基于小波分析的激光雷达信号消噪和气溶胶粒子谱反演算法研究

发布时间：2018-07-01 20:41

本文选题：激光雷达 + 气溶胶　；参考：《北方民族大学》2017年硕士论文

【摘要】：本文主要研究Mie散射激光雷达信号消噪和反演气溶胶粒子谱分布。由于Mie散射激光雷达系统现有的缺陷,使得所测得的回波信号信噪比不高,严重影响对消光系数、水汽混合比和偏振的研究。另外,由于气溶胶光学厚度和粒子谱的关系式属于第一类Fredholm积分方程,求解出的气溶胶粒子谱rn)(通常是病态、不唯一的。为了解决这两个问题,本文主要用小波方法对激光雷达信号消噪和研究气溶胶粒子谱分布。回波信号消噪算法的研究主要是用自适应BP小波神经网络消噪算法来实现。用正交小波函数作为隐含层结点函数,搜索算法选取网络中最优的参数、阈值和隐含层结点个数,下降速度最快的Levenberg-Marquardt算法作为自适应小波神经网络梯度算法。通过对比拟合输出的均方误差和给定的均方误差,来调节整个BP小波神经网络,直到参数最优为止。与其它小波神经网络不同的是,本文的BP小波神经网络隐含层结点个数并不是给定的,而是通过搜索算法获得的。这种方法构造出的自适应BP小波神经网络可随外界环境的改变而进行自适应的调整。在气溶胶粒子谱反演算法的研究中,首先假设粒子是球形的,通过Mie理论求出消光系数和散射系数;其次,用CE-318测得所需的光学厚度;最后用小波Galerkin与Tikhonov正则化相结合的算法来求解。与传统的迭代算法相比,本文提出的方法方便简洁、易于计算,即用小波Galerkin方法把光学厚度和气溶胶粒子谱的关系式离散成线性方程组的形式,再用Tikhonov正则化法求出正则解。这样便克服了方程解的不稳定、不唯一的缺点。为了验证两种方法的可行性,前者与小波阈值消噪算法进行对比,后者选取2012年11月-2016年10月银川地区这四年中典型天气的光学厚度数据反演气溶胶粒子谱,并与实际的天气状况对比分析。实验结果表明:自适应BP小波神经网络消噪方法优于小波阈值消噪方法,并且用小波Galerkin方法反演气溶胶粒子谱分布来分析的银川地区天气状况与当地气象局提供的历史数据相吻合。
[Abstract]:In this paper, the de-noising of Mie scattering lidar signal and the inversion of aerosol particle spectrum distribution are studied. Due to the defects of Mie scattering lidar system, the signal to noise ratio (SNR) of the measured echo signal is not high, which seriously affects the study of extinction coefficient, water vapor mixing ratio and polarization. In addition, because the relation between aerosol optical thickness and particle spectrum belongs to the first kind Fredholm integral equation, the aerosol particle spectrum rn) (is usually ill-conditioned and not unique. In order to solve these two problems, the wavelet method is mainly used to de-noising the laser radar signal and to study the aerosol particle spectrum distribution. The research of echo signal de-noising algorithm is mainly realized by adaptive BP wavelet neural network de-noising algorithm. The orthogonal wavelet function is used as the hidden layer node function, and the optimal parameters, threshold and the number of hidden layer nodes in the network are selected by the search algorithm. Levenberg-Marquardt algorithm, which has the fastest descent speed, is used as the adaptive wavelet neural network gradient algorithm. The whole BP wavelet neural network is adjusted by comparing the mean square error of fitting output with the given mean square error until the parameters are optimal. Different from other wavelet neural networks, the number of hidden layer nodes in BP wavelet neural network is not given, but is obtained by searching algorithm. The adaptive BP wavelet neural network constructed by this method can be adjusted adaptively with the change of the external environment. In the research of aerosol particle spectrum inversion algorithm, the extinction coefficient and scattering coefficient are calculated by Mie theory, and the required optical thickness is measured by CE-318. Finally, wavelet Galerkin and Tikhonov regularization algorithm are used to solve the problem. Compared with the traditional iterative algorithm, the method presented in this paper is simple and convenient to calculate. The relation between optical thickness and aerosol particle spectrum is discretized into linear equations by wavelet Galerkin method, and the regular solution is obtained by Tikhonov regularization method. In this way, it overcomes the instability and ununiqueness of the solution of the equation. In order to verify the feasibility of the two methods, the former is compared with the wavelet threshold denoising algorithm, and the latter selects the optical thickness data of typical weather in Yinchuan area from November 2012 to October 2016 to retrieve aerosol particle spectrum. And compared with the actual weather conditions. The experimental results show that the adaptive BP wavelet neural network denoising method is better than the wavelet threshold denoising method. The analysis of the aerosol particle spectrum distribution using the wavelet Galerkin method coincides with the historical data provided by the local meteorological bureau.
【学位授予单位】：北方民族大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN957.51

【参考文献】