Research On Collaborative Optimization Of Passenger Flow Control And Train Skip-Stopping In Urban Rail Transit Based On Reinforcement Learning

With the continuous accelerating of urbanization,the demand for passenger travel has surged.Due to its convenience and punctuality,urban rail transit attracts and carries a large number of urban traffic trips,which makes the passenger flow congestion problem of urban rail transit follow.Especially in peak hours,when the limited transportation capacity of train lines is not enough to meet the travel demand,many passengers will gather on the platform and be unable to get on the train.If the number of passengers waiting on the platform exceeds the design capacity limit of the platform,the station will have potential operational safety hazards.In order to relieve the pressure on the platform and protect the safety of the station,the way of passenger flow control is often adopted to limit the entry;For particularly crowded stations,the train skip-stopping is often used to reserve surplus space for crowded stations,so as to achieve the purpose of rapid transportation of concentrated passenger flow.However,the existing station passenger flow control and train skip-stopping strategies are usually based on the subjective experience judgment of the operation management personnel,and lack the help of scientific methods to determine the passenger flow control time,place,intensity and train skip-stopping situation.In order to provide a scientific basis for subway operation and management personnel,this paper proposes a method based on reinforcement learning double deep Q network to optimize the passenger flow control rate and train skip-stopping at each station within a certain period of time,so as to minimize the average waiting time of passengers when the number of passengers waiting at the platform is controlled under the platform safety threshold,it is helpful to alleviate the passenger congestion in some stations and ensure the travel efficiency of passengers.The main contents include the following aspects:(1)Investigate the passenger flow characteristics of urban rail transit stations,lines and networks,analyze the reasons for the formation of large passenger flow aggregation of urban rail transit,give the steps of passenger flow control and train skip-stopping collaborative optimization to dredge large passenger flow,and summarize the dynamic nonlinear characteristics of collaborative optimization and the random uncertainty characteristics of passenger arrival and travel.(2)Construct the collaborative optimization model of passenger flow control and train skip-stopping.The objective is to minimize the comprehensive index of the average waiting time of passengers(waiting time outside the station due to passenger flow control and waiting time on the platform)and the number of people exceeding the limit on the platform,take the passenger flow control rate and train skip-stopping as decision variables,and take the travel demand,flow control,train capacity as constraints.A multiobjective and multi-stage decision-making passenger flow coordination control model is constructed.(3)Design and implementation of the solution algorithm based on double deep Q network.Based on the reinforcement learning framework,the action set design suitable for multi station passenger flow control and train skip-stopping combination optimization is carried out;The reinforcement learning simulation environment considering passenger travel uncertainty is constructed to realize the simulation and optimization target value calculation of passenger arrival,enter station,boarding,alighting and transfer.(4)Case analysis of peak passenger flow inflow control of urban rail transit partial network: Taking Beijing rail transit operation line as the background,line 13 and Changping line are selected for on-site investigation and experiment.This paper uses Python language programming to realize the simulation environment and optimize the algorithm solution process,and calls the deep neural network of tensorflow framework to realize reinforcement learning training.The experimental results show that the iterative convergence of the algorithm proposed in this paper is stable,the model optimization results can control the number of waiting on the platform within the safety threshold and ensure the passenger travel efficiency.There are 42 figures,12 tables and 81 references.