An Research Of Beijing Metro Passenger Travel And Abnormal Behavior Identification: A Survey For "In-Out" At The Same Subway Station

With the investment increase ofmetro construction,the mileage of metro operations has increased significantly,and metro has entered the stage of network operation.Metro network operation is increasingly demandingmetro network service quality and refined management.Taking Beijing metro as an example,the metro network is extending in all directions,applying information technology,passengers can achieve barrier-free transfers and improve service experience.But at the same time,related problems also arise,because the ticket is not real-name and barrier-free transfer,it is difficult for the operating unit to accurately obtain the passenger travel path,which restricts the further improvement of the level of refined management.In this paper,in the pre-processing of 70 million passenger travel data of Beijing metro,some abnormal ride behaviors were found.These abnormal ride behaviors will cause hidden dangers to operational safety and will also affect metro network services and related company benefits.In this paper,one kind of abnormal riding behavior called “in-out at the same station”is counted.In-out at the same station,that is,in a short time,passengers swipe in-out at the same station.Statistics show that the ratio of in-out at the same station in Beijing Railway station is about 3.5‰,and in Tiantongyuan station is about 7‰;some stations with small passenger flow,such as Liangxiang station,even reach 9‰.What are the causes of the above-mentioned abnormal riding behavior? How to effectively identify and strengthen management? This paperwill analyze the reasons with the data,provide analysis methods and models for the operation management unit,and make preliminary management recommendations.The main research work is following.By analyzing Beijing metro OD(Origin-Destination)transaction data,based on data visualization analysis and machine learning algorithms,construct a metro co-station in-out analysis model;using cluster analysis,passenger feature analysis and machine learning methods,On the basis of the verification analysis of passenger portraits,identify passengers with abnormal behaviors,and provide decision-making recommendation for metro management.Specifically,the main content of this study includes five aspects: Firstly,based on data analysis,clarify the concept of in-out at the same station,and use Python language to design the analysis algorithm of in-out at the same station for verification analysis.The verification results show that the algorithm studied has good structural characteristics and can analyzelarge-scale data.Secondly,it is to analyze the characteristics of metro passengers in-out at the same station from the four dimensions of station,line,time and passenger.According to the results of the bucket analysis,it is found that,unlike the total result of in-out at the same station,the stations with a large proportion of in-out at the same station are Jingtai station,Liangxiang University Town station and other remote stations with few passenger flow,and the time interval of in-out at the same station is mainly concentrated within 5 minutes,individual tickets appeared 120 times at the same station.Thirdly,it is to carry out tracking analysis and time distribution analysis on passengers,and on this basis,study the behavior clustering of passengers.Fourthly,it is to use data visualization technology to analyze in-out at the same station habits of typical tickets in various categories,and to classify related abnormal behavior into three categories,namely,suspected theft,suspected begging and selling,and suspected illegal advertisements.The latter two can be called welfare act.Fifthly,it is to further discuss the problems of suspected theft,suspected begging and selling,and suspected illegal advertisements in the abnormal behavior,and based on the random forest algorithm,classify and manage the abnormal behavior to provide a predictable method for the operation management unit.The innovations of the paperis following.(1)Propose one kind of analytical methods for intensive data.In order to solve the problem of similarity or dissimilarity between two data objects in the process of cluster analysis of OD transaction data,the concept of dissimilarity measurement is introduced.The dissimilarity between data objects is completed before the cluster analysis of OD transaction data.Generally,the dissimilarity matrix(or similarity matrix)is used to express its dissimilarity.Traditional OD transaction data clustering analysis algorithms also need to solve the problems of low data mining efficiency and low accuracy in data-intensive computing environments.This paper proposes an improveddata-intensive Density Base Distributed Clustering(IDBDC)algorithm in the data-intensive computing environment,and introduces the Map Reduce programming model under the open source Hadoop project,which combines cloud computing and data stream clustering technology and integrates clustering algorithms into the Map Reduce model to effectively solve data analysis and mining in data-intensive computing environments.(2)Recognition methods and portraits of abnormal passenger.Perform clustering algorithm and data visualization analysis on tickets with abnormal in-out at the same station,analyzein-outat the same station habits of typical tickets in various categories,and conduct classification research based on related abnormal behavior,based on an data-intensive density base distributed clustering algorithm in the computing environment classifies abnormal behaviors into three categories,namely suspected theft behavior,suspected begging and selling,and suspected illegal advertisements behavior.Suspected theft generally has frequent transfers,random stops,and frequent short trips;suspected begging and selling generally have a relatively stable begging route every day and have long time duration stayed at the station;suspected illegal advertisements generally have fixed distribution time,and is distributed by the gang.(3)Passenger behavior prediction and verification.Due to management cost considerations,the operation management unit cannot fully verify the behavior characteristics of each ticket,that is,whether these tickets are suspected theft,suspected begging and selling,and suspected illegal advertisements behavior.This paper uses learning sample methods to predict the behavior characteristics of various types of tickets.This paper designs a predictive analysis method based on the random forest method,uses bootsrap resampling method to extract multiple samples from the original sample,model the decision tree for each bootsrap sample,and then combine the predictions of multiple decision trees,which are obtained by voting final prediction result.The research data set shows that the description of the predicted behavior is basically consistent with the behavior of suspected theft,suspected begging and selling,and suspected illegal advertisements.After the actual sampling test,the prediction model proposed in this paper is relatively accurate,which improves the accuracy and efficiency of the management of abnormal behavior.Figure 115,Table 63,and Reference148.