Research On Analysis And Prediction Methods For Highway Crash Risk Based On Historical Data

Statistical analysis on traffic safety is an important methodology for investigating the traffic accident data. Its purposes are to understand the inherent regularity between traffic accident phenomena and the influential factors from the macroscopic perspective. It also makes it possible to predict the traffic safety risk of certain traffic facilities. It is of great theoretical and practical meaning to study the traffic accident data using statistical analysis method, and it could also contribute to the traffic safety and management work. Based on the results of current research, the thesis analyzes the shortcomings of previous studies, and discusses relevant traffic safety problems on traffic accident risks and the impact of different influential factors from the three levels, which are intersections, segments and traffic analysis zones. In the research,knowledge from different fields including traffic safety engineering, statistical theory and method are used synthetically. The characteristics of real traffic database are considered, Geographic Information System and statistical analysis software like ArcGIS, SPSS and R are used. By using the established statistical models, traffic safety risks are evaluated quantitatively. Furthermore, inherent regularities of the relationship between accident and influential factors are discussed initially. Specifically, the research work in this paper mainly has the following aspects:(1) In the previous studies on traffic safety evaluation at the intersections, the influential area of the intersection is usually chosen as a round with a radius of 250 feet.But the radius of 250 feet may be not appropriate for all cases. Based on the consideration, this paper discusses how to set the radius of the influential area of the intersection by setting the buffer size increasingly using the GIS platform. The study found that the most appropriate range for the influential area should be set bewteen 90 feet and 150 feet. In real traffic safety planning, it is suggested that the reliability of the traffic safety risk assessment should be improved for different scenarios, according to the specific conditions of the local specific analysis, or as an alternative, by increasing the scope of influence gradually. In addition, this paper also analyzes the influence of number of lanes, ramp length, AADT and speed limit on the crash accidents frequencies using the negative binomial regression model.(2) According to the degree of hazard caused by traffic accidents, the Accident Hazard Index is introduced to evaluate the hazard of traffic accidents in different severity levels. The quasi-Poisson model is employed to analyze the Accident Hazard Index and several important independent variables. The independent variables are identified which have significant influence on accident hazard, and the related quantitative results are obtained. Through statistical analysis, it is found that rural roads have a larger accident hazard, and more dense intersections will increase the hazard of traffic accidents too. As the state/federal highway safety conditions and protective measures are better, such roads have a smaller traffic accident hazards. This paper also draws a similar result to the previous studies.(3) The occurrence of traffic safety accidents is often a complex process controlled by many influential variables, including geometry design, traffic flow characteristics,driver attributes, etc. It may be highly relevant in nature. The collinearity may exist between the independent variables, which could not be ignored. As in previous studies,collinearity between the independent variables has not been paid much attention. Basd on the motivation, the discriminant analysis method by the minimizing the Bayesian risk is introduced. By minimizing the misclassification error, the sample according to the severity of the accident is categorized into several different partitions. Under the proposed rule, the validity of the proposed method is illustrated by the combination of simulated data and experimental observation data.(4) According to the distribution characteristics of the crash frequency of the vehicle, the data is analyzed by using the kernel density and the Q-Q diagram. It is found that the logarithmic normal distribution is the best choice for the distribution of the crash frequency. Based on the right-skewed feature and the heterogeneous variance of the data, the corresponding logarithmic normal Hurdle model is established, and the performance of the gamma-Hurdle model and the Weibull-Hurdle model is compared and analyzed to confirm the advantages of the proposed model. Since the mean,variance, and skewness of the lognormal distribution are all dependent on the scale parameters, the extended extension model can be further relaxed, and the scale parameters can be regarded as sample independent variables for regression analysis to enhance the modeling flexibility. By comparing the performance of the Tobit model regression method, it is found that the logarithmic normal Hurdle model can extract more information from the data and thus have more advantages. In addition, by comparing the Poisson model and the negative binomial regression model instead, the similarities and differences between the collision rate and the accident occurrence counting method are analyzed, and the validity of the model is described.(5) Evaluating the risk of vehicle collision traffic accident from the traffic analysis zones, and measuring the traffic safety level of the traffic analysis zones by introducing a number of comprehensive indicators, and the safety risk level of each traffic analysis zones on the Geographic Information System platform. For those traffic analysis zones with higher risks was pointed out and visualized. In addition, the quantitative relationship between the crash risks and the different influential variables associated with the traffic analysis zones is analyzed by using the negative binomial regression analysis model. It is found that the total length of the lane and the AADT in the traffic area are positively correlated with the traffic safety risk of the traffic analysis zones, and the average free flow speed of the traffic analysis zones is negatively correlated with the traffic safety risk. At the economic and social factors of the traffic analysis zones, more higher income peoples in the traffic analysis zones, the less the employees in the retail and service industries, the higher the education, the less the population, the lower the level of traffic safety risk. In addition, the results of these studies can not only reflect the traffic safety level of a traffic district, but also to predict security risks of the analysis zones, and could guide the correct direction of traffic safety planning in some sense in the future.