Research On Algorithms Of Railway Passenger Flow Prediction Integrating Passenger Flow Relationship And Spatial-temporal Similarity

China’s railway passenger transport has maintained rapid development.Accurate railway passenger flow prediction can help formulate reasonable passenger transport plans,make full use of railway passenger transport resources and meet the travel needs of passengers.The current railway passenger flow forecasting methods generally analyze and design for a single station,which separates the connection between stations in the railway network.Based on the flow relationship and similarity of railway passenger flow,this paper constructs a railway passenger flow prediction model integrating passenger flow and temporal and spatial similarity,which achieves higher prediction accuracy than the single station model.The main work and innovations of this paper are as follows:(1)Aiming at the problem that the current railway passenger flow prediction method does not deeply explore the passenger flow connection between stations,it is proposed to construct passenger flow diagram and spatial-temporal similarity diagram to represent the relationship between railway passenger flow.Based on the OD relationship of railway passenger flow,this paper constructs a passenger flow diagram to represent the passenger flow relationship between stations;Based on the similarity relationship of passenger flow at each station,dynamic time warping(DTW)algorithm and shape distance algorithm are proposed to measure the similarity relationship of passenger flow sequence,and a spatialtemporal similarity map is constructed.(2)Combined with the station connection in passenger flow diagram and spatial-temporal similarity diagram,a passenger flow prediction model F-SAGCN based on passenger flow diagram and spatial-temporal similarity diagram is proposed.Firstly,the weight of interaction between different stations at different times is determined through the attention mechanism,and then the characteristics of other stations are aggregated by using the station relationship in the passenger flow diagram and the spatialtemporal similarity diagram through the graph convolution module,and the time characteristics are extracted on this basis.Finally,using the idea of residuals,the aggregation features obtained by graph convolution are regarded as the impact of other stations on their own passenger flow,and the extracted time series features of stations are fused for prediction.Through the comparative experiment using the railway passenger flow data of a railway bureau,the average absolute percentage error of 46 stations in the data set is reduced to 7.93%.The experiment shows that the railway passenger flow prediction model integrating passenger flow and spatial-temporal similarity proposed in this paper improves the prediction accuracy.