Short-term Passenger Flow Prediction And Dynamic Regulation Method Of Train Operation In Urban Rail Transit

With the acceleration of urbanization,the increasing passenger flow has increased the burden on urban rail transit.Due to the uneven distribution of passengers’ travel demand in time and space,the passenger flow of the subway shows a dynamic trend,which seriously affects the train operation.The dynamic regulation of train operation according to the passenger flow of the station can make the operation of the train better match the passenger flow characteristics and reduce the passengers’ travel time.However,the existing passenger flow-based train operation regulation methods all regard the passenger flow as a static parameter,and cannot obtain the real-time passenger flow data of the station,so the regulation strategy cannot match the real-time passenger flow distribution.Therefore,it is of great significance to study the short-term passenger flow prediction method of urban rail transit and adjust the train operation according to the realtime passenger flow dynamically.This dissertation studies the short-term passenger flow prediction of urban rail transit and the dynamic regulation method of train operation.In terms of short-term passenger flow prediction,this dissertation uses Generative Adversarial Networks(GAN)to enhance the original passenger flow data set,then deeply mines the spatiotemporal distribution characteristics of subway passenger flow,and establishes a deep neural network model to accurately predict the passenger flow of each station from the time and space dimensions.In terms of dynamic regulation of train operation,in order to ensure the real-time performance of train operation regulation,this dissertation converts the problem of train operation regulation into a Markov decision process and uses reinforcement learning to solve it.The main contents of this paper are as follows:(1)The existing algorithms of urban rail transit passenger flow prediction and train operation regulation are studied.The principle and implementation of the current passenger flow forecasting methods and train operation regulation methods are studied respectively.Combined with the characteristics of urban rail transit,the advantages and disadvantages of the existing algorithms are analyzed.(2)Urban rail transit space-time passenger flow is predicted accurately.A GAN network is built to enhance the data of the original passenger flow data set to improve the generalization ability of the prediction model;deep neural network models are established to predict the passenger flow from the space-time dimension accurately.The simulation results show that the prediction effect has been greatly improved compared with existing methods.(3)A real-time regulation model of urban rail transit train operation based on dynamic passenger flow is established.According to the running state of the train,a dynamic model of the arrival time of the train,the number of people on the train,and the number of people who cannot get on the train due to the train capacity are established,and the optimization objectives and constraints are designed.On this basis,the process of train operation regulation is transformed into a Markov decision process,and a solution method based on reinforcement learning is designed.(4)Based on the reinforcement learning Q-learning and DQN algorithms,the train operation regulation model is simulated and verified in the ATS simulation system.The reinforcement learning algorithm can not only achieve good optimization results in multiconstrained optimization problems but also meet the real-time requirements of train operation regulation.The verification results in the ATS simulation system also prove the effectiveness of the algorithm.The simulation results show that the passenger flow prediction method and the train operation regulation method based on dynamic passenger flow proposed in this dissertation can accurately predict the passenger flow of each station,so that the train operation regulation strategy can effectively match the passenger flow of the station and improve the urban rail transit service quality.The dissertation includes 67 figures,19 tables,and 87 references.