Lane-level Traffic Flow Prediction Based on Dynamic Spatio-temporal Convolution Network
A lane-level traffic flow prediction model(DSTCNN)based on dynamic spatio-temporal con-volution network was proposed.By constructing the topological structure of the lane section node net-work,the dynamic graph convolution network was used to extract the traffic flow state of the adjacent lane section and the upstream and downstream lane sections where the lane section is predicted at the same time,and the spatial feature information of the predicted lane section and different lane sections was obtained.The extended causal convolution and gating mechanism were used to form a gated time convolution network to extract the time dynamic characteristics of traffic flow sequences.Through the fully connected(FC)network layer,the temporal and spatial characteristics of traffic flow were inte-grated to realize the prediction.The results show that DSTCNN model can better capture the tempo-ral and spatial characteristics of lane traffic flow synchronously,and the prediction accuracy is signifi-cantly improved compared with other classical models,so it has better prediction performance.
万方数据
维普
