【作者】 蓝翔；

【导师】 张亚平；

【作者基本信息】 哈尔滨工业大学 ， 交通运输规划与管理， 2009， 硕士

【摘要】 交通事故给人类社会造成巨大损失,为提高道路安全水平,世界各国均投入了大量人力、物力和财力进行交通事故研究,最大限度地防止交通事故发生。因此,揭示交通事故发生机理,建立交通事故预测模型,把握未来交通事故的发展趋势,对制定减少交通事故的对策,提高交通安全水平,具有重要的理论和现实意义。采用科学的预测方法建立道路事故预测模型,才能准确预计或推算交通事故的发展趋向、水平和程度。本文采用非参数统计方法进行交通事故预测,侧重于理论体系和量化模型的构建,利用非参数统计方法定性预测交通事故形态和时空分布变化趋势,引入非参数回归模型定量地进行宏观事故预测。在剖析事故的形态分布、时空分布特性基础上,采用适合交通事故分布趋势预测的Mann-Whitney、Cox-Stuart等模型进行预测黑龙江省近三年交通事故形态及时空分布的趋势。以事故总数和总死亡人数之间的关系为研究对象,构建了基于核回归、多项式回归、稳健估计的交通事故非参数回归预测模型。模型特点在于,进行交通事故预测时无需过份关注变量和因变量之间的关系,突破了函数关系形式需要假定的限制。此外,考虑到不同的密度估计方法将产生不同的非参数回归预测函数,且不同的小波基也将分解出不同的非参数回归预测函数,提出了基于小波分析的非参数回归交通事故预测建模方法。最后,根据近三年交通事故数据对未来年黑龙江省交通事故进行了非参数回归预测,通过聚类分析,给出相应的交通安全水平等级划分标准,对黑龙江省近、中、远期交通安全状况趋势进行了预测评价。更多还原

【Abstract】 Nowadays, traffic crashes cause many death and great economic loss all over the world. And every country, especially in the western developed countries, invested great number of money, human recourses and energy to research on the traffic crashes in order to reduce the number of people died in the accidents and to improve the level of safety. It is necessary to predict the trend of future traffic safety to make effective transportation safety policy. Also Crash prediction model plays important role in explaining the reason why crashes’happening. So social organizer and government administer systems can benefit from research on traffic safety prediction model. As we all know, it is the scientific prediction models and methods that help us to get precise trend of development of traffic crashes and to acquire the level of safety.This paper focuses on traffic prediction after introducing nonparametric statistics. At first, the future trend of time, space and forms of crashes distribution are discussed. Then nonparametric regression models are used for macro traffic prediction in Hei Longjiang province.First of all, some useful and suitable nonparametric models are intruding into timely distribution of crashes. These power tools are used for crashes in different time and location. After that, some nonparametric test models, such as Mann-Whitney, Cox-Stuart and etc, are used for statistical analysis on the crashes. Finally, last 3 years crash data in Heilongjiang is taken for an example.Secondly, the dissertation discusses the relationship between the number of crashes and the number of death caused by traffic accidents. In order to break through the bottleneck that sometimes function formation or the requirement of presuppose, nonparametric regression methods by kernel estimation of Nadaraya-Watson, kernel estimation by multinomial, and Locally Weighted EstimationScatter plot Smoothing (LOWESS) are introduced into the death prediction. Besides, considering different nonparametric regression methods can be dispatched by different wavelets, such as Schauder, Haar, and Daubechies, three nonparametric regression methods bases on wavelet analysis are put forward to predict the crashes.Finally, the future safety situation in Hei Longjiang province is predicted according the recent 3 years data. Furthermore, the scale of traffic safety is made out according K-M clustering analysis.更多还原