Research On Dangerous Driving Behavior Identification Method At Entrance And Exit Of Highway Tunnel

As an important node of the highway network,the tunnel section greatly improves the efficiency of road traffic and brings great convenience to residents ’ travel.The tunnel entrance and exit sections are the high accident-prone locations of the tunnel.The existing research on driving safety at tunnel entrances and exits mainly focuses on the physiological changes of drivers ’ hearts and the changes of driving speed caused by the changes of driving environment.Few studies have studied the safety of drivers ’ driving behavior from the perspective of vehicle trajectory.Therefore,it is necessary to analyze the vehicle driving characteristics from the short-distance vehicle trajectory data,deeply and comprehensively grasp the vehicle driving mechanism of the tunnel entrance and exit section,propose the dangerous driving behavior spectrum applicable to the driving environment of the tunnel entrance and exit section,and construct the dangerous driving behavior recognition model,so as to realize the active prevention and control of traffic accidents on the tunnel entrance and exit section,and further combine with the intelligent information release equipment to warn dangerous drivers and ensure the safety of driving on the tunnel entrance and exit section.The study collected the original vehicle trajectory within 250 m of the entrance/exit section of the tunnel,processed the vehicle trajectory data to obtain the trajectory parameter set.Based on the trajectory parameter set,the variation characteristics and distribution rules of vehicle speed,acceleration,lateral offset value and heading angle are analyzed,and four main dangerous driving behavior types under the driving environment of entrance and exit section are determined.The risk measurement method is selected to quantify the severity of four kinds of dangerous driving behaviors.The threshold value is determined by the quartile difference method to determine whether the driver has some kind of dangerous driving behavior.Combined with the four driving behavior weights calculated by the CRITIC weight method,the total score of the driver ’s dangerous driving behavior spectrum eigenvalue is calculated.The driving state of drivers is divided into normal and dangerous by the total scores,taking 36 characteristic parameters of driving behavior which have high correlation with driving behavior as the input indexes of the model,combining with feature engineering,the characteristic parameters are optimized,and finally,8 characteristic parameters are determined as the input indexes of the model,combined with oversampling-ensemble learning algorithm,multiple dangerous driving behavior recognition models are established.Main results are as follows.1.The driving speed of the vehicle at the entrance section shows a trend of increasing first and then decreasing,while the driver driving out of the tunnel shows accelerated behavior.The higher the vehicle speed,the larger the fluctuation range of the lateral offset value,and the fluctuation range of the lateral offset value outside the tunnel is significantly higher than that inside the tunnel.The driver ’s lane change demand in the exit section is higher than that in the entrance section,and the driver in the exit section focuses on completing the lane change behavior within 50 m outside the hole.2.The weights of the four driving behaviors of rapid change,snake driving,dangerous car-following and dangerous lane-changing are 0.392,0.287,0.139 and 0.182,respectively.93.1% of the drivers’ total driving behavior spectrum characteristic scores were concentrated in the(0,0.15)range,and the proportion of drivers with a high level of danger was at a low level.3.The model established by SMOTE-LGBM algorithm has the best comprehensive performance,in which the F1-score is 0.917 and the AUC value is 0.910,which can effectively identify the dangerous drivers in the tunnel entrance and exit section in real time,providing a new method for the safety research of the tunnel entrance/exit section and laying a theoretical foundation for the active safety prevention and control technology of traffic accidents.