Research On Short-term Passenger Flow Prediction Of Urban Rail Transit Stations Based On Passenger Flow Component Decomposition

With expanding city scale in urbanization,travel demand of residents is getting stronger,and as an important public transit travel mode,rail transit has been developing quickly,which has developed into the network operation stage in many cities' rail transit systems,and making urban rail transit networked is facing a circumstance where demand from passengers is increasing,network structure and ridership's variation is more complicated,thus it brings problems and challenges to operation managers.Analyzing passenger flow's characteristics deeply and predicting the urban rail transit's short-term ridership based on analysis are essential for managers to formulate manage plans and relevant policies,for operation department to improve the service level of rail transit,and for rail transit travels to be guaranteed.Therefore,based on previous research and smart card data of Beijing rail transit,this paper analyzed passenger flow's timevarying rule and passenger's travel characteristics from the network aspect and station aspect,and then based on these characteristics,the passengers flow was divided into stable flow and random flow.By comparing and analyzing several short-term passenger volume predicting methods,combining different ridership's characteristics,corresponding predicting models were chosen for both kinds of flow,and short-term passenger volume predicting method was built based on decomposing passenger components,and the prediction of workdays' entrance passenger flow of urban rail transit was realized,and the model provided was verified by applying ridership data of different kinds of stations.Research details are described as below:(1)By using the smart card data of Beijing on Nov.2015 and structuring passenger flow time series by setting time interval as 15 min,passenger flow's time-varying rules of the whole network and stations were analyzed,and passengers' travel characteristics of the whole network and stations were analyzed from passengers' perspectives.(2)With applying several predicting methods on several days' passenger flow data of four different kinds of stations,different methods' traits were compared to provide reference to subsequent model.(3)According to the analysis of passenger flow's time-varying rules and passengers' travel time characteristics of the whole network and stations,the number of days of travel and the range of travel time were the two variates to evaluate the passenger travel stability,and relevant ridership decomposition rule and algorithm were made on these two variates.(4)Based on the provided ridership components decomposition algorithm,ridership components of four stations were decomposed,and the characteristics of both kinds of flows were analyzed.According to different ridership characteristics and different predicting models' traits,it was decided that the historical average(HA)model was used on stable flow,while the relevance vector machine(RVM)under the frame of Bayes was applied on random flow,so the short-term predicting method based on flow components decomposition was built.(5)By instance analysis and comparison with other models,it was verified the model provided by this paper had favorable predictive ability and generalization ability.