Research On Traffic Flow Imputation And Prediction Based On Spatiotemporal Pattern

With the deepening of urbanization,the contradiction between the growing number of motor vehicles and the slow speed of road construction has become increasingly prominent,and the number of motor vehicles exceeds the maximum carrying capacity of the road,causing traffic congestion and affecting the further development of the city.Intelligent transportation system is a cost-effective and efficient traffic management system that effectively alleviates traffic congestion by rationalizing urban roads and vehicles based on collected high-quality traffic data and accurate traffic flow prediction results.However,due to factors such as data acquisition equipment failure and network communication interruption,there are missing values in the collected data,or even incorrect traffic information,and the data quality is generally low.At the same time,the complex spatiotemporal patterns in low-quality traffic data and data further pose challenges to the accurate prediction of traffic conditions.This dissertation focuses on the above two issues.Aiming at the problem of missing data,this dissertation fills in the missing data based on cooperative adversarial learning,evaluates the credibility of the completed data,and provides complete traffic data and additional quality evaluation information for the next prediction task,which helps to improve the accuracy of the prediction results.At the same time,this dissertation uses the spatiotemporal convolution model to capture complex spatio-temporal patterns in traffic data to accurately predict future traffic conditions.The main research contents and contributions of this dissertation are as follows:(1)Reliable filling of missing data:In order to complete the data with high quality and test the credibility of the data,this dissertation designs a cooperative adversarial completion model,which consists of a generative model and an evaluation model,in which the generative model learns the spatiotemporal pattern of the real data to generate forged data with a similar spatiotemporal distribution to the real data,and is used to fill in the missing values,while the evaluation model tests the quality and reliability of the forged data and outputs the confidence of the data.In addition,the bidirectional cyclic spatiotemporal convolutional network is used to capture the spatio-temporal variation patterns of history and future in traffic data,which further improves the completion performance of the model.High-quality complete filling data and additional confidence assessment results can provide rich and reliable data support for traffic prediction tasks and help improve the accuracy of prediction results.Through a variety of experiments,the completion effect of the model is verified,and the improvement effect of the completion data and additional confidence information on the accuracy of the prediction results is tested.(2)Accurate prediction of traffic state:Aiming at the problem of traffic state prediction,in order to capture the complex spatiotemporal patterns between data and improve the prediction effect,this dissertation designs a spatiotemporal graph convolutional prediction model based on reliable traffic flow data,uses three adjacency matrices to describe different types of spatial connections between roads,and designs a spatiotemporal module to capture multi-level temporal dependence and dynamic spatial correlation.The spatiotemporal module consists of two parts,one is the multi-attention time convolution model,which combines the time convolution network with the multiattention mechanism to jointly extract the multi-level temporal dependence in the data and filter and filter relevant information,and the other part is adaptive gating spatial convolution,which uses adaptive matrix information and gating mechanism to capture the dynamically changing spatial correlation.Finally,the three types of spatiotemporal characteristics are effectively integrated to accurately predict the future traffic state.Finally,through a large number of experiments,the effectiveness of the prediction model and its spatiotemporal module is verified.