Monitoring And Estimating Of Sudden Large Passenger Flow In Urban Rail Transit

With the rapid progress of area economy,the scale of urban rail transit(URT)has been expanded year by year in recent years,and the operation mode and passenger flow(PFL)distribution have become a network pattern.Accurate and stable short-term PFL prediction plays a guiding role in the safe and stable operation of stations and trains.At the same time,due to the demand for spiritual life,various competitions and concerts are often held in some economically developed cities,which makes the transportation and evacuation of such large PFL become a brand new problem.Based on this,this paper takes individual travel chain as the research material,and through analyzing the generation mechanism and characteristics of sudden large PFL,puts forward the online PFL monitoring-prediction system,which provides the basis for the management department to make decisions in advance.Firstly,the individual card data of URT is used to capture the travel trajectory,and the occurrence mechanism of regular and sudden large PFL,as well as the differences and connections between the characteristics of PFL are explained from a micro perspective.By analyzing the size and time effect of the inbound and outbound PFL,it is concluded that outbound PFL was monitored in real time,and inbound PFL was estimated according to the monitoring results.Secondly,although the monitoring target is the outbound PFL,due to the PFL characteristics of URT,the time series decomposition method STL is applied to eliminate the periodic items,which eliminates the influence of periodic large PFL on the detection model.At the same time,due to the existence of outliers in the historical data,HTM anomaly detection model is used to carry out rolling detection modeling,which overcomes the influence of outliers.By combining the advantages of the two models,a STL-HTM detection framework was obtained,and an alarm system was established according to the detection results.Thirdly,based on test results of STL-HTM detection system,combined with the historical travel characteristics of individuals,the prediction period of the sudden PFL into the station is studied.By analyzing the composition of sudden PFL,the forecasting strategies of grouping and superposition are put forward.For the first part of the regular PFL,the recursive and direct prediction strategies are applied to multi-step prediction.For second part PFL of special purpose,by extracting the distribution of the time difference between inbound and outbound swiping,the distribution is carried out according to the additional outbound total amount.The third part transfer PFL is obtained by error fitting.Finally,taking Shanghai urban rail transit system as an example,the application of STL-HTM online monitoring,and the forecast model for the sudden inbound PFL of station is estimated.The results show that,compared with traditional methods,the proposed detection and estimated model has significant improvement in each index.Through the model parameter analysis,the maximum difference between MAE and RMSE in peak hours is within 20 person-times.At the same time,most of the difference between MAE and RMSE in the whole period was within 10 person-times.