一种融合Akaike信息检验的多粗差识别方法  

A method for multiple outlier identification combining with Akaike information criterion

作　　者：余航 宁一鹏 李黎 赵伟 YU Hang;NING Yipeng;LI Li;ZHAO Wei(School of Geography Science and Geomatics Engineering,SUST,Suzhou 215009,China;Research Center of Beidou Navigation and Environmental Remote Sensing,SUST,Suzhou 215009,China;School of Surveying Mapping and Geographic Information,Shandong Jianzhu University,Jinan 250101,China)

机构地区：[1]苏州科技大学地理科学与测绘工程学院,江苏苏州215009 [2]苏州科技大学北斗导航与环境感知研究中心,江苏苏州215009 [3]山东建筑大学测绘地理信息学院,山东济南250101

出　　处：《苏州科技大学学报(自然科学版)》2025年第1期66-73,共8页Journal of Suzhou University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金项目(42204011);江苏省科技计划项目(BK20230660)。

摘　　要：对于多粗差的识别问题,可通过建立多重备选假设模型,构建局部检验统计量,并以最大值统计量作为识别粗差的依据。然而,当粗差量级较大时,由于数值计算精度的问题,最大统计量反而无法区分两个或多个备选假设模型,导致“湮没”现象频发。本文提出将Akaike信息检验与多粗差识别方法进行融合应用,在不改变多粗差识别方法中探测、识别与调节改正框架的基础上,通过以Akaike信息最小为准则确定备选模型,避免上述“湮没”现象的发生。数值实验结果表明,相较于迭代数据探测法,融合Akaike信息检验的多粗差识别方法对于量级较大的多粗差问题具有较好的识别成功率,且粗差识别过程中该方法不受整体虚警率的影响。For the problem of identifying multiple outliers,multiple alternative hypotheses models can be established,local test statistics constructed,and the maximum value of the test statistics used to pinpoint the position of outliers.However,when the magnitude of the gross error is large,due to issues with numerical calculation precision,the maximum statistic is unable to distinguish between two or more alternative hypothesis models,leading to frequent occurrences of the“swamping”phenomenon.This paper proposes the integration of Akaike information criterion with the multiple outliers testing.On the basis of not changing the detection,identification,and adjustment framework within multiple outliers testing,the paper determines the alternative model based on the criterion of minimum Akaike information,thereby avoiding the occurrence of the aforementioned“swamping”phenomenon.Numerical experiments show that compared with the iterative data snooping method,the proposed method has a better rate in identifying gross errors of larger magnitude,and that this method is not affected by the overall false alarm rate during the gross error identification process.

关 键 词：多重备选假设 Akaike信息检验 粗差探测 迭代数据探测法 

分 类 号：P207[天文地球—测绘科学与技术]