Research On Traffic Flow Prediction Model Of Urban Area Based On Traffic Bayonet Data

In many cities,the construction of urban transportation infrastructure cannot meet the increasing traffic demand,and the growth in motor vehicles is much faster than the construction progress of transportation facilities,leading to an imbalance between urban traffic supply and demand.Traffic congestion has become one of the important reasons for hindering urban development.To deal with urban traffic congestion problems,it is important to accurately determine the traffic status based on the realtime traffic information from the road network and accordingly take scientific and reasonable control measures.Therefore,it has been the focus of the intelligent transportation research field about how to achieve real-time traffic flow prediction and identify the traffic status,and eventually provide effective data support for traffic control.With the development of traffic electronic equipment and image processing technology,traffic bayonet data can accurately identify the information of each motor vehicle,calculate the traffic flow,and has the advantages of easy maintenance and strong applicability.It has become an important data source for urban intelligent transportation and can be widely used in traffic flow prediction and traffic status identification.However,the existing methods for traffic flow prediction based on traffic bayonet data only consider simple road scenes without the spatial correlation characteristics in the road network.It only applies to a single bayonet prediction and cannot be used for urban areas.It is therefore urgent to carry out comprehensive experimental research on the analysis of spatial-temporal characteristics and improvement of network structure.In this thesis,based on the traffic bayonet data in the urban road network,a data model of the traffic bayonet is studied and the spatial-temporal distribution characteristics of traffic data are analyzed.Considering the characteristics of different bayonets,an improved model for traffic flow prediction is constructed,which realizes the simultaneous prediction of multiple bayonets in urban areas and achieves good accuracy.With the help of traffic state recognition methods,the predicted traffic flow data is converted into the traffic state,realizing the prediction of traffic congestion,and providing effective data support for dynamic road management and scheduling.The main research work and achievements of this paper are as follows:(1)Spatial and temporal characteristics analysis of traffic bayonet data.In this thesis,different traffic bayonet data organization plans and preprocessing methods are analyzed.The data is cyclical,trended and continuous in the time dimension,while correlated in the spatial dimension.It is found that there is a one-day versus one-week cycle characteristic of in the time dimension,and the traffic flow changes on the road network downstream in the adjacent space are similar,while the impact on traffic flow caused by events such as traffic congestion is also spatially propagated.(2)Construction of prediction models for traffic flow in the urban area.In this thesis,the influencing factors of the traffic flow are analyzed to address the problem that the existing traffic flow prediction model can be only applied to a single road condition scenario.Considering the spatial-temporal correlation characteristics of different bayonets,a deep learning model for simultaneous traffic flow prediction at multiple chokepoints is constructed,which is composed of Conv LSTM,Bi LSTM,and attention modules.The model can learn the common features of traffic flow in the urban area and realize efficient traffic flow prediction with a single model for multiple bayonets.(3)Research on traffic flow prediction in the area.Aiming at Nanhai District of Foshan City whose congestion is more serious,a training set of traffic flow is constructed with the characteristics of temporal periodicity and spatial correlation of traffic flow.The proposed method selects a prediction loss function and the dropout mechanism is applied to mitigate the overfitting phenomenon.It realizes the dynamic real-time prediction of traffic flow at multiple bayonets in urban areas.Through the analysis of MAPE error,it is shown that the proposed traffic flow forecasting model has good prediction accuracy,which can not only realize the synchronization and accurate prediction of real-time data,but also reveal the spatialtemporal correlation of traffic flow.(4)Traffic state recognition based on traffic flow variabilityCombined with the theoretical methods of traffic flow variation degree calculation,the research of traffic status identification method is proposed.Based on the similarity and difference characteristics of traffic flow at bayonets,the proposed method realizes the traffic status prediction.A comparative experiment is carried out with the real-time traffic data of Gaode,and the results show that the traffic congestion recognition is consistent with Gaode,which can provide an application way for high-precision traffic flow prediction.