Comparison of Several Methods for Outlier Estimation

• Main Author: LU Tieding
• Format: Article
• Language: zho
• Published: Surveying and Mapping Press 2016-06-01
• Series: Acta Geodaetica et Cartographica Sinica
• Subjects: mean shift data snooping; outlier estimation; the method of simultaneous locating and evaluating multidimensional gross errors(LEGE); quasi-detection of gross errors(QUAD); partly least squares(PLS)
• Online Access: http://html.rhhz.net/CHXB/html/2016-6-656.htm

Computational problem of gross errors estimation is discussed based on the mean shift model, and the gross errors estimation formulas of the observed statistical correlation data snooping method are given. The relationships of gross errors estimation of the data snooping method, the method of simultaneous locating and evaluating multidimensional gross errors (LEGE), quasi-accurate detection of gross errors (QUAD) method and the partial least-squares (PLS) method are discussed. It is proved that ① in the case of correlated observations, calculation of gross errors estimation of the PLS method and the QUAD method are equivalent. However, these two methods are different with the data snooping method and the LEGE method; ② in the case of uncorrelated and unequal weight observations, calculation of gross errors estimation of the QUAD method, the PLS method and the data snooping method are equivalent, but these three methods are different with the LEGE method; ③ in the case of uncorrelated and equal weight observations, calculation of gross errors estimated value of these four methods are equivalent. Finally, the case studies verify the conclusions.