1、硕 士 学 位 论 文基于机会连接的移动数据流量卸载策略研究专业名称 ： 信息与通信工程基于机会连接的移动数据流量卸载策略研究I摘要随着移动互联网的快速发展和智能手机、平板电脑、个人电脑等移动终端设备的普及，移动数据流量急剧增长，从而给蜂窝网络带来了巨大的流量压力。建设更多基站或者升级蜂窝网络配置等传统的解决途径，难以有效应对数据流量的急剧增长。当前研究热点表明，移动数据流量卸载是解决蜂窝网络负载压力的有效措施之一。基于机会连接的流量卸载利用用户之间建立机会连接，是一种高效灵活的应对流量激增的解决方案。本文基于真实的中国移动用户上网记录数据，研究了基于机会连接的移动数据流量卸载策略，针对完全信
2、息和不完全信息两种应用场景分别设计了流量卸载算法，主要研究内容如下：1. 根据用户相遇关系进行网络建模，分析了网络拓扑结构、用户移动性和用户上网行为特征，然后采用随机森林算法对用户相遇的机会连接进行预测，并将该算法应用在真实数据集上，取得了较好的预测效果。2. 提出了一种在完全信息应用场景中基于 Stackelberg 博弈的流量卸载算法，在机会连接预测的基础上，将运营商、内容提供用户和内容需求用户三方之间的交互行为建模为 Stackelberg 博弈过程，分析了构建的博弈模型中 Nash均衡的存在性，然后利用迭代算法设计流量卸载策略，实验结果表明该算法可以有效减轻运营商的流量负载。3. 提出
3、了一种在不完全信息应用场景中基于强化学习的流量卸载算法，在分析用户流量需求的基础上，设计了系统效用函数，然后利用强化学习Gradient Bandit 算法设计使总效用最大化的流量卸载策略，并通过实验验证了该算法的有效性。本文利用用户之间建立机会连接，针对完全信息和不完全信息两种应用场景提出的流量卸载算法可以有效减轻蜂窝网络的流量负载压力，能够为两种应用场景中的流量卸载决策提供指导，高效应对数据流量的持续增长。本文的研究工作不仅具有应用价值，也为基于机会连接的移动数据流量卸载策略研究提供了一种新思路。关键词 ：流量卸载，机会连接，博弈论，强化学习硕士学位论文IIAbstractWith the
4、 rapid development of mobile Internet and the popularity of mobile terminal devices such as smart phones, tablet computers, personal computers and so on, the rapid growth of mobile data traffic has brought tremendous traffic pressure to cellular networks. Traditional solutions such as building more
5、base stations or upgrading cellular network configurations have been unable to cope with the continuous growth of data traffic. The current research hotspot shows that mobile data offloading is one of the effective measures to solve the load pressure of cellular networks. Mobile data offloading base
6、d on opportunistic connection is an effective solution to cope with the surge of mobile data, taking advantage of opportunistic connection between users. Based on the real data provided by the mobile communication operator, this paper study the strategy of mobile data offloading based on opportunist
7、ic connection. We design the mobile data offloading algorithms respectively for two application scenarios of complete information and incomplete information. The main achievements are listed as follows:1. We construct the network based on the encounter relations between users. The network topology f
8、eatures, user mobility characteristics and user internet behavior characteristics are analyzed. We exploit random forest algorithm to predict the opportunistic connection between users. The prediction model based on random forest is applied to the real data and has achieved great performances.2. We
9、propose a mobile data offloading algorithm based on Stackelberg game in a complete information scenario. Based on the results of the opportunistic connection prediction, we model the interaction characteristics among the operator, the content provider and demander as a Stackelberg game. We analyze t
10、he existence of Nash equilibrium in the Stackelberg game model constructed in this paper. An iterative algorithm is exploited to design mobile data offloading strategies. Experiment results show that the algorithm proposed in our paper can effectively reduce the traffic load in the cellular network.
11、3. We propose a mobile data offloading algorithm based on the reinforcement learning in an incomplete information scenario. Based on the analysis of users mobile data traffic demand, we design the system utility function. The 基于机会连接的移动数据流量卸载策略研究IIIGradient Bandit algorithm in reinforcement learning
12、is exploited to solve the optimal mobile data offloading strategies which maximize the system utility function. The experiments verified the effectiveness of the algorithm proposed in our paper.Taking advantage of the opportunistic connection between users, this paper propose mobile data offloading
13、algorithms in the complete and incomplete scenarios respectively. The algorithms can effectively relieve the traffic load pressure of the operator, provide guidance for mobile data offloading decisions in two application scenarios, and well handle the continuous growth of data traffic. The research
14、in this paper not only has practical significance, but also provides a new insight of mobile data offloading based on opportunistic connection.Keywords：Mobile Data Offloading, Opportunistic Connection, Game Theory, Reinforcement Learning硕士学位论文IV基于机会连接的移动数据流量卸载策略研究I目录摘要 .IAbstract.II1 绪论 .11.1 研究背景及意
15、义 .11.2 国内外研究现状 .21.2.1 基于互补网络的流量卸载研究现状 .21.2.2 基于机会连接的流量卸载研究现状 .31.3 本文的主要内容及结构安排 .42 用户行为统计分析 .62.1 数据集信息 .62.2 用户移动行为统计分析 .82.3 用户上网行为统计分析 .112.4 本章小结 .123 基于用户行为分析的机会连接预测模型 .143.1 网络模型 .143.2 网络结构和用户行为特征提取 .173.3 机会连接预测模型 .253.3.1 随机森林算法 .253.3.2 仿真实验与分析 .263.4 本章小结 .304 完全信息场景中的流量卸载策略 .314.1 流量卸载模型与算法 .314.1.1 Stackelberg 博弈模型 .324.1.2 算法设计与实现 .354.2 仿真实验与分析 .394.3 本章小结 .435 不完全信息场景中的流量卸载策略 .445.1 流量卸载模型与算法 .445.1.1 用户流量需求分析 .455.1.2 基于强化学习的流量卸载算法 .465.2 仿真实验与分析 .525.3 本章小结 .556 总结与展望 .566.1 全文总结 .566.2 研究展望 .57参考文献 .58附录 A 攻读硕士学位期间发表的学术论文 .63附录 B 攻读硕士学位期间申请的发明专利 .64致谢 .65硕士学位论文II
