基于BIRCH-LKD的在站车辆中时异常检测算法

发布时间：2018-06-23 11:16

  本文选题：车辆中时 + 异常检测　； 参考：《北京理工大学学报》2017年11期

【摘要】：针对铁路车辆在站中转作业异常较多的情况,提出基于BIRCH-LKD的在站车辆中时异常检测算法.该算法以车辆中时序列为研究对象,不考虑异常值的具体形式,对序列分组,引入中时序列特征向量,做类球形簇转化;采用基于划分的显性异常检测方法得到中时序列特征向量的聚类特征树,查找序列显性异常,缩小异常检测范围;利用隐性异常检测算法计算剩余数据对象的K距离,根据距离差值变化规律,筛选序列隐性异常;最后,利用中时序列中位数异常判定条件,排除下界异常,实现中时序列的异常检测.实验结果表明,该算法检出率高,能够快速识别中时序列异常值,有效率达85%以上,去除异常值后的中时序列符合实际情况的趋势且更加平稳.
[Abstract]:A BIRCH-LKD algorithm based on BIRCH-LKD is proposed to detect abnormal railway vehicles in station. The algorithm takes the vehicle time series as the research object, does not consider the concrete form of the outliers, and introduces the middle time sequence feature vector to transform the sequence into spherical clusters. Based on the explicit anomaly detection method based on partition, the clustering feature tree of middle time sequence feature vector is obtained to find the sequence dominant anomaly and narrow the range of anomaly detection, and to calculate the K distance of the remaining data object by using the hidden anomaly detection algorithm. According to the variation rule of distance difference, the hidden anomaly of the sequence is screened. Finally, the anomaly detection of the middle time sequence is realized by using the condition of the median anomaly of the middle time sequence to exclude the lower bound anomaly. The experimental results show that the algorithm has a high detection rate and can quickly identify the outliers of middle time series, and the effective rate is more than 85%. The time series after removing the outliers are in line with the trend of the actual situation and are more stable.
【作者单位】： 北京交通大学交通运输学院;中国铁路总公司信息技术中心;
【基金】：中国铁路总公司(省部级)科技研究开发计划课题(2014X009-A)
【分类号】：TP311.13;U292.11
，

本文编号：2056942