Research On Traffic Risk Prediction Method In Highway Diversion Area Based On Vehicle Trajectory

With the vigorous development of expressway infrastructure in my country,the total mileage of expressways across the country has been increasing,and the network has become increasingly perfect.As an important part of the expressway,the operation state of the diversion area also directly affects the operation efficiency of the expressway network.In the diversion area,there are frequent lane changes in the traffic flow,resulting in disordered operation of the traffic flow and frequent traffic conflicts,making the diversion area a major accident black spot on the expressway.Therefore,it is particularly important to effectively and reasonably evaluate and predict the driving safety of vehicles in the diversion area,clarify the impact mechanism of driving risks,and strengthen the management and control of highway driving safety.In order to improve the traffic safety in the diversion area,the paper takes the high-speed diversion area of Nanjing Airport as an example to effectively predict the vehicle driving risk through the operation status of the traffic flow,and analyze and interpret the impact mechanism.The vehicle trajectory is extracted from the video data of the aerial photography of the drone,and the operation state of the traffic flow in the diversion area is obtained by statistical analysis,and the safety surrogate index(Extended Time-to-Collision,ETTC)is calculated to determine the safety state of the vehicle;consider the heterogeneity and time period instability,build different random parameter Logit models,compare the goodness of fit to select the optimal model,and identify and analyze the significant influencing factors of driving risk.Construct and optimize non-parametric machine learning models(decision tree(DT),random forest(RF),Adaboost,Gradient Boosting Decision Tree(GBDT),XGBoost)for risk identification and effective prediction of traffic safety in the diversion area;finally,combined with the prediction and analysis results,the optimization measures and suggestions of traffic safety management and control in the diversion area are put forward.The results of the case analysis show that there are differences in the operation status of the traffic flow in the morning and noon in the diversion area.The vehicle speed is higher and the traffic flow is larger in the morning,and the impact mechanism of the driving risk is more complicated.The traffic conflict between vehicles in the diversion area is persistent.The mean ETTC of the two-period hazard samples are 2.91 s and 2.67 s,respectively,and the TET(Time Exposed Time to Collison)are 1.06 s and 0.91 s,respectively.Vehicle-traffic clashes are more severe during midday,but the danger duration is shorter.The intensity of traffic conflicts at the front and end of the diversion area is relatively high,and the traffic conflicts of vehicles traveling on the deceleration lane at the end are the most serious.In addition,the random parameters logit model with heterogeneity in mean and variances has the best fitting performance,which can not only reflect unobserved heterogeneity,but also identify more significant influencing factors.The results show that the driving risks in the morning and noon are significantly different.The average vehicle speed,the average speed of the vehicle in the diversion area and the headway are the key factors for the analysis of vehicle driving safety in the diversion area.The standard deviation of acceleration and the average vehicle acceleration in the split area can be considered as secondary key factors.The tree-based machine learning algorithm has high accuracy and good stability for the prediction of driving risk in the diversion area.The average prediction accuracy of random forest is the highest,reaching 96.56%,and it has good generalization ability.At the same time,the paper combines the research results with engineering practice,and proposes management and control optimization measures and suggestions for vehicles and safety facilities,which provides effective method support and reference for highway safety management.