【作者】 徐江；

【导师】 钱焕延；

【作者基本信息】 南京理工大学 ， 计算机应用技术， 2016， 博士

【摘要】 无线传感器网络(Wireless Sensor Networks,WSNs)是随着半导体技术、微系统技术、通信技术等技术的发展而产生和迅速发展起来的。无线传感器网络由各类集成化的微型传感器节点协同感知、采集和处理网络覆盖的地理区域中感知对象的数据,通过嵌入式系统对数据进行处理,并通过随机自组织无线通信网络将这些数据传送给基站,最后通过互连网或卫星网络到达管理节点。无线传感器网络目前广泛地应用于国防军事、国家安全、环境监测和医疗卫生等领域。而在这些应用研究中,节点定位和网络覆盖是无线传感器网络应用的两个研究热点。节点的位置信息对于实现无线传感器网络众多应用起到至关重要的作用;而网络覆盖则决定了无线传感器网络所能提供的服务范围,也在很大程度上影响了网络的成本和各种具体应用的性能。本文针对无线传感器网络中的节点定位与网络覆盖技术进行了比较深入的研究与探讨。在节点定位方面,首先分析改进了传统DV-Hop定位算法,然后以核方法为研究手段,结合主流学习算法,提出了两种新型定位算法;在网络覆盖方面,主要针对静态覆盖中基于虚拟势场的覆盖算法做了研究,提出了基于静电场理论的移动传感器网络部署算法。本文的主要研究内容以及创新点如下:1.提炼出了 DV-Hop算法产生误差的几个原因,在此基础上提出使用粗定位到精确定位的递增式算法,并且在三边测量法计算阶段引入共线度概念,选择定位质量好的单元进行计算,从而使定位算法得到了较高的定位精度;2.借助移动信标节点,使其按照规划的路径移动,从而产生若干个虚拟信标节点,这样可以有效减少真实信标节点的数量。同时将由这些虚拟信标节点与监控区域未知节点交流获得的信号向量作为直推支持向量机中训练的有标签数据样本。根据训练样本的特点,提出一种多类对多类的分类法,据此推断未知节点的位置。实验与仿真结果表明,该方法获得了较高的定位精度;3.提出了一种基于节点跳数和核方法的无线传感器网络定位算法,该算法的基本思想是利用高斯核函数度量节点间的相似性。通过收集和利用实际距离和节点间跳数信息,将信标节点间的跳数信息和距离信息作为训练数据,使用偏最小二乘法学习并构建其间的最优模型,并用此模型预测未知节点到已知节点距离。实验结果表明,本定位算法定位精度较高,受信标节点的数目影响较小,并且具有将强的环境适应性,适合不同的部署环境等特点;4.提出了一种基于虚拟势场,分布式、自适应、可扩展的移动传感器网络部署算法,该算法把部署区域内的障碍物、节点等看作是带电荷的粒子,粒子受到其它障碍物和粒子的库仑力作用而产生运动,最终所有节点在力的相互作用下自动扩散到整个网络而完成部署。仿真结果表明,该算法在各种场景下的性能指标表现良好。更多还原

【Abstract】 The wireless sensor network was generated with the development of semiconductor,microsystem and communication technologies and has been developing rapidly.The wireless sensor network uses various integrated micro-sensor nodes for the collaborative sensing,collection and processing of data of sensed objects in the geographic area covered by the network,processes the data with an embedded system,and then transfers the data to the base station through a random self-organizing wireless communication network from which the data finally reach management nodes through internet or satellite network.Currently,the wireless sensor network has been widely applied to various fields,such as national defense and military,national security,environment monitoring and medial treatment&public health.In the wireless sensor network application research,node location and network coverage are two focuses.The location information of node plays a role of great importance in various applications of wireless sensor network;network coverage decides the scope of services offered by the wireless sensor network,and largely affects the cost of network and the performance of specific application.The paper makes an in-depth study and exploration on the node location and network coverage technology of wireless sensor network.With regard to node location,the paper first analyzes and improves traditional DV-Hop location algorithm,and then proposes two new location algorithms combined with the mainstream learning algorithm using the research approach of kernel method.With regard to network coverage,the research mainly focuses on the virtual-potential-field-based coverage algorithm in static coverage and proposes an adaptive and extensible deployment algorithm.Specific research procedures are as follows:1.Define the error causes of DV-Hop algorithm,and,on this basis,propose an incremental algorithm from rough location to accurate location,and then introduce the concept of collinearity degree at the stage of trilateration calculation stage and choose the units of good location quality for the calculation so that the location algorithm can get a high location accuracy.2.Generate some virtual beacon nodes with the help of mobile beacon nodes by making them move in given paths,to reduce the number of real beacon nodes effectively;meanwhile,consider the signal vectors obtained from the communication of the virtual beacon nodes and the unknown nodes in monitored area as the training labeled data samples of TSVM.According to the characteristics of training samples,propose a multiclass-vs.-multiclass classification method,on which basis to infer the position of unknown node.Experiment and simulation results show that the method has the high location accuracy.3.Propose a wireless sensor network location algorithm based on hop count and kernel method.The algorithm’s basic idea is to measure the similarity of nodes using the Gaussian kernel function.Collect and use the information of real distance and hop count between nodes and use the data of hop count and distance between beacon nodes as training data,and then study with PLS and build the optimal model to predict the distance from the unknown node to the known node.Experiment results show that the location algorithm has features of the high location accuracy,little influence from the number of beacon nodes,the strong environmental adaption and the suitability for different deployment environments.4.Propose a virtual-potential-field-based distributed,adaptive and extensible mobile sensor network deployment algorithm.The algorithm considers the obstacles and nodes in deployment area as charged particles.The particles move under the Coulombic force of other obstacles and particles,and finally all nodes spread to the entire network automatically due to the interaction of force,in which case the deployment is completed.According to simulation results,the algorithm’s performance indexes show good performance in various scenes.更多还原