Spiral Tunnel Based On Driver Characteristics Driving Behavior Risk Study

The "white hole effect" caused by changes in the driving environment at the exit stage of the tunnel has exacerbated the difficulty and uncertainty of driver information acquisition and manipulation behavior,bringing hidden dangers to traffic safety.In-depth study of the inner formation mechanism of driving behavior,the construction of driving behavior model,the formation of accurate identification,indepth analysis,fusion prediction of driving behavior risk analysis system,to ensure the tunnel exit traffic safety and the future intelligent development of tunnel monitoring and early warning system is of great significance.This research addresses the complex characteristics of driving behavior in the tunnel exit phase,takes the risk of driving at the exit of a long spiral tunnel as the research content,carries out real-world driving experiments with the Jinjiazhuang spiral tunnel,collects vehicle braking information,illumination and driver behavior characteristics data,and conducts research on the behavioral risk of drivers in the tunnel exit section through a neural network approach.The research has a guiding role for the establishment of tunnel exit risk warning system and also provides new ideas for tunnel driving safety research.The detail research information and the result are as follows.(1)A study based on deep learning for predicting the vehicle speed during the tunnel exit phase and visualizing the driver’s comfort area.A vehicle speed prediction model based on deep convolutional neural network is proposed,and the degree of contribution of the driving area in the tunnel exit phase is presented as a heat map by the Grad-CAM algorithm.The results show that the overall model mean square error is 0.221 smaller error and the model accuracy is high.,the visualization of the convolutional neural network using Grad-CAM algorithm to analyze the features learned by the network improves the model performance,and the visualization of the tunnel exit gaze area yields the distribution of the tunnel exit phase gaze area at different locations and vehicle speeds.(2)The study of clustering of driving behaviors based on driver characteristics and integration of driving environment analyzes driver behavior types.By proposing a quantitative evaluation method of driver behavior types at the tunnel exit stage,comparing different clustering methods,we found that the classical K-means model evaluation index effect is better than other clustering methods.The driving behavior type discrimination experiments show that the K-means driving behavior type discrimination algorithm has optimal driving behavior data clustering results,and clusters the driver data in the tunnel exit phase into three categories: conservative,smooth and aggressive,and analyzes the feature parameters in the driving behavior type clusters to prove that the driving behavior type labels show good consistency with trends in driving safety characteristics.(3)Research on tunnel exit risk prediction model based on driver behavior features.The quantitative model of driving behavior risk is constructed by learning the deep features of driving behavior data through long and short-term memory network model.The experimental results show that the model accuracy reaches 80%prediction effect is significant,the tunnel exit stage is divided into three types of risk levels,in which five risk points obtained in the driving behavior characteristics are different,the gaze area difference is obvious.