Research On Vehicle Trajectory Prediction Method Based On Sectional Traffic Data In Super Long Tunnel

With the rapid development of China’s socio-economic and transportation undertakings,the number of the super long tunnels has been increasing as an important part of urban expressways and highways and other traffic arteries.As of 2021,China has built 7810 long tunnels,including 1599 super long tunnels and 6211 long tunnels.Due to the influence of various combinations of geometric lines,poor lighting,space closure and other factors,the driving environment of main line super long tunnels is more complex,especially the super long underwater tunnels,trunk tunnels in mountainous cities,highway tunnels in mountain area.At the same time,with the rising traffic flow of trunk tunnels,trucks,buses,cars and other types of vehicles,their operating characteristics are different,coupled with the long tunnels are prone to driving fatigue,traffic accidents,resulting in reduced efficiency,emergency rescue difficulties,may also lead to fire,secondary accidents and other mass casualties of serious accidents.Therefore,there is an urgent need to strengthen the long tunnel traffic operation refinement of real-time monitoring and early warning control,in order to enhance tunnel traffic efficiency,reduce safety risks.The vehicle trajectory monitoring and prediction is one of the key and basic conditions to realize the refined real-time monitoring and early warning control of tunnel traffic operation.This thesis takes trunk super long tunnels as a typical example.It takes advantage of side-mounted video to collect cross-sectional traffic data such as vehicle speed,vehicle passing time and headway time distance/spacing.Then,it analyzes the traffic operation characteristics and vehicle trajectory characteristics of the tunnel,and establishes vehicle operation rules and FVD（Full Velocity Difference Model）optimization model under different driving conditions.Finally,we propose the prediction model of vehicle trajectory based on real-time multi-sectional traffic data.The main contents of this thesis are as follows:（1）Analysis of environmental characteristics and traffic operation characteristics of super long tunnels.The general characteristics of traffic operation in super long tunnels are analyzed in terms of their environmental characteristics（visual characteristics,ambient light illumination,road conditions,noise pollution,air quality,weather conditions）and traffic operation characteristics（road conditions characteristics,vehicle movement characteristics,driver characteristics）.Focusing on trunk super long tunnels,the Jiaozhou Bay Tunnel in Qingdao,as an example,the traffic flow characteristics are analyzed based on the traffic flow,average speed,headway time distance and other parameters of each cross-section.（2）Vehicle driving state differentiation and operation rule analysis based on sectional traffic data.In this study,the distinction thresholds between collision danger zone,braking avoidance zone,following driving zone,and free driving zone are defined according to the cross-sectional traffic data such as speed and headway time distance/spacing.And the distance threshold is converted into the headway distance threshold to establish the Wiedemann model following domain based on headway time distance data.The vehicle running performance is divided by the headway data,and the vehicle running rules under the free driving condition,following running condition and emergency braking condition are analyzed to establish the vehicle operation rules under different driving conditions.（3）Parameter optimization of FVD following model.According to the cross-sectional traffic data with speed and headway/spacing,the correlation between vehicle speed and space headway is fitted,and a desired headway spacing fitting model is constructed to replace the safe headway spacing of the FVD tracking model.It obtains the optimal range of the time required for vehicles to reach the stable following state in the super long tunnel through the statistical analysis of the time required for vehicles to reach the stable following state,which is used as the criterion for optimizing the FVD following model by adopting the iterative idea,the speed and position coordinates of the rear vehicle at t+T₀ are iterated through the speed and acceleration of the rear vehicle at t+T₀.The numerical simulation is carried out based on the vehicle section data.The driver sensitivity coefficientαis adjusted according to the simulation effect to gain the improved FVD following model.（4）Build a whole-process vehicle trajectory prediction model based on real-time sectional traffic data for super long tunnels.By dividing the vehicle driving states based on the calibrated Wiedemann model following domain,the trajectory prediction equations under different driving states including free driving state,following driving state and emergency braking state are established,and finally the whole-process vehicle trajectory prediction model based on multi-sectional traffic data of long tunnels is constructed.In this thesis,a real-time vehicle trajectory prediction model with the whole proces is constructed by the limited cross-sectional traffic flow data of the super long tunnel,and the average accuracy of trajectory prediction is 95.45%when verified by the vehicle velocity and position coordinate data measured in the Jiaozhou Bay Undersea Tunnel in Qingdao.The results show that the vehicle trajectory prediction model with multiple cross-sectional traffic data can predict the vehicle trajectory in super long tunnels more accurately with limited cross-sectional traffic flow data,which has important theoretical and application values for the digital twin of traffic flow operation and real-time traffic operation monitoring and early warning control.