Investigating Method Of Influence Factors And Spatial Distribution Characteristics For Drunkdriving Crashes

In recent years,the number of motor vehicles and total alcohol consumption in China continue to grow rapidly,and the illegal behaviors of drunk-driving of motor vehicles(hereinafter referred to as drunk-driving events)occur frequently,resulting in traffic crashes(hereinafter referred to as drunk-driving crashes)which account for a high proportion of the overall accidents data.Taking Beijing as the research object,this paper is committed to the analysis of the influencing factors of drunk driving crashes and the research on the mining method of spatial distribution characteristics,aiming at reducing the frequency and the severity of drunk driving crashes.Firstly,this paper proposes a two-stage method for investigating influence factors of drunk driving crashes based on cluster-based regression model.In the first stage,complex group classification is achieved based on latent class analysis.In the second stage,significant influence factors and their heterogeneity are analyzed based on random parameter logit model.This two-stage method can be used to solve the problem of identifying the factors account for the severity of drunk-driving events and drunkdriving crashes.The results show that factors influencing the severity of drink-driving events(which measures whether an accident occurs or not)vary across the city as a whole and among different groups,age of driver,season,time interval,functional zones are significant factors with heterogeneity,gender of driver,junction type are significant hidden factors which are hidden in latent class.Factors influencing the severity of drinkdriving crashes(measured by death/injury or not)also varied across the city as a whole and across different groups,Quasi-driving type,time interval,road type,functional zones and day type are the significant factors with heterogeneity,day type and speed limit are significant hidden factors which are hidden in latent class.Secondly,this paper proposes a two-stage spatial heterogeneity analysis method based on LISA-GWR.In the first stage,cold-hot spots and outliers are detected based on local indicators of spatial association(LISA),and in the second stage,spatial influencing factors are identified based on geographical weighted regression model(GWR).Thhis two-stage method can be used to find spatial distribution characteristics of drunk-driving crashes and identify spatial influence factors of drunk-driving accidents.The results show that drunk-driving crashes are mainly distributed in the suburbs of Beijing,while the non-crashes drunk-driving events are mainly distributed in the urban areas of Beijing,which showed the distribution characteristics of outliers.The influence factors of drunk-driving crashe have spatial non-stationary characteristics,at the districr scale,the density of drunk driving events,the density of companies,restaurants,intersection density,scientific research and education density are significantly correlated with the density of drunk driving crashes,which can explain the above spatial distribution differences.Finally,this paper puts forward the policy of treating drunk-driving crashes from multiple perspectives.Measures to reduce the severity of drunk-driving crashes include paying attention to the distance between intersections in the planning process and using new detection technology to prevent drunk-driving behavior.Measures to reduce the severity of drunk-driving crashes include improving electronic cameras in blind areas such as low-speed intersections,curves and bridges,lower speed limit strategies should be implemented in accident black spots on higher speed limit roads such as urban expressways.The research of this paper is helpful to promote the development of basic methods and engineering applications of accident data analysis,and provide some policy support for the treatment of drunk driving.