Research On Collaborative Control Of Rail Transit Passenger Flow And Congestion At Multiple Stations Based On Short-Term Prediction

With the rapid development of urban rail transit,the problem of the large demand for travel during peak passenger flow periods not matching the limited capacity of rail transit has become increasingly prominent.In order to predict the evolution of passenger flow and develop corresponding passenger flow control schemes to ensure the operational safety of urban rail transit during peak passenger flow periods,this paper conducted research on shortterm passenger flow prediction for rail transit networks and multi-station collaborative control methods for passenger flow congestion based on the following steps:Firstly,the Automatic Fare Collection(AFC)data was used to obtain passenger travel OD;based on the Dijkstra algorithm,the passenger travel path information was extracted;from the perspective of the rail transit network,line,and station,the spatial and temporal characteristics of rail transit passenger flow demand under normal operating conditions were explored.Secondly,from multiple perspectives of space,time,and environmental factors that affect passenger flow distribution,a combination of qualitative and quantitative methods was used to analyze the impact of different features on passenger flow changes.Based on the Kmeans clustering method,rail transit stations were divided according to passenger flow change characteristics.In addition,the Spearman correlation coefficient and autocorrelation function were used to test the degree of correlation between historical passenger flow,environmental factors,and previous passenger flow and current passenger flow changes.Finally,multi-dimensional feature selection and fusion of passenger flow changes were achieved.Thirdly,the deep learning theory was introduced,and based on the spatial and temporal characteristics of passenger flow changes,a network-level short-term passenger flow prediction model based on multi-feature fusion was constructed using residual networks and attention-based long short-term memory networks.Through self-contrasting experiments,hyper-parameter selection was completed,and the performance of the short-term passenger flow prediction model under different model structures was compared with the baseline model from both horizontal and vertical perspectives,verifying the advantages of the short-term passenger flow prediction model based on multi-feature fusion and obtaining future passenger flow evolution prediction results to provide data support for passenger flow control scheme development.Finally,the passenger flow congestion propagation mechanism at rail transit stations was analyzed.On the basis of the existing passenger flow control methods,taking into account the constraints of passenger flow travel demand and train operation dynamics,and balancing service fairness and rail transit operation efficiency,a multi-station collaborative control model for passenger flow congestion was constructed,aiming to minimize the total passenger flow congestion and maximize the utilization rate of the remaining subway capacity,and a self-adaptive genetic algorithm was designed to solve the problem.Based on the future passenger flow evolution prediction results obtained from the above research,a passenger flow control scheme was developed for the early peak period.The experimental results show that compared with the predicted scenario and the actual scenario,the total number of delayed passengers decreased by 30.4% and 24.5%,respectively,while slightly improving the utilization rate of the remaining train capacity,effectively alleviating the congestion backlog of platform passengers,and achieving rational allocation of train capacity resources to ensure the safety and efficient operation of rail transit.