连续压实质量检测参数单点异常值识别及处理  

Identification and processing method of single outlier of continuous compaction quality measured value

作　　者：聂志红[1] 阚常壮 谢扬[1] NIE Zhihong;KAN Changzhuang;XIE Yang(School of Civil Engineering,Central South University,Changsha 410075,China)

机构地区：[1]中南大学土木工程学院,长沙410075

出　　处：《哈尔滨工业大学学报》2019年第3期150-157,共8页Journal of Harbin Institute of Technology

基　　金：国家自然科学基金(51478481)

摘　　要：为解决现行连续压实检测参数异常值识别及处理方法未考虑数据空间分布特征的问题,结合地统计学的半变异函数,提出基于自相关距离的近邻加权估计识别法,并定义单点异常值判定指标:异常指数α_i,将异常值剔除后,利用普通克里金插值法对原异常值点处的数据进行估计,并通过沪昆高铁娄底试验段进行了连续压实质量检测试验验证.结果表明:当某点的异常指数α_i>0.2时,可判定其为单点异常值点;相比于现行的拉依达准则(3σ准则)识别方法,基于自相关距离的近邻加权估计法具有更高的准确度和识别效率;普通克里金插值法能够更为准确地估计单点异常值处的数据,降低数据的变异系数,提高连续压实质量检测参数的均匀性.To take the spatial distribution into account when identifying and processing single outlier continuous compaction quality, the near neighbor weighted estimation and identification method was developed based on autocorrelation distance. Judge index of single outlier was defined as abnormal index α i . After eliminating the outliers, the original outlier data were estimated using ordinary Kriging interpolation method. Then continuous compaction experiments were carried out in Loudi construction site of Shanghai-Kunming high-speed railway. Results show that the test value can be determined as a single outlier when the abnormal index α i is greater than 0.2. Compared with the current Pauta criterion recognition method, the near neighbor weighted estimation and identification method based on autocorrelation distance has higher accuracy and recognition efficiency. The ordinary Kriging interpolation method could provide a more accurate estimation of the single outlier data, reduce the coefficient variation of the data, and improve the uniformity of continuous compaction quality measured value.

关 键 词：连续压实质量检测 单点异常值 自相关距离 异常指数 普通克里金插值 

分 类 号：U213.1[交通运输工程—道路与铁道工程]