Analysis Of Distribution Characteristics And Influencing Factors Of Warning Behavior Based On Vehicle Network Data

With the continuous development of China’s transportation industry and the continuous expansion of freight volume,freight safety issues have become increasingly important,and road traffic accidents are mostly caused by dangerous driving behavior of drivers.Therefore,it is necessary to study the dangerous driving behavior of freight drivers.In addition,with the improvement of intelligent degree and function expansion of vehicle equipment,it is possible to analyze dangerous driving behavior based on vehicle networking big data.Accordingly,based on the vehicle networking warning and vehicle trajectory data,and integrating the weather and road data,this paper studies the distribution law of dangerous driving behaviors such as overspeed warning,fatigue driving warning and high-risk warning occurred by freight vehicle drivers in the driving process,and analyzes the obvious and indigenous factors affecting the occurrence and types of warning behaviors,so as to provide theoretical support for improving freight safety.The main work of this paper is as follows :(1)Distribution characteristics of warning behaviors.Based on the visual analysis of 121823 warning data provided by a freight company in Guangxi,this paper explores the spatial and temporal distribution characteristics of warning frequency and types,and compares the differences of warning types under different factors.The results show that the occurrence of warning behavior increases year by year,and has a certain spatial aggregation,mainly in Nanning,Liuzhou,Guilin and other economically developed cities;the chi-square test results of the distribution of warning types under different factors show that there are significant differences in the occurrence frequency of warning types under different temperature,horizontal visibility,precipitation,time period,road grade and speed.(2)Analysis of influencing factors of drivers’ warning behavior.The binomial Logit model and Probit model are constructed to explore the significant influencing factors of drivers’ warning behavior.The results show that the validity and goodness of fit of the two models are good.Among them,the increase in speed,the increase in cumulative driving time of a single trip,the increase in precipitation,the time period of6 to 18,and the road grade of two or three will increase the probability of warning.The probability of warning will decrease with the increase of acceleration,horizontal visibility,time interval from 18 to 24 and road grade 4.(3)Analysis of influencing factors of driver warning type.Random parameter Logit model is constructed based on discrete selection model,and XGBoost model,random forest model and BP neural network model are constructed based on machine learning model to explore the main influencing factors of driver warning type.The results show that the random parameter Logit model can better explain the heterogeneous selection of various factors on the type of warning.XGBoost model performs best in three machine learning models.The main influencing factors of drivers’ warning occurrence types are speed,acceleration,road grade,precipitation,cumulative driving time of single stroke,and horizontal visibility.Different variables have different impacts on different types of warning.Overspeed warning is mainly positively affected by acceleration and road grade.Fatigue driving warning is mainly positively affected by cumulative driving time of a single stroke.High-risk warning is mainly positively affected by speed and precipitation and negatively affected by horizontal visibility.