Research On Traffic Road Network Modeling Based On Graph Neural Network

At present,the problems of traffic flow control and traffic congestion control accompanying the development of road traffic have become the key problems in the operation and development of large and medium-sized cities in China,and they are also the hot research direction in the field of traffic and transportation.Urban road traffic has shown network development,and there is often a large gap between the analysis of a single road section and the actual situation.Therefore,the research of traffic network should take the characteristics of traffic flow on each road section and the structural relationship between road sections as the starting point.The continuous development of deep learning theory,combined with traffic flow theory and graph topology theory,has brought new research ideas to the modeling of traffic network.This paper will take the graph neural network modeling of traffic network as the goal,combine the graph structure data of road network traffic flow with graph neural network theory,and realize the learning of the latent characteristics of traffic network,so as to effectively complete the two downstream tasks of road network congestion estimation and road network structure prediction,and provide reference and basis for practical tasks such as road traffic network operation state evaluation,management and control,construction and planning.Firstly,starting from the research background and significance,this paper summarizes and analyzes the relevant research status at home and abroad from three aspects: the research of road traffic network,the research of graph structure data based on deep learning theory and the research of deep learning method of generative model,combs out the research ideas for traffic network modeling,and determines the research content of this paper based on the research goal and technical route.Secondly,this paper expounds the graph structure data construction method of road network traffic flow,and preprocesses the collected original data to meet the needs of modeling.At the same time,in order to obtain the distribution of road network traffic flow data,this paper analyzes the characteristics of road network traffic flow data,The analysis results provide data support for the subsequent traffic network modeling,and provide a theoretical basis for the analysis of the example verification results of the modeling.Then,this paper defines the road network congestion estimation problem,applies the graph neural network to the graph structure data of the traffic network,establishes the Multi-layer Graph Convolutional Neural Network Model(Multi-GCN)as the feature extractor,clarifies the road network traffic flow feature extraction method,and extracts the features by using the nonlinear dimension reduction capability of the neural network model,so as to estimate the congestion of the road network according to the highdimensional traffic flow characteristic parameters.Through experiments,on the one hand,the accurate estimation of the congestion of the road network in different traffic periods is realized,on the other hand,the effectiveness of the model is verified,and as an integral part of the subsequent modeling,for which it provides a research basis.Finally,this paper defines the problem of road network structure prediction,combines the above research results with the Auto-Encoder architecture,and establishes a Graph AutoEncoder based on Multi-layer Graph Convolution Model(Multi-GCN-GAE).The model can learn the latent characteristic relationship between the features and structure of traffic network;Further,combined with the inference generative theory,a Variational Graph Auto-Encoder based on Multi-layer Graph Convolution Model(Multi-GCN-VGAE)is established.The model can sample new feature relationship samples from this latent characteristic relationship and generate a new road network structure on the basis of the original road network structure.Through experiments,the above model can well estimate the road network topology structure through the traffic flow parameter characteristics of the road network.