Research On Traffic Flow Prediction Of Urban Road Network Based On Spatio-Temporal Convolutional Network

With the rapid growth of society and economy,the stock of vehicles is increasing rapidly.As a comprehensive application of advanced science and technology,the intelligent transportation system covers the application of Internet of Things technology and big data analysis.It comprehensively collects,analyzes,analyzes,predicts,and makes decisions about traffic road network conditions.The intelligent transportation system has improved the management and control level of the traffic management department.It also provided convenience information for people to travel healthily.However,many current traffic condition forecasts are not very accurate.The main reason is that the traffic flow prediction as a basic technology is not accurate enough.The traffic flow prediction methods currently applied are mainly based on the historical time series of traffic flow to make predictions.This method ignores a lot of information and leads to inaccurate prediction results.Some methods based on deep learning consider the spatiotemporal correlation of traffic flow,but they do not consider the influence of other external factors,resulting in inaccurate prediction results.In recent years,after the research of graph convolutional networks has become hot.How to use deep learning models to make more accurate predictions of traffic flows in urban road networks has attracted many scholars’ research and attention.We model and analyze the traffic flow of the urban road network to study and predict the traffic flow.We propose a Spatio-Temporal Dynamic Graph Convolutional Network that considers external factors to predict the average speed of traffic flow in urban road networks.The traffic flow in the urban road network is temporal dynamics and spatially correlated.In this paper,the road network in the urban area is regarded as a graph structure.The multi-loop sensors on the road are the nodes of the graph.As edges,the data collected by the sensors is considered as the attributes of the nodes.This paper comprehensively considers the spatio-temporal correlation characteristics of traffic flow.It models the traffic flow of the urban road network as a random walk on the graph.The graph convolutional network is used to obtain the spatial correlation characteristics.On the other hand,the traffic flow information at each moment is regarded as a graph data,and the historical data of each node is regarded as a time series.The time dynamics of traffic flow data can be handled by GRU network and encoder-decoder structure.In addition,the attention vector is used as a competitive choice on the encoder-decoder structure to measure the similarity between the historical state and the predicted state.This paper also considers the influence of external factors on the traffic flow in order to model the uncertainty characteristics of the traffic flow as much as possible.This article adds three external factors: time period,weekday mark,and traffic accident mark.Through the external factors,focus more on the significant correlation between different nodes.Experiments show that external factors have a positive effect on traffic flow prediction.Experimental verification on two real-world urban road network datasets shows that the proposed model in this paper can improve the average speed prediction accuracy of urban road network traffic flow,and multi-step The accuracy of long prediction tasks has good results.