变系数回归的GNSS时间序列异常值探测方法  

Outlier detection in GNSS position time series based on variable coefficient regression model

作　　者：宣滨[1] 胡倩伟[2,3] 吴超超 XUAN Bin;HU Qianwei;WU Chaochao(Jiangxi V&T College of Communications,Nanchang 330013,China;Chinese Academy of Surveying and Mapping,Beijing 100036,China;Beijing Geo-Vision Tech.Co.,Ltd.,Beijing 100070,China;Beijing Tianshi Hongtu Technology Co.,Ltd.,Beijing 102600,China)

机构地区：[1]江西交通职业技术学院,南昌330013 [2]中国测绘科学研究院,北京100036 [3]北京四维远见信息技术有限公司,北京100070 [4]北京天时宏图科技有限公司,北京102600

出　　处：《测绘科学》2022年第6期51-56,共6页Science of Surveying and Mapping

摘　　要：针对传统的基于谐波模型的异常值探测算法效率较低的问题,该文引入变系数回归模型替代谐波模型分离GNSS坐标时间序列的信号与噪声(残差),并提出了一种基于变系数回归(VCR)模型与四分位距(IQR)统计量的组合异常值探测算法VCR_IQR。该算法首先采用VCR分离时间序列中的信号与噪声,接着采用IQR准则探测噪声中的异常值。将VCR_IQR与传统基于谐波模型的异常值探测算法—最小二乘探测法(LS_IQR)和最小一乘探测法(L1_ModIQR)进行模拟实验对比分析,结果表明,VCR_IQR能够探测到92%的异常值,而LS_IQR和L1_ModIQR仅能探测到81%和85%的异常值,表明VCR_IQR的探测效率优于LS_IQR和L1_ModIQR。Aiming at the problem of low efficiency of traditional outlier detection algorithms based on harmonic models,this paper introduced a variable coefficient regression model instead of harmonic models to separate the signal and noise(residuals)of GNSS coordinate time series,and proposed a combined outlier detection algorithm named VCR_IQR based on the variable coefficient regression(VCR)model and the interquartile range(IQR)statistic.The algorithm first used VCR to separate the signal and the noise in the time series,and then uses the IQR criterion to detect outliers in the noise.To evaluate the performance of the proposed approach,the VCR_IQR with the conventional approaches which were based on the harmonic model(LS_IQR and L1_ModIQR)with the simulation experiments were compared.The results showed that VCR_IQR could detect 92%of the outliers,while LS_IQR and L1_ModIQR could only detect 81%and 85%of the outliers,indicating that VCR_IQR was more efficient than LS_IQR and L1_ModIQR.

关 键 词：GNSS坐标时间序列 变系数回归模型 异常值探测 四分位距统计量 

分 类 号：P228[天文地球—大地测量学与测量工程]