煤系地层接收弹性波信号的盲源分离方法研究

发布时间：2019-04-22 08:35

【摘要】：本课题来源于国家自然科学基金项目“基于连续震源的煤层反射式槽波精细探测理论”。针对国家自然科学基金项目中,需要在对煤系地层所接收的弹性波信号的混合模型和源信号无法精确获知的情况下,从观测信号中分离出各源信号。本课题旨在研究设计适于煤系地层下接收弹性波信号的盲源分离算法。(1)详细叙述了盲源分离的国内外发展历程并对不同算法的不同应用进行分类;阐述了盲源分离的基础知识和分离要求,确定煤系地层的噪声环境和弹性波的传输特性,明确了煤系地层接收弹性波信号盲源分离算法的设计要求。(2)选取合适的基于负熵的FastICA算法的非线性函数和正交化公式,与自然梯度算法分别处理煤系地层接收到的混合弹性波信号,经过Matlab仿真实验成功实现混合信号分离。验证非高斯性判据和似然度判据与互信息判据的本质相同性,从而证明基于负熵的FastICA算法、自然梯度算法和基于互信息的FastICA算法均能实现煤系地层下接收弹性波信号的盲源分离,并验证了独立分量分析算法存在的分离顺序和分离信号幅值不确定的两个固有问题。(3)提出一种新的非正交分解算法。传统经典算法独立分量分析等需要满足观测信号数不少于源信号、独立成分必须是非高斯分布等前提条件,大大降低了独立分量分析在实际煤系地层下弹性波信号盲源分离的实用性。提出的新的非正交分解算法无需满足以上前提条件,利用相关分析从单一观测信号中选择初等函数,然后,用这些功能函数作为非正交信号分解算法的基从1个混合信号中逐一分离出不同的源信号。利用该算法对由方波、正弦波、衰减波调制信号和随机噪声合成的单个典型观测信号进行了仿真实验,实验表明,该算法不仅可以从单一观测信号中准确提取出所有源信号,且相比目前普遍认可的独立分量分析具有分离顺序确定、分离信号能量符号确定等更为优良的性能。
[Abstract]:This subject comes from the project of National Natural Science Foundation of China "the fine detection theory of coal bed reflection trough wave based on continuous source". In the project of National Natural Science Foundation of China, it is necessary to separate the source signals from the observed signals when the mixed model and the source signals of the elastic wave signals received by the coal measures stratum cannot be accurately obtained. The purpose of this paper is to study and design a blind source separation algorithm which is suitable for receiving elastic wave signals in coal measure strata. (1) the development of blind source separation at home and abroad is described in detail and the different applications of different algorithms are classified. The basic knowledge and separation requirements of blind source separation are expounded, and the noise environment of coal measure strata and the transmission characteristics of elastic waves are determined. The design requirements of blind source separation algorithm for receiving elastic wave signals in coal measure strata are defined. (2) the nonlinear function and orthogonalization formula of FastICA algorithm based on negative entropy are selected. The mixed elastic wave signal received by coal measure stratum is processed separately with natural gradient algorithm, and the mixed signal separation is successfully realized by Matlab simulation experiment. It is proved that the non-Gaussian criterion, likelihood criterion and mutual information criterion are essentially identical to each other, so that the FastICA algorithm based on negative entropy is proved. Both the natural gradient algorithm and the FastICA algorithm based on mutual information can realize the blind source separation of receiving elastic wave signals in coal measure strata. Two inherent problems of the separation sequence and the uncertainty of the amplitude of the separated signals are verified. (3) A new non-orthogonal decomposition algorithm is proposed. The traditional independent component analysis (ICA) algorithm needs to satisfy the premise that the number of observed signals is no less than that of the source signals, and the independent components must be non-Gao Si distribution and so on. It greatly reduces the practicability of independent component analysis in blind source separation of elastic wave signals in actual coal measure strata. The proposed new non-orthogonal decomposition algorithm does not need to satisfy the above prerequisites. The correlation analysis is used to select the elementary function from a single observed signal, and then, These function functions are used as the basis of non-orthogonal signal decomposition algorithm to separate different source signals from one mixed signal one by one. Simulation experiments are carried out on a single typical observation signal composed of square wave, sine wave, attenuation wave modulation signal and random noise. The experiments show that the algorithm can extract all the source signals accurately from a single observation signal. Compared with the independent component analysis (ICA), which is generally accepted at present, it has better performance such as separation sequence determination, separation signal energy symbol determination and so on.
【学位授予单位】：山东科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN911.7

【参考文献】