摘要
考虑到传统谐波模型难以精确地描述GNSS坐标时间序列的非线性时变季节性变化,进一步影响了异常值探测.本文提出了一种基于经验模态分解(Empirical Mode Decomposition,EMD)和四分位距统计量(Interquartile Range,IQR)的组合异常值探测算法.新算法的基本思想是:先采用EMD算法将GNSS坐标时间序列分解为若干不同频率的IMF(Intrinsic Mode Function)分量,并采用相关系数法进行信噪分离,最后采用IQR准则对残差序列进行异常值探测.模拟实验分析结果表明,传统算法仅能探测到83%的异常值,而新算法能够探测到91%的异常值.陆态网络实测数据分析结果进一步验证了新算法较传统算法能更加有效地探测出GNSS坐标时间序列的异常值.
Considering that the signals extracted by traditional harmonic model are not adequate to describe the non-linear time-varying seasonal variation in GNSS position time series, which will further affect the performance of outliers detection. For this reason, the paper proposed a new approach for outliers detection in GNSS position time series which combined the EMD and IQR statistic. The basic idea of the new algorithm is: firstly, the EMD algorithm is employed to decompose the position time series into several components with different frequencies, and then the correlation coefficient method is used to separate the signal and noise, lastly, the outliers in residual time series are detected by the IQR criterion. The effectiveness of new algorithm is verified by the simulated and real time series analysis and the results show that the new approach performs better than the traditional approach.
作者
刘丹丹
LIU DanDan(School of Civil and Architecture Engineering,Panzhihua University,Panzhihua 617067,China)
出处
《地球物理学进展》
CSCD
北大核心
2021年第5期1865-1873,共9页
Progress in Geophysics
基金
四川省测绘地理信息学会科技开放基金课题“金沙流域(攀枝花段)三维景观格局脆弱性分析”(ccx202119)资助。
