基于强化学习与均衡决策的物联网不确定环境路由选择策略研究

Routing, which is a critical technology for interconnection among sensing information, has become a key issue to be solved for research and development of the future large-scale IoT. In view of uncertain environmental constraints and routing demand, a routing scheme based on reinforcement learning and value density weighted transfer function is built for networks without comprehensive information and routing metrics. Taking the situation of inaccurate routing demand and non-quantifiable routing metrics into account, a balanced routing algorithm based on fuzzy multi- objective decision is put forward. Under the circumstances of dynamic networks, equilibrium hierarchical routing model is established upon definition of node changing rate and link stability expression. The research is expected to find a most satisfactory routing path in the uncertain terms of not comprehensive, fuzzy and dynamic environments with balanced consideration of routing cost, energy consumption, link stability, transmission delay and so forth. On the basis of seamless connection among heterogeneous terminals of IoT, tradeoff between energy consumption and optimal routing selection in uncertain environment is to be achieved. The work is expected to build low-power routing model of multi-factor balancing decisions concerning network features and demand constrains. Research results could be extended to a variety of environments and applications of IoT, as well as offer theoretical foundations and schemes for large scale IOT building-up.

路由技术，作为实现感知信息互联互通的关键技术，已成为未来物联网研究与发展中亟待解决的重点问题。本项目将从物联网不确定环境约束与路由需求入手，针对网络信息与路由度量不全面的情况，构建基于强化学习与值密度加权转移函数的未知环境路由策略；面向路由需求不准确和路由度量非量化的网络环境，提出基于模糊多目标决策的均衡路由决策算法；在网络节点动态变化的环境下，建立基于节点变化率与链路稳定因子定义的分级均衡路由模型。以期实现在物联网未知、模糊、动态的不确定环境下，根据网络需求均衡路由代价、网络能耗、链路稳定、传输时延等度量因素，选取最满意路径。在实现物联网异构终端间无缝连接的基础上，寻求大规模物联网不确定环境下最优路由选择与网络能耗间的折中解。该项目的研究成果有望针对网络特性与需求限制，建立低能耗多因素均衡决策的路由模型，可扩展至物联网的多种环境与应用中，为全面构建大规模物联网提供理论依据与实现方案。

路由技术作为实现感知信息互联互通的关键技术，已成为未来物联网研究与发展中亟待解决的关键问题。物联网复杂多样、模糊、动态的网络环境为路由选择带来了诸多的不确定性。项目组首先从物联网未知环境下的路由约束与路由需求入手，通过强化学习建立加权均衡的路由选择模型，并通过改进RFID的防碰撞技术提升了物联网的数据识别性能。其次，针对物联网的异构接入环境，提出了协同定位方法与动态网络环境下的路由机制，通过相位旋转、误差距离加权、位置指纹以及归一化门限的方法完成异构物联网中的协同定位工作。最后，针对模糊动态环境下的物联网异构网络切换展开研究，利用二元法建立相对优属度矩阵并根据最大隶属度原则选取最满意的路径，通过邻居节点变化率表征节点与网络的变化速率，对不同等级的变化率采取不同的路由策略，建立变化率分级路由选择模型，利用马尔可夫决策过程解决异构网络中的垂直切换问题。项目组研究的基于强化学习与均衡决策的物联网不确定环境路由选择策略，在性能上可实现分组投递率提高15.02%，网络能耗降低约20.57%，端到端时延降低约12.53%。在项目研究期限内，已发表SCI/EI检索学术论文12篇，申请发明专利3项，已获授权1项，参加国内外学术交流4次。综上，本项目在物联网网络特性以及路由代价、网络能耗、链路稳定性、端到端时延等性能需求分析的基础上，建立了未知环境下的多因素均衡路由模型，性能良好，为全面构建大规模物联网提供了理论依据与可靠的实现方案。