| Statistical analysis of road crashes plays an important role for mastering the distribution law of crashes,predicting the development trend of crashes,formulating crash prevention measures and optimizing the traffic safety planning scheme.The modeling and the analysis of crashes from macro and micro spatial scales has become one of the important research topics in the field of traffic safety.In recent years,the theory and the practice of traffic safety management show that road crash data have the characteristics of spatial correlation,temporal correlation and excessive zero.If these characteristics are ignored in the modeling,the fitting performance and the prediction accuracy of crash models will be seriously affected,and even wrong conclusions may be drawn.The existing research of crash modeling and analysis is often limited to propose solutions to one of the above characteristics.Much research work is needed in the modeling theory and method to consider the spatial and temporal correlation of crash from the macro and micro perspectives,and special attention needs to be paid to solve excessive zero from the micro perspective in the crash data.This thesis studies the relationship between the road crash frequency and its influencing factors at the macro and micro levels,comprehensively considering the spatial and temporal correlation of crashes,and the characteristics of excessive zero and potential heterogeneity in the data at the micro level are especially considered,thus,we establishe the corresponding crash frequency model.Based on the results of those models we quantitatively analyze the impact of relevant factors on crashes.In order to reduce the occurrence of road crashes,the corresponding improvement measures are put forward.The main research work of this paper is as follows:(1)Modeling and analysis of crash frequency considering the spatial effect at the macro level from cross-section dataBased on the cross-section data,the city under study is divided into 100 traffic analysis zones by grid generation method.The crash frequency and travel interest points in each traffic analysis zone are extracted.Considering the spatial correlation of the crash frequency,the spatial lag model and the spatial error model of the crash frequency at the macro level are constructed.The results show that the spatial error model considering the spatial correlation is better than spatial lag model and the general linear model;crash frequency at the macro level of traffic analysis zone has spatial correlation,and this correlation is more a result of the correlation of error terms.The results show that the number of hospitals and schools has a significant positive impact on the crash frequency.Therefore,in order to reduce the occurrence of crashes,traffic management departments should strengthen the management and the control of mixed traffic flows of pedestrians,non-motor vehicles and motor vehicles in and around hospitals and schools.(2)Modeling and analysis of crash frequency considering spatio-temporal effects at the macro level from panel dataBased on the panel data,from the macro county level,factors,for example,population,regional area,motor vehicle ownership,income,and proportion of uninsured population are extracted.Considering the spatial correlation and time effect of crashes,we constructed the spatial autocorrelation model of variable intercept time-fixed effect and the spatial error model of variable intercept time fixed effect.Based on the models,the influence of spatial correlation and socio-economic factors on crash frequency are analyzed.Moran's I and Geary's C test results confirm that there is a significant spatial correlation among crash frequencies of counties.The results of Chow test,LR test and Hausmann test show that the individual effect and time effect should be added into the crash frequency model,and the time effect should be in the form of a fixed effect.The model estimation results show that the spatial autocorrelation model of variable intercept time fixed effect is the best,which reveals that the spatial correlation of crashes mainly comes from the correlation of spatial error terms,and seldomly comes from the correlation of crash frequencies.In addition,the increase in population,motor vehicle ownership,the proportion of households with an income lower than the poverty line,and the proportion of uninsured population will cause more crashes,while the increase of median family income will lead to fewer crashes.Therefore,the traffic management departments should enhance the investment in public transport in low-income areas and improve the travel sharing rate of public transport,so as to reduce the dependence of lowincome people on car travel and to reduce the proportion of motor vehicles with low safety standards on the road.(3)Modeling and analysis of critical crash frequency considering the excessive zero problem and spatial effects at the micro levelAt the micro road segment level,road characteristic factors,for example,road grade,horizontal curve are extracted.Their influence on critical crash frequency is analyzed.Aiming at solving both the spatial correlation of critical crash and excessive zero in the data,our study established the zero-inflated negative binomial spatial model and hurdle negative binomial spatial model of crash frequency.The results show that the zeroinflated and hurdle dual-state models outperform the standard Poisson model and the standard negative binomial model.Secifically,the zero-inflated model is better than the hurdle model in dealing with excessive zero,and the zero-inflated negative binomial model considering spatial correlation and heterogeneity has the best overall performance.In addition,there is a significant positive correlation between the setting of horizontal curve warning signs and critical crash frequency,while the road grade has a significant negative correlation with critical crash frequency.Therefore,motor vehicle drivers should keep vigilant and drive cautiously in the road segment with warning signs,and traffic safety managers should add relevant warning signs to remind drivers of safety according to historical casualty data.(4)Modeling and analysis of crash frequency based on negative binomial Lindley spatio-temporal effects model at the micro levelIn view of the problem that the dual-state models are only suitable for critical crashes,a negative binomial Lindley space-time effect model of the total crash frequency is established at the micro level to deal with the characteristics and problems of the spatial correlation,temporal correlation and excessive zero of the total crash frequency.Further,our research analyzes the influence of road characteristic factors on the total crash frequency.The results suggest that there is a significant spatial correlation in the total crash frequency;compared with the standard negative binomial model,the negative binomial Lindley model can better deal with excessive zero characteristics in crash data;the model considering spatial correlation and temporal 1st order random walk effect appears to be the optimal model.In addition,for the segment where overtaking is allowed,the total crash frequency is less because of the better sight distance and road conditions.The increase of average turning angle of horizontal curve and intersection density corresponds to the increase of total crash frequency.Therefore,it is suggested that the traffic department should limit the horizontal curve turning angle and intersection density within the scope of conditions in the road planning and design stage to raise the road traffic safety level. |
PDF Full Text Request