The Study On Peak Duration Prediction Of Urban Railway Station Passenger Flow

In recent years,urban rail transit has been developing rapidly in China.Many large cities have built a certain scale but not fully mature rail network,so it has caused many problems.Among those problems,accurately prediction of the duration of the peak period of passenger flow in railway station is one of the problems that the railway management department pays special attention to.Firstly,The AFC data was processed and analyzed by using python and matlab in the paper and the rail station flow（in）was got,then the theory of passenger service ability of congested easily parts（security,ticket gate,staircase+escalator）was got by surveying,and critical threshold of peak passenger flow——minimum capacity of 85%of the passenger flow value,was calculated according to the V₈₅ theory.In addition,the passenger flow per minute and the duration of the peak period of passenger flow were analyzed by setting judgement conditions for the starting and ending points of the peak period of passenger flow.At last,the variation characteristics of daily passenger flow and peak period of passenger flow are analyzed,and the starting point of peak period is predicted.Secondly,three models are selected to predict the duration of the peak period of passenger flow in urban railway station,and Nanping station is taken as a case study.1)the traditional multiple linear regression model（MLR model）was established,and the factors with linear relations were selected as the independent variables of the model,and the application result of test data shows that the model RMSE=28,MAE=20,and the prediction result is discretized according to the discretization condition of 30min,and the overall accuracy is poor,the prediction effect is good only within the interval of（60,90]min.Therefore,the traditional model has a great limitation in the prediction of the duration of the peak period of passenger flow in urban railway station.2）the bayesian network model（BN model）was constructed,the influencing factors were discretized and imported into the model for structure learning and parameter learning,and the structure diagram of bayesian network was obtained.The structure diagram shows that there are 5 factors that directly affect the predicted results.Duration of the test data in accordance with the 15 min,30 min,60 min three conditions of discretization and application,the result shows that the three conditions of the overall classification accuracy is above 90%,and part of the prediction accuracy of 100%,the prediction of extreme effect is good,and the prediction effect of 15 min is best,the prediction effect of 60 min is worst.As a whole,the BN model can predict the duration of urban rail peak passenger flow in a better way.3)the support vector regression model（SVR model）and combination model（BN-SVR model）were constructed.The experimental result shows that the overall prediction effect of SVR model is good,RMSE=13,MAE=7,but the prediction effect is poor when the duration is around75min、110min and150min.RMSE=9 and MAE=6 of BN-SVR model,although the prediction accuracy of some periods is slightly less than that of SVR model,the prediction accuracy of about 75min、110min and150min is much higher than that of SVR model.Therefore,BN model and BN-SVR model established in this paper can scientifically and accurately predict the duration of peak passenger flow in railway stations,and provides a basis for the management and operation of the railway station.