Study On Short-Term Forecasting Method For Boarding Passenger Flows In Urban Rail Transit

The development of the urban rail transit system has accumulated massive amounts of card swiping data of AFC passengers,which provides a basis for analysis of rail transit passenger flow and short-term forecasting.On the basis of analyzing the characteristics of passenger flow,a short-term passenger flow prediction model was constructed and the prediction performance of the model was analyzed in this dissertation.The contents are as follows:(1)A description of AFC data and line station data,and initial data Extraction and processing;(2)Analyzing the factors affecting the passenger flow distribution of stations from different dimensions,and classify the stations,and finally Analyzed the characteristics of the temporal and spatial distribution of rail transit passenger flow;(3)Select representative methods-ARIMA and LSTM,and construct a short-term model based on ARIMA and LSTM.Hourly inbound passenger flow prediction model,and an example calculation;(4)According to the calculation results,first compare the prediction performance of ARIMA and LSTM,and then analyze the prediction performance of LSTM model for different time granularity and different prediction stations.The results show that:(1)The LSTM prediction model has greater advantages than the ARIMA prediction model.(2)For the time granularity,when the time granularity is 5min or15 min,the LSTM prediction effect is stable,but the calculation speed will obviously decrease with the complexity of the network structure;the time granularity is 30 min,the LSTM prediction effect is unstable,but the calculation speed increases with the complex changes in the network structure are minor.(3)For the station type,stations with regular travel distribution,15 min or 30 min can reflect the law of passenger flow.Stations where travel is affected by many factors,30 min can reflect the law of passenger flow;Under the same time granularity,the prediction performance of the LSTM model of the stations with concentrated inbound passenger flow distribution is always better than that of the stations with large inbound passenger flow fluctuations.