Traffic Flow Data Filling And Prediction Based On Deep Learning

With the rapid development of social economy,the number of cars is gradually rising,and traffic congestion is followed,which seriously affects the urban development and travel efficiency of travelers,and brings great pressure to the modern transportation system.As an important part of intelligent transportation system,traffic flow prediction plays an indispensable role in travel guidance and traffic congestion relief.However,accurate traffic flow prediction is often inseparable from complete data,and there is a problem of insufficient feature information extraction in current data filling research.There are some problems in traffic flow prediction research,such as insufficient spatio-temporal information mining and less consideration of long-term dependence.Therefore,a traffic flow data filling and prediction method based on deep learning is proposed in this paper.The specific work content is as follows:Condering that most existing generative adversarial neural network filling models have the problem of insufficient extraction of historical time features of traffic flow data,inspired by the self-attention mechanism,this paper builds a traffic data filling model based on SelfAttention Generative Adversarial Network(SA-GAN),which adaptively allocates the feature weights of traffic flow data,and effectively captures the hidden feature information in historical periods.Experiments on open data sets show that SA-GAN is superior to the existing advanced methods,which lays a good foundation for traffic flow prediction.Inspired by feature extraction and codec of existing graph data,a traffic flow prediction model based on Gate Recurrent Graph Attention network(STGGAT)is proposed to consider the spatiotemporal correlation and long-term dependence of traffic flow.A spatio-temporal feature extraction layer combining graph attention network and gated recurrent neural network was constructed to accurately extract the spatio-temporal correlation of traffic flow data.The codec component unit is constructed to achieve the purpose of single-step and multi-step prediction.The experimental results show that the proposed STGGAT model has lower mean absolute error,root mean square error and mean absolute percentage error than most existing models in single-step and multi-step prediction.Considering the traffic condition and the actual demand of planning management department,this paper designs the traffic flow forecasting system in detail and uses the idea of object-oriented programming to develop it,which has the functions of system management,data management and analysis and prediction.It can check the traffic flow prediction results visually,and realize the traffic flow prediction model and algorithm proposed in this paper.The research results of this paper show that the traffic flow data filling and prediction method proposed in this paper based on deep learning has achieved good results in traffic flow filling and prediction.It has carried out reasonable planning for travelers’ travel routes,improved travel efficiency and further alleviated traffic congestion,which has certain construction significance for the development of intelligent transportation industry.