Theory And Method Of Rail Station Design Ridership Prediction Considering Peak Deviation

In the face of frequent station congestions and regular peak flow restrictions in most of China’s stations,which have resulted in loss of passenger benefits and negative social impacts contrary to the overall requirement of high quality and efficiency in the "Outline for the Construction of a Strong Transportation State",it has become a national strategic need to strengthen the scientific research on station design passenger forecasting methods.Conventional forecasting techniques have the obvious drawback of placing the emphasis on the line rather than the station,often neglecting one of the key issues of the station itself: the noncoincidence between the peak hours of the station and the peak hours of the entire line,i.e.the "station and line peak deviation",and thus the forecasting Station traffic,which is limited to the peak hour of the line,is used in practice by designers to design station traffic,resulting in some stations being undersized,under-served and overcrowded.In fact,the station peak and the line peak are two separate concepts,and the inconsistency of station and line peaks has been revealed in major cities around the world.In this regard,the latest Code of Practice for Urban Rail Transit Passenger Flow Prediction(GB/T 51150-2016)has added new content and requirements for station peak prediction,i.e."when the station’s peak passenger flow does not occur during the morning and evening peak hours,the station peak occurrence time and station passenger flow should be predicted and analysed".Based on this background,this study focuses on the problem of peak deviation at stations and lines,and systematically improves the theory and methodology for predicting station peak hours and design passenger flow,and reveals the mechanism of peak deviation from multiple perspectives in detail.The main research content includes the following four aspects:(1)Starting from the spatial and temporal patterns of rail passenger flow,the study investigates the causes of station peak deviations and paves the way for the subsequent chapters.To address the multidimensional properties of passenger flow data,a non-negative tensor decomposition method is introduced to unify the spatio-temporal characteristics of multidimensional data,and to systematically analyse the spatio-temporal characteristics of passenger flow from the passenger flow itself to the source of passenger flow generation.At the level of the passenger flow itself,the underlying pattern patterns and interaction structures of the day,time and station dimensions are extracted and interpreted at the physical and realistic levels,so as to fully understand the intrinsic laws of the heterogeneity of station peak patterns and the formation of station peak deviations;at the level of the source of passenger flow generation,K-Medoids clustering and multinomial logistic regression are used to link the passenger flow patterns at the station as a link.At the source level,K-Medoids clustering and multinomial logistic regression are used to explore the driving mechanisms behind the different peak characteristics of stations by four types of external factors: land use,network structure,accessibility and station attributes.(2)A methodological framework for predicting station peak hours that encompasses station and line peak deviations is proposed,while simultaneously exploring the mechanisms of peak deviation times.The deviation time variables are defined to measure the deviation of peak occurrence times between stations and lines,and are incorporated into the general framework of the conventional method to correct for individual station peak occurrence time forecasts at the micro level.Regarding the construction of the deviation time relationship model,considering the possible spatial dependence problem,three types of spatial measurement models combined with four spatial weighting schemes are proposed to form a total of twelve spatial model specifications to effectively capture the spatial dependence form of deviation time and the decay characteristics of spatial spillover effects;at the same time,considering the actual layout of the metro network,the network-based distance is used instead of the Euclidean-based distance to set the spatial The optimal spatial model scheme for each type of peak is determined by using the index system such as the red pool information quantity and likelihood rate test to achieve the prediction and explanation of deviation time.(3)A methodological framework for predicting station design passenger flow based on the passenger flow scale correction factor is proposed,and the dynamic change mechanism of the correction factor is also revealed.The scale correction factor is defined as a measure of the difference in scale between the station passenger flow during its own peak hours and the peak hours of the line,and it is incorporated as a deviation factor in the general framework of the traditional method to correct the forecast results of station design passenger flow.A state-ofthe-art non-linear technique(polar gradient boosting tree model),which does not require a priori assumptions and has a flexible model structure,is used to fit and explain the complex relationship between the deviation coefficients and the influencing factors.In addition,the relative importance of individual and collective factors on the deviation coefficient at different times and directions is compared and analysed,and relevant policy advice is proposed to provide reference for the formulation of priority strategies for land use planning around stations in order to enhance line operational efficiency.(4)Considering that changes in land use planning or network patterns along rail transit lines may potentially affect the peak deviation and design passenger flow of stations in the course of urban spatial development,a model for predicting the peak hours and design passenger flow of stations under the planning environment is constructed as a useful supplement to the proposed improvement method.The general idea is to treat the station peak formation mechanism as a weighted superposition of passenger travel probability curves for different purposes in terms of station peak time distribution.The model framework consists of two major parts: the daily station passenger flow prediction model is constructed based on the land use incidence method,incorporating the network index system in graph theory;the multi-peak Gaussian curve is used to fit the probability distribution curve of passenger flow travel time for each purpose,and the land use is used as a bridge to divide the passenger flow attributes and determine the passenger flow weights for each purpose,so as to construct the station passenger flow time probability distribution prediction model;finally,the station design with time variables is integrated to form the Finally,the model is integrated to form a station design prediction model with time variables.The paper aims to fill the research gap in the theoretical method of predicting station peak hours and design passenger flow in the context of peak deviation,provide more reliable prediction results for designers to determine the design scale of stations and operators to formulate train operation plans,and promote the scientific development and sustainable development of rail transit systems.At the same time,by systematically and comprehensively revealing the influencing mechanism of peak deviation,it provides an important reference for planners to avoid peak deviation in too many stations with a forward-looking planning perspective,thereby saving capacity resources and improving operational efficiency.