基于BIRCH-LKD的在站车辆中时异常检测算法

Anomaly Detection Algorithm of On-Station Wagon Operation Time Based on BIRCH-LKD

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.

In consideration of the on-station wagon operation anomaly situation,an anomaly detection algorithm of on-station wagon operation time（WOT）was proposed based on BIRCHLKD.This algorithm was focusing on the WOT sequence and converting the sequence into the spherical cluster with WOT feature vector,without considering the peculiar form of the anomaly.A feature clustering tree of WOT feature vector was developed based on the classification rule to be utilized in the dominant anomaly detection and shortening survey range.After computing the K-distance of the other data object,according to the different variation,the appropriate sequence values were selected as the recessive anomaly.At last,the lower bound of WOT was excluded by using the median anomaly condition.The results show that the algorithm has a high detection rate.The anomaly of the WOT sequence can be identified quickly,and the accuracy can be more than 85%.The WOT sequence becomes more smooth and unbroken after the removal of the anomalies,and conforms to the actual development trend.