大气光学湍流预报模式研究

发布时间：2018-05-25 05:29

本文选题：大气光学湍流强度 + 预报模式　；参考：《中国科学技术大学》2017年硕士论文

【摘要】：大气光学湍流是大气中的一种重要现象,当它发生时,会对在其中传输的光束产生影响,因此长期观测大气光学湍流强度有助于地面激光设备的使用和天文台选址的搭建,但实际中因为搭载观测平台的成本较高,长期大范围观测大气光学湍流强度难以实现,因此找到一种能够准确预报大气光学湍流强度的方法有助于解决这一问题。本论文基于在成都和德令哈两种不同地区所做的实验,介绍了后向传播神经网络与支持向量机两种预报大气光学湍流的办法,建立两个模型,并比较了两种模型的效果,主要结论如下:(1)经过训练好的后向传播神经网络模型能够基本准确的表现出成都地区的大气光学湍流强度Cn2的日变化特征,而且在夜间的预报也更贴近观测值,平均相对误差率为3.03%,但是与观测结果相比,预报结果的峰值会有一小时的超前;在德令哈地区的实验表明,BP模型也能表现出该地区的基本日变化特征,平均相对误差率为3.53%,该地区Cn2的转换时刻明显,尤其是在18:00大气光学湍流强度会出现突然下降,后向传播神经网络模型能够准确的表现出这一变化。(2)支持向量机模型通过循环确定关键参数后也被证明可以用来估算近地面的大气光学湍流强度Cn2,成都地区的实验表明通过支持向量机模型能够表现出该地区的大气光学湍流的日变化特征,平均相对误差率为2.81%;在德令哈地区也进行了相应的实验进行验证,使用支持向量机建立模型在该地区做出了9天的预测结果与观测值吻合基本较好,能够明显表现出该地区大气光学湍流强度Cn2的日变化特征,平均相对误差率为3.38%,Cn2的频数分布图表明其与观测值的分布相似,均满足高斯分布。(3)通过在成都和德令哈地区的两次实验表明,经过训练的两种模型均能够通过一天的数据得到随后6至9天的大气光学湍流强度Cn2的预报,相关分析、平均绝对误差和相对误差等统计量的分析均表明,这两种模型能够准确表现出这两个地区近地面的大气光学湍流强度的变化,后向传播神经网络模型的相对误差率等统计量略大于支持向量机模型,但两者差距不大,均能够表现出良好的非线性特征。
[Abstract]:Atmospheric optical turbulence is an important phenomenon in the atmosphere. When it occurs, it will affect the beam which is transmitted in it. Therefore, the long-term observation of atmospheric optical turbulence intensity is helpful to the use of ground laser equipment and the establishment of the site selection of the observatory. However, in practice, the high cost of carrying the observation platform is high and the atmospheric light is observed for a long time in a large range. It is difficult to realize the turbulence intensity, so finding a method to predict the intensity of atmospheric optical turbulence is helpful to solve this problem. Based on the experiments in two different regions in Chengdu and Delingha, this paper introduces the two methods of predicting atmospheric optical turbulence in the back propagation neural network and support vector machine, and establishes two methods. The results of the two models are compared and the main conclusions are as follows: (1) the trained back propagation neural network model can basically accurately show the diurnal variation characteristics of atmospheric optical turbulence intensity Cn2 in Chengdu, and also more close to the night forecast, the average relative error rate is 3.03%, but the results are compared with the observation results. The peak value of the forecast results will be ahead of an hour, and the experiment in Delingha shows that the BP model can also show the basic diurnal variation of the region. The average relative error rate is 3.53%. The transition time of Cn2 in this area is obvious, especially at 18:00, the atmospheric optical turbulence intensity will drop suddenly, and the backward propagation neural network model is used. The model can accurately show this change. (2) the support vector machine model is also proved to be used to estimate the atmospheric optical turbulence intensity Cn2 in the near ground. The experiment in Chengdu region shows that the support vector machine model can show the diurnal variation characteristics of atmospheric optical turbulence in this area. The error rate is 2.81%, and the corresponding experiments are also carried out in Delingha. The support vector machine model has been established in the area for 9 days and the results are in good agreement with the observed values. It can obviously show the diurnal variation characteristics of the atmospheric optical turbulence intensity Cn2 in this area, the average relative error rate is 3.38%, the frequency of Cn2. The distribution diagram shows that it is similar to the distribution of the observed values and satisfies Gauss distribution. (3) through two experiments in Chengdu and Delingha, the trained two models can obtain the prediction, correlation analysis, mean absolute error and relative error of the atmospheric optical turbulence intensity of 6 to 9 days after one day of data. The quantitative analysis shows that the two models can accurately show the change of atmospheric optical turbulence intensity near the ground in the two regions, and the relative error rate of the back propagation neural network model is slightly larger than the support vector machine model, but the gap between the two models is not large, and all of them can show good nonlinear characteristics.
【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P427.1

【参考文献】