基于WBS2的一种多维数据多变点检验方法及其应用

A Change-point Detection Scheme and Application for Multivariate Data Based on WBS2

提出了一种新的多维数据变点检验方法,它适用于具有高度相关性的多维数据并具有快速检测的能力。该方法先通过结合主成分分析和线性投影,将多维数据有效地降成一维数据。利用阶梯状模拟数据进行对比实验,实验结果表明具有轨道交通行人流数据分布特征的数据适用于WBS2.SDLL方法,由于SDLL仅使用阈值作为次要模型选择标准且不基于惩罚项,使其不需要事先得知数据的最大变点数量。WBS2.90和WBS2.95分别对应检验的置信水平为90%和95%的情况。因此利用WBS2.SDLL方法对降维后的上海地铁一号线上25个站点进出闸机口的行人数据进行变点检验,从而找出地铁上行人数据的突变规律及人流的高峰时间段。同时,一号线上单个站点闸机进出口行人流的变点检验结果也验证了该方法的有效性。

Inthis paper,a multi-dimensional data change point test method was presented,which is suitable for multi-dimensional data with high correlation.The method effectively reduced multidimensional data into one-dimensional data by combining principal component analysis and linear projection.Using ladder-like simulation data for comparison experiments,the simulation results showed that the data with the characteristics of pedestrian flow data distribution is suitable for WBS2.SDLL method.Since SDLL only used the threshold as the secondary model selection criterion and was not based on a penalty approach,it didn′t need to know the maximum number of change-points present in the data,in either theory or practice.WBS2.90 and WBS2.95 were respectively correspond to the confidence level of the test are 90%and 95%.Therefore,after reducing the dimension of pedestrian data at the gate of 25 stations on Shanghai metro line 1,the WBS2.SDLL method was used to test the change points to find out the mutation law of pedestrian data and the peak time of people flow.At the same time,the effectiveness of the method was also verified bythe change point inspection results of the pedestrian flow at the gate of the single station on Line 1.