Urban Traffic Flow Prediction Method Based On Multi-Graph Neural Network

In the process of urban development,the problem of traffic congestion becomes more and more serious because of the rapid growth of cars and the incomplete transportation infrastructure.Intelligent transportation system(ITS)is an effective means to solve the problem of traffic congestion.Traffic flow prediction,as a key problem in ITS,becomes a hot research domain in the field of transportation.Accurate and real-time traffic flow prediction information can help drivers to make route planning,help traffic managers improve traffic efficiency and safety,and then avoid or alleviate traffic congestion.This paper studies and analyzes the research status and relevant knowledge of traffic flow prediction.Based on the analysis of the shortcomings of the existing research methods and the complexity of traffic flow data,the traffic flow prediction problem is studied.The main contributions are as follows:First,this paper proposes a traffic flow prediction method that combines spatiotemporal correlation with social events.On the one hand,for the non-European spatial correlation of the road network and the temporal correlation of the time series data of traffic flow,a feature extraction sub-network that combines graph convolutional neural network and recurrent neural network is designed.On the other hand,in view of the potential impact of social events on traffic flow,a social event feature extraction sub-network based on convolutional neural networks is designed.Finally,a temporal and spatial correlation feature extraction sub-network and a social event feature extraction sub-network are combined to implement a traffic flow prediction model.Secondly,this paper proposes a traffic flow prediction method based on Multi-graph Convolution.Through further research on traffic flow data,it is found that in addition to the spatial associations generated by the road network topology,there are many other forms of semantic spatial associations(such as city function zoning,historical data performance,etc.).Therefore,graph convolutional networks are used to learn potential semantic features from multi-source associations,and the topology of the road network is combined to optimize the previously proposed model to achieve better performance.Thirdly,experimental evaluation.A series of experiments were performed on the proposed method on two real datasets.First of all,the method proposed in this paper is compared with eight baseline methods at different time intervals.The experimental results strongly confirm the performance of the method proposed in this paper.Compared with the best baseline method,it reduces the prediction error by about 3% to 6%.Secondly,the internal elements of the method and the performance of the method in different scenarios are studied in order to analyze the method proposed in this article more comprehensively.