摘要
针对海量接触网检测数据中存在异常值的现象,研究数据的智能识别与预处理技术。对莱因达准则、格拉布斯准则、狄克逊准则、肖维勒准则、绝对均值法以及分位数法几种异常值判别准则进行分析,应用接触线高度动态检测数据进行实例验证,比较和分析各种异常值检测方法对误差辨识的效果。结果表明,几种方法对于离散粗大误差均有较高的识别率,但对于连续出现的异常值,绝对均值法与分位数法效果更佳。
Aiming at abnormal values which exist in massive test data of overhead contact line, this paper carried out research on the intelligent recognition and preprocessing technique of data. It analyzed several discrimination criteria for abnormal values, including PauTa criterion, Grubbs criterion, Dixon criterion, Chauvenet criterion, absolute mean value method and quantile approach. It carried out verification by applying dynamic test data of contact wire height as examples. The paper then compared and analyzed the error identification performance of different test methods for abnormal values. Results show that all these methods have high recognition rate on discrete gross error. However, for abnormal values which appear continuously, absolute mean value method and quantile approach have better pertbrmance.
作者
汪海瑛
周威
孙刚
付浩月
WANG Haiying;ZHOU WeiI;SUN Gang;FU Haoyue(l. Infrastructure Inspection Research Institute, China Academy of Railway Sciences Corporation Limited, Beijing 100081, China;Power Supply Inspection Depot, China Railway Shenyang Group Co Ltd, Shenyang Liaoning 110001, China)
出处
《中国铁路》
2018年第6期93-97,共5页
China Railway
基金
中国铁路总公司科技研究开发计划项目(J2017J006)
中国铁道科学研究院科技研究开发计划项目(2016YJ136)
中国铁道科学研究院基础设施检测研究所基金项目(2017JJXM22)
关键词
接触线高度
检测数据
异常值
判别准则
contact wire height
test data
abnormal value
discrimination criterion
