基于FNN的Dual-Pol气象雷达降水粒子分类技术研究
发布时间：2018-08-21 10:47
【摘要】：云内降水粒子合理的分类具有重要的应用价值,其不仅可以提高定量降水的精确测量,而且能为人工影响天气的运行决策和评估提供重要的参考依据。论文利用双极化气象雷达对降水粒子分类技术进行研究,主要工作内容如下:1、研究了气象回波和非气象回波的微物理特性,并重点分析了气象回波中各降水类型在双极化气象雷达中的极化特性。在对气象回波和非气象回波的微物理特性研究中,主要对气象回波中降水粒子的尺寸、形状、取向等方面进行研究分析,对非气象回波中的生物回波和地杂波的强度和径向速度进行了研究分析。通过对双极化气象雷达的极化参量研究来解释气象回波中各降水粒子的极化特性。2、针对双极化气象雷达降水分类研究中各极化参量隶属函数的建立往往采用经验值,不能准确对降水粒子分类的问题,提出一种基于T-S(Takagi-Sugeno)模型的FNN(Fuzzy Neural Network)有导师监督降水粒子分类方法。该方法结合模糊逻辑思想和神经网络学习训练思想,建立了一种自适应的修正隶属函数参数的模糊神经网络。首先对双极化气象雷达接收的极化参量进行模糊化、规则计算、退模糊处理。其次,利用FNN误差反馈的学习特点对模糊化过程中的不同降水类型各极化参量隶属函数参数计算,并重新建立新的隶属函数,保证了降水粒子分类精度。通过对S波段、C波段、X波段双极化气象雷达实测数据的处理结果证明了该方法的有效性。3、针对存在地杂波情况下的降水粒子分类问题,提出一种基于FNN-CM(Fuzzy Neural Network-C Mean)的无导师监督降水粒子分类方法。该方法首先利用FNN对晴空模式下地杂波训练学习,计算得到地杂波各极化参量的隶属函数参数,并利用其对降雨模式下的地杂波进行抑制。其次,对杂波抑制后的降水粒子进行分类研究。通过计算每种降水类型的聚类中心和每个降水粒子隶属于每种降水类型的隶属度来构造降水粒子隶属度的代价函数。当降水粒子代价函数满足条件时,对计算得到的模糊隶属度矩阵进行退模糊处理,得到每个降水粒子的类型。该方法可以有效消除地杂波对降水粒子分类精度的影响。通过对C波段、S波段的双极化气象雷达实测数据的处理结果证明了该方法的有效性。
[Abstract]:The reasonable classification of precipitation particles in the cloud has important application value. It can not only improve the accurate measurement of quantitative precipitation, but also provide an important reference for artificial weather decision making and evaluation. In this paper, the precipitation particle classification technology is studied by using dual-polarization weather radar. The main work is as follows: 1. The microphysical characteristics of meteorological echo and non-meteorological echo are studied. The polarization characteristics of different precipitation types in meteorological echo in dual polarization weather radar are analyzed. In the study of microphysical characteristics of meteorological echo and non-meteorological echo, the size, shape and orientation of precipitation particles in meteorological echo are studied and analyzed. The intensity and radial velocity of biological echo and ground clutter in non-meteorological echo are studied and analyzed. The polarization characteristics of the precipitation particles in the meteorological echo are explained by studying the polarization parameters of the dual-polarization weather radar. The membership function of each polarization parameter in the precipitation classification research of the dual-polarization meteorological radar is usually established by using the empirical value. In order to solve the problem of precipitation particle classification, a new FNN (Fuzzy Neural Network) supervised precipitation particle classification method based on T-S (Takagi-Sugeno) model is proposed. This method combines the idea of fuzzy logic and the learning and training idea of neural network to establish an adaptive fuzzy neural network which modifies the parameters of membership function. Firstly, the polarization parameters received by dual polarimetric weather radar are fuzzy, regular calculation and deblurring processing are carried out. Secondly, the membership function parameters of different precipitation types and polarization parameters are calculated by using the learning characteristics of FNN error feedback, and a new membership function is established to ensure the precision of precipitation particle classification. The processing results of S band C band and X band dual polarization meteorological radar data show that the method is effective. 3. Aiming at the precipitation particle classification problem in the presence of ground clutter. An unsupervised precipitation particle classification method based on FNN-CM (Fuzzy Neural Network-C Mean is proposed. The method firstly uses FNN to train and study ground clutter in clear sky mode, and calculates the membership function parameters of each polarization parameter of ground clutter, and uses it to suppress ground clutter under rainfall mode. Secondly, the precipitation particles after clutter suppression are classified. The cost function of each precipitation particle membership degree is constructed by calculating the cluster center of each precipitation type and the membership degree of each precipitation particle belonging to each precipitation type. When the cost function of precipitation particle satisfies the condition, the fuzzy membership matrix obtained by the calculation is de-fuzzy, and the type of each precipitation particle is obtained. This method can effectively eliminate the influence of ground clutter on the accuracy of precipitation particle classification. The validity of the proposed method is proved by processing the measured data of the C-band / S-band dual-polarization weather radar.
【学位授予单位】：中国民航大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P412.25
[Abstract]:The reasonable classification of precipitation particles in the cloud has important application value. It can not only improve the accurate measurement of quantitative precipitation, but also provide an important reference for artificial weather decision making and evaluation. In this paper, the precipitation particle classification technology is studied by using dual-polarization weather radar. The main work is as follows: 1. The microphysical characteristics of meteorological echo and non-meteorological echo are studied. The polarization characteristics of different precipitation types in meteorological echo in dual polarization weather radar are analyzed. In the study of microphysical characteristics of meteorological echo and non-meteorological echo, the size, shape and orientation of precipitation particles in meteorological echo are studied and analyzed. The intensity and radial velocity of biological echo and ground clutter in non-meteorological echo are studied and analyzed. The polarization characteristics of the precipitation particles in the meteorological echo are explained by studying the polarization parameters of the dual-polarization weather radar. The membership function of each polarization parameter in the precipitation classification research of the dual-polarization meteorological radar is usually established by using the empirical value. In order to solve the problem of precipitation particle classification, a new FNN (Fuzzy Neural Network) supervised precipitation particle classification method based on T-S (Takagi-Sugeno) model is proposed. This method combines the idea of fuzzy logic and the learning and training idea of neural network to establish an adaptive fuzzy neural network which modifies the parameters of membership function. Firstly, the polarization parameters received by dual polarimetric weather radar are fuzzy, regular calculation and deblurring processing are carried out. Secondly, the membership function parameters of different precipitation types and polarization parameters are calculated by using the learning characteristics of FNN error feedback, and a new membership function is established to ensure the precision of precipitation particle classification. The processing results of S band C band and X band dual polarization meteorological radar data show that the method is effective. 3. Aiming at the precipitation particle classification problem in the presence of ground clutter. An unsupervised precipitation particle classification method based on FNN-CM (Fuzzy Neural Network-C Mean is proposed. The method firstly uses FNN to train and study ground clutter in clear sky mode, and calculates the membership function parameters of each polarization parameter of ground clutter, and uses it to suppress ground clutter under rainfall mode. Secondly, the precipitation particles after clutter suppression are classified. The cost function of each precipitation particle membership degree is constructed by calculating the cluster center of each precipitation type and the membership degree of each precipitation particle belonging to each precipitation type. When the cost function of precipitation particle satisfies the condition, the fuzzy membership matrix obtained by the calculation is de-fuzzy, and the type of each precipitation particle is obtained. This method can effectively eliminate the influence of ground clutter on the accuracy of precipitation particle classification. The validity of the proposed method is proved by processing the measured data of the C-band / S-band dual-polarization weather radar.
【学位授予单位】：中国民航大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P412.25
【参考文献】
相关期刊论文 前8条
1 黄钰;阮征;郭学良;何晖;嵇磊;;垂直探测雷达对北京地区夏季降水分类统计[J];高原气象;2016年03期
2 刘黎平;胡志群;吴，
本文编号：2195466
本文链接：https://www.wllwen.com/shoufeilunwen/benkebiyelunwen/2195466.html
