基于图卷积网络的车联网资源管理  

Resource management of vehicle network based on graph convolutional neural network

作　　者：王晓昌 朱文星 孙彦赞 吴雅婷[1,2,3] 王涛 Wang Xiaochang;Zhu Wenxing;Sun Yanzan;Wu Yating;Wang Tao(Shanghai Institute for Advanced Communication and Data Science,Shanghai University,Shanghai 200444,China;Key Laboratory of Specialty Fiber Optics and Optical Access Networks,Shanghai University,Shanghai 200444,China;Joint International Research Laboratory of Specialty Fiber Optics and Advanced Communication,Shanghai University,Shanghai 200444,China)

机构地区：[1]上海大学上海先进通信与数据科学研究院,上海200444 [2]上海大学特种光纤与光接入网重点实验室,上海200444 [3]上海大学特种光纤与先进通信国际合作联合实验室,上海200444

出　　处：《电子测量技术》2021年第3期114-119,共6页Electronic Measurement Technology

基　　金：国家重点研发计划(2017YFE0121400);国家自然科学基金(61501289,61671011,61420106011)项目资助。

摘　　要：车辆到一切(V2X)通信是有效地提高交通安全性和移动性的解决方案。为了解决深度学习在功率分配中存在的需要大量训练数据和泛化性问题,减少车辆网络信道干扰,提出了基于图卷积神经网络(GCN)的总用户速率最大化,总用户能效最大化的两种准则下的功率分配框架。所提出的框架首先将无线干扰信道转化为图数据结构,证明了干扰信道的无序性;其次根据不同功率分配准则的特点,构建了GCN网络结构,同时提出相应的损失函数。通过与基于加权最小均方误差(WMMSE)算法训练的多层感知器(MLP)网络对比,仿真数据表明,在小样本训练、可扩展性、可泛化性几个方面,所提出方案优于对比算法。Vehicle to everything(V2 X)communication is an effective solution to improve traffic safety and mobility.In order to solve the problems of deep learning that requires a large amount of training data and generalization in power allocation,this paper proposes a power distribution framework that based on graph convolutional network(GCN)under the three criteria of maximizing the total user rate,maximizing the total user energy efficiency.The proposed framework first converts the wireless interference channel into a graph data structure,which proves the disorder of the interference channel.Secondly,according to the characteristics of different power distribution,the GCN network structure is constructed,and the corresponding loss function is proposed.Compared with the multilayer perceptron(MLP)network trained based on the weighted minimum mean square error(WMMSE)algorithm,the simulation data shows that the proposed scheme is better than the comparison algorithm in terms of small sample training,scalability and generalizability.

关 键 词：车辆网络 能源效率 图卷积神经网络 功率分配 深度学习 

分 类 号：TN929.5[电子电信—通信与信息系统]