Prediction And Application Of Regional Traffic Flow State Of Expressway In Shaanxi Province Based On Graph Convolutional Network Model

In recent years,the accelerated urbanization has intensified the conflict between traffic supply and demand.Using spatial-temporal data in the region to predict future traffic status accurately and in real time,effective monitoring and analysis of regional characteristics can provide a convenient and effective reference for traffic travel,road planning and management.However,road networks are intricate and complex,and traffic distribution is spatially non-directional and irregular in structure.The state of traffic flow within the road network also shows obvious fluctuation and cycle characteristics along with time changes.Accurate and rapid prediction of regional traffic status becomes an important research topic.In this paper,we first obtain four months of flow data from 186 toll stations covering 3000 km of highways in Shaanxi Province.In order to estimate the traffic information of any cross-section of highways without vehicle detectors installed,a method of toll data imputation flow is designed.This can be useful for obtaining traffic information on freeways without installing any additional regular maintenance equipment.At the same time,an innovative fusion model is proposed to predict the traffic volume,which can realize the traffic analysis of toll booths.The main research and results of this paper are as follows:(1)For the problem of dense noise during peak hours of the obtained traffic data,the dense point distribution is reduced by autocorrelation analysis method and decomposition noise reduction.The set of methods can quickly determine the location of dense points.For the traffic data containing missing values,the tensor decomposition method is used to correlate the coupled location information.This framework can effectively deal with the missing problem in spatial-temporal data.(2)A fused temporal convolutional-temporal graph convolutional network(TCN-STGCN)is proposed and the above pre-processed traffic data are combined with the spatial convolution property of spectral domain graph convolution to extract the site spatial structure.The temporal information of sites is stored using the features of deep learning methods such as temporal convolutional network(TCN)with large expansion field and short training period.The advantages of these two models are combined to improve the network structure.The accuracy of predicting regional traffic flow is obtained to be higher and the training time is shorter compared with existing common deep learning models.At the same time,this paper expands from single-point prediction to regional prediction,and performs parallel calculations on the traffic status of the overall road network,matching the existing toll station points,which can achieve more accurate and efficient prediction and management.(3)Taking the application of improving toll station capacity as an example,the calculation of different lane queuing models is derived using the above predicted medium and long-time traffic.At the same time,it simulates the traffic conditions of duplex lanes,and transforms the toll lanes into duplex lanes to improve the capacity to a certain extent.The method can provide theoretical and engineering practice reference for high-speed toll collection departments,and also provides a possibility for intelligent traffic control.