摘 要

车联网的主要应用依赖于实时有效的通信数据传输，对车联网通信流量进行分析和预测是促使车联网通信资源合理高效的承载智能交通系统业务的关键。由于车联网系统具有高速移动、通信环境复杂多变的特点，使用传统技术难以完成预测。本文基于机器学习方法完成真实交通场景下的全天候交通流量预测和仿真场景下的车联网通信流量预测。

对交通流量的预测，基于首届大学生交通数据建模大赛公开数据中的延安高架发布段流量速度数据库，对数据进行处理，再选择几类机器学习算法进行模型训练和优化，得出XGBoost模型效果良好、性能更优，能很好地应用于本课题的交通流量预测场景。

对通信流量的预测，以武汉市洪山区书城路与文荟街交汇处至书城路与文秀街交汇处路段为实验场景，在Sumo平台上搭建交通仿真场景获得交通数据，连接NS3，搭建基于DSDV协议的车联网通信仿真场景得到车联网通信数据。在Sumo平台上修改交通配置，再进行Sumo-NS3联合仿真，得到了不同交通仿真场景下的实时通信数据。研究交通流量与通信流量的相关性，结果表明交通流量与通信流量具有较高的相关性，将交通数据与通信数据混合作为数据集。选择几类机器学习算法进行模型训练和优化，得出决策树模型效果良好、性能更优，能很好地应用于本课题的通信流量预测场景。

关键词：车联网；流量预测；机器学习；Sumo；NS3

Abstract

The main application of the Internet of vehicles depends on the real-time and effective communication data transmission. The analysis and prediction of the communication flow of the Internet of vehicles is the key to make the communication resources of the Internet of vehicles carry the business of the intelligent transportation system reasonably and efficiently. Due to the features of high-speed mobile and intricate communication environment, it is hard to accomplish the prediction using traditional technology. Based on the machine learning method, this paper completes the all-weather traffic flow prediction in the real traffic scenario and the communication flow prediction in the simulation scenario.

For the traffic flow prediction, based on the traffic speed database of Yan'an viaduct release section in the first college students traffic data modeling competition, the data is integrally processed, and then several kinds of machine learning methods are selected for model training and optimizing. The consequences show that the XGBoost model has terrific effect and performance, and can be well applied to the traffic flow prediction scene of this topic.

For the prediction of communication flow, take the section from the intersection of Shucheng road and Wenhui street to the intersection of Shucheng road and Wenxiu street in Hongshan District of Wuhan City as the experimental scenario, build a traffic simulation scenario on SUMO platform to obtain traffic data, connect ns-3, and build a vehicle networking communication simulation scenario based on DSDV protocol to obtain vehicle networking communication data. Modify the traffic configuration on SUMO platform, and conduct sumo-ns3 joint simulation to get the real-time communication data in different traffic simulation scenarios.

Studying the correlation of traffic flow with communication flow, and it turns out that traffic flow and communication flow have high correlation. Traffic data and communication data are mixed as data sets. Several kinds of machine learning algorithms are selected for model training and optimizing, and the consequence shows that the decision tree model has good effect and better performance, which can be well applied to the communication traffic prediction scenario of this topic.

Keywords：VANET;Flow prediction;Machine learning;Sumo;NS3

目 录

摘要 I

Abstract II

第1章 绪论 1

1.1 研究背景及意义 1

1.2 国内外研究现状 1

1.3 论文工作及结构 2

第2章 相关技术 4

2.1 主要模型算法概述 4

2.1.1 决策树模型 4

2.1.2 随机森林模型 5

2.2 VANET路由协议分析 5

2.3 仿真软件工具 6

2.3.1 Sumo简介 6

2.3.2 NS3及其可视化工具NetAnim简介 6

第3章 基于XGBoost的交通流预测 8

3.1 数据集及模型评价方法 8

3.1.1 数据准备 8

3.1.2 模型评价方法 9

3.2 交通流预测流程 9

3.3 数据集特性分析 10

3.3.1 确认数据规模 10

3.3.2 缺失值处理 10

3.3.3 数据集统计分布 10

3.4 属性与目标的相关性分析 12

3.5 数据处理 16

3.6 特征选择 18

3.7 模型训练 20

3.8 模型优化 20

3.9 提交结果 22

第4章 基于决策树的通信流量预测 24

4.1 数据来源 24

4.2 仿真实验 24

4.2.1 Sumo仿真过程 24

4.2.2 NS3仿真过程 28

4.3 流量相关性分析 31

4.4 通信流量预测 32

总结与展望 36

本文工作总结 36

未来工作展望 37

参考文献 38

致谢 39

第1章 绪论

1.1 研究背景及意义

车载式自组织网络技术(Vehicular-Ad-hoc-Networks，VANET)是一类应用于城市交通系统中的移动自组织网络（Mobile-Ad-hoc-Networks，MANET，是MANET在智能交通系统领域(Intelligent Transportation System，ITS)的应用和延伸。车载式自组织网络技术通过通信、云计算、信息安全等技术的融合实现了车与车、车与人、车与道路基础设施、车与网络服务平台的通信互联。

1、在车联网系统中车与车之间通过共享车辆位置、行驶速度等状态信息，实现车辆辅助驾驶、紧急制动等功能。