Research On Traffic Flow Detection And Prediction Scheme For Urban Road Networks

As urbanization continues to accelerate,vehicle ownership explodes in recent years,and a series of traffic-related problems such as traffic congestion and environmental pollution have emerged.Intelligent Transportation Systems(ITS)is currently regarded as an important way to improve road traffic management and urban traffic conditions around the world.The acquisition of traffic flow data and the prediction of future traffic situations are two important parts of ITS.On the one hand,the acquisition of traffic flow data is the basis for building it,and traffic flow detection technology research is of great significance.On the other hand,traffic flow changes rapidly and swiftly,and traffic flow prediction technology can sense the future traffic status of urban road networks in advance,which is an important way to realize traffic guidance and path planning and alleviate traffic congestion.This thesis aims to propose a lane-level traffic flow detection scheme that is cost-effective and suitable for largescale deployment.And based on this,traffic flow prediction of urban road networks is implemented,thus contributing to the construction of ITS.Therefore,this thesis is divided into the following two parts for the above.First,the existing traffic flow detection schemes rarely focus on the acquisition of lane-level traffic flow information,making it difficult to obtain richer road information.These schemes generally have problems such as complex systems and high costs.Aiming at this problem,this thesis proposes a lane-level traffic flow detection scheme based on highly economical microwave radar sensors.This scheme acquires the raw signals of vehicles passing by microwave radar sensors and achieves the extraction of vehicle motion features through data pre-processing algorithms.The scheme designs lane discrimination,traffic flow detection,vehicle speed detection,and adaptive threshold adjustment algorithms respectively to achieve lane-level traffic flow detection.The experimental results show that the scheme can realize lane-level traffic flow and vehicle speed detection,has the advantages of low costeffectiveness,and is suitable for large-scale deployment,meeting the needs of large-scale deployment in urban ITS scenarios.Second,due to the complex spatiotemporal correlation of traffic flow and the susceptibility to external factors,it is impossible to optimize the traffic of urban road networks by using historical data only.Therefore,this thesis investigates the traffic flow prediction problem of urban road networks based on the implementation of traffic flow detection and proposes an Attention-based Feature Fusion Temporal Graph Convolutional Networks(AFFT-GCN)traffic flow prediction model.The model achieves comprehensive acquisition of spatial and temporal correlation of traffic flow data by Graph Convolutional Network(GCN),Gated Recurrent Unit(GRU),and Attention Model.A feature fusion unit is designed in the model to achieve the fusion of external factors.The simulation experiments prove that the model can effectively improve the performance of traffic flow prediction.