基于历史数据的道路交通事故风险分析与预测方法研究

发布时间：2018-05-29 21:35

本文选题：交通安全 + 负二项回归模型　；参考：《北京交通大学》2017年博士论文

【摘要】：交通安全统计分析是对交通事故数据进行分析的重要方法,它能从宏观角度上认识事故现象与影响因素之间的内在规律性,从而使得预测某种交通设施的交通安全风险成为可能。通过统计分析方法研究交通事故数据,对交通安全和管理工作有着重要的理论意义和应用价值。本论文在分析国内外研究现状的基础上,针对以前研究存在的不足,综合运用交通安全工程学、统计学理论与方法,借助于地理信息系统平台和统计分析软件,结合真实交通数据库本身的特点,通过建立合适的统计模型,从交叉口、路段、交通小区三个层面对相关交通安全问题进行了分析,研究了交通事故与不同影响因素之间的相关关系,从数量层面对交通安全风险进行了讨论,并初步揭示事故发生的内在规律性。具体来讲,本论文研究工作主要有如下几个方面:(1)匝道可以看做特殊形式的无信号交叉口。以往在匝道相关交通安全研究中,通常将匝道的安全影响范围选作半径为250英尺的圆形区域,而250英尺的半径长度并不一定适用于所有匝道处的安全风险评估。基于这种思考,本论文通过在GIS平台下设计逐步增加缓冲区大小的方法,讨论了如何设置匝道安全影响范围半径的问题。研究发现,在研究匝道相关的交通安全风险分析时,设置影响区域半径的最合适范围是90英尺到150英尺。而在实际的交通安全规划工作中,建议针对不同的地区,根据当地的具体条件具体分析,或者采用逐步增大影响范围的方法,来提高交通安全风险评估的可靠性。此外,本论文还针对匝道附近的交通安全数据,采用负二项回归模型,分析了匝道车道数、匝道长度、匝道日均交通量、限速对交通安全事故数的影响作用。(2)针对交通事故的危害程度,提出了事故危害指数作为综合评价不同严重程度的交通安全事故危害性的指标。针对事故数据过度离散的特点,采用了拟泊松模型对事故危害指标和若干重要的自变量进行了回归分析,识别对事故危害有显著性影响的自变量,并进一步得出了相关数量结果。通过统计分析发现,乡村道路上面的事故危害性更大,而更密集的交叉口也会增大交通事故的危害性。由于州/联邦所有的道路的安全条件和防护措施较好,这类道路也有更小的交通事故危害性。对于交通流量、铺面等级等经常讨论的影响因素,本论文也得出了与以前研究类似的结果,即交通流量越大、铺面状况越差,其交通事故危害性就越大。(3)交通安全事故的发生往往是由很多复杂的影响变量控制的复杂过程,这些自变量包括道路几何设计、交通流特征、驾驶员属性等等,而这些因素本质上可能是高度相关的,因此自变量之间可能存在的多重共线性不可忽略。针对以往在对交通安全数据的研究中,自变量之间的多重共线性并未受到足够重视的研究现状,提出了基于极小化贝叶斯风险的判别分析方法,在已知先验信息的条件下,通过极小化最小分类误差,将样本按照事故严重程度划分为若干个不同类别。在所提出划分规则下,通过模拟数据和实验观测数据相结合的方法,说明了所提出方法的有效性。(4)针对交通事故率的分布特征,通过核密度图和Q-Q图对数据进行了初步的分析和检验,发现对数正态分布能够较好的符合事故率的分布规律。在同时考虑数据的偏度非零特征及异方差特性的基础上,建立了相应的对数正态Hurdle模型,并且通过对比分析了其与γ-Hurdle模型和Weibull-Hurdle模型的表现,进一步确认了所提出模型的优势。由于对数正态分布的均值、方差和偏度均取决于尺度参数,因此可进一步放松限制推广模型,将尺度参数视为样本自变量进行回归分析,以增强建模灵活性。通过对比基于Tobit模型回归方法的表现,发现这类参数可调的对数正态Hurdle模型能够从数据中提取更多信息,从而具有更强的优势。此外,通过对比泊松模型和负二项回归模型,分析了基于事故率与事故发生计数方法的异同,说明了所建立模型的有效性。(5)从交通小区层面对交通安全事故的风险进行了评估,通过提出若干综合指标来衡量交通小区的交通安全水平,并在地理信息系统平台上对各交通小区的安全风险水平进行了可视化处理,对那些风险较高的交通小区进行了特别标注。此外,还利用负二项回归分析模型,对交通安全风险与交通小区相关的不同自变量之间的定量关系进行了分析。研究发现,交通小区内的车道总长度与交通流量与该小区的交通安全风险呈正相关关系,而交通小区的平均自由流车速与小区交通安全风险呈负相关关系。在交通小区的经济和社会层面来讲,交通小区内的平均收入越高、零售业和服务业从业人员越少、教育越高、人口数量越少,其交通安全风险水平就越低。这些研究结果不仅能反映某个交通小区的交通安全水平,同时也能预测未来交通小区的安全风险,从而指导正确的交通安全规划方向。
[Abstract]:Traffic safety statistical analysis is an important method to analyze traffic accident data. It can understand the inherent regularity between the accident phenomenon and the influencing factors from the macroscopic angle, thus making it possible to predict the traffic safety risk of a traffic facility. On the basis of analyzing the present situation at home and abroad and aiming at the shortcomings of previous research, this paper comprehensively uses traffic safety engineering, statistics theory and method, with the help of geographic information system platform and statistical analysis software, combined with the characteristics of real traffic database itself. After establishing a suitable statistical model, this paper analyzes the related traffic safety problems from three layers of intersection, section and traffic district, studies the relationship between traffic accidents and different influence factors, discusses the risk of traffic safety from the quantitative level, and preliminarily reveals the inherent regularity of the accident. The main research work is as follows: (1) the ramp can be regarded as a special type of non signal intersection. In the past study of ramp related traffic safety, the safety impact of the ramp is usually selected as a circular area with a radius of 250 feet, and the radius length of 250 feet does not necessarily apply to the safety risk assessment at all ramps. Based on this thinking, this paper discusses how to set the radius of the ramp safety impact range by designing a method of gradually increasing the size of the buffer zone under the GIS platform. It is found that the most appropriate range of the radius of the affected area is 90 feet to 150 feet in the study of the ramp related traffic safety risk analysis. In the work of traffic safety planning, it is suggested that the reliability of the risk assessment of traffic safety should be improved according to the specific conditions of local conditions, or the method of increasing the scope of influence gradually. In addition, this paper also uses negative two regression models to analyze the ramp vehicle for the traffic safety data near the ramp. The number, the length of the ramp, the daily traffic volume of the ramp and the influence of the speed limit on the number of traffic accidents. (2) in view of the damage degree of the traffic accidents, the accident hazard index is put forward as an index to evaluate the hazards of traffic safety accidents with different severity. Therefore, the hazard index and some important independent variables are regressive analysis to identify the independent variables which have a significant impact on the accident damage, and further obtain the relevant quantitative results. Through statistical analysis, it is found that the accidents on the rural roads are more dangerous and the more dense intersection will increase the danger of traffic accidents. The safety and protection measures of all roads in the state are better, and these roads also have smaller traffic accidents. For traffic flow, pavement grade and other factors often discussed, this paper also draws a similar result to previous research, that is, the greater the traffic flow, the worse the pavement condition, the greater the danger of traffic accidents. (3) (3) The occurrence of a general safety accident is often a complex process controlled by a number of complex variable variables. These variables include road geometric design, traffic flow characteristics, driver attributes and so on. These factors may be highly correlated in nature, so the possible multiple collinearity between the independent variables can not be ignored. In the study of secure data, the multiple collinearity between independent variables has not been paid much attention to. A discriminant analysis method based on minimization Bayesian risk is proposed. Under the condition of known prior information, the sample is divided into several different categories by minimizing the minimum classification error by the minimum classification error. Under the proposed division rules, the effectiveness of the proposed method is illustrated by the combination of simulated data and experimental observation data. (4) according to the distribution characteristics of traffic accident rate, the data are preliminarily analyzed and tested through the nuclear density map and Q-Q diagram, and it is found that the lognormal distribution can better conform to the distribution law of the accident rate. On the basis of the non zero zero characteristic and the heteroscedasticity of the data, the corresponding logarithmic normal Hurdle model is established, and its performance with the gamma -Hurdle model and the Weibull-Hurdle model is compared and analyzed. The advantages of the proposed model are further confirmed. The mean, variance and bias of the lognormal distribution depend on the mean, variance and bias of the lognormal distribution. In order to improve the modeling flexibility, the scaling parameters can be further relaxed and the scale parameters are considered as the sample independent variables to enhance the flexibility of the modeling. By comparing the performance of the Tobit model regression method, it is found that the lognormal Hurdle model of this kind of parameters can extract more information from the data and thus has a stronger effect. In addition, by comparing the Poisson model and the negative two item regression model, the similarities and differences between the accident rate and the accident counting method are analyzed, and the validity of the model is illustrated. (5) the risk of traffic safety accidents is evaluated from the traffic zone layer, and some comprehensive indexes are put forward to measure the traffic safety of the traffic district. On the platform of geographic information system, the level of safety risk of each traffic district was visualized, and the traffic areas with higher risk were specifically tagged. In addition, the negative two regression analysis model was used to determine the quantitative relationship between the traffic safety risk and the different independent variables. It is found that the total lane length and traffic flow in the traffic district have a positive correlation with the risk of traffic safety in this district, and the average free flow speed of the traffic district is negatively related to the risk of traffic safety in the residential district. The fewer employees in the service industry, the higher the education, the less the population, the lower the risk level of traffic safety. These results can not only reflect the level of traffic safety in a certain traffic district, but also predict the safety risk of future traffic areas, thus guiding the correct direction of traffic safety planning.
【学位授予单位】：北京交通大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：U491.31

【参考文献】