Using Trajectory Data To Explore The Relationships Between Traffic Flow Operation Conditions And Traffic Safety On Expressway Merging Areas

As the aorta of urban traffic,expressways provide fast,safe and comfortable long-distance intra-city travel services.However,many expressways have frequent congestion,and the traffic flow in the congested sections is complicated,leading to high crash risks.Especially in the expressway merging area,where vehicles change their lanes frequently,making the traffic flow chracteristics complicated and associated with a large number of traffic conflicts,which brings serious crash risks.An accurate understanding of the relationship between the traffic flow dynamic chracteriscs and traffic safety in the expressway merging area can provide a basis for the development of real-time crash risk prediction models and traffic safety control strategies.Domestic studies on the impact of traffic flow on expressway merging areas usually use traffic flow parameters at the segment level.The discussion on the impact of macroscopic traffic flow status and microscopic driving characteristics on traffic safety is still not deep enough to fully capture the impact of dynamic traffic flow characteristics on traffic safety.To this end,this article takes the Nanjing Yingtian Street Expressway Merging Area as the research section,and uses drone aerial photography to extract highresolution n trajectory data to conduct a systematic and quantitative study on the relationship between dynamic traffic flow operation conditions and traffic conflicts analysis.First,the traffic state under the three-phase traffic flow framework is divided based on the macroscopic aggregated traffic flow parameter criteria,and the impact of the traffic state on the crash risk is analyzed.aggregate trajectory data in time and space dimensions,explore the method of transforming microscopic trajectory data into macroscopic traffic flow characteristics,and divide macroscopic traffic state under threephase theory framework based on the commonly used macroscopic criteria.The macroscopic criteria are mainly based on the correlation between the flow rate density and value of average vehicle speed.The traffic conflict technique is used to compare and analyze the crash risk under different traffic states.At the same time the effects of dangerous traffic flow parameters on the crash risks are statistically analyzed.The results show that the more congested the traffic state is,the higher the crash risk will be,and the introduction of traffic state can better use traffic flow chracteristics to describe the crash risk.Secondly,the phase change perception model based on trajectory data is used to divide the traffic state under the three-phase traffic flow framework,and the traffic safety in each traffic state is evaluated.The wavelet transform method is used to identify phase transition points and calculate model parameter values.Using the traffic conflict technique,accident risks in different traffic states under the three-phase traffic flow theory are compared and analyzed.The traffic flow characteristics under different traffic states are statistically explored,and the correlation between these traffic flow characteristics and the accident risks are calculated.By making accident risk hot spots diagram,the location characteristics of accident risk are studied.The study found that high accident risk is associated with more congested traffic conditions,and the introduction of traffic state can improve the fitness for crash risk regression.In addition,the high accident risk in the merging area is concentrated in the upstream of the lane drop location.Finally,Support Vector Machine is used to establish a short-term traffic safety prediction model based on trajectory data,which mainly includes two parts: using Support Vector Machine Classification Model to predict the accident risk possibility based on the traffic conflict severity;and using Support Vector Machine Regression Model to predict the energy-based consequence severity of latent crash caused by the serious traffic conflicts The possibility of the accident risk is whether there will be a serious conflict,which is determined by the comparasion with the value of severity threshold defined by the cumulative distribution function,and the consequence of the traffic conflict is evaluated by the destructive energy created by collision associated with serious traffic conflicts.Besides,different kernel functions and the impact of macroscopic traffic states,time series characteristics,and different time steps on prediction performance of both Support Vector Machine Classification model and Regression Model are analyzed.The results show that considering the traffic state and time series characteristics has a positive effect on the improvement of prediction performance,and the shorter the prediction time step is,the better the prediction performance will be.