Passenger Flow Analysis And Forecasting Of Urban Rail Transit System Based On Multisource Data

As one of the main transportation modes to undertake public passenger transportation services in the metropolis,urban rail transit is an important channel to ensure concentrated and on-time travel for residents.With the slow downward speed of urban rail transit construction,urban rail transit operation management enters a period of rapid improvement.Improving operation management helps to improve the passenger transportation sharing rate and economic efficiency of rail transit.Passenger flow analysis and forecasting of the urban rail transit system can provide the basis for passenger flow organization,abnormal passenger flow detection,and rail train schedule,which is important for improving the management efficiency and service level of the urban rail transit system.Firstly,the passenger flows of each transportation mode are extracted based on automatic fare collection data from the rail transit system,bus data,and taxi order data.The passenger flow of bus stations is obtained by fusing automatic fare collection data from the bus system and bus entry data.The change trends in passenger flow of each transportation mode are analyzed through data visualization.The characteristics of time distribution and spatial distribution of rail transit passenger flow are summarized.The results show that the fluctuation trend of passenger volume of different rail transit lines is different.The passenger flow of some lines shows an obvious growth trend.There is no station where the inbound passenger flow is consistently greater than the outbound passenger flow or the outbound passenger flow is greater than the inbound passenger flow.Secondly,based on the service scope of rail transit and other transportation modes to passengers,the buffer zone of rail transit stations is delineated,and the bus and taxi passenger flows in the buffer zone of stations are identified to provide the data basis for passenger flow correlation analysis.The maximum information coefficient is used to measure the correlation between rail transit inbound passenger flow and bus departure flow and taxi departure flow respectively,and the transfer entropy is used to estimate the causal relationship between taxi arrival flow in the buffer zone of stations and rail transit inbound passenger flow.The results show that there is a significant correlation between inbound passenger flow and bus departure passenger flow at most rail transit stations,and a significant correlation between inbound passenger flow and taxi departure passenger flow at individual stations.At some rail transit stations,taxi arrival passenger flow with a fixed time lag has a stronger causal relationship with the inbound passenger flow of rail transit stations.Then,the inbound and outbound passenger peaks of rail transit stations are extracted based on the peak hour.The time and spatial patterns of passenger peaks are defined and extracted.The frequency periods and the maximum flow direction of passenger peaks of rail transit stations are analyzed.The frequency distribution of time and spatial patterns of all stations are statistically analyzed.The overall time and spatial pattern distribution characteristics are explored.The abnormal passenger flow peaks are analyzed and identified.The results show that some stations have frequent peak hours and frequent maximum flow directions;the overall time pattern of the rail transit system is usually concentrated in the morning peak hours and the spatial pattern is usually concentrated in the train station direction.Compared with the peak hour passenger flow,the duration of the abnormal passenger flow peak is short and the volume of passenger flow is large,and some stations have a sudden increase and decrease in passenger flow in a short period.Finally,the inter-station passenger interactions of rail transit are modeled based on the analysis of inter-station passenger interaction characteristics,and the inter-station passenger interactions of rail transit are divided into the interaction based on walking time,based on the line connection,and based on the correlation coefficient.The inter-station interactions are estimated and represented as interaction graphs by using the web map interface to obtain data such as walking time and walking distance between stations.A short-time passenger flow forecasting model with interaction graphs and historical passenger flow as inputs is proposed.The graph convolutional neural network is applied to capture the non-Euclidean spatial interaction features between stations.The multi-task learning structure is used to forecast the short-time passenger flow of multiple stations simultaneously.The experimental results show that the proposed model has higher forecasting accuracy and a better fit than other classical comparison models.