Research On Railway Passenger Volume Forecast Model Based On Spatial Temporal Graph Convolutional Networks

With the rapid development of China’s market-oriented economy and the continuous improvement of high-speed rail technology,the number of railway passengers in China keeps increasing,and it shows a strong correlation with the economic development of urban areas and other factors.How to forecast railway passenger flow effectively,and provide data support for the management department to allocate railway capacity layout and arrange running trains,has become the problem that needs to be solved to improve railway operation efficiency and economic benefits.Railway passenger volume is affected by many factors and the composition is complex.The current railway passenger volume prediction methods are mostly based on statistical methods and time series prediction methods.The spatial characteristic information of railway stations and the influence of external factors between stations are not considered enough,which is difficult to meet the demand of prediction.In order to solve the above problems,this paper proposes a prediction model of railway passenger flow based on spatio-temporal data to solve the annual railway volume prediction problem,and explores and improves the model to solve the fine-grained railway passenger volume prediction problem.The specific research content is as follows.(1)In view of the influence of different external factors such as the spatial characteristics of railway stations and the characteristics of economy and population on the traffic volume prediction,a multi-factor dynamic graph convolution railway passenger volume prediction model is proposed.Using time attention mechanism and gated time convolution to get the time correlation of traffic data,using dynamic graph convolution to get the spatial correlation between stations based on railway passenger flow and external characteristics,and finally using the fusion module of learnable parameters to fuse the predictions that get the correlation of different characteristics between sites,then,the railway passenger flow prediction integrating external characteristics and dynamic correlation is obtained.The validity of the algorithm is validated by comparison experiments with other models on some urban railway passenger flow datasets in the Changjiang Delta.(2)To solve the problem of establishing correlation between spatio-temporal fine-grained passenger volume prediction and external characteristics of different granularity,an improved method is proposed.A DTW algorithm is used to calculate the similarity between different features of a site,and then generate the external feature adjacency matrix.Then,the railway passenger volume data is convoluted by dynamic graph and static graph based on external feature adjacency matrix,respectively,to obtain the spatial correlation based on passenger flow and the spatial correlation brought by external feature.Finally,the results are fused by the fusion module of learnable parameters to obtain the predicted results.The validity of the model is verified on the 2022 Spring Festival railway passenger volume data set of some cities in the Changjiang Delta,and discussed the different acquisition methods of external factors.(3)Design and implement a railway traffic flow visualization platform that integrates different prediction algorithms and data analysis.Combined with Spring Boot and Vue framework,the development of front and back end separation is carried out.The platform has realized the visual query for a variety of economic and social development indicators,and the prediction effect of the two models has been visualized.