Research On Optimization Of Passenger Flow Collaborative Control In Urban Rail Transit Network

With the increasing scale of urban rail transit network,the rail transit network of many large and medium-sized cities during peak hours presents the characteristics of over-saturation of regional passenger flow,uneven distribution of passenger flow in time and space,and difficult transportation organization In order to deal with this problem,line-level and network-level passenger flow collaborative control is an effective means to alleviate the congestion of passenger flow and improve the safety operation level as well as the transportation efficiency.Research on passenger flow distribution and dynamic evolution is an important prerequisite for developing collaborative passenger flow control strategies.However,the existing research has seldom fully considered the dynamics and complexity of passenger flow distribution under the condition of passenger flow control.In addition,most urban rail lines and stations implement passenger flow control strategies based on static historical operating data and manual experience.There is a lack of corresponding scientific basis and theoretical support for the refined collaborative control of passenger flow.In response to these problems,this dissertation has done the following work:(1)Dynamic estimation of OD distribution of urban rail transit passenger flow based on computing graph.Based on the analysis of the characteristics of rail transit passenger flow distribution during peak periods,a framework for dynamic estimation of passenger flow OD distribution and route screening is established from the perspective of data mining.Using AFC data and alternative route information,a neural network model for estimating the OD distribution of passenger flow is proposed to try to reveal the relationship between passenger departure time,arrival time and route selection reflected in passenger flow data.A solution algorithm based on the chain rule is designed,and the“loss error” between the estimated traffic and the real data is obtained based on back-propagated mechanism to achieve the calibration of the OD distribution ratio of the entire network.This framework attempts to build a modeling and optimization method that integrates AFC data and passenger travel decision-making process with the aid of data-driven and model-driven models,and enhances the accuracy of the OD distribution estimation of urban rail transit passenger flow.A case study based on Beijing subway network is analyzed.The results show that the proposed model can effectively perform dynamic estimation of OD distribution and path selection.(2)Passenger flow evolution of urban rail transit network based on space-time state network.Facing the complexity and randomness of the temporal and spatial evolution of passenger flow in multiple scenarios,combined with the results of OD distribution estimation,the evolution model of rail transit network passenger flow based on space-time state network is proposed to try to solve the dynamic matching relationship between the passenger flow of urban rail network and routes and trains in multiple scenarios.Firstly,based on the physical topology network of urban rail transit,the station node is expanded into a small network composed of multiple nodes such as entry/exit,platform,channel,etc.Then the time dimension is introduced to construct a passenger travel space-time network that includes ride arc,waiting arc,virtual arc,and "block space-time domain" to describe the behaviors of passengers retaining,delaying and transferring under the conditions of current restriction and station closure.On this basis,considering the impact of passenger congestion in the station aisle on the speed of passengers,the state dimension reflecting the degree of congestion is introduced,and an extended space-time-state network is constructed.Secondly,with the goal of minimizing the generalized cost of the system,considering the constraints of train capacity and station passage capacity,a temporal and spatial evolution model of passenger flow suitable for multiple scenarios of current limitation and station closure is established,and a heuristic decomposition algorithm based on Lagrangian relaxation is designed to solve the model.Finally,an empirical analysis is carried out with the Beijing subway network as a case.The results show that this model can effectively accurately portray people-road-vehicles on the urban rail network in multiple scenarios,realize the accurate deduction of the passenger flow distribution,and further provide control strategy evaluation basis for subsequent passenger flow control optimization.(3)Optimization of passenger flow collaborative control of urban rail transit network based on deep reinforcement learning.On the basis of clarifying the optimization goal of the network passenger flow collaborative control and the meaning of the passenger flow control strategy,a network-level passenger flow collaborative control optimization nonlinear programming model is proposed.On this basis,the decision variables,objective functions,and constraints in the model are abstracted into the "action","reward",and normalized "state" in the reinforcement learning problem,respectively,and a reinforcement learning model for network-level collaborative control of passenger flow is constructed based on the Actor-critic deep learning framework.Combined with the post-experience replay strategy,the model is solved by the deep deterministic strategy gradient algorithm.A case study based on Beijing subway network is implemented and analyzed.The results show that the model can provide refined passenger flow control strategies from the three dimensions of time,space,and intensity,and achieve coordinated passenger flow control at the rail transit network level.