Research On Passenger Flow Forcasting Method For Urban Rail Tansit During Massive Gatherings

At present,China's urban rail transit has stepped into a period of rapid development,with the rapid expansion of networks in major cities.Rail transit has gradually become the core of urban public transportation system and its efficient works become an important driving force for urban development.In recent years,various large-scale cultural and sports activities and celebrations become more and more frequent,which greatly affects the normal operation of urban rail transit,causing a sudden increase in traffic volume and aggravating the unbalance of network operation and the potential operation risk.In order to deal with the shocks by large-scale activities,it is necessary to accurately predict the passenger demand the scenario of large-scale activities,so as to properly organize the adjustment of transport capacity and avoid local congestion caused by the event.In addition,station closure control is more and more applied in the passenger flow organization of large-scale activities,so it is necessary to conduct in-depth research on the law of passenger flow under the condition of station closure intervention.In this study,based on the spatiotemporal correlation characteristics of urban rail transit network and the passenger flow characteristics,the passenger flow prediction methods suitable for normal conditions,large-scale activities and station closure intervention conditions were proposed respectively,to support the passenger flow prediction of large-scale activities in all aspects.The main research contents are as follows:Firstly,based on the massive historical passenger flow data,the laws of passenger flow in network under normal conditions,large-scale activities and station closure intervention are compared and analyzed,and the key factors influencing the evolution of passenger flow under different scenarios are clarified,so as to provide reference for the full discovery of the demand for passenger flow prediction and the design of scientific and effective passenger flow prediction methods.Secondly,aiming the the problem of short-term passenger flow prediction under normal conditions,a GCGRU model combining cyclic Gating Unit and Graph Convolutional neural Network is proposed.It combines GCN(Graph Convolution Network)to extract the spatial correlation of passenger flows in different stations,and describes the time-varying characteristics of passenger flows in stations with GRU(Gated Recurrent Unit).Compared with ARIMA(Autoregressive Integrated Moving Average),SVR(Support Vector Regression),BPNN(Backward Propogating Neural Network),the prediction accuracy was significantly improved.Thirdly,aiming at the passenger flow prediction problem in large-scale activities,the Grey Prediction model based on the passenger flow component division is constructed to predict the change value of activity volume and background volume respectively.The OD distribution under the influence of large-scale activities was predicted by analyzing the historical law of activity inbound volume and OD volume.Based on the case results of the Autumn Canton Fair,it was shown that the model had high accuracy and good adaptability,and could meet the decision-making needs of transportation organization and passenger service.Fourthly,in view of the station closure intervention in large-scaled activities,a combination forecast model is proposed based on Symbolic Aggregate Approximation(SAX)and Dynamic Factor Model(DFM).SAX algorithm was used to identify the potential impact range of station closure control,and DFM was used to decompose the general and fluctuation characteristics of passenger flow.The case study based on National Day 70 th anniversary celebration shows that this method can effectively identify the affected stations.Compared with ARIMA model,the root-mean-square(RMS)error decreases by 26.1% and the MAE(Mean Absolute Error)decreases by 20.78%.