Multi-dimensional Data Change-point Detection And Its Application In Intelligent Transportation

Variable point statistical inference is one of the focus problems in statistics,which is mainly used in the fields of fault diagnosis,medical diagnosis,industrial control and meteorology.In this paper,we study the problem of multi-dimensional chang point inspection and proposing two kinds of algorithm: principal component analysis combined with weighted projection and Lt + WBS2,who are both used to test changed points and peak period of pedestrian flow through multiple entrances and exits of Shanghai subway.Inferring the cause of the change point and provide data support for the decision-making of the rail transit department.This paper mainly studies the multi-dimensional data multi-point detection problem from two aspects: data reduction and construction of test statistics:1.Proposing a method of data dimensionality reduction:principal component analysis combined with weighted projection.Then use WBS2.SDLL to detect change points.First,the performance of WBS2.SDLL is verified by generating signal data and step-like simulation data,besides multi-dimensional normal data is regenerated for simulation experiments.Contrasting the change-point detection ways based on different dimensionality reduction methods.The experimental results show that the test error of proposed algorithm is smaller.Through the change point test of pedestrian flow data at the gate entrances and exits of multiple stations of Shanghai Metro Line 1,obtained the peak hours.2.Constructing a change point detection method which combining the deviation statistic Lt and WBS2.Lt is a non-parametric deviation correction statistic for change point detection of high-dimensional data.WBS2 is an adaptive data segmentation algorithm,which can recursively generate complete data subsegments.Different from all data subsections in WBS that are pre-segmented,the position of the change point tested in each group of data in WBS2 determines the position of the next group of data subsections,and the segmentation's speed is faster.The process of constructing test statistics combines the temporal and spatial dependence of the data,and can estimate the change points located on the boundary of the time series.Through generating multi-dimensional random samples among a multivariate linear process,and conducting comparative simulation experiments on mutually independent and interdependent data sets.The experimental results verify the robustness and efficiency of the proposed method.Finally,the method is used to detect the peak pedestrian flow at the entrance and exit of Shanghai subway turnstiles.Through comparative experiments with the state-of-the-art algorithms,which provides solutions to the problem of multi-dimensional data change-point detection.In the meantime applies the proposed algorithm to empirical research,which also provides theoretical support for relevant decision makers.