Anomaly Patterns Detection And Visualization Research On Travel Behaviors With Subway Transit Data

The development of public transport is an important way to alleviate urban traffic congestion and the rail transit in Beijing plays an important role in the city's public transport as it is developing rapidly.With the increment of rail traffic flow,a large number of passengers who are not real passengers naturally join the rail transit passengers and they were called anomaly passengers in this paper.The presence of abnormal passengers seriously affects the safety of the passengers.So the trip characteristics of anomaly passengers should be known well to put forward a more effective way to manage the rail transit,improve the service level of rail transit,and increase the satisfaction degree of the passengers.The automatic fare collection system collects the data of the historical data of rail transit passengers,which provides a large number of high quality data for the study of the characteristics of rail transit passengers.The existing researches on smart card data are mainly concerned with the routine trips paying less attention to anomaly passengers.Under this background,this paper deeply explores the traffic trip law of anomaly passengers based on the data of Beijing urban rail transit records.We derived the longitudinal activity sequences from multi-day smart card data as representations of passengers and we clustered passengers to find anomaly passengers.Furthermore,we defined the distance between travel patterns of different anomaly passengers considering the spatiotemporal factors and clustered anomaly passengers based on the distance to detect anomaly travel groups.Finally,we validated the abnormal travel groups using the social network data.Generally speaking,the specific content of this paper is divided into the following three points:(1)Firstly,we proposed method of recognizing anomaly passengers based on smart card data of rail transit.We derived the longitudinal activity sequences from multiday smart card data as representations of passengers.Then PCA(Principal Component Analysis)was used to extract statistical trends in the sequences of all passengers and the low-dimensionality representation of the sequences was got.We clustered the low-dimensionality data using k-Means algorithm to recognize the anomaly passengers.(2)Secondly,we detected anomaly travel groups from the recognized anomaly passengers.Part of the abnormal travel passengers were in the form of groups,known as anomaly travel groups.We defined the distance between travel patterns of different anomaly passengers considering the spatiotemporal factors and cluster anomaly passengers based on the distance using DBSCAN to detect anomaly travel groups,those participated into the same group were considered to haverelationships.For anomaly passenger groups,we use social network data to verify the abnormal travel groups.(3)Thirdly,we analyzed the travel characteristics of anomaly passengers and anomaly passenger groups are analyzed using visualization method.We analyzed the temporal and spatial characteristics of anomaly passengers and the temporal and spatial characteristics of anomaly passenger groups.Also namomaly high incidence area were also analyzed.Overall,this paper recognized anomaly passengers and anomaly passenger groups based on smart card data and analyzed the temporal and spatial characteristics using visualization method.According to the results,we collected the stations that anomaly passengers occurred frequently and that is basically match with relevant reports.