Study On Lane-level Traffic State Estimation Of Expressway Based On Intelligent Connected Vehicle Data

Urban expressways are fast traffic arteries that carry large volume of traffic in cities and undertake most of the daily commuting traffic tasks in cities.With the development of China’s economic society and urbanization,the contradiction between traffic supply and demand becomes prominent,and the traffic congestion of freeways arises at the historic moment.Fine traffic management is an effective means to solve this problem,and high-precision road traffic status information is a prerequisite for achieving refined management,so it is of great significance to study lane-level traffic state estimation of urban freeways.With the development of communication perception technology and algorithm,intelligent connected vehicles(ICVs)have entered a mature commercial stage.An intelligent connected vehicle(ICV)is capable of collecting microscopic trajectories of itself and ambient vehicles using on-board sensors(IMU/GPS,lidar,radar,camera),which provides a new data source for traffic state estimation.It is foreseeable that the proportion of ICVs on the road will gradually increase in the future.So based on the intelligent connected vehicle data obtained by simulation,two methods of estimating the traffic state at the lane level are proposed: 1)Based on the data assimilation framework of extended Kalman filter,combined with intelligent networked vehicle data,fixed detector data and probe vehicle data,a multi-lane macro traffic flow model and measurement model are established,respectively to fit their different characteristics,and the lane-level traffic state estimation is realized.2)Based on graph neural network,the studied road is divided into spatial cells.The traffic operation between the cells is represented by the topology graph.And the intelligent connected vehicle data is substituted into the spatio-temporal graph convolutional neural network to achieve lane-level traffic state estimation.Finally,the simulation intelligent connected vehicle data based on NGSIM dataset is used to verify the estimation effect of the above two methods.The results demonstrate that by utilizing only 3-5% ICVs in the mixed traffic,the two proposed methodology could produce accurate estimate of lane-level traffic speed and density,and they perform better in terms of speed estimation than density estimation.In addition,the estimation accuracy improves with the growth of ICVs’ penetration rate in the mixed traffic.