基于统计与假设检验的高速公路交通事故数据分布特性

为了研究高速公路基本路段上交通事故数据的分布特征, 将事故数、伤亡事故数、事故死亡人数与事故受伤人数归类为离散型事故数据, 将事故间隔时间与平均每年每公里事故数归类为连续型事故数据; 对于离散型事故数据, 采用均匀划分法、动态聚类法与滑动窗法划分高速公路统计区段, 运用泊松分布、负二项分布、零堆积泊松分布与零堆积负二项分布对事故数据进行拟合; 对于连续型事故数据, 以收费区间为路段划分标准, 用正态分布、负指数分布进行事故数据拟合; 运用皮尔逊卡方值对各种拟合结果进行拟合优度检验。研究结果表明: 在各种区段上, 事故数均服从负二项分布, 有些情况下会同时服从负二项分布与泊松分布, 伤亡事故数与事故死亡人数主要服从零堆积泊松分布或零堆积负二项分布, 拟合优度检验中的概率均大于0.05;平均每年每公里的事故数比较符合正态分布, 而事故间隔时间则主要服从负指数分布, 拟合优度检验中的概率也均大于0.05;交通事故数据的统计分布特征是建立事故预测模型与事故多发点鉴别的前提条件之一, 而事故间隔时间可作为安全可靠度的度量指标。 

Distribution characteristics of traffic crash data of freeway based on statistics and hypothesis test

In order to analyze the distribution characteristics of traffic crash data on the basic sections of freeway, traffic crash number, fatal and injury crash number, death and injury numbers of traffic crash were taken as discrete random variables, and crash interval time and average annual crash number per kilometer were taken as continuous random variables.For discrete crash data, the sections of freeway were divided by using equally divided method, dynamic clustering method and sliding window method, and crash data were fitted by using Poisson distribution, negative binomial distribution, zero-inflated Poisson distribution and zeroinflated negative binomial distribution.For continuous crash data, the sections were divided based on the toll intervals, crash data were fitted by using normal distribution and negativeexponential distribution.The goodness-of-fit tests of various fitting results were performed by using Pearson's square.Analysis result shows that in all sections, crash numbers are subject to negative binomial distribution, and in some cases, obey negative binomial distribution and Poisson distribution at the same time.Fatal and injury crash number and death number of traffic crash mainly obey zero-inflated Poisson distribution or zero-inflated negative binomial distribution.The probabilities of goodness-of-fit test are all greater than 0.05.Average annual crash number per kilometer is more subject to normal distribution, while crash interval time mainly obeys negative exponential distribution, and the probabilities of goodness-of-fit test are also greater than 0.05.The statistical distribution characteristic of traffic crash data is one of the prerequisites for establishing crash prediction model and the identification of crash black spots, and crash interval time can be used as the measurement indicator of safety reliability.