Fluctuation And Forecast Of Urban Rall Transit Passenger Flow Under Rainfall

Forecasting passenger flow is the basis of the operation decision,while the operation decision is the further evaluation.Accurate prediction plays an important role in ensuring the normal realization of urban transportation system functions,and operation departments should strengthen the application of prediction technology to improve the service level of urban rail transportation and the quality of public travel.In this paper,the existing clustering method is improved to refine the station clustering.The spatial and temporal fluctuation characteristics of passenger flow in and out of stations,OD and sensitive line cross-sections in rainy days and the interaction with land use indicators were analyzed for each type of stations.Finally,the refined rainfall characteristics and indicators of passenger flow fluctuation patterns were selected to construct the prediction model.The main research results are as follows:(1)To better study the characteristics of passenger flow fluctuations and the generation mechanism under the rainfall environment of different functional stations,so as to improve the prediction accuracy of the passenger flow prediction,a refined classification of stations is required.The POI-K-Means++clustering model is established by selecting three types of indicators:POI data,station attributes and passenger flow characteristics.Using the elbow method to determine the optimal number of clusters,so that the stations in each subclass not only have similar passenger flow trends but also station attribute characteristics,which improves the deficiency of the existing model in defining mixed stations.It also enhances the interpretation of the clustering results on the influence of rainfall on each station.(2)The nine-term moving average method is used to obtain the smoothed values of normal weather passenger flow,and the concept of passenger flow volatility is introduced to characterize the deviation under the influence of rainfall.The fluctuation of passenger flow at various types of rail transit stations,between stations,sections,and the characteristics of peak heavy passenger flow sections during rainy days are compared.At the station level,rainfall magnitude and other time-varying characteristics produce persistent passenger flow fluctuations.There are differences in the magnitude of passenger flow fluctuations,rainfall sensitivity at different moments and temporal stacking at various types under different rainfall intensities.At the level of interaction between different types of stations,there is an interactive relationship between the OD passenger flow fluctuations and POI indicators and influenced by the coupling.From the sensitive line,the sections that produce continuous positive and opposite fluctuations in the up and down sections are located in the area of the continuous residential station,and of regional center type stations and mixed residential-employment type stations with supporting facilities in the morning peak.In addition,the area of continuous residential stations produces a large passenger flow section and directional imbalance.Accordingly,station layout and land use planning recommendations are proposed along the line.(3)Based on the fluctuation characteristics of passenger flow under the rainfall and the station clustering,we introduce the passenger flow fluctuation that can reflect the instantaneous and continuous impact patterns of rainfall.The refined time-varying characteristics of rainfall are corresponded to the fluctuation of passenger flow on the time axis,and a prediction model with universal applicability to both normal and rainfall environments is established,so it has better prediction accuracy than existing models.The rainfall is treated with ordinal variables,and the rainfall level,occurrence time,duration and corresponding passenger flow fluctuation indicators are input into the optimized LSTM and Bi LSTM models in the time axis for prediction.The evaluation metrics of MAPE and R~2 show that the optimized Bi LSTM model has higher prediction accuracy for all types of stations,MAPE values were reduced by 0.78%on average and R~2 improved by 0.0128.The Bi LSTM model can encode back-to-back information and sense the impact of weather on passenger travel at future moments when training the model,which has promising applications in operation scheduling,safety management,passenger travel guidance and smart city planning.This thesis is supported by the General Project of Shaanxi Provincial Natural Science Foundation(Youth),No.2022JQ-345.