Research On Short-term OD Passenger Flow Prediction Model Between Stations And Lines Of Urban Rail Transit

With the rapid expansion of urban rail transit network scale and dramatic increase of passenger flow,the adaptability of operation scheme to passenger flow time-varying is obviously insufficient to meet the existing passenger flow demand.Real-time and accurate rail transit OD passenger flow prediction is helpful for the operation department to better understand and respond to passenger flow changes,and provide strong support to improve the operation efficiency of rail transit system.A short-time OD passenger flow prediction model for urban rail transportation is constructed in this paper using rail AFC data,by combining the spatial and temporal characteristics of OD passenger flow.The main research contents are as follows:Firstly,the original data of rail AFC is processed by My SQL and Python to obtain OD passenger flow data with different time granularity,analyze the passenger flow characteristics from two aspects of periodicity and fluctuation,combine the OD attractiveness,apply K-means clustering algorithm to divide the rail transit OD passenger flow into five categories,and analyze the short-time OD passenger flow from three aspects of operation time,preceding passenger flow and external weather OD passenger flow influencing factors are analyzed and summarized.Secondly,taking the advantage that Bi GRU can handle passenger flow time series in both directions,the grid search method is applied to optimize the model parameters,and the Bi GRU prediction model is constructed to predict different types of OD on passenger flow,and the model prediction effect is analyzed in three aspects: different prediction models,different time granularity,and working days and non-working days.The MAE is reduced by 0.0286～1.6278 and the RMSE is reduced by 0.2821～4.2345;in terms of different types of OD pairs,the prediction model is less effective in predicting the unpeaked OD pairs with relatively unstable and less regular passenger flow;in terms of prediction time granularity,the Bi GRU model has the best prediction effect at 15 minutes time granularity;on weekdays,the model has the best prediction effect.In terms of prediction time granularity,the Bi GRU model has the best prediction effect at 15-minute time granularity;in terms of weekdays and non-working days,the Bi GRU model has better prediction effect than non-working days on weekdays.Finally,considering the temporal and spatial characteristics of OD passenger flow and combining the advantages of GCN and Bi GRU,the optimal combination of parameters of the model is determined by the grid search method,and the combined GCNBi GRU model is constructed.The OD passenger flow of Chongqing rail transit line 3 is selected for example analysis,and the prediction effect of the combined model is analyzed from different prediction models and two aspects of weekdays and non-weekdays,and the results show that the prediction effect of the combined GCN-Bi GRU model is better than that of the GCN,GRU,Bi GRU and GCN-GRU models,and the MAE is reduced by0.0082～0.1220 and RMSE is reduced by 0.0118～0.2381;the combined GCN-Bi GRU model has better prediction effect in both working days and non-working days.