与本文提出的重力固体潮信号正交分解模型一致,分别对应于模型中的正交谐波分量。最后将本文提出的算法用在实测重力固体潮信号中,过对重力固体潮信号中包含的长周期波的分析,研究长周期波的时序变化特征,从中读取其包含的震颤异常波从而提取地震前兆信息。通过对云南地区的实际震例分析表明,长周期波在地震前后的确存在异常变化特征量,此类异常变化常出现在地震前和地震后大约30天左右。本文得到的时序特征量的变化容易观察,对于地震前兆信息的分析具有明确的物理意义。
[Abstract]:With the development of instrument technology and the more complete understanding of the earth, more and more scientists believe that earthquakes can be predicted. Therefore, the study of seismic precursor information has become a hot topic, and the analysis of solid tide signal to extract seismic precursor information has become an important way. The solid tide is mainly divided into the earth tilt tide and the gravity tide, and the gravity tide can reflect the change of the earth's gravity. The inversion of the inner structure of the earth and the prediction of the future motion can provide a most intuitive method. The gravity solid tide signal contains a large number of harmonic components, which are generated by the earth rotation. The gravitational action of the moon and the sun. In the analysis of gravity solid tide signals, only the harmonic components contained therein are analyzed separately. In this paper, an orthogonal decomposition model of gravity solid tide is introduced, which can be used to integrate the daily wave in the gravity solid tide signal. The semidiurnal wave and the long-period wave are separated in three directions of the three-dimensional orthogonal space in order to realize the decomposition of the signal. In this paper, we introduce the independent component analysis (ICA) algorithm, which is independent Component Analysis. ICA) is an algorithm developed to solve the problem of blind source separation, which mainly solves the problem that the source signal contains multiple independent components to separate each independent component. The traditional ICA algorithm is sensitive to the initial weight of the unmixing matrix. However, the desired separation effect can not be achieved. In this paper, the distribution estimation algorithm is proposed to optimize the calculation of the unmixing matrix in the ICA algorithm. In order to obtain a better and more suitable unmixing matrix, the distribution estimation Distribution Algorithm is obtained. EDAs are developed from genetic algorithms, which abandon the steps of crossover and recombination in genetic algorithms and update the probability model of the population itself. In this paper, EDA algorithm and ICA algorithm are proposed to deal with the problem, and the simulation results are verified, and better separation results are obtained. The feasibility of this algorithm is verified. Then the theoretical value of gravity tide is analyzed and calculated by using the independent component analysis algorithm based on EDA, which verifies the correctness of the orthogonal decomposition model of gravity tide. Three kinds of harmonic components are extracted, which are the daily wave, the semi-diurnal wave and the long-period wave, and the separated independent components are consistent with the orthogonal decomposition model of gravity tide signal proposed in this paper. At last, the proposed algorithm is used to analyze the long-period wave in the measured gravity solid tide signal. The time-series variation characteristics of long-period waves are studied, and the seismic precursor information is extracted by reading the tremor abnormal waves from them. Long period waves do exist abnormal variation characteristics before and after earthquakes, which often occur about 30 days before and after the earthquake. The time series characteristic quantities obtained in this paper are easy to be observed. The analysis of earthquake precursor information is of clear physical significance.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P312.4;TN911.7

【相似文献】