Long Term Prediction on Urban Traffic Flow Based on Multi-source Spatio-temporal Graph Convolutional Neural Network Model
Traffic flow prediction is an important part of intelligent transport system.Accurate,timely and effective prediction information are of great significance for urban traffic control and guidance.However,due to the fact that urban road network traffic flow is affected by various external factors such as land use properties and weather changes,the traffic flow prediction faces enormous challenges.In order to predict the traffic flow of urban road networks effectively,multi-source spatio-temporal graph convolutional neural network model is proposed.The external factors affecting urban traffic flow are divided into static and dynamic categories.A clear and structured classifiication basis to understand the various external factors affecting traffic flow is provided.Then,the static factors are coded into multiple graphs,specifically distance matrix,functional similarity matrix,and connectivity matrix.The three-channel input spatio-temporal correlation modeling module is composed of multiple graphs.The graph convolution network is used to model the spatial correlation of traffic flow,learning node features,and adjacency information.The door control cycle single element is used to model the temporal correlation of traffic flow for capturing dynamic changes and periodic patterns of traffic flow data.Finally,the fusion layer is used to fuse the multi-channel output with dynamic factors as the final predicted output.In order to verify the model effectiveness,SZ-TAXI dataset is used for comparing with 7 benchmark models.The result shows that the multi-source spatio-temporal graph convolutional neural network model integrated multiple external factors achieves optimal performance than benchmark model in the evaluation indicators of MAE and RMSE.The ablation experiment shows that the multi-source method for handling static factors and the method for integrating dynamic factors effectively improve the long-term prediction performance of urban road network traffic flow.
NETL
NSTL
万方数据
