时间序列异常值探测的Bayes方法及其GNSS数据质量控制中的应用

发布时间：2018-06-06 15:46

本文选题：验后残差 + 异常值探测　；参考：《中国惯性技术学报》2016年01期

【摘要】：GNSS卫星精密定轨定位的观测量验后残差分析是数据质量控制的重要组成部分。目前的方法都是根据经验设定异常值判断的阈值,没有利用验后残差序列的时间序列特性。引入基于识别变量的AR模型异常值探测的Bayes方法对验后残差序列中的异常值进行探测;将异常值探测问题转化为识别变量后验概率值的计算问题,并给出明确的判别阈值,通过后验概率值与事先给定的阈值进行比较判别出异常值的位置;运用Gibbs抽样设计了识别变量后验概率值的计算方法和异常值的估算方法。实测数据对该算法的验证表明该算法能够准确地探测出异常值的位置,将异常值剔除之后的观测值序列应用到精密单点定位中,结果表明该算法的使用提高了精密单点定位的定位精度。
[Abstract]:The posterior residual analysis of GNSS satellite precise orbit determination is an important part of data quality control. The current methods are based on experience to set the threshold of outliers, without using the time series characteristics of posteriori residuals. The Bayes method of detecting the outliers of AR model based on identifying variables is introduced to detect the outliers in the posteriori residual sequence, the problem of detecting outliers is transformed into the calculation of the posteriori probabilities of identifying variables, and the definite threshold is given. The position of outliers is determined by comparing the posteriori probability value with the pre-given threshold value, and the calculation method and the estimation method of the outlier value are designed by using Gibbs sampling. The verification of the algorithm by measured data shows that the algorithm can accurately detect the position of outliers and apply the outliers sequence to precise single point positioning. The results show that the accuracy of precision single point positioning is improved by using this algorithm.
【作者单位】：解放军信息工程大学地理空间信息学院;北京卫星导航中心;
【基金】：国家自然科学基金(41104047,41174026)
【分类号】：P228.4

【相似文献】