【作者】 杨涛；

【导师】 狄慧鸽； 白水成；

【作者基本信息】 西安理工大学 ， 电子信息（专业学位）， 2023， 硕士

【摘要】 云是大气中不同相态水凝物的集合体,根据温度的不同,云粒子的相态分为冰相、液相以及混合相态。不同相态云粒子对光具有不同的散射和吸收效应,这将直接作用到云的形成和发展过程,进而影响全球气候变化。目前可用于探测和分析的云相态设备主要有被动、主动遥感探测仪器,包括多通道成像仪、激光雷达、毫米波雷达以及微波辐射计等。但在云相态的识别中存在无真值、阈值划分模糊、相态误判等问题。基于上述问题,本论文研究了用于云相态识别的k-means算法和模糊逻辑算法,在分析这两个算法优势与关联的基础上建立了一套可自适应的云相态分类程序,该方法称为“聚类-模糊逻辑”算法。聚类k-means算法能够在无真值参与算法训练的条件下分析样本群中潜在的规律;模糊逻辑算法则可在阈值模糊的条件下进行精准分类。“聚类-模糊逻辑”算法将聚类k-means算法与模糊逻辑算法的优势进行有机融合,利用聚类k-means算法分析输入参数潜在规律的能力为模糊逻辑算法建立具有针对性的隶属函数,克服了模糊逻辑算法在云相态领域无可靠隶属函数的问题。基于聚类k-means算法建立的隶属函数能够根据不同环境、不同设备条件进行自我调整,有效避免了实验条件的差异可能带来的云相态误判。“聚类-模糊逻辑”算法同时适用于联合观测,该算法基于数据进行分类并不限制所使用探测设备。为证明算法的合理性,利用毫米波云雷达的观测结果,使用成熟的毫米波雷达产品数据进行相态分类并对比“聚类-模糊逻辑”算法的分类结果,在考虑两种设备时空分辨率差异的情况下证明“聚类-模糊逻辑”算法能够精准分类云相态。为了识别更完整云层的相态提出了拼接法,将毫米波雷达分类结果与激光雷达分类结果拼接融合,充分发挥两台设备的探测优势。本论文采用西安理工大学激光雷达遥感研究中心研发的云系降水潜力探测激光雷达与无线电探空仪、微波辐射计进行同步观测,将激光雷达输出的后向散射系数、退偏比、雷达比以及温度数据作为算法输入进行云相态识别研究与统计。结合探测结果进行了个例分析,并对2022年秋季的连续观测结果进行相态识别。本次连续观测累计观测云层22642 min,采集25组连续的云层演化数据,通过“聚类-模糊逻辑”算法对云相态进行分析。统计结果显示,西安地区秋季云相态主要以冰云为主,占比为48.54%。更多还原

【Abstract】 Cloud is a collection of different phases hydrocondensates in the atmosphere.According to the different temperature,the phase of cloud particles can be divided into ice phase,liquid phase and mixed phase.Different phase cloud particles have different scattering and absorbing effects on light,which will directly affect the formation and development of clouds,and then change the global climate.At present,the cloud phase equipment that can be used for detection and analysis mainly includes passive and active remote sensing detection instruments,including multi-channel imager,Lidar,millimeter wave radar,and microwave radiometer.However,there are some problems in cloud phase identification,such as no true value,fuzzy threshold partition and misjudgment of phase state.Based on the above problems,this paper studies the k-means algorithm and fuzzy logic algorithm for cloud phase identification,and establishes a set of adaptive cloud phase classification program on the basis of analyzing the advantages and correlation of the two algorithms,which is called "clustering-fuzzy logic" algorithm.The clustering k-means algorithm can analyze the potential rules in the sample group without true algorithm training;the fuzzy logic algorithm can accurately classify under fuzzy threshold condition.The "clustering-fuzzy logic" algorithm integrates the advantages of clustering k-means algorithm and fuzzy logic algorithm,and uses the ability of clustering k-means algorithm to analyze the potential rules of input parameters to establish targeted affiliation functions for fuzzy logic algorithm,which overcomes the problem that fuzzy logic algorithm has no reliable affiliation function in the field of cloud phase.The affiliation function based on the clustering k-means algorithm can self-adjust according to different environments and different equipment conditions,effectively avoid the possible miscalculation of cloud phase states caused by the difference of experimental conditions.The "clustering-fuzzy logic"algorithm is also applicable to joint observations,and the classification based on data does not limit the detection equipment used.In order to prove the rationality of the algorithm,the observation results of millimeter wave cloud radar and mature product data of millimeter-wave radar were used to classify phase states,and compared with the results of "clustering-fuzzy logic" algorithm.It was proved that the "clustering-fuzzy logic" algorithm can accurately classify the cloud phase states when considering the difference in spatial and temporal resolution of the two devices.In order to identify the phase state of more complete clouds,a splicing method is proposed,which combines the classification results of millimeter wave radar and lidar,and gives full play to the detection advantages of the two devices.In this paper,the cloud precipitation potential detection lidar developed by the Lidar Remote Sensing Research Center of Xi’an University of Technology is used for synchronous observation with radiosonde and microwave radiometer.The backscatter coefficient,declination ratio,radar ratio and temperature data of lidar output are used as algorithm inputs for cloud phase identification research and statistics.The individual case was analyzed with detection results,and the phase identification of the continuous observations in autumn 2022 was conducted.In this continuous observation,clouds were observed for 22642 min,25 sets of continuous cloud evolution data were collected,and cloud phase states were analyzed by the"clustering-fuzzy logic" algorithm.The statistical results show that the autumn cloud phase in Xi’an is mainly dominated by ice clouds,accounting for 48.54%of the total.更多还原