Research On Traffic Flow Prediction Based On Graph Convolutional Neural Network

With the rapid development of technology and the great progress of society,modern cities are developing in the direction of smart cities,and intelligent transportation is undoubtedly an integral part of smart cities.An efficient Intelligent Transportation System(ITS)can bring huge economic benefits to the cities,greatly reduce travel costs for people,and highly enhance the well-being of residents.The study of traffic flow prediction is an insurmountable step in the realization of ITS.An advanced and effective traffic flow prediction model can accurately predict the traffic flow trend in the future period,thus helping people to better plan their travel routes and providing important references for traffic resources.The traffic flow prediction problem is a complex spatio-temporal data prediction problem.The prediction accuracy of current mainstream traffic flow prediction models based on mathematical statistics and traditional machine learning is still low.To this end,this thesis explores traffic flow prediction models based on graph neural networks,especially graph convolutional neural networks,to further improve traffic flow prediction accuracy.Firstly,in order to solve the problem that existing traffic flow prediction models fail to fully exploit the spatial correlation between traffic flows,this thesis proposes ACE-GCRN,a traffic flow prediction model based on adaptive community enhanced graph convolutional recurrent neural network,which enhances the model’s ability to capture the spatial dependence of traffic flow by introducing a community division algorithm to mine the community structure of road networks.In addition,the model’s prediction accuracy of traffic flow is enhanced by constructing an adaptive graph convolutional network module and fully study of the feature embedding of each node in the road network and closely fit the downstream prediction task.Experiments show that the model outperforms other benchmark models in the short-time traffic flow prediction task.Secondly,considering the influence of weather on traffic flow and the unique periodic characteristics of traffic flow,this thesis proposes a model named W-PAGCRN(Periodic Adaptive Graph Convolutional Recurrent Network with Weather),which explores the intrinsic correlation between weather and traffic flow changes by introducing the corresponding historical weather data of the traffic flow dataset collection area.In addition,the model’s ability to capture the temporal correlation of traffic flow is enhanced by additionally constructing a module to capture the daily and weekly cycle characteristics of traffic flow to fully exploit the cycle characteristics of traffic flow.The experiments show that the prediction performance of the model achieves a large improvement on several traffic flow data sets.There are 31 figures,7 tables,and 69 citations in this thesis.