| Accurate prediction of urban traffic speed can help traffic management departments accurately grasp urban road network information.Precise information can more effectively support control decisions and better guide drivers to make travel choices,thereby improving traffic flow efficiency and reducing traffic congestion.For commuters,accurate short-term speed prediction can improve travel efficiency and experience,reduce travel time and costs,and improve urban quality of life.To more accurately predict short-term traffic speed in urban road networks,this paper first analyzes the spatiotemporal characteristics of urban road network speed,including the spatiotemporal characteristics of intersections in the road network.Then,a novel shortterm speed prediction model is proposed that considers the spatiotemporal characteristics of the road network as well as the impact of intersections.The advantage of the model lies in its ability to better consider the abstract relationship between intersections and speed in the road network.Finally,the characteristics of the model and other models in different road network scenarios are analyzed and studied through three evaluation indicators,proving that the model proposed in this paper is more suitable for predicting short-term road network speed than other models.The main work is as follows:The area road network composed of 50 roads around Fuzhong Road in Futian District is taken as the research object,and the spatiotemporal correlation characteristics of the average speed of the road network are analyzed from the aspects of time and space.In the time dimension,the similarity,periodicity,and uncertainty of urban road network speed are analyzed.It is found that the average speed characteristics of the previous period can affect the characteristics of the road network in the next period.When predicting road network speed,the characteristics of road network speed in the time dimension need to be fully considered.In the spatial dimension,the similarity,spatial autocorrelation,and spatial heterogeneity of urban road network speed are analyzed.It is found that the spatial characteristics of road network speed can be transmitted to adjacent road sections and intersections,so the characteristics of road network speed in the time dimension and the influence of intersections on road network speed need to be considered in subsequent predictions.This paper defines a new factor in the road network called the intersection influence factor,based on the intersection delay time.In the spatial dimension,two spatial convolutional blocks are constructed to respectively extract the spatial characteristics of road network speed and the abstract relationship between intersections and speed in the road network.Finally,a distance and intersection influenced graph convolutional network model(DI-GCN)is constructed by combining a temporal convolutional block with a spatial convolutional block.Depending on the arrangement of the spatial convolutional blocks,the DI-GCN model can be divided into DI-GCN(parallel)and DI-GCN(stacked)models.Finally,the speed data of the area road network is used to prove that the DI-GCN(stack)model is more suitable for shortterm speed prediction of urban road networks.When predicting the speed in the next15 minutes,compared with the STGCN model,the MAE,RMSE,and MAPE errors of the DI-GCN(stack)model decrease by 7.688%,8.599%,and 6.437%,respectively.By comparing the speed prediction data of the road network composed of small areas and 331 road sections in the administrative region,the performance of the DIGCN(parallel)model and DI-GCN(stacked)model was compared.When predicting the speed for the next 15 minutes,compared to the STGCN model,the MAE,RMSE,and MAPE errors of the DI-GCN(stacked)model decreased by 5.872%,2.133%,and6.317%,respectively,performing better than the DI-GCN(parallel)model.When predicting the speed for the next 45 minutes,compared to the STGCN model,the MAE,RMSE,and MAPE errors of the DI-GCN(parallel)model decreased by 20.635%,22.771%,and 11.440%,respectively,performing better than the DI-GCN(stacked)model.The DI-GCN(stacked)model performed better in predicting the road network of small areas and the speed within 30 minutes in the administrative region.The DIGCN(parallel)model performed better in predicting the speed of the road network in the administrative region for the next 45 minutes. |
