Temporal And Spatial Distribution Short-term Prediction Of Passenger Flow In High-speed Railway Stations Based On Graph Embedding LSTM

With the rapid development of high-speed railway construction in China,the travel demand of passengers on high-speed railway is constantly increasing,which puts forward higher requirements for the development of high-speed railway with high quality and high standards.High-speed railway station,as the passenger distribution hub in the high-speed railway network,is under unprecedented pressure due to the significantly increasing passenger flow demand.On the one hand,the huge passenger flow gathered in the relatively closed area of the station is not only a test of the station operation and control ability,but also an impact on the safety of passengers.On the other hand,as the station is a transportation commercial complex,with the increase of passenger flow,the demand for personalized service of passengers is also gradually rising.How to effectively perceive and predict the distribution state of the passenger flow in the station,and according to the situation of the passenger flow,optimize the layout of the facilities in the station,improve the service level of the station,and ensure the safety of the station operation has become a bottleneck problem for the station operation and management personnel.In order to effectively solve the problem of passenger flow distribution prediction in high-speed railway stations,this paper conducts a deep research on it.The specific contents are as follows:(1)Passenger flow data statistics and spatiotemporal characteristics analysis.In this paper,the spatiotemporal distribution of passenger flow in the station is described and the research ideas are determined.The data of passenger flow in different regions are analyzed by using the monitoring video data which highly reflects the change of passenger flow in the station.At the same time,combined with the station passenger statistics data and passenger operation data,the station multi-source passenger flow data are deeply analyzed,and the passenger flow distribution characteristics are extracted from the time characteristics,space characteristics and spatiotemporal correlation.(2)Establish embedded model of network graph structure of passenger flow in station based on random walk.Summarize and analyze the layout of the waiting layer,streamline composition and passenger flow network characteristics.By introducing graph structure data,the passenger flow network is modeled as topological graph structure to express its spatial structure explicitly.The random walk sampling and skip-gram neural network model were used to embed the graph structure as the node vector,and the similar nodes in the passenger flow network were selected as the basic input of the prediction model using the vector cosine distance as the metric standard,which effectively extracted and utilized the spatial distance characteristics and spatial and temporal logic characteristics of the passenger flow network.(3)Based on feature embedded LSTM,establish the prediction model of spatial-temporal distribution of passenger flow in station.Considering the factors that influence the spatio-temporal variation of passenger flow,"functional attraction degree" and "proximity degree of departure time" are put forward to quantify the influence degree of different functional areas and train arrival and departure time on the passenger station distribution respectively.Based on LSTM neural network,different input variables were set according to different influencing factors,and the passenger flow prediction model of LSTM based on feature embedding is constructed.(4)Verify the prediction model by actual data.The video statistics of Tianjin West Railway Station are used as the basic input of the passenger flow prediction model to verify the model.The passenger flow prediction results show that compared with the traditional ARIMA prediction model and the basic LSTM model,the prediction effect of the graph-based LSTM passenger flow prediction model constructed in this paper is significantly improved,in which MAE and RMSE are reduced by 0.3-0.6 and MAPE is reduced by 3.6%.Moreover,it can realize the simultaneous prediction of passenger flow in multiple regions,which can effectively reflect the change trend of future short-term passenger flow in different regions of the station.With 65 figures,22 tables,and 90 reference literatures.