论文总字数：39670字

摘 要

欧盟每10年制定一次道路交通安全行动计划，其委员会在2010年确立了2020年道路交通死亡人数下降为2010年的50%的总体目标。虽然该方案制定了欧盟各国共同的行动准则，但是由于欧盟各国在人口结构、经济水平、社会发展等方面存在着许多差异，这些影响因素都导致了它们道路交通安全发展趋势的差异。鉴于此，本文选取了老龄化、城镇化和机动化三个指标作为待研究的影响因素，应用ARIMA和ARIMAX模型对欧盟各国道路交通死亡人数及其年变化率进行了时间序列分析。

首先，本文采集了欧盟各国1991年至2015年间的道路交通死亡人数、老龄化、城镇化和机动化这四个变量的实际值以及它们的年变化率序列，分为实际值序列组和变化率序列组。在此基础之上对两组序列进行了描述性统计分析，探究了两组序列历年来的变化情况以及它们在长期趋势和短期波动性上的区别。

然后，本文应用ARIMA模型对欧盟各国的道路交通死亡人数的实际值序列和变化率序列进行了建模和预测，并对比了该模型在不同国家和不同数据集上的预测精度。结果表明，ARIMA模型可以较好地预测出序列的长期趋势，但无法预测出短期波动；它在实际值序列上的预测精度整体上高于其在变化率序列上的预测精度，而且不同国家之间预测精度差异较大。

最后，本文加入了老龄化、城镇化和机动化这三个外生变量的实际值序列和变化率序列，分别与道路交通死亡人数的实际值序列和变化率序列匹配成新的数据集，对欧盟各国建立了含有外生变量的ARIMAX模型。在此基础之上，对比了ARIMAX模型在不同国家和不同数据集上的预测精度，进一步对比了ARIMA和ARIMAX模型的效果并探究了各外生变量的重要性。结果表明，ARIMAX模型可以较好地预测出序列的短期波动，它在实际值序列上的预测精度高于ARIMA模型；仅考虑含有单个外生变量的模型，机动化的作用大于城镇化而老龄化作用最小；仅考虑含有多个外生变量的模型，老龄化的作用大于城镇化，而机动化的作用最小。

关键词：欧盟国家，道路交通安全，ARIMA模型，ARIMAX模型，外生变量

ABSTRACT

EU formulates a road traffic safety action every 10 years, and in 2010 it set an overall goal of reducing 50% road traffic fatalities by 2020. Although the scheme has formulated common guidelines for all EU members, there are still many differences in aspects such as population structure, economic and social progress among EU countries, which may lead to different trends of road traffic safety. In view of this, this paper chose three indicators: aging, urbanization and motorization as the exogenous variables to be studied, and then applied time series analysis to study the trend of road traffic fatalities as well as its annual change rate.

Firstly, this paper collected relevant data of EU countries from 1991-2015 including road traffic fatality, aging, urbanization and motorization. The data is divided into two different sets: real value series of these four variables and their annual change rate series. On this basis, the descriptive statistical analysis of the two data sets was carried out, and their differences in long-term trend and short-term volatility were explored.

Secondly, this paper used ARIMA model to model and forecast the real value series and the change rate series of road traffic fatalities in EU countries, then compared the prediction accuracy of the models in different countries and different data sets. The results showed that ARIMA model could predict the long-term trend of the series well, but performed badly on predicting short-term fluctuations; its prediction accuracy in the real value series was higher than that in the change rate series on the whole, and the prediction accuracy varies greatly among different countries.

Finally, this paper added three exogenous variables into established ARIMA models and thus built ARIMAX models with exogenous variables for EU countries. On this basis, the prediction accuracy of ARIMAX model in different countries and different data sets was compared. The effects of ARIMAX model and ARIMAX model were further compared and the importance of exogenous variables was explored. The results showed that ARIMAX models could predict the short-term fluctuations well, and its prediction accuracy was higher than that of ARIMA model in the real value series. Considering models with single exogenous variable, motorization had the greatest influence, next was urbanization and aging was least powerful; considering models with multiple exogenous variables, aging had the greatest influence, followed by urbanization and motorization.

KEY WORDS: EU countries, road traffic safety, ARIMA model; ARIMAX model; exogenous variable

目 录

摘 要 I

ABSTRACT II

第一章 绪论 1

1.1 研究背景与意义 1

1.2文献综述 2

1.2.1 道路交通风险模型的产生 2

1.2.2 不同的建模技术 3

1.2.3 时间序列模型的类型 4

1.3 研究内容和技术路线 7

1.4 本章小结 8

第二章 数据介绍 9

2.1 欧盟整体数据情况 9

2.2 欧盟各国数据情况介绍 13

2.3 本章小结 16

第三章 ARIMA时间序列分析 17

3.1 ARIMA模型介绍 17

3.2 ARIMA模型的建立 18

3.3 不同数据集的预测结果比较 23

3.4 本章小结 25

第四章 ARIMAX时间序列分析 26

4.1 ARIMAX模型介绍 26

4.2 ARIMAX模型的建立 27

4.3 不同数据集的预测效果比较 32

4.4 ARIMA与ARIMAX预测效果比较 33

4.5 各外生变量重要性分析 35

4.6 本章小结 36

第五章 总结与展望 37

5.1 研究内容与主要结论 37

5.2 不足与展望 37

致 谢 39

参考文献 40

附录A：欧盟各国时间序列预处理及可视化R语言代码 43

附录B：ARIMA以及ARIMAX模型R语言代码 45

附录C：欧盟各国ARIMAX模型参数选择结果 48

剩余内容已隐藏，请支付后下载全文，论文总字数：39670字