Research On OD Inference And Intelligent Prediction Of Public Transport Passenger Flow

With the rapid development of Intelligent Transportation Systems(ITS),the collection,analysis,and prediction of traffic data has become more and more important in road planning,vehicle operation and dispatching.However,the one-ticket credit card scheme is still adopted in most cities at present,that is,the passenger swipes the card when getting on the bus,and does not have to swipe the card when getting off the bus,so it is difficult to directly obtain the passenger’s full travel Origin-Destination(OD).In addition,some existing research do not consider the impact of passenger transfer behavior between rail transit and bus on inferred passenger flow OD,resulting in low accuracy of inference.In public traffic station passenger flow prediction,the existing methods often ignore the periodic characteristics of the station passenger flow and the spatio-temporal correlation of the traffic graph data.Aiming at the above problems,this thesis proposes the OD inference method of public transport passenger flow based on Trip Chain-Attraction Weight and the Spatio-Temporal Graph Convolutional Neural Network Prediction Model Based on the Periodic Component(Periodic ST-GCN).The main research of this thesis are as follows:First,this thesis proposes a passenger flow OD inference model based on Trip Chain and Attraction Weight.The model combines all bus and rail transit card swipe data to infer the OD of the entire traffic network.In the Trip Chain Method,this thesis considers the passenger’s transfer behavior between bus and rail transit,making the passenger’s travel trajectory more complete.Using the Trip Chain Method,more than60% of the passenger flow data throughout the day can be inferred,resulting in more accurate passenger OD inference results.In the Station Attraction Weight Method Based on Trip Chain OD,this thesis uses the Trip Chain Method to infer the historical OD data of the last 10 days to predict the possible getting-off stations for passengers,and this thesis uses the Station Attraction Weight Method Based on Point of Interest(POI)and Station Transfer Capabilities as supplement,which not only considers the attraction weight based on the environment around the station and the transfer capabilities of the station,but also improves the inferred matching rate of the getting-off station.According to the actual survey results,it is found that the accuracy of the passenger flow OD inference results is more than 90%.Second,this thesis proposes Spatio-Temporal Graph Convolutional Neural Network Model Based on Periodic Component to predict the passenger flow at public transportation stations.The model not only captures the spatio-temporal characteristics of traffic data through the spatio-temporal convolution block with a sandwich structure composed of one spatial-dimensional convolution and two time-dimensional convolutions,but also effectively considers the periodicity of passenger flows at public transportation stations through the recent,daily and weekly periodic components.In addition,because the graph convolution in the spatial dimension uses pure convolution operations,which makes the model less training parameters and converges faster.Through the experiment of predicting the OD of passenger flow at public transportation stations in Chongqing,it is found that Periodic ST-GCN achieves more excellent results in two evaluation indicators: MAE(Mean Absolute Errors)and RMSE(Root Mean Squared Errors).Third,using Unity Platform to Realize Chongqing Public Transport Passenger Flow OD Visualization System.This thesis analyzes the congestion of bus stations,rail transit stations and rail transit lines based on the calculated passenger flow OD,and adjust the departure strategy of rail transit at different time periods.