Research On Forecasting Of Urban Rail Transit Passenger Flow Based On Big Data

With continuous development in recent years,the rail transit has become the main mode of transportation for residents.With the continuous improvement and development of rail transit lines,the passenger flow is increasing,which may also easily cause the passenger flow congestion at rail transit stations.It is not conducive to residents’ travel and rail transit operation management.Therefore,the short-term forecast for the passenger flow in rail transit stations can help optimize the operation and management and ensure citizens’ safe and efficient travel.Through collecting the card data of all rail transit stations from Hangzhou Auto Fare Collection(AFC)System,the paper establishes the short-term passenger flow prediction model after processing the data.First of all,the author conducts data processing to turn the original data into passenger flow data in every ten minutes of each station,and analyzes passenger flow data characteristics from the station and time aspects.From station aspect,distribution characteristics of passenger flow of different stations are analyzed,and the stations are classified into five types,then the passenger flow distribution of each type of stations are analyzed.And from the time aspect,the paper analyzes the characteristics of the passenger flow on workdays and holidays.It is concluded that the passenger flow on weekdays presents different degrees of peak,while the passenger flow on weekends is relatively more complex and is greatly affected by random factors.This paper establishes the LightGBM prediction model of the passenger flow on workdays and weekends respectively.Taking Jinshahu Station as an example,the paper measures the feature importance of the passenger flow on workday and weekend respectively,and optimizes the model parameters by using grid search algorithm.At last,RMSE is used to evaluate the prediction results of the passenger flow.RMSE results show that LightGBM prediction model has good effect on workday prediction,and the RMSE value of passengers entering and exiting the station is 18.26 and 13.67 respectively,while LightGBM prediction model has a relatively poor effect on weekend prediction,and the RMSE value of passengers entering and exiting the station is 23.36 and 28.14 respectively.In order to optimize the prediction model of weekend passenger flow,this paper,based on a new weight fusion of tree model and neural network model,establishes the LSTM model to predict weekend passenger flow.Then the author conducts weight fusion of LightGBM model and LSTM model,which outputs the RMSE value of passengers entering and exiting the station at 19.46 and 25.97 respectively.The prediction accuracy of weekend passenger flow is therefore improved.