The Research Of Traffic Safety Evaluation On Mainline Toll Plaza's Diverging Area Using Microcosmic Trajectory Data

As a main accident-prone location on highway,much attention has been paid to the safety of toll plaza,especially in the diverging area at upstream toll plaza.The limited road space,complex lane configuration,and different kinds of toll collection types could easily confuse drivers when they are driving at the diverging area,also increase the vehicles' crash risks.Although,the traffic management department has attached importance to the above problem and implement specific measures,for example,change the charging method,these measures were not amount to much for traffic safety improvement.In addition,the facilities and toll collection type update has also led to the failure of the existing safety evaluation system,so it is necessary to fix the traditional model and make the safety warning system of toll plaza cloud remain valid in the rapidly updated traffic environment.Therefore,it is particularly important to effectively and reasonably evaluate the safety of vehicles in the toll plaza diverging area,clarify the influence mechanism of vehicle crash risk,and establish a widely applicable safety evaluation system of this area.To this end,the present thesis founded by the National Natural Science Foundation of China(No.51778141)and the Fundamental Research Funds for the Central Universities and the Postgraduate Research & Practice Innovation Program of Jiangsu Province(KYCX17-0148),took the toll plaza diverging areas in Nanjing city as examples,explored the vehicle's crash risk characteristics and influence mechanism,and improved the traffic safety of toll plaza diverging area.The main research contents of this paper include the following five aspects:Firstly,the updated definition of mainline toll plaza diverging area was proposed,the research scope of vehicle diverging behavior at toll plaza was extended to the main lane where vehicle usually take preliminary diverging behavior.Moreover,the driving process,lane changing and speed characteristics of vehicles in the diverging area are summarized from the theoretical perspective.In addition,this paper introduced an automated video data analysis system from the aspects of system framework and function,object detection,object tracking,error elimination and coordinate system transformation,and analyzed how to obtain the complete vehicle trajectory and improve the accuracy in detail.On this basis,took the east diverging area of Hu-Rong expressway Nanjing toll station as an example,explored the traffic flow characteristics from the aspects of vehicle type,driving time,driving speed,speed change and lane choice etc.Secondly,the traffic conflict technology and the safety surrogate measures were expounded in detail,and the limitations of the traditional conflict proximal surrogate indicators were expounded emphatically.To eliminate limitation of conventional Time-To-Collision(TTC),this study proposed an extended TTC(ETTC)is to evaluate the traffic risk in toll plaza diverging area,which can be utilized for the more general case of two-dimensional movements.Furthermore,the unconstrained feature of vehicle movement in the toll plaza diverging area was analyzed and confirmed,the traffic conflict in this area was also defined and classified,and the process and influencing factors of traffic conflict were introduced in detail.Based on the microcosmic vehicle trajectory data and ETTC,the traffic conflicts of vehicles in the toll plaza diverging area were extracted and the safety status was distinguished.Subsequently,the characteristics of traffic conflicts were described and the differences of vehicle crash risks were compared.Thirdly,this study develops the logistic regression(LR)model and five typical nonparametric models including,K-Nearest Neighbor(KNN),Artificial Neural Networks(ANN),Support Vector Machines(SVM),Decision Trees(DT),and Random Forest(RF)to examine the relationship between influencing factors and vehicle collision risk.The model accuracy and results of these models were compared,and the best fit model was selected for traffic safety evaluation in the toll plaza diverging area.Subsequently,the random parameter logistic(RPLR)model based on Bayesian approach was developed to investigate the effects of explanatory factors on the collision risk of vehicles diverging or adjusting their lane position,which can get rid of the IIA hypothesis constraint and take the heterogeneity across observations into consideration.Based on the model results,this study explored the relationship between vehicle type,toll collection lane choice,driving speed,driving position,traffic flow and vehicle safety.The research results could provide valuable information for accurate assessment of collision risk,and provide a theoretical basis and empirical guidance for traffic safety managers.Fourthly,by exploring the continuous dynamic changes of vehicle diverging behavior,and analyzed the spatial-temporal dynamic differences of vehicle collision risk in the toll plaza diverging area,this study emphasized the importance of considering the travel time/distance varying influence on vehicle collision risk.Based on the RPLR model,time-varying mixed logit model and distance-varying mixed logit model were developed to examine the time/distance varying effects of influencing factors on vehicle collision risk.The model was validated in model performance comparison of six models.In addition,this study further explored the effects of mixed traffic on vehicle collision risk in the toll plaza diverging area,vehicles were divided into four mixed traffic categories and their time-varying influence mechanism of traffic safety were compared.Based on the different effects of them on vehicle collision risk,the adverse impacts of mixed traffic on the safety of toll plaza diverging area was confirmed.Finally,based on the discussion of the limitations of offline static estimation in the safety management practice and for the dynamic update data analysis,this study simulated the discrete characteristics of data by six kinds of data sampling approaches.Subsequently,the Bayesian dynamic logistic regression model was employed to estimate the vehicle collision risk in the toll plaza diverging area,which could implement the adaptive fix with the data update,the effectiveness of model adaptive modification and dynamic estimation were also validated by model performance comparison.In addition,the grayscale clustering evaluation method was adopted to build a pre-classification model of vehicle safety in the toll plaza diverging area.The vehicle was divided into four safety grades,and the ETTC threshold and crash probability threshold of the four grades were also optimized.Based on the above efforts,a vehicle safety warning system with dynamic update and subdivision of vehicle safety levels for the toll plaza diverging area was constructed,which can be applied into safety management directly.