Study On The Real Time Passenger Flow Prediction Of Urban Rail Transit Stations

Urban rail transit has gradually become the skeleton of urban public transport.It is important to master the real-time station passenger flow to adjust the operation chart.Based on realizing dynamic traffic management,improving the rail transit travel experience,a more applicable prediction method is explored.Using big data technology to dig out the information of the automatic ticket collection system,and generate the short-term station passenger flow time series.According to the statistical theory with qualitative and quantitative analysis,it is found that the time series passenger flow equals with periodicity,trend and autocorrelation and influenced by site location and scale,land use characteristics,external traffic conditions,passenger attributes and other factors.Average daily short-term passenger flow,maximum short-term passenger flow,peak passenger flow,peak time and convexity are put forward to describe the law and as the basis for site classification.Combined the LSTM neural network with HP filter,through the improved PSO algorithm,gains filtering smoothing parameters,the number of hidden layers,the time step and the number of iterations to suit for f each station.Simulation results show that the combination model can greatly improve the prediction accuracy.The urban rail transit site real-time prediction model has high accuracy,applicability and practical significance,which are able to provide effective theory as well as method of forecasting,strengthen the safety protection,and provide scientific basis for dynamic management scheme.