The Studies And Application Of Online Monitoring For Passenger Flow Surveillance In Urban Transport Systems Based On Statistical Process Control

The online passenger flow monitoring of modern public transportation is an important issue in the intelligent transportation system,and it is also a significant part in the construction of smart cities.It is essential in maintaining social stability and security and preventing unexpected group events and system failure.Due to the rapid advancement in information technology and data acquisition,it is possible to carry out efficient and accurate anomaly detection for the traffic passenger flow by using data collected in real time.But at the same time,the data of huge capacity and complex form also put forward a problem for the statistical analysis.How to accurately and effectively detect the anomalies of complex passenger flow in the modern transportation system has become a highly concerned and valuable issue.This paper aims to provide scientific monitoring schemes by using the statistical process control theory and constructing effective control charts.Statistical process control(SPC)charts are useful methods in the field of quality management,and find broad applications in intelligence manufacturing,environmental monitoring,social network,healthcare and many other applications.SPC has become an effective tool for monitoring the stability of various processes.The analysis shows that traditional statistical process control methods are limited by the application background,and are usually based on some common distributions and data forms,which may be difficult to apply to the complex traffic passenger flow datastream monitoring.For instance,in the passenger flow monitoring of single important station studied in this paper,the underlying distribution can be complex.It is difficult to model the data under most commonly used distribution assumptions in conventional control chart studies.Moreover,due to the uncertainty of the shift sizes of passenger flow processes,it is difficult for the control chart to maintain the detection sensitivity of various shifts while the underlying distribution is difficult to identify.Therefore,it is urgent to develop a new change point detection procedure more accurate and effective under the background of passenger flow monitoring by refering the existing theoretical framework of control chart.Another example is the passenger flow monitoring of multiple stations or the whole line.Because the passenger features at different stations vary greatly,the common method is to sort the collected passenger flow data into tensors for lower information loss.Most existing tensor datastream control schemes are proposed to monitor image processes,whose features are quite different from the passenger flow tensors.Thus the existing schemes may not be well applicable to this scenario.How to monitor the passenger flow tensor datastream more scientifically and efficiently is also one focus of this paper.Concerning the above problems,this paper made three main contributions.In the first part,we focus on the problem of anomaly detection of passenger flow in a few important stations.For sequential collected passenger flow data that may be modelled by complex mixture distributions,inspired by the robust likelihood ratio method,this paper proposes a new minimum distance criterion.Then,this paper discusses the principle of solving the likelihood ratio function for several common distance criteria,and constructs the corresponding cumulative sum type control chart based on the criterion.Based on the real application scenarios,a nonparametric version of the proposed control chart without distribution assumption is given.In the second part,we focus on the monitoring of multiple stations or a whole line.In this scenario,the form of the data to be monitored tends to be much more complex.Therefore,we propose to use the tensor method to analyse the data.Different from traditional tensor methods which directly use the tensor of original structure for analysis,we suggest verifying whether the data should be fitted by the tensor model in the first step.Based on the tensor normal assumption,a new modelling and a fast online estimation are constructed.Finally,we design several control charts for different types of process shifts for corresponding online monitoring purposes.In the third part,for the special scenario of passenger flow monitoring,we further studied the longitudinal feature of the studied tensor datastream.We analyzed the problems of the traditional change-point detection model in passenger flow data monitoring,proposed the idea of using dynamic tensor method for monitoring,and accordingly presented a more applicable longitudinal tensor modelling and analysis method.The monitoring process based on the predicted residual tensor.For each part of the above research,we have verified the actual performance of the proposed method through comprehensive and in-depth numerical simulation and comparison,and provided broad and detailed practical examples to show the application value of the proposed methods.