基于分位数回归的高速公路交通事故预测

发布时间：2018-06-05 07:11

本文选题：高速公路 + 交通事故　；参考：《东南大学》2015年硕士论文

【摘要】：由于近年来高速公路里程与交通量的迅速增长,交通事故明显增加,给人民的生命财产造成了严重危害。因此,对交通事故进行深入分析,预测交通事故的发展趋势,对于道路交通安全评价、规划以及决策具有重要意义。基于此目的,本文提出了基于分位数回归的高速公路交通事故预测方法。由于国内在交通安全方面的研究起步较晚,在交通事故预测方法的研究上,特别是在回归模型方法上,无论是研究对象的深度和广度都与国外存在一定的差距,有待于进一步深化研究。本文首先,对交通事故数据常见特征(偏大离差、偏小离差、低样本均值和小样本量)以及可能存在的问题(未记录、时变性变量、内生变量、遗漏变量)进行分析和考虑,结合回归模型法、神经网络法、时间序列法、灰色预测法的基本原理以及优缺点,确定了本文对比的事故预测方法。然后,从道路环境因素角度,详细调研了道路线形(平面线形、纵断面线形和平纵线形组合)、特殊路段(桥梁、隧道)和交通环境(交通量、交通组成、车速)对交通安全的影响,以此确定本文事故预测时的变量。最后,将普通分位数回归扩展到事故数据的应用上,用分位数回归的两个指标(边际效应和灵敏度)解释变量,并基于分位数回归的一系列参数估计,提出了本文交通事故的预测方法：历史数据法和概率法,与负二项回归法进行对比。实际案例证明了分位数回归在交通事故预测上的优越性。在分析过程中,分位数回归给出了任意分位点下的参数估计、边际效应和灵敏度,相比负二项回归的一组估计更加全面；在事故预测上,历史数据法准确度高达74%,均方根误差为0.618。基于分位数回归的两种预测方法在准确度和误差上都要优于负二项回归模型。
[Abstract]:Due to the rapid increase of highway mileage and traffic volume in recent years, traffic accidents have obviously increased, which has caused serious harm to people's life and property. Therefore, it is of great significance for road traffic safety evaluation, planning and decision-making to deeply analyze traffic accidents and predict the development trend of traffic accidents. For this purpose, a method of expressway traffic accident prediction based on quantile regression is proposed in this paper. Due to the late start of the research on traffic safety in China, there is a certain gap between the depth and breadth of the research object, especially the regression model method, in the study of traffic accident prediction method. Further research is needed. First of all, the common characteristics of traffic accident data (large deviation, small deviation, low sample mean and small sample size) and possible problems (unrecorded, time-varying, endogenous, missing) are analyzed and considered. Combined with the basic principles, advantages and disadvantages of regression model method, neural network method, time series method and grey prediction method, the accident prediction method compared in this paper is determined. Then, from the point of view of road environmental factors, the road alignment (plane line, vertical section line and vertical line combination), special sections (bridges, tunnels) and traffic environment (traffic volume, traffic composition) are investigated in detail. The influence of speed on traffic safety is used to determine the variables of accident prediction in this paper. Finally, the ordinary quantile regression is extended to the application of accident data, and the variables are explained by two indicators of quantile regression (marginal effect and sensitivity), and a series of parameter estimates based on quantile regression are presented. This paper puts forward the methods of traffic accident prediction: historical data method and probability method, which are compared with negative binomial regression method. A case study shows the superiority of quantile regression in traffic accident prediction. In the process of analysis, the quantile regression gives the parameter estimation, marginal effect and sensitivity at any locus, which is more comprehensive than the negative binomial regression. The accuracy of historical data method is as high as 74 and the root mean square error is 0. 618. The accuracy and error of the two prediction methods based on quantile regression are better than that of the negative binomial regression model.
【学位授予单位】：东南大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U491.31

【相似文献】