Research On Schemes Of Traffic Data Imputation And Prediction Based On Deep Learning

In recent years,following with continuous promotion of urbanization construction,the car ownership of urban and rural residents is increasing,and the traffic congestion problem is becoming more and more serious.As an important junction of urban roads,traffic congestion on arterial roads is particularly serious.The implementation of Intelligent Transportation Systems(ITS)can help alleviate this problem.ITS relies on abundant traffic data,but affected by the external environment and internal factors,the collected traffic data is usually incomplete.A large number of data missing leads to the deterioration of data quality and seriously affects traffic data mining.Therefore,the imputation of missing traffic data is one of the main research contents of ITS.In addition,as one of the keys to ensuring the accurate implementation of traffic guidance and control,the prediction of traffic information is also an indispensable part of ITS.Accurate traffic prediction can provide a reference for traffic guidance and traffic control technology,and then improve road capacity and alleviate traffic congestion.At present,there are still some shortcomings in the research on traffic data imputation and prediction.Based on the analysis of existing traffic data imputation and prediction schemes,this paper proposes a traffic data imputation scheme based on deep learning and a traffic prediction framework based on attention graph fusion,respectively.The main research work of this paper is as follows:(1)For the problem of the existing imputation schemes,such as the imputation precision of traffic data is not high in combinational missing pattern,or the imputation precision decreases due to the increase of data missing rate,this paper proposes a traffic data imputation scheme based on deep learning considering the combinational missing pattern.The scheme is mainly c omposed of seven network layers that can be effectively used for missing value imputation of traffic data.Experiments on a real arterial road traffic data set show that the scheme is superior to other baseline schemes in the imputation error,model stability,and imputation data fitting ability.(2)For the problem of the multi-step prediction error accumulation,insufficient mining of dynamic spatial features and representation of graph structure data for the spatial relationship,and the decrease of prediction accuracy of traffic information caused by the complex road network structure of arterial roads,this paper presents a traffic prediction framework based on attention graph fusion.This framework is composed of three modules: multiscale temporal feature fusion,multi-graph convolution,and dynamic spatialtemporal prediction,which can fully mine the temporal features and spatial information of traffic data.The comparative experiments and ablation experiments are carried out on a real arterial road traffic data set.The experimental results not only prove the superiority of the prediction framework but also verify the ability of each module to extract the spatial-temporal information of arterial road traffic data.