【作者】 秦岭；

【导师】 胡荣强；

【作者基本信息】 武汉理工大学 ， 交通信息工程与控制， 2009， 博士

【摘要】 传感器网络是基于应用的网络,与传统的无线通信网络相比,它具有节点规模大、自组织多跳、无人值守、无通信基础设施等特点。而能量约束始终是限制传感器网络技术发展和应用的瓶颈。有效提高网络能效,降低节点能耗,并延长网络生存期,是本文研究工作的核心。围绕着在路由中兼顾无线传感器网络能量有效性和均衡性的主题,以人工智能(Artificial Intelligence,AI)为指导,采用神经网络的数学模型作为分析工具,本文综合并讨论了有关无线传感器网络能耗问题的国内外研究现状。同时,论文以研究传感器网络的体系架构为前提,分层剖析了传感器节点的硬件结构和设计、各层次的主要算法和协议、常用传感器网络的节能技术和策略。从硬件和协议栈两个层面分析了传感器节点能耗产生的主要原因。针对影响节点能耗的主要因素,分析了传感器网络的整体结构、节点结构、通信方式和网络覆盖,本文并给出了无线传输能耗的数学模型。在对传感器网络关键节能技术,诸如单节点节能技术、数据融合技术以及垮层节能算法总结的基础上,着重讨论和比较了传感器网络中几类经典的媒体访问控制算法、协议;分析了目前无线传感器网络节能算法、协议和策略亟待解决的问题。论文的主要成果和贡献在于将人工智能引入到无线传感器网络的路由算法中。首先,结合传感器网络路由层次上的数据融合技术,考虑到在平面式路由算法中网络能量不均衡,路由缺乏可靠的服务质量的特点,提出了一种基于神经网络自组织映射模型的平面路由算法——自组织映射融合(Self-OrganizingMap and Data Fusion,SOMDF)算法,它可降低节点间的数据传输量,规避了因数据冗余给网络带来的数据冲突和耗能增大的风险,并解决了因网络有效性和均衡性相互制约产生的矛盾。其次,与传统的无线网络相比,无线传感器网络的路由是以数据为中心。为了保证数据传输路径的可靠性、稳定性和实时性,同时兼顾网络能量有效性和均衡性特点,依托SOMDF算法设计了一种新的动态路由选择(DynamicalRouting Selection,DRS)策略,该策略将路由按照由SOMDF算法给出的链接评估质量(Connectivity Evaluation Quality,CEQ)的量值大小,依次划分为主传输路由(Trunk Routing,TR)和备用传输路由(Brach Routing,BR),采用“轮唤”方式工作,有效地降低了节点能耗,并延长了网络的生存期。最后,鉴于无线传感器网络的分族路由算法具有一定的聚类和自组织特性,将一种带有短期记忆效应和非线性动态的人工神经网络回声状态网络模型融合到分族路由算法中,提出了带有动态记忆效应的分族路由算法——回声状念路由选择(Echo State Networks Routing Selection,ESNRS)算法,它可以有效降低传感器节点的通信能耗,减少因族头的选择和分族周期的确定所带来的时耗,平衡了传感器网络的总能量,达到了有效延长网络生存期的目的。为了验证论文提出的算法优劣,本文分别从算法的收敛性、传输延时、节点能耗和网络的能量均衡性四个方面,对SOMDF算法、DRS策略和ESNRS算法进行了性能测试和评估,最后以算法仿真的性能曲线说明了测评的结果。更多还原

【Abstract】 Sensor networks is developed based on applications, which has such features as big scale, self-organized hop, no watching and no communicational infrastructure compared with traditional wireless communication networks. However, energy constraint becoming bottlenecks always restricts developments and applications in sensor networks.Increasing networks efficiency, decreasing nod’s power consumption and prolonging network lifetime play a very important role on the research of sensor networks, which is the essence in this paper. To balance on networks power-efficiency and power-equation in routing paths is the main topic in this paper. Moreover, the guidance is by way of applying artificial intelligence technique, especially using mathematical models of neural networks as tools for analyzing sensor networks in the paper. We colligate and discuss researching issues at home and abroad about power consumption in warless sensor networks. Then we make a deep research on popular algorithms and protocols , design on hardware in each layer of OSI model, analyze the main reason on sensors nod’s power consumption in both hardware layer and protocols layer on the promise of considering systematical structure of sensor networks.In this paper, we put the mathematical model for warless transmitting consumption forward by analyzing nod’s inner structure, communicational way and coverage on account of impulsions on nod’s consumption.we discuss and compare some typical Middle Access Control algorithms and protocols in sensor networks, analyze the present problems about power-saving algorithms, protocols and problem been not still dealt, which is all based on considering and analyzing key techniques in energy-saving, such as power-saving in single sensor, data fusion and power-saving algorithms in crossing layers.Conclusions and contributions are applying artificial intelligence technique to routing algorithms in sensor networks. Firstly, considering non-equation in networks’ general energy in and non-reliable Quality of Service. Self-Organizing Map and Data Fusion algorithms-SOMDF is presented based on both neural networks models and data aggression. This algorithm can decrease data transmission on sensor nods and avoid the risk of data conflict and more power consumption. In addition, Dynamical Routing Selection strategy is put forth in the paper based on SOMDF algorithm. This Strategy can ensure reliability, stability, real time and balancing networks energy efficiency and equation in routing compared with traditional wireless networks, which has property of routing base on data. DRS divide routing paths into two types: one is Trunk Routing and another is Branch Routing according to Connectivity Evaluation Quality computed by SOMDF. Finally, in view of such properties as assembling and self-organizing of cluster routing algorithms in sensor networks, we apply Echo State Networks model in neural networks to cluster routing algorithms. Echo State Networks Routing Selection algorithm--ESNRS is specially designed for clustering in routing paths. ESNRS has short-term memory, which can decrease power in communicating and prolong the lifetime of networks.In order to measure merits and drawbacks, we evaluate and test SOMDF, DRS and ESNRS from converge, transmission delay, nod’s power consumption and networks’ energy equation.更多还原