Research On Spatiotemporal Series Modeling And Prediction Technology For Traffic Dat

Traffic forecasting is a key part of an advanced ITS,which is essential for traffic planning,traffic management and traffic control.With accurate forecasting,it not only provides a scientific basis for traffic managers when avoiding traffic congestion and limiting vehicle flow,but also provides efficiency for travelers in making travel plans and choosing travel routes.At this stage,there are three main challenges that traffic forecasting still faces: First,it is a challenge to model the dynamically spatial-temporal correlations from traffic data.Second,it is a challenge to describe the spatial structure dynamically.Third,it is a challenge to reduce the transfer of errors at long-term traffic forecasting.To deal with the above problems,we propose three spatial-temporal series-level traffic prediction models for spatial-temporal data processing on this paper,where the superiority of the proposed models is verified on real datasets.The main contributions of this paper are as follows:(1)Addressing the first problem,we propose a novel short-term traffic prediction model based on a multi-view attention mechanism(MH-GCN).It adopts an attentionbased encoder-decoder structure.Firstly,MH-GCN uses a spatial-temporal attention mechanism in encoder to model dynamic spatial and nonlinear temporal correlations.Then,the decoder utilizes Convolution Group,Pooling Group,and Dilation Group to extract different hierarchical of characteristics from the already modeled features,and then the fused results are used for predicting future traffic conditions.Experiments on two real datasets demonstrate that the proposed MH-GCN obtains improvements over the state-of-the-art baselines at short-term.(2)Regarding the second problem,we propose a novel dynamic graph learningbased spatial-temporal data modification and prediction model(DSGCN).To equip the graph convolution network with a flexible and practical graph structure,DSGCN designs a graph learning module to model the varying relations among nodes in the road network.In particular,we first provide an inception structure cooperated with the dilated convolution to capture the temporal dependencies.Second,a dual-stage graph convolution module is proposed to capture the complex spatial dependencies.Experiments results demonstrate that DSGCN outperforms the state-of-the-art baselines,especially for long-term traffic prediction.(3)Aiming at the third problem,we propose a novel long-and short-term forecasting model based on multi-scale convolutional networks(MSCN),an end-to-end solution to solve traffic forecasting problem.MSCN first employs an encoder with spatial-temporal attention mechanism to model both spatial and temporal correlations.Then,the decoder utilizes multiple convolutions and form groups of units to extract spatial-temporal features for different resolutions.In particular,we propose a soft thresholding fusion mechanism to adaptively adjust the transfer of spatial and temporal correlations.Experiments on real traffic datasets demonstrate that the proposed MSCN obtains improvements over the state-of-the-art baselines.