通信学报 ›› 2022, Vol. 43 ›› Issue (6): 133-142.doi: 10.11959/j.issn.1000-436x.2022122

• 学术论文 • 上一篇    下一篇

基于动态图嵌入的车联网拓扑控制

孙雁飞1,2, 尹嘉峥3, 亓晋2, 胡筱旋2, 陈梦婷2, 董振江4   

  1. 1 江苏省高性能计算与智能处理工程研究中心，江苏 南京 210023
    2 南京邮电大学物联网学院，江苏 南京 210003
    3 南京邮电大学通信与信息工程学院，江苏 南京 210003
    4 南京邮电大学计算机学院，江苏 南京 210023
  • 修回日期:2022-05-09 出版日期:2022-06-01 发布日期:2022-06-01
  • 作者简介:孙雁飞（1976- ），男，山东济南人，博士，南京邮电大学研究员，主要研究方向为人工智能在工业互联网、能源互联网、智慧供应链等领域的关键技术与应用
    尹嘉峥（1998- ），男，山东济南人，南京邮电大学硕士生，主要研究方向为车联网、复杂网络等领域的技术与应用
    亓晋（1983- ），男，山东济南人，博士，南京邮电大学副教授，主要研究方向为人工智能在工业互联网、能源互联网、智慧供应链等领域的关键技术与应用
    胡筱旋（1992- ），女，江苏南京人，博士，南京邮电大学讲师，主要研究方向为机器学习、云计算
    陈梦婷（1994- ），女，安徽桐城人，博士，南京邮电大学讲师，主要研究方向为系统辨识与深度学习
    董振江（1970- ），男，江苏南京人，博士，南京邮电大学研究员，主要研究方向为计算机视觉、知识图谱在车联网、高精度定位等领域的关键技术与应用
  • 基金资助:
    国家自然科学基金资助项目(62172235);南京邮电大学引进人才自然科学研究启动基金资助项目(NY221136)

Topology control based on dynamic graph embedding in Internet of vehicles

Yanfei SUN1,2, Jiazheng YIN3, Jin QI2, Xiaoxuan HU2, Mengting CHEN2, Zhenjiang DONG4   

  1. 1 Jiangsu Engineering Research Center of HPC and Intelligent Processing, Nanjing 210023, China
    2 School of Internet of Things, Nanjing University of Posts and Telecommunications, Nanjing 210003, China
    3 School of Communications and Information Engineering, Nanjing University of Posts and Telecommunications, Nanjing 210003, China
    4 College of Computer, Nanjing University of Posts and Telecommunications, Nanjing 210023, China
  • Revised:2022-05-09 Online:2022-06-01 Published:2022-06-01
  • Supported by:
    The National Natural Science Foundation of China(62172235);Natural Science Research Startup Founda-tion of Recruiting Talents of Nanjing University of Posts and Telecommunications(NY221136)

在线阅读

36

PDF下载

575

摘要：

目的：随着汽车市场规模的增长，道路运载负荷压力也不断增加，但由于车联网的动态性、复杂性及较差的通信环境，以及车辆间快速变化的相对距离及互相遮挡，导致网络节点间出现频繁的断连和信号的衰落，使得网络拓扑的控制十分困难。为了构建更加稳定合理的车联网，本文利用模糊推理等方法提取驾驶员特征，提出针对车联网环境的图嵌入技术，充分利用车辆特征构建网络，实现车联网的拓扑发现及控制。

方法：利用本文提出的范围标签图嵌入（LRGE）方法发现并控制车联网拓扑。首先，建立车辆网络模型，按照路测单元合理地划分路段为多个子网络，使用驾驶员辅助系统获得车辆的相关历史信息，利用傅里叶变换、模糊推理方法，提取驾驶员的驾驶特征，并处理车辆信息得到低维特征向量。然后对新加入网络的车辆采用所提出的基于范围的冷启动方法，根据目标车辆加入区域的车辆特征，更新修正其特征向量，使新加入的车辆节点与新网络中车辆节点更易建立连接。最后基于车联网的应用环境，使用LRGE游走方式进行车辆节点间的随机游走，根据LRGE图嵌入模型对建立的随机游走进行优化，确定邻接矩阵，并根据历史序列信息进行动态更新。该方法针对实际车联网的特征，融合特征向量与驾驶员驾驶特征标签信息，进行拓扑发现及控制。使用平坦衰落复合信道模型仿真实际通信环境，并在 NGSIM 真实车辆数据集上进行实际效果的测试。

结果：从NGSIM数据集及平坦衰落复合信道的通信条件下的仿真结果可以看出，LRGE方法基本可以实现较为合理的网络拓扑控制，驾驶风格较为激进的车辆更偏向于同前向车辆集群建立连接，相似驾驶风格的车辆可以更长时间地建立稳定的连接。与随机网络、深度游走（DeepWalk）图嵌入、Node2Vec图嵌入方法及动态生长（DN）算法进行对比。测试建立的连边数目、通信断链概率和网络可达节点间的平均跳数指标，发现Node2Vec方法及LRGE方法建立的网络较为合理，断链概率较低，网络冗余较少，构建的网络拓扑更切实可行。为了体现主要目标属性网络连通性和稳健性优劣，对比所建立网络拓扑的连通概率、PageRank表示的重要性分布及割点占比，LRGE方法建立的网络连通概率高，网络重要性分布平坦，割点占比较少，在连通性和稳健性方面，具有优势。变更通信条件，对比不同通信条件下的相对关系，进一步验证了实验的结果。

结论：虽然车联网的高度动态性及复杂性使得网络难以合理稳定构建，但是车辆及驾驶员本身具有不同的特征，因此，可以提取并利用这些特征来辅助网络拓扑的控制。通过模糊推理等方法提取驾驶员特征，使用可以充分利用车辆特征信息的针对车联网特征的图嵌入方法，可以更加合理有效地构建网络拓扑。并且图嵌入方法计算简单，建立的网络性能优秀，能够对动态性较强的车联网做出快速地反应，及时更新网络拓扑，最终可以实现具有良好动态性、连通性以及健壮性的车联网拓扑控制。

关键词: 车联网, 图嵌入, 拓扑控制, 复杂网络, 模糊推理

Abstract:

Objectives:With the growth of the automotive market, the road carrying pressure is increasing. However, due to the dynamics, complexity and poor communication environment of Internet of vehicles (IoV), as well as the rapidly changing distance and occlusion between vehicles, frequent chain scission and signal fading among the nodes of the network. It is difficult to control the topology of network. In order to build a more stable and reasonable IoV, fuzzy inference and other methods was used to extract vehicle features, and a graph embedding method for the IoV environment was proposed to make full use of vehicle features to build a network, so as to realize the topology discovery and control of IoV.

Methods:The proposed label-range graph embedding (LRGE) method was used to discover and control the topology of IoV. The first thing is to establish the vehiclar network model. The road was reasonably divided into several sub networks according to the road side unit (RSU). The driver assistance system was used to obtain the relevant historical information of the vehicle. Fourier transform and fuzzy inference methods were used to extract the driving features of the drivers, and the low-dimensional feature vector was obtained by processing the vehicle information. Then, proposed range based cold boot method was applied to the vehicles newly added to the network. According to the vehicle features of the target vehicle in join area, the feature vector was updated and modified to make it easier to establish the connections with the vehicle nodes in the new network. Finally, based on the application environment of the IoV, the LRGE method was used to perform random walk among vehicle nodes, and then the established random walk was optimized according to the LRGE model to determine the adjacency matrix. This matrix was dynamically updated according to the historical sequence information. According to the characteristics of the actual IoV, this method fused the feature vector and the driver feature label to realize topology discovery and control. The flat fading composite channel model was used to simulate the actual communication environment and the actual effect was tested on the NGSIM dataset.

Results:From the simulation results under NGSIM dataset and flat fading composite channel, it can be seen that LRGE method could basically achieve more reasonable network topology control. Vehicles with more aggressive driving styles prefer to establish connections with forward vehicle clusters, and vehicles with similar driving styles are able to establish stable connections for a longer time. It is contrasted with random network, DeepWalk methods,Node2Vec methods and dynamic growth (DN) algorithm. By testing the established connection number, chain scission probability and average hops between reachable nodes, it is found that the networks established by Node2Vec method and LRGE method are more reasonable, with low chain scission probability, less network redundancy and more practical network topology. In order to reflect the difference of the main target, connectivity and robustness of the network, the connected probability, the importance distribution represented by PageRank and the proportion of cut-vertices were contrasted. The network established by LRGE method has higher connected probability. Its centrality distribution of nodes is flat, and the proportion of cut-vertices is relatively small, so it has advantages in connectivity and robustness. The experimental results were further verified by comparing the relative relations under different communication environment.

Conclusions:Although the high dynamics and complexity of IoV make it difficult to build a reasonable and stable network, vehicles and drivers have different features. Therefore, these features can be extracted and used to assist in controlling the topology of network. Driver features are extracted through fuzzy inference and other methods, and the graph embedding method for IoV can make full use of vehicle feature information. Hence, the network topology can be constructed more reasonably and effectively. Moreover, the graph embedding method is simple to calculate, and the performance of established network is better. It can make a rapid response to the dynamic IoV, update the network topology in time, and finally realize the topology control of IoV with good dynamics, connectivity and robustness.

Key words: Internet of vehicles, graph embedding, topology control, complex network, fuzzy inference

中图分类号: 

No Suggested Reading articles found!