Research Of Internet Of Vehicles Traffic Flow Prediction Model Based On Deep Learning

The problem of road traffic congestion brings a lot of hidden dangers to people’s travel,and also becomes a factor that restricting the development of the city.The prediction of traffic flow can make the traffic management department guide the traffic in time according to the prediction results,so as to allocate the traffic resources reasonably and improve the traffic environment.The traffic flow data In this paper is taken from Nanming District,Guiyang City,Guizhou Province,so the paper takes this area as the research object.Firstly,the basic analysis is carried out based on the electronic map data and traffic flow data of the Nanming District road network,the traffic flow data is denoised according to the wavelet threshold denoising theory.At the same time,the relevant research on the regional road network structure and the spatial distribution of the important roads is carried out for the future optimization of traffic network structure and the formulation of traffic guidance strategy.Then the paper did some research on the construction of traffic flow prediction model of Nanming road network based on the deep learning algorithm.At present,the deep learning framework of traffic flow prediction is divided into two kinds:decentralized modeling and centralized modeling.The efficiency of centralized model is more higher,which is suitable for large-scale network traffic flow prediction scenarios,but the prediction accuracy is not as good as decentralized model.Therefore,this paper extends the research point from decentralized modeling to centralized modeling,and deep learning algorithms RNN,LSTM and GRU are selected as the main tools to build the model.The results shows that the prediction accuracy of the three algorithms is similar in the decentralized modeling scenario,but in the centralized modeling scenario,there is a certain gap between RNN and the other two algorithms in prediction accuracy;the prediction accuracy of the decentralized model is higher than the centralized model,but the decentralized model needs to train the data for each road section,which greatly reduces the training and prediction effect of the model.The centralized model,by contrast,is more efficient which trains all data of the road network during a training session.So in order to improve the prediction accuracy of the centralized model,this paper proposes a centralized model based on attention mechanism,which can help the model focus attention on more important and effective information for the prediction target in the prediction process.In this paper,the working principle of attention mechanism is described in detail,the core weight matrix is visualized,and the architecture and operation mechanism of the proposed model are fully explained.Then,the improvement of the accuracy of the proposed scheme is verified by experiments.The research results provide a reference for the construction of traffic flow prediction model for the road network.