改进极移预报的研究

发布时间：2018-02-21 05:17

本文关键词： 极移预报最小二乘支持向量机 GM(1 1) 经验模式分解大气角动量海洋角动量　出处：《中南大学》2013年硕士论文　论文类型：学位论文

【摘要】：摘要：高精度的地球定向参数(EOP)具有重要的科学意义和实际应用价值。EOP是天球参考框架和地球参考框架之间转换的必要条件；同时,深空探测和卫星导航等领域对EOP的预报值呈现出日益增长的需求。现代测地技术(VLBI, GPS, SLR等)是获取EOP的主要手段,但是由于复杂的数据处理过程,获取EOP往往存在时间延迟。为了满足EOP使用的实时性需求,国内外广泛开展了EOP预报的相关研究。目前,EOP中的UT1-UTC、日长变化和岁差章动模型都取得了较为实用的预报成果。但是在极移预报方面,由于其自身激发机理的复杂性,预报结果并不很理想。因此,作为EOP预报的重要组成部分,极移预报是一项值得深入研究的工作。本文主要进行改进极移预报的相关研究,主要从两方面着手,一是改进模型的预报,主要改善神经网络中最新的一种模型——最小二乘支持向量机；二是完善物理建模的预报,引入大气、海洋激发源。主要研究内容如下： (1)将最小二乘支持向量机应用于极移序列的预报中。地球定向参数包含复杂的非线性因素,应用非线性模型进行预报是较好的选择途径之一。本文将新的机器学习模型——最小二乘支持向量机应用于极移预报,该模型能够更好的处理包含非线性因素的数据。实验结果证明了将该模型应用于极移预报中具有可行性和有效性。 (2)由于单一模型对于极移残差序列预报的改善有限。GM(1,1)因具有简单有效,易于编程实现等优点被广泛应用于预测领域。但是该模型适合短期预报。本文尝试将最小二乘支持向量机和GM(1,1)模型的组合模型应用于极移残差序列的预报中。实验结果证明了该组合模型对1-10天的超短期预报精度有改善。 (3)将经验模式分解应用到极移短期预报中。考虑到极移包含的高频信号对于极移短期预报有阻碍作用。经验模式分解能够充分保留信号本身拥有的非平稳和非线性特征；具有自适应能力强；对信号类型没有限制等特点。本文采用经验模式分解对极移序列进行分解,去除高频信号,然后基于最小二乘外推模型和最小二乘支持向量机模型的组合模型对去除高频信号的重构极移序列进行1-30天的短期预报。实验结果表明,将该模型应用到极移短期预报具有可行性,预报精度有明显改善。 (4)考虑大气、海洋和极移具有相关性,将大气和海洋角动量χ1、χ2序列通过积分转换到极移域中,获得由大气、海洋激发的极移序列；在预报模型中分别加入这两个激发序列。实验结果表明,加入激发的极移序列以后,预报精度有改善。同时,考虑大气和海洋角动量既有方向又有大小,首次将大气和海洋角动量看作矢量,进行矢量和计算。实验结果表明,在预报模型中加入由大气和海洋联合激发的极移序列以后,极移的预报精度有改善。但是,对于激发源和加入方式的选择,并没有获得明确的结论,这一定程度也说明了极移激发的复杂性。
[Abstract]:Abstract: the earth orientation parameters with high precision (EOP) has important scientific significance and practical application value of.EOP is a necessary condition for transformation between the celestial reference frame and earth reference frame; at the same time, the forecast of EOP in deep space exploration and satellite navigation value showing a growing demand.
Modern geodetic techniques (VLBI, GPS, SLR) is the main means of access to EOP, but due to the complexity of data processing, to obtain EOP time delay often exists. In order to meet the needs of real-time EOP used widely at home and abroad, to carry out related research EOP prediction. At present, EOP in UT1-UTC, changes in length of day and nutation models have more practical forecasting results. But the pole shift in forecasting, because of its complexity and excitation mechanism, the forecasting result is not very satisfactory. Therefore, as an important part of EOP forecast, forecast the pole shift is a worthy of further study.
This paper mainly research the pole shift improved forecast, mainly from two aspects, one is to improve the model prediction, mainly to improve a new model of neural network and least squares support vector machines; the two is to improve the physical modeling forecast, into the atmosphere, ocean excitation source.
The main contents are as follows:
(1) the least squares support vector machine is used to shift sequence prediction. Earth orientation parameters contain complex nonlinear factors, the application of nonlinear model prediction is one of the best approaches. This paper will choose the new machine learning model, least squares support vector machine for polar motion prediction, the model can better handle contain data nonlinear factors. The experimental result shows that the model is applied to the pole shift is feasible and effective in forecasting.
(2) due to the single model for the pole shift residuals prediction (1,1) for improving.GM Co. which is simple and effective, easy programming is widely used in the field of forecasting. But the model is suitable for short-term forecasting. This paper attempts to apply the least squares support vector machine (1,1) model and GM combined model is applied to the pole shift error sequence prediction. Experimental results show that the combined model has improved on the 1-10 day of the ultra short term forecast accuracy.
(3) the application of empirical mode decomposition to the shift in the short-term forecasting. Considering the high frequency shift signal contains the pole shift for short-term forecasting hinders. EMD can fully retain the signal itself has the nonlinear and non-stationary characteristics; has strong adaptive ability; no restrictions on the signal characteristics of the type. Empirical mode decomposition of the pole shift sequence, removing the high frequency signal, then the combination model of least square extrapolation model and least squares support vector machine model to remove the high frequency signal reconstruction shift sequence forecast based on 1-30 day. The experimental results show that the model is applied to the feasibility of the pole shift forecast, forecast accuracy is obviously to improve.
(4) considering the atmosphere, oceans and the pole shift will have correlation, atmospheric and oceanic angular momentum x 1, x 2 sequence conversion to the shift in the integral domain, obtained by the atmosphere, the pole shift sequence of oceanic excitations; these two sequences were added to stimulate in the forecasting model. The experimental results show that adding excitation after the pole shift sequence, the prediction accuracy has improved.
At the same time, considering the atmospheric and oceanic angular momentum is the direction and size of the atmospheric and oceanic angular momentum as a vector, vector and calculation. Experimental results show that adding in the forecast models inspired by the atmosphere and ocean with the pole shift sequence after the pole shift, forecast precision has improved.
However, there is no definite conclusion for the choice of excitation source and the way of joining, which also explains the complexity of the pole shift excitation to some extent.

【学位授予单位】：中南大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P127.4

【参考文献】